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 #include <linux/personality.h>
  38 #include <linux/swap.h>
  39 #include <linux/utsname.h>
  40 #define __NO_VERSION__
  41 #include <linux/module.h>
  42
  43 #include <asm/uaccess.h>
  44 #include <asm/pgalloc.h>
  45 #include <asm/mmu_context.h>
  46
  47 #ifdef CONFIG_KMOD
  48 #include <linux/kmod.h>
  49 #endif
  50
  51 int core_uses_pid;
  52 char core_pattern[65] = "core";
  53 /* The maximal length of core_pattern is also specified in sysctl.c */
  54
  55 static struct linux_binfmt *formats;
  56 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  57
  58 int register_binfmt(struct linux_binfmt * fmt)
  59 {
  60         struct linux_binfmt ** tmp = &formats;
  61
  62         if (!fmt)
  63                 return -EINVAL;
  64         if (fmt->next)
  65                 return -EBUSY;
  66         write_lock(&binfmt_lock);
  67         while (*tmp) {
  68                 if (fmt == *tmp) {
  69                         write_unlock(&binfmt_lock);
  70                         return -EBUSY;
  71                 }
  72                 tmp = &(*tmp)->next;
  73         }
  74         fmt->next = formats;
  75         formats = fmt;
  76         write_unlock(&binfmt_lock);
  77         return 0;
  78 }
  79
  80 int unregister_binfmt(struct linux_binfmt * fmt)
  81 {
  82         struct linux_binfmt ** tmp = &formats;
  83
  84         write_lock(&binfmt_lock);
  85         while (*tmp) {
  86                 if (fmt == *tmp) {
  87                         *tmp = fmt->next;
  88                         write_unlock(&binfmt_lock);
  89                         return 0;
  90                 }
  91                 tmp = &(*tmp)->next;
  92         }
  93         write_unlock(&binfmt_lock);
  94         return -EINVAL;
  95 }
  96
  97 static inline void put_binfmt(struct linux_binfmt * fmt)
  98 {
  99         if (fmt->module)
 100                 __MOD_DEC_USE_COUNT(fmt->module);
 101 }
 102
 103 /*
 104  * Note that a shared library must be both readable and executable due to
 105  * security reasons.
 106  *
 107  * Also note that we take the address to load from from the file itself.
 108  */
 109 asmlinkage long sys_uselib(const char * library)
 110 {
 111         struct file * file;
 112         struct nameidata nd;
 113         int error;
 114
 115         error = user_path_walk(library, &nd);
 116         if (error)
 117                 goto out;
 118
 119         error = -EINVAL;
 120         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 121                 goto exit;
 122
 123         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC);
 124         if (error)
 125                 goto exit;
 126
 127         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 128         error = PTR_ERR(file);
 129         if (IS_ERR(file))
 130                 goto out;
 131
 132         error = -ENOEXEC;
 133         if(file->f_op && file->f_op->read) {
 134                 struct linux_binfmt * fmt;
 135
 136                 read_lock(&binfmt_lock);
 137                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 138                         if (!fmt->load_shlib)
 139                                 continue;
 140                         if (!try_inc_mod_count(fmt->module))
 141                                 continue;
 142                         read_unlock(&binfmt_lock);
 143                         error = fmt->load_shlib(file);
 144                         read_lock(&binfmt_lock);
 145                         put_binfmt(fmt);
 146                         if (error != -ENOEXEC)
 147                                 break;
 148                 }
 149                 read_unlock(&binfmt_lock);
 150         }
 151         fput(file);
 152 out:
 153         return error;
 154 exit:
 155         path_release(&nd);
 156         goto out;
 157 }
 158
 159 /*
 160  * count() counts the number of arguments/envelopes
 161  */
 162 static int count(char ** argv, int max)
 163 {
 164         int i = 0;
 165
 166         if (argv != NULL) {
 167                 for (;;) {
 168                         char * p;
 169
 170                         if (get_user(p, argv))
 171                                 return -EFAULT;
 172                         if (!p)
 173                                 break;
 174                         argv++;
 175                         if(++i > max)
 176                                 return -E2BIG;
 177                 }
 178         }
 179         return i;
 180 }
 181
 182 /*
 183  * 'copy_strings()' copies argument/envelope strings from user
 184  * memory to free pages in kernel mem. These are in a format ready
 185  * to be put directly into the top of new user memory.
