4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * #!-checking implemented by tytso.
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.
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.
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
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/key.h>
38 #include <linux/personality.h>
39 #include <linux/binfmts.h>
40 #include <linux/swap.h>
41 #include <linux/utsname.h>
42 #include <linux/module.h>
43 #include <linux/namei.h>
44 #include <linux/proc_fs.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/rmap.h>
50 #include <linux/acct.h>
51 #include <linux/cn_proc.h>
53 #include <asm/uaccess.h>
54 #include <asm/mmu_context.h>
57 #include <linux/kmod.h>
61 char core_pattern[65] = "core";
62 int suid_dumpable = 0;
64 EXPORT_SYMBOL(suid_dumpable);
65 /* The maximal length of core_pattern is also specified in sysctl.c */
67 static struct linux_binfmt *formats;
68 static DEFINE_RWLOCK(binfmt_lock);
70 int register_binfmt(struct linux_binfmt * fmt)
72 struct linux_binfmt ** tmp = &formats;
78 write_lock(&binfmt_lock);
81 write_unlock(&binfmt_lock);
88 write_unlock(&binfmt_lock);
92 EXPORT_SYMBOL(register_binfmt);
94 int unregister_binfmt(struct linux_binfmt * fmt)
96 struct linux_binfmt ** tmp = &formats;
98 write_lock(&binfmt_lock);
102 write_unlock(&binfmt_lock);
107 write_unlock(&binfmt_lock);
111 EXPORT_SYMBOL(unregister_binfmt);
113 static inline void put_binfmt(struct linux_binfmt * fmt)
115 module_put(fmt->module);
119 * Note that a shared library must be both readable and executable due to
122 * Also note that we take the address to load from from the file itself.
124 asmlinkage long sys_uselib(const char __user * library)
130 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
135 if (!S_ISREG(nd.dentry->d_inode->i_mode))
138 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
142 file = nameidata_to_filp(&nd, O_RDONLY);
143 error = PTR_ERR(file);
149 struct linux_binfmt * fmt;
151 read_lock(&binfmt_lock);
152 for (fmt = formats ; fmt ; fmt = fmt->next) {
153 if (!fmt->load_shlib)
155 if (!try_module_get(fmt->module))
157 read_unlock(&binfmt_lock);
158 error = fmt->load_shlib(file);
159 read_lock(&binfmt_lock);
161 if (error != -ENOEXEC)
164 read_unlock(&binfmt_lock);
170 release_open_intent(&nd);
176 * count() counts the number of strings in array ARGV.
178 static int count(char __user * __user * argv, int max)
186 if (get_user(p, argv))
200 * 'copy_strings()' copies argument/environment strings from user
201 * memory to free pages in kernel mem. These are in a format ready
202 * to be put directly into the top of new user memory.
204 static int copy_strings(int argc, char __user * __user * argv,
205 struct linux_binprm *bprm)
207 struct page *kmapped_page = NULL;
216 if (get_user(str, argv+argc) ||
217 !(len = strnlen_user(str, bprm->p))) {
228 /* XXX: add architecture specific overflow check here. */
233 int offset, bytes_to_copy;
236 offset = pos % PAGE_SIZE;
238 page = bprm->page[i];
241 page = alloc_page(GFP_HIGHUSER);
242 bprm->page[i] = page;
250 if (page != kmapped_page) {
252 kunmap(kmapped_page);
254 kaddr = kmap(kmapped_page);
257 memset(kaddr, 0, offset);
258 bytes_to_copy = PAGE_SIZE - offset;
259 if (bytes_to_copy > len) {
262 memset(kaddr+offset+len, 0,
263 PAGE_SIZE-offset-len);
265 err = copy_from_user(kaddr+offset, str, bytes_to_copy);
271 pos += bytes_to_copy;
272 str += bytes_to_copy;
273 len -= bytes_to_copy;
279 kunmap(kmapped_page);
284 * Like copy_strings, but get argv and its values from kernel memory.
286 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
289 mm_segment_t oldfs = get_fs();
291 r = copy_strings(argc, (char __user * __user *)argv, bprm);
296 EXPORT_SYMBOL(copy_strings_kernel);
300 * This routine is used to map in a page into an address space: needed by
301 * execve() for the initial stack and environment pages.