 186  */
 187 int copy_strings(int argc,char ** argv, struct linux_binprm *bprm)
 188 {
 189         struct page *kmapped_page = NULL;
 190         char *kaddr = NULL;
 191         int ret;
 192
 193         while (argc-- > 0) {
 194                 char *str;
 195                 int len;
 196                 unsigned long pos;
 197
 198                 if (get_user(str, argv+argc) ||
 199                                 !(len = strnlen_user(str, bprm->p))) {
 200                         ret = -EFAULT;
 201                         goto out;
 202                 }
 203
 204                 if (bprm->p < len)  {
 205                         ret = -E2BIG;
 206                         goto out;
 207                 }
 208
 209                 bprm->p -= len;
 210                 /* XXX: add architecture specific overflow check here. */
 211                 pos = bprm->p;
 212
 213                 while (len > 0) {
 214                         int i, new, err;
 215                         int offset, bytes_to_copy;
 216                         struct page *page;
 217
 218                         offset = pos % PAGE_SIZE;
 219                         i = pos/PAGE_SIZE;
 220                         page = bprm->page[i];
 221                         new = 0;
 222                         if (!page) {
 223                                 page = alloc_page(GFP_HIGHUSER);
 224                                 bprm->page[i] = page;
 225                                 if (!page) {
 226                                         ret = -ENOMEM;
 227                                         goto out;
 228                                 }
 229                                 new = 1;
 230                         }
 231
 232                         if (page != kmapped_page) {
 233                                 if (kmapped_page)
 234                                         kunmap(kmapped_page);
 235                                 kmapped_page = page;
 236                                 kaddr = kmap(kmapped_page);
 237                         }
 238                         if (new && offset)
 239                                 memset(kaddr, 0, offset);
 240                         bytes_to_copy = PAGE_SIZE - offset;
 241                         if (bytes_to_copy > len) {
 242                                 bytes_to_copy = len;
 243                                 if (new)
 244                                         memset(kaddr+offset+len, 0,
 245                                                 PAGE_SIZE-offset-len);
 246                         }
 247                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 248                         if (err) {
 249                                 ret = -EFAULT;
 250                                 goto out;
 251                         }
 252
 253                         pos += bytes_to_copy;
 254                         str += bytes_to_copy;
 255                         len -= bytes_to_copy;
 256                 }
 257         }
 258         ret = 0;
 259 out:
 260         if (kmapped_page)
 261                 kunmap(kmapped_page);
 262         return ret;
 263 }
 264
 265 /*
 266  * Like copy_strings, but get argv and its values from kernel memory.
 267  */
 268 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 269 {
 270         int r;
 271         mm_segment_t oldfs = get_fs();
 272         set_fs(KERNEL_DS);
 273         r = copy_strings(argc, argv, bprm);
 274         set_fs(oldfs);
 275         return r;
 276 }
 277
 278 /*
 279  * This routine is used to map in a page into an address space: needed by
 280  * execve() for the initial stack and environment pages.
 281  *
 282  * tsk->mmap_sem is held for writing.
 283  */
 284 void put_dirty_page(struct task_struct * tsk, struct page *page, unsigned long address)
 285 {
 286         pgd_t * pgd;
 287         pmd_t * pmd;
 288         pte_t * pte;
 289
 290         if (page_count(page) != 1)
 291                 printk(KERN_ERR "mem_map disagrees with %p at %08lx\n", page, address);
 292         pgd = pgd_offset(tsk->mm, address);
 293
 294         spin_lock(&tsk->mm->page_table_lock);
 295         pmd = pmd_alloc(tsk->mm, pgd, address);
 296         if (!pmd)
 297                 goto out;
 298         pte = pte_alloc(tsk->mm, pmd, address);
 299         if (!pte)
 300                 goto out;
 301         if (!pte_none(*pte))
 302                 goto out;
 303         lru_cache_add(page);
 304         flush_dcache_page(page);
 305         flush_page_to_ram(page);
 306         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(page, PAGE_COPY))));
 307         tsk->mm->rss++;
 308         spin_unlock(&tsk->mm->page_table_lock);
 309
 310         /* no need for flush_tlb */
 311         return;
 312 out:
 313         spin_unlock(&tsk->mm->page_table_lock);
 314         __free_page(page);
 315         force_sig(SIGKILL, tsk);
 316         return;
 317 }
 318
 319 int setup_arg_pages(struct linux_binprm *bprm)
 320 {
 321         unsigned long stack_base;
 322         struct vm_area_struct *mpnt;
 323         int i;
 324
 325         stack_base = STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE;
 326
 327         bprm->p += stack_base;
 328         if (bprm->loader)
 329                 bprm->loader += stack_base;
 330         bprm->exec += stack_base;
 331
 332         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 333         if (!mpnt)
 334                 return -ENOMEM;
 335
 336         down_write(&current->mm->mmap_sem);
 337         {
 338                 mpnt->vm_mm = current->mm;
 339                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
 340                 mpnt->vm_end = STACK_TOP;
 341                 mpnt->vm_page_prot = PAGE_COPY;
 342                 mpnt->vm_flags = VM_STACK_FLAGS;
 343                 mpnt->vm_ops = NULL;
 344                 mpnt->vm_pgoff = 0;
 345                 mpnt->vm_file = NULL;
 346                 mpnt->vm_private_data = (void *) 0;
 347                 insert_vm_struct(current->mm, mpnt);
 348                 current->mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 349         }
 350
 351         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 352                 struct page *page = bprm->page[i];
 353                 if (page) {
 354                         bprm->page[i] = NULL;
 355                         put_dirty_page(current,page,stack_base);
 356                 }
 357                 stack_base += PAGE_SIZE;
 358         }
 359         up_write(&current->mm->mmap_sem);
 360
 361         return 0;
 362 }
 363
 364 struct file *open_exec(const char *name)
 365 {
 366         struct nameidata nd;
 367         struct inode *inode;
 368         struct file *file;
 369         int err = 0;
 370
 371         err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_POSITIVE, &nd);
 372         file = ERR_PTR(err);
 373         if (!err) {
 374                 inode = nd.dentry->d_inode;
 375                 file = ERR_PTR(-EACCES);
 376                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 377                     S_ISREG(inode->i_mode)) {
 378                         int err = permission(inode, MAY_EXEC);
 379                         if (!err && !(inode->i_mode & 0111))
 380                                 err = -EACCES;
 381                         file = ERR_PTR(err);
 382                         if (!err) {
 383                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 384                                 if (!IS_ERR(file)) {
 385                                         err = deny_write_access(file);
 386                                         if (err) {
 387                                                 fput(file);
 388                                                 file = ERR_PTR(err);
 389                                         }
 390                                 }
 391 out:
 392                                 return file;
 393                         }
 394                 }
 395                 path_release(&nd);
 396         }
 397         goto out;
 398 }
 399
 400 int kernel_read(struct file *file, unsigned long offset,
 401         char * addr, unsigned long count)
 402 {
 403         mm_segment_t old_fs;
 404         loff_t pos = offset;
 405         int result = -ENOSYS;
 406
 407         if (!file->f_op->read)
 408                 goto fail;
 409         old_fs = get_fs();
 410         set_fs(get_ds());
 411         result = file->f_op->read(file, addr, count, &pos);
 412         set_fs(old_fs);
 413 fail:
 414         return result;
 415 }
 416
 417 static int exec_mmap(void)
 418 {
 419         struct mm_struct * mm, * old_mm;
 420
 421         old_mm = current->mm;
 422         if (old_mm && atomic_read(&old_mm->mm_users) == 1) {
 423                 mm_release();
 424                 exit_mmap(old_mm);
 425                 return 0;
 426         }
 427
 428         mm = mm_alloc();
 429         if (mm) {
 430                 struct mm_struct *active_mm;
 431
 432                 if (init_new_context(current, mm)) {
 433                         mmdrop(mm);
 434                         return -ENOMEM;
 435                 }
 436
 437                 /* Add it to the list of mm's */
 438                 spin_lock(&mmlist_lock);
 439                 list_add(&mm->mmlist, &init_mm.mmlist);
 440                 mmlist_nr++;
 441                 spin_unlock(&mmlist_lock);
 442
 443                 task_lock(current);
 444                 active_mm = current->active_mm;
 445                 current->mm = mm;
 446                 current->active_mm = mm;
 447                 task_unlock(current);
 448                 activate_mm(active_mm, mm);
 449                 mm_release();
 450                 if (old_mm) {
 451                         if (active_mm != old_mm) BUG();
 452                         mmput(old_mm);
 453                         return 0;
 454                 }
 455                 mmdrop(active_mm);
 456                 return 0;
 457         }
 458         return -ENOMEM;
 459 }
 460
 461 /*
 462  * This function makes sure the current process has its own signal table,
 463  * so that flush_signal_handlers can later reset the handlers without
 464  * disturbing other processes.  (Other processes might share the signal
 465  * table via the CLONE_SIGNAL option to clone().)