303 * vma->vm_mm->mmap_sem is held for writing.
305 void install_arg_page(struct vm_area_struct *vma,
306 struct page *page, unsigned long address)
308 struct mm_struct *mm = vma->vm_mm;
312 if (unlikely(anon_vma_prepare(vma)))
315 flush_dcache_page(page);
316 pte = get_locked_pte(mm, address, &ptl);
319 if (!pte_none(*pte)) {
320 pte_unmap_unlock(pte, ptl);
323 inc_mm_counter(mm, anon_rss);
324 lru_cache_add_active(page);
325 set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
326 page, vma->vm_page_prot))));
327 page_add_new_anon_rmap(page, vma, address);
328 pte_unmap_unlock(pte, ptl);
330 /* no need for flush_tlb */
334 force_sig(SIGKILL, current);
337 #define EXTRA_STACK_VM_PAGES 20 /* random */
339 int setup_arg_pages(struct linux_binprm *bprm,
340 unsigned long stack_top,
341 int executable_stack)
343 unsigned long stack_base;
344 struct vm_area_struct *mpnt;
345 struct mm_struct *mm = current->mm;
349 #ifdef CONFIG_STACK_GROWSUP
350 /* Move the argument and environment strings to the bottom of the
356 /* Start by shifting all the pages down */
358 for (j = 0; j < MAX_ARG_PAGES; j++) {
359 struct page *page = bprm->page[j];
362 bprm->page[i++] = page;
365 /* Now move them within their pages */
366 offset = bprm->p % PAGE_SIZE;
367 to = kmap(bprm->page[0]);
368 for (j = 1; j < i; j++) {
369 memmove(to, to + offset, PAGE_SIZE - offset);
370 from = kmap(bprm->page[j]);
371 memcpy(to + PAGE_SIZE - offset, from, offset);
372 kunmap(bprm->page[j - 1]);
375 memmove(to, to + offset, PAGE_SIZE - offset);
376 kunmap(bprm->page[j - 1]);
378 /* Limit stack size to 1GB */
379 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
380 if (stack_base > (1 << 30))
381 stack_base = 1 << 30;
382 stack_base = PAGE_ALIGN(stack_top - stack_base);
384 /* Adjust bprm->p to point to the end of the strings. */
385 bprm->p = stack_base + PAGE_SIZE * i - offset;
387 mm->arg_start = stack_base;
388 arg_size = i << PAGE_SHIFT;
390 /* zero pages that were copied above */
391 while (i < MAX_ARG_PAGES)
392 bprm->page[i++] = NULL;
394 stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
395 stack_base = PAGE_ALIGN(stack_base);
396 bprm->p += stack_base;
397 mm->arg_start = bprm->p;
398 arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
401 arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
404 bprm->loader += stack_base;
405 bprm->exec += stack_base;
407 mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
411 memset(mpnt, 0, sizeof(*mpnt));
413 down_write(&mm->mmap_sem);
416 #ifdef CONFIG_STACK_GROWSUP
417 mpnt->vm_start = stack_base;
418 mpnt->vm_end = stack_base + arg_size;
420 mpnt->vm_end = stack_top;
421 mpnt->vm_start = mpnt->vm_end - arg_size;
423 /* Adjust stack execute permissions; explicitly enable
424 * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
425 * and leave alone (arch default) otherwise. */
426 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
427 mpnt->vm_flags = VM_STACK_FLAGS | VM_EXEC;
428 else if (executable_stack == EXSTACK_DISABLE_X)
429 mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
431 mpnt->vm_flags = VM_STACK_FLAGS;
432 mpnt->vm_flags |= mm->def_flags;
433 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
434 if ((ret = insert_vm_struct(mm, mpnt))) {
435 up_write(&mm->mmap_sem);
436 kmem_cache_free(vm_area_cachep, mpnt);
439 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
442 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
443 struct page *page = bprm->page[i];
445 bprm->page[i] = NULL;
446 install_arg_page(mpnt, page, stack_base);
448 stack_base += PAGE_SIZE;