 466  */
 467
 468 static inline int make_private_signals(void)
 469 {
 470         struct signal_struct * newsig;
 471
 472         if (atomic_read(&current->sig->count) <= 1)
 473                 return 0;
 474         newsig = kmem_cache_alloc(sigact_cachep, GFP_KERNEL);
 475         if (newsig == NULL)
 476                 return -ENOMEM;
 477         spin_lock_init(&newsig->siglock);
 478         atomic_set(&newsig->count, 1);
 479         memcpy(newsig->action, current->sig->action, sizeof(newsig->action));
 480         spin_lock_irq(&current->sigmask_lock);
 481         current->sig = newsig;
 482         spin_unlock_irq(&current->sigmask_lock);
 483         return 0;
 484 }
 485
 486 /*
 487  * If make_private_signals() made a copy of the signal table, decrement the
 488  * refcount of the original table, and free it if necessary.
 489  * We don't do that in make_private_signals() so that we can back off
 490  * in flush_old_exec() if an error occurs after calling make_private_signals().
 491  */
 492
 493 static inline void release_old_signals(struct signal_struct * oldsig)
 494 {
 495         if (current->sig == oldsig)
 496                 return;
 497         if (atomic_dec_and_test(&oldsig->count))
 498                 kmem_cache_free(sigact_cachep, oldsig);
 499 }
 500
 501 /*
 502  * These functions flushes out all traces of the currently running executable
 503  * so that a new one can be started
 504  */
 505
 506 static inline void flush_old_files(struct files_struct * files)
 507 {
 508         long j = -1;
 509
 510         write_lock(&files->file_lock);
 511         for (;;) {
 512                 unsigned long set, i;
 513
 514                 j++;
 515                 i = j * __NFDBITS;
 516                 if (i >= files->max_fds || i >= files->max_fdset)
 517                         break;
 518                 set = files->close_on_exec->fds_bits[j];
 519                 if (!set)
 520                         continue;
 521                 files->close_on_exec->fds_bits[j] = 0;
 522                 write_unlock(&files->file_lock);
 523                 for ( ; set ; i++,set >>= 1) {
 524                         if (set & 1) {
 525                                 sys_close(i);
 526                         }
 527                 }
 528                 write_lock(&files->file_lock);
 529
 530         }
 531         write_unlock(&files->file_lock);
 532 }
 533
 534 /*
 535  * An execve() will automatically "de-thread" the process.
 536  * Note: we don't have to hold the tasklist_lock to test
 537  * whether we migth need to do this. If we're not part of
 538  * a thread group, there is no way we can become one
 539  * dynamically. And if we are, we only need to protect the
 540  * unlink - even if we race with the last other thread exit,
 541  * at worst the list_del_init() might end up being a no-op.
 542  */
 543 static inline void de_thread(struct task_struct *tsk)
 544 {
 545         if (!list_empty(&tsk->thread_group)) {
 546                 write_lock_irq(&tasklist_lock);
 547                 list_del_init(&tsk->thread_group);
 548                 write_unlock_irq(&tasklist_lock);
 549         }
 550
 551         /* Minor oddity: this might stay the same. */
 552         tsk->tgid = tsk->pid;
 553 }
 554
 555 int flush_old_exec(struct linux_binprm * bprm)
 556 {
 557         char * name;
 558         int i, ch, retval;
 559         struct signal_struct * oldsig;
 560
 561         /*
 562          * Make sure we have a private signal table
 563          */
 564         oldsig = current->sig;
 565         retval = make_private_signals();
 566         if (retval) goto flush_failed;
 567
 568         /*
 569          * Release all of the old mmap stuff
 570          */
 571         retval = exec_mmap();
 572         if (retval) goto mmap_failed;
 573
 574         /* This is the point of no return */
 575         release_old_signals(oldsig);
 576
 577         current->sas_ss_sp = current->sas_ss_size = 0;
 578
 579         if (current->euid == current->uid && current->egid == current->gid)
 580                 current->mm->dumpable = 1;
 581         name = bprm->filename;
 582         for (i=0; (ch = *(name++)) != '\0';) {
 583                 if (ch == '/')
 584                         i = 0;
 585                 else
 586                         if (i < 15)
 587                                 current->comm[i++] = ch;
 588         }
 589         current->comm[i] = '\0';
 590
 591         flush_thread();
 592
 593         de_thread(current);
 594
 595         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 596             permission(bprm->file->f_dentry->d_inode,MAY_READ))
 597                 current->mm->dumpable = 0;
 598
 599         /* An exec changes our domain. We are no longer part of the thread
 600            group */
 601
 602         current->self_exec_id++;
 603
 604         flush_signal_handlers(current);
 605         flush_old_files(current->files);
 606
 607         return 0;
 608
 609 mmap_failed:
 610 flush_failed:
 611         spin_lock_irq(&current->sigmask_lock);
 612         if (current->sig != oldsig) {
 613                 kmem_cache_free(sigact_cachep, current->sig);
 614                 current->sig = oldsig;
 615         }
 616         spin_unlock_irq(&current->sigmask_lock);
 617         return retval;
 618 }
 619
 620 /*
 621  * We mustn't allow tracing of suid binaries, unless
 622  * the tracer has the capability to trace anything..
 623  */
 624 static inline int must_not_trace_exec(struct task_struct * p)
 625 {
 626         return (p->ptrace & PT_PTRACED) && !(p->ptrace & PT_PTRACE_CAP);
 627 }
 628
 629 /*
 630  * Fill the binprm structure from the inode.
 631  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 632  */
 633 int prepare_binprm(struct linux_binprm *bprm)
 634 {
 635         int mode;
 636         struct inode * inode = bprm->file->f_dentry->d_inode;
 637
 638         mode = inode->i_mode;
 639         /*
 640          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 641          * vfs_permission lets a non-executable through
 642          */
 643         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 644                 return -EACCES;
 645         if (bprm->file->f_op == NULL)
 646                 return -EACCES;
 647
 648         bprm->e_uid = current->euid;
 649         bprm->e_gid = current->egid;
 650
 651         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 652                 /* Set-uid? */
 653                 if (mode & S_ISUID)
 654                         bprm->e_uid = inode->i_uid;
 655
 656                 /* Set-gid? */
 657                 /*
 658                  * If setgid is set but no group execute bit then this
 659                  * is a candidate for mandatory locking, not a setgid
 660                  * executable.
 661                  */
 662                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
 663                         bprm->e_gid = inode->i_gid;
 664         }
 665
 666         /* We don't have VFS support for capabilities yet */
 667         cap_clear(bprm->cap_inheritable);
 668         cap_clear(bprm->cap_permitted);
 669         cap_clear(bprm->cap_effective);
 670
 671         /*  To support inheritance of root-permissions and suid-root
 672          *  executables under compatibility mode, we raise all three
 673          *  capability sets for the file.
 674          *
 675          *  If only the real uid is 0, we only raise the inheritable
 676          *  and permitted sets of the executable file.
 677          */
 678
 679         if (!issecure(SECURE_NOROOT)) {
 680                 if (bprm->e_uid == 0 || current->uid == 0) {
 681                         cap_set_full(bprm->cap_inheritable);
 682                         cap_set_full(bprm->cap_permitted);
 683                 }
 684                 if (bprm->e_uid == 0)
 685                         cap_set_full(bprm->cap_effective);
 686         }
 687
 688         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 689         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 690 }
 691
 692 /*
 693  * This function is used to produce the new IDs and capabilities
 694  * from the old ones and the file's capabilities.
 695  *
 696  * The formula used for evolving capabilities is:
 697  *
 698  *       pI' = pI
 699  * (***) pP' = (fP & X) | (fI & pI)
 700  *       pE' = pP' & fE          [NB. fE is 0 or ~0]
 701  *
 702  * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
 703  * ' indicates post-exec(), and X is the global 'cap_bset'.