450 up_write(&mm->mmap_sem);
455 EXPORT_SYMBOL(setup_arg_pages);
457 #define free_arg_pages(bprm) do { } while (0)
461 static inline void free_arg_pages(struct linux_binprm *bprm)
465 for (i = 0; i < MAX_ARG_PAGES; i++) {
467 __free_page(bprm->page[i]);
468 bprm->page[i] = NULL;
472 #endif /* CONFIG_MMU */
474 struct file *open_exec(const char *name)
480 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
484 struct inode *inode = nd.dentry->d_inode;
485 file = ERR_PTR(-EACCES);
486 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
487 S_ISREG(inode->i_mode)) {
488 int err = vfs_permission(&nd, MAY_EXEC);
489 if (!err && !(inode->i_mode & 0111))
493 file = nameidata_to_filp(&nd, O_RDONLY);
495 err = deny_write_access(file);
505 release_open_intent(&nd);
511 EXPORT_SYMBOL(open_exec);
513 int kernel_read(struct file *file, unsigned long offset,
514 char *addr, unsigned long count)
522 /* The cast to a user pointer is valid due to the set_fs() */
523 result = vfs_read(file, (void __user *)addr, count, &pos);
528 EXPORT_SYMBOL(kernel_read);
530 static int exec_mmap(struct mm_struct *mm)
532 struct task_struct *tsk;
533 struct mm_struct * old_mm, *active_mm;
535 /* Notify parent that we're no longer interested in the old VM */
537 old_mm = current->mm;
538 mm_release(tsk, old_mm);
542 * Make sure that if there is a core dump in progress
543 * for the old mm, we get out and die instead of going
544 * through with the exec. We must hold mmap_sem around
545 * checking core_waiters and changing tsk->mm. The
546 * core-inducing thread will increment core_waiters for
547 * each thread whose ->mm == old_mm.
549 down_read(&old_mm->mmap_sem);
550 if (unlikely(old_mm->core_waiters)) {
551 up_read(&old_mm->mmap_sem);
556 active_mm = tsk->active_mm;
559 activate_mm(active_mm, mm);
561 arch_pick_mmap_layout(mm);
563 up_read(&old_mm->mmap_sem);
564 if (active_mm != old_mm) BUG();
573 * This function makes sure the current process has its own signal table,
574 * so that flush_signal_handlers can later reset the handlers without
575 * disturbing other processes. (Other processes might share the signal
576 * table via the CLONE_SIGHAND option to clone().)
578 static int de_thread(struct task_struct *tsk)
580 struct signal_struct *sig = tsk->signal;
581 struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
582 spinlock_t *lock = &oldsighand->siglock;
583 struct task_struct *leader = NULL;
587 * If we don't share sighandlers, then we aren't sharing anything
588 * and we can just re-use it all.
590 if (atomic_read(&oldsighand->count) <= 1) {
591 BUG_ON(atomic_read(&sig->count) != 1);
596 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
600 if (thread_group_empty(current))
601 goto no_thread_group;
604 * Kill all other threads in the thread group.
605 * We must hold tasklist_lock to call zap_other_threads.
607 read_lock(&tasklist_lock);
609 if (sig->flags & SIGNAL_GROUP_EXIT) {
611 * Another group action in progress, just
612 * return so that the signal is processed.
614 spin_unlock_irq(lock);
615 read_unlock(&tasklist_lock);
616 kmem_cache_free(sighand_cachep, newsighand);
619 zap_other_threads(current);
620 read_unlock(&tasklist_lock);
623 * Account for the thread group leader hanging around:
626 if (!thread_group_leader(current)) {
629 * The SIGALRM timer survives the exec, but needs to point
630 * at us as the new group leader now. We have a race with
631 * a timer firing now getting the old leader, so we need to
632 * synchronize with any firing (by calling del_timer_sync)
633 * before we can safely let the old group leader die.