 704  *
 705  */
 706
 707 void compute_creds(struct linux_binprm *bprm)
 708 {
 709         kernel_cap_t new_permitted, working;
 710         int do_unlock = 0;
 711
 712         new_permitted = cap_intersect(bprm->cap_permitted, cap_bset);
 713         working = cap_intersect(bprm->cap_inheritable,
 714                                 current->cap_inheritable);
 715         new_permitted = cap_combine(new_permitted, working);
 716
 717         if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
 718             !cap_issubset(new_permitted, current->cap_permitted)) {
 719                 current->mm->dumpable = 0;
 720
 721                 lock_kernel();
 722                 if (must_not_trace_exec(current)
 723                     || atomic_read(&current->fs->count) > 1
 724                     || atomic_read(&current->files->count) > 1
 725                     || atomic_read(&current->sig->count) > 1) {
 726                         if(!capable(CAP_SETUID)) {
 727                                 bprm->e_uid = current->uid;
 728                                 bprm->e_gid = current->gid;
 729                         }
 730                         if(!capable(CAP_SETPCAP)) {
 731                                 new_permitted = cap_intersect(new_permitted,
 732                                                         current->cap_permitted);
 733                         }
 734                 }
 735                 do_unlock = 1;
 736         }
 737
 738
 739         /* For init, we want to retain the capabilities set
 740          * in the init_task struct. Thus we skip the usual
 741          * capability rules */
 742         if (current->pid != 1) {
 743                 current->cap_permitted = new_permitted;
 744                 current->cap_effective =
 745                         cap_intersect(new_permitted, bprm->cap_effective);
 746         }
 747
 748         /* AUD: Audit candidate if current->cap_effective is set */
 749
 750         current->suid = current->euid = current->fsuid = bprm->e_uid;
 751         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 752
 753         if(do_unlock)
 754                 unlock_kernel();
 755         current->keep_capabilities = 0;
 756 }
 757
 758
 759 void remove_arg_zero(struct linux_binprm *bprm)
 760 {
 761         if (bprm->argc) {
 762                 unsigned long offset;
 763                 char * kaddr;
 764                 struct page *page;
 765
 766                 offset = bprm->p % PAGE_SIZE;
 767                 goto inside;
 768
 769                 while (bprm->p++, *(kaddr+offset++)) {
 770                         if (offset != PAGE_SIZE)
 771                                 continue;
 772                         offset = 0;
 773                         kunmap(page);
 774 inside:
 775                         page = bprm->page[bprm->p/PAGE_SIZE];
 776                         kaddr = kmap(page);
 777                 }
 778                 kunmap(page);
 779                 bprm->argc--;
 780         }
 781 }
 782
 783 /*
 784  * cycle the list of binary formats handler, until one recognizes the image
 785  */
 786 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
 787 {
 788         int try,retval=0;
 789         struct linux_binfmt *fmt;
 790 #ifdef __alpha__
 791         /* handle /sbin/loader.. */
 792         {
 793             struct exec * eh = (struct exec *) bprm->buf;
 794
 795             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
 796                 (eh->fh.f_flags & 0x3000) == 0x3000)
 797             {
 798                 struct file * file;
 799                 unsigned long loader;
 800
 801                 allow_write_access(bprm->file);
 802                 fput(bprm->file);
 803                 bprm->file = NULL;
 804
 805                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 806
 807                 file = open_exec("/sbin/loader");
 808                 retval = PTR_ERR(file);
 809                 if (IS_ERR(file))
 810                         return retval;
 811
 812                 /* Remember if the application is TASO.  */
 813                 bprm->sh_bang = eh->ah.entry < 0x100000000;
 814
 815                 bprm->file = file;
 816                 bprm->loader = loader;
 817                 retval = prepare_binprm(bprm);
 818                 if (retval<0)
 819                         return retval;
 820                 /* should call search_binary_handler recursively here,
 821                    but it does not matter */
 822             }
 823         }
 824 #endif
 825         /* kernel module loader fixup */
 826         /* so we don't try to load run modprobe in kernel space. */
 827         set_fs(USER_DS);
 828         for (try=0; try<2; try++) {
 829                 read_lock(&binfmt_lock);
 830                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 831                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
 832                         if (!fn)
 833                                 continue;
 834                         if (!try_inc_mod_count(fmt->module))
 835                                 continue;
 836                         read_unlock(&binfmt_lock);
 837                         retval = fn(bprm, regs);
 838                         if (retval >= 0) {
 839                                 put_binfmt(fmt);
 840                                 allow_write_access(bprm->file);
 841                                 if (bprm->file)
 842                                         fput(bprm->file);
 843                                 bprm->file = NULL;
 844                                 current->did_exec = 1;
 845                                 return retval;
 846                         }
 847                         read_lock(&binfmt_lock);
 848                         put_binfmt(fmt);
 849                         if (retval != -ENOEXEC)
 850                                 break;
 851                         if (!bprm->file) {
 852                                 read_unlock(&binfmt_lock);
 853                                 return retval;
 854                         }
 855                 }
 856                 read_unlock(&binfmt_lock);
 857                 if (retval != -ENOEXEC) {
 858                         break;
 859 #ifdef CONFIG_KMOD
 860                 }else{
 861 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
 862                         char modname[20];
 863                         if (printable(bprm->buf[0]) &&
 864                             printable(bprm->buf[1]) &&
 865                             printable(bprm->buf[2]) &&
 866                             printable(bprm->buf[3]))
 867                                 break; /* -ENOEXEC */
 868                         sprintf(modname, "binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
 869                         request_module(modname);
 870 #endif
 871                 }
 872         }
 873         return retval;
 874 }
 875
 876
 877 /*
 878  * sys_execve() executes a new program.
 879  */
 880 int do_execve(char * filename, char ** argv, char ** envp, struct pt_regs * regs)
 881 {
 882         struct linux_binprm bprm;
 883         struct file *file;
 884         int retval;
 885         int i;
 886
 887         file = open_exec(filename);
 888
 889         retval = PTR_ERR(file);
 890         if (IS_ERR(file))
 891                 return retval;
 892
 893         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 894         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
 895
 896         bprm.file = file;
 897         bprm.filename = filename;
 898         bprm.sh_bang = 0;
 899         bprm.loader = 0;
 900         bprm.exec = 0;
 901         if ((bprm.argc = count(argv, bprm.p / sizeof(void *))) < 0) {
 902                 allow_write_access(file);
 903                 fput(file);
 904                 return bprm.argc;
 905         }
 906
 907         if ((bprm.envc = count(envp, bprm.p / sizeof(void *))) < 0) {
 908                 allow_write_access(file);
 909                 fput(file);
 910                 return bprm.envc;
 911         }
 912
 913         retval = prepare_binprm(&bprm);
 914         if (retval < 0)
 915                 goto out;
 916
 917         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
 918         if (retval < 0)
 919                 goto out;
 920
 921         bprm.exec = bprm.p;
 922         retval = copy_strings(bprm.envc, envp, &bprm);
 923         if (retval < 0)
 924                 goto out;
 925
 926         retval = copy_strings(bprm.argc, argv, &bprm);
 927         if (retval < 0)
 928                 goto out;
 929
 930         retval = search_binary_handler(&bprm,regs);
 931         if (retval >= 0)
 932                 /* execve success */
 933                 return retval;
 934
 935 out:
 936         /* Something went wrong, return the inode and free the argument pages*/
 937         allow_write_access(bprm.file);
 938         if (bprm.file)
 939                 fput(bprm.file);
 940
 941         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 942                 struct page * page = bprm.page[i];
 943                 if (page)
 944                         __free_page(page);
 945         }
 946
 947         return retval;
 948 }
 949
 950 void set_binfmt(struct linux_binfmt *new)
 951 {
 952         struct linux_binfmt *old = current->binfmt;
 953         if (new && new->module)
 954                 __MOD_INC_USE_COUNT(new->module);
 955         current->binfmt = new;
 956         if (old && old->module)
 957                 __MOD_DEC_USE_COUNT(old->module);
 958 }
 959
 960 #define CORENAME_MAX_SIZE 64
 961
 962 /* format_corename will inspect the pattern parameter, and output a
 963  * name into corename, which must have space for at least
 964  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
 965  */
 966 void format_corename(char *corename, const char *pattern, long signr)
 967 {
 968         const char *pat_ptr = pattern;
 969         char *out_ptr = corename;
 970         char *const out_end = corename + CORENAME_MAX_SIZE;
 971         int rc;
 972         int pid_in_pattern = 0;
 973
 974         /* Repeat as long as we have more pattern to process and more output
 975            space */
 976         while (*pat_ptr) {
 977                 if (*pat_ptr != '%') {
 978                         if (out_ptr == out_end)
 979                                 goto out;
 980                         *out_ptr++ = *pat_ptr++;
 981                 } else {
 982                         switch (*++pat_ptr) {
 983                         case 0:
 984                                 goto out;
 985                         /* Double percent, output one percent */
 986                         case '%':
 987                                 if (out_ptr == out_end)
 988                                         goto out;
 989                                 *out_ptr++ = '%';
 990                                 break;
 991                         /* pid */
 992                         case 'p':
 993                                 pid_in_pattern = 1;
 994                                 rc = snprintf(out_ptr, out_end - out_ptr,
 995                                               "%d", current->pid);
 996                                 if (rc > out_end - out_ptr)
 997                                         goto out;
 998                                 out_ptr += rc;
 999                                 break;
1000                         /* uid */
1001                         case 'u':
1002                                 rc = snprintf(out_ptr, out_end - out_ptr,
1003                                               "%d", current->uid);
1004                                 if (rc > out_end - out_ptr)
1005                                         goto out;
1006                                 out_ptr += rc;
1007                                 break;
1008                         /* gid */