636 spin_unlock_irq(lock);
637 if (hrtimer_cancel(&sig->real_timer))
638 hrtimer_restart(&sig->real_timer);
641 while (atomic_read(&sig->count) > count) {
642 sig->group_exit_task = current;
643 sig->notify_count = count;
644 __set_current_state(TASK_UNINTERRUPTIBLE);
645 spin_unlock_irq(lock);
649 sig->group_exit_task = NULL;
650 sig->notify_count = 0;
651 spin_unlock_irq(lock);
654 * At this point all other threads have exited, all we have to
655 * do is to wait for the thread group leader to become inactive,
656 * and to assume its PID:
658 if (!thread_group_leader(current)) {
659 struct task_struct *parent;
660 struct dentry *proc_dentry1, *proc_dentry2;
661 unsigned long ptrace;
663 leader = current->group_leader;
665 * If our leader is the child_reaper become
666 * the child_reaper and resend SIGKILL signal.
668 if (unlikely(leader == child_reaper)) {
669 write_lock(&tasklist_lock);
670 child_reaper = current;
671 zap_other_threads(current);
672 write_unlock(&tasklist_lock);
676 * Wait for the thread group leader to be a zombie.
677 * It should already be zombie at this point, most
680 while (leader->exit_state != EXIT_ZOMBIE)
683 spin_lock(&leader->proc_lock);
684 spin_lock(¤t->proc_lock);
685 proc_dentry1 = proc_pid_unhash(current);
686 proc_dentry2 = proc_pid_unhash(leader);
687 write_lock_irq(&tasklist_lock);
689 BUG_ON(leader->tgid != current->tgid);
690 BUG_ON(current->pid == current->tgid);
692 * An exec() starts a new thread group with the
693 * TGID of the previous thread group. Rehash the
694 * two threads with a switched PID, and release
695 * the former thread group leader:
697 ptrace = leader->ptrace;
698 parent = leader->parent;
699 if (unlikely(ptrace) && unlikely(parent == current)) {
701 * Joker was ptracing his own group leader,
702 * and now he wants to be his own parent!
703 * We can't have that.
708 ptrace_unlink(current);
709 ptrace_unlink(leader);
710 remove_parent(current);
711 remove_parent(leader);
713 switch_exec_pids(leader, current);
715 current->parent = current->real_parent = leader->real_parent;
716 leader->parent = leader->real_parent = child_reaper;
717 current->group_leader = current;
718 leader->group_leader = leader;
720 add_parent(current, current->parent);
721 add_parent(leader, leader->parent);
723 current->ptrace = ptrace;
724 __ptrace_link(current, parent);
727 list_del(¤t->tasks);
728 list_add_tail(¤t->tasks, &init_task.tasks);
729 current->exit_signal = SIGCHLD;
731 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
732 leader->exit_state = EXIT_DEAD;
734 write_unlock_irq(&tasklist_lock);
735 spin_unlock(&leader->proc_lock);
736 spin_unlock(¤t->proc_lock);
737 proc_pid_flush(proc_dentry1);
738 proc_pid_flush(proc_dentry2);
742 * There may be one thread left which is just exiting,
743 * but it's safe to stop telling the group to kill themselves.
750 release_task(leader);
752 BUG_ON(atomic_read(&sig->count) != 1);
754 if (atomic_read(&oldsighand->count) == 1) {
756 * Now that we nuked the rest of the thread group,
757 * it turns out we are not sharing sighand any more either.
758 * So we can just keep it.
760 kmem_cache_free(sighand_cachep, newsighand);
763 * Move our state over to newsighand and switch it in.