1009                         case 'g':
1010                                 rc = snprintf(out_ptr, out_end - out_ptr,
1011                                               "%d", current->gid);
1012                                 if (rc > out_end - out_ptr)
1013                                         goto out;
1014                                 out_ptr += rc;
1015                                 break;
1016                         /* signal that caused the coredump */
1017                         case 's':
1018                                 rc = snprintf(out_ptr, out_end - out_ptr,
1019                                               "%ld", signr);
1020                                 if (rc > out_end - out_ptr)
1021                                         goto out;
1022                                 out_ptr += rc;
1023                                 break;
1024                         /* UNIX time of coredump */
1025                         case 't': {
1026                                 struct timeval tv;
1027                                 do_gettimeofday(&tv);
1028                                 rc = snprintf(out_ptr, out_end - out_ptr,
1029                                               "%ld", tv.tv_sec);
1030                                 if (rc > out_end - out_ptr)
1031                                         goto out;
1032                                 out_ptr += rc;
1033                                 break;
1034                         }
1035                         /* hostname */
1036                         case 'h':
1037                                 down_read(&uts_sem);
1038                                 rc = snprintf(out_ptr, out_end - out_ptr,
1039                                               "%s", system_utsname.nodename);
1040                                 up_read(&uts_sem);
1041                                 if (rc > out_end - out_ptr)
1042                                         goto out;
1043                                 out_ptr += rc;
1044                                 break;
1045                         /* executable */
1046                         case 'e':
1047                                 rc = snprintf(out_ptr, out_end - out_ptr,
1048                                               "%s", current->comm);
1049                                 if (rc > out_end - out_ptr)
1050                                         goto out;
1051                                 out_ptr += rc;
1052                                 break;
1053                         default:
1054                                 break;
1055                         }
1056                         ++pat_ptr;
1057                 }
1058         }
1059         /* Backward compatibility with core_uses_pid:
1060          *
1061          * If core_pattern does not include a %p (as is the default)
1062          * and core_uses_pid is set, then .%pid will be appended to
1063          * the filename */
1064         if (!pid_in_pattern
1065             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1066                 rc = snprintf(out_ptr, out_end - out_ptr,
1067                               ".%d", current->pid);
1068                 if (rc > out_end - out_ptr)
1069                         goto out;
1070                 out_ptr += rc;
1071         }
1072       out:
1073         *out_ptr = 0;
1074 }
1075
1076 int do_coredump(long signr, struct pt_regs * regs)
1077 {
1078         struct linux_binfmt * binfmt;
1079         char corename[CORENAME_MAX_SIZE + 1];
1080         struct file * file;
1081         struct inode * inode;
1082         int retval = 0;
1083
1084         lock_kernel();
1085         binfmt = current->binfmt;
1086         if (!binfmt || !binfmt->core_dump)
1087                 goto fail;
1088         if (!current->mm->dumpable)
1089                 goto fail;
1090         current->mm->dumpable = 0;
1091         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1092                 goto fail;
1093
1094         format_corename(corename, core_pattern, signr);
1095         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW, 0600);
1096         if (IS_ERR(file))
1097                 goto fail;
1098         inode = file->f_dentry->d_inode;
1099         if (inode->i_nlink > 1)
1100                 goto close_fail;        /* multiple links - don't dump */
1101         if (d_unhashed(file->f_dentry))
1102                 goto close_fail;
1103
1104         if (!S_ISREG(inode->i_mode))
1105                 goto close_fail;
1106         if (!file->f_op)
1107                 goto close_fail;
1108         if (!file->f_op->write)
1109                 goto close_fail;
1110         if (do_truncate(file->f_dentry, 0) != 0)
1111                 goto close_fail;
1112
1113         retval = binfmt->core_dump(signr, regs, file);
1114
1115 close_fail:
1116         filp_close(file, NULL);
1117 fail:
1118         unlock_kernel();
1119         return retval;
1120 }