765 spin_lock_init(&newsighand->siglock);
766 atomic_set(&newsighand->count, 1);
767 memcpy(newsighand->action, oldsighand->action,
768 sizeof(newsighand->action));
770 write_lock_irq(&tasklist_lock);
771 spin_lock(&oldsighand->siglock);
772 spin_lock(&newsighand->siglock);
774 rcu_assign_pointer(current->sighand, newsighand);
777 spin_unlock(&newsighand->siglock);
778 spin_unlock(&oldsighand->siglock);
779 write_unlock_irq(&tasklist_lock);
781 if (atomic_dec_and_test(&oldsighand->count))
782 sighand_free(oldsighand);
785 BUG_ON(!thread_group_leader(current));
790 * These functions flushes out all traces of the currently running executable
791 * so that a new one can be started
794 static void flush_old_files(struct files_struct * files)
799 spin_lock(&files->file_lock);
801 unsigned long set, i;
805 fdt = files_fdtable(files);
806 if (i >= fdt->max_fds || i >= fdt->max_fdset)
808 set = fdt->close_on_exec->fds_bits[j];
811 fdt->close_on_exec->fds_bits[j] = 0;
812 spin_unlock(&files->file_lock);
813 for ( ; set ; i++,set >>= 1) {
818 spin_lock(&files->file_lock);
821 spin_unlock(&files->file_lock);
824 void get_task_comm(char *buf, struct task_struct *tsk)
826 /* buf must be at least sizeof(tsk->comm) in size */
828 strncpy(buf, tsk->comm, sizeof(tsk->comm));
832 void set_task_comm(struct task_struct *tsk, char *buf)
835 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
839 int flush_old_exec(struct linux_binprm * bprm)
843 struct files_struct *files;
844 char tcomm[sizeof(current->comm)];
847 * Make sure we have a private signal table and that
848 * we are unassociated from the previous thread group.
850 retval = de_thread(current);
855 * Make sure we have private file handles. Ask the
856 * fork helper to do the work for us and the exit
857 * helper to do the cleanup of the old one.
859 files = current->files; /* refcounted so safe to hold */
860 retval = unshare_files();
864 * Release all of the old mmap stuff
866 retval = exec_mmap(bprm->mm);
870 bprm->mm = NULL; /* We're using it now */
872 /* This is the point of no return */
874 put_files_struct(files);
876 current->sas_ss_sp = current->sas_ss_size = 0;
878 if (current->euid == current->uid && current->egid == current->gid)
879 current->mm->dumpable = 1;
881 current->mm->dumpable = suid_dumpable;
883 name = bprm->filename;
885 /* Copies the binary name from after last slash */
886 for (i=0; (ch = *(name++)) != '\0';) {
888 i = 0; /* overwrite what we wrote */
890 if (i < (sizeof(tcomm) - 1))
894 set_task_comm(current, tcomm);
896 current->flags &= ~PF_RANDOMIZE;
899 /* Set the new mm task size. We have to do that late because it may
900 * depend on TIF_32BIT which is only updated in flush_thread() on
901 * some architectures like powerpc
903 current->mm->task_size = TASK_SIZE;
905 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
906 file_permission(bprm->file, MAY_READ) ||
907 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
909 current->mm->dumpable = suid_dumpable;
912 /* An exec changes our domain. We are no longer part of the thread
915 current->self_exec_id++;
917 flush_signal_handlers(current, 0);
918 flush_old_files(current->files);
923 put_files_struct(current->files);
924 current->files = files;
929 EXPORT_SYMBOL(flush_old_exec);
932 * Fill the binprm structure from the inode.
933 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
935 int prepare_binprm(struct linux_binprm *bprm)
938 struct inode * inode = bprm->file->f_dentry->d_inode;
941 mode = inode->i_mode;
943 * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
944 * generic_permission lets a non-executable through
946 if (!(mode & 0111)) /* with at least _one_ execute bit set */
948 if (bprm->file->f_op == NULL)
951 bprm->e_uid = current->euid;
952 bprm->e_gid = current->egid;
954 if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
956 if (mode & S_ISUID) {
957 current->personality &= ~PER_CLEAR_ON_SETID;
958 bprm->e_uid = inode->i_uid;
963 * If setgid is set but no group execute bit then this
964 * is a candidate for mandatory locking, not a setgid
967 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
968 current->personality &= ~PER_CLEAR_ON_SETID;
969 bprm->e_gid = inode->i_gid;
973 /* fill in binprm security blob */
974 retval = security_bprm_set(bprm);
978 memset(bprm->buf,0,BINPRM_BUF_SIZE);
979 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
982 EXPORT_SYMBOL(prepare_binprm);
984 static int unsafe_exec(struct task_struct *p)
987 if (p->ptrace & PT_PTRACED) {
988 if (p->ptrace & PT_PTRACE_CAP)
989 unsafe |= LSM_UNSAFE_PTRACE_CAP;
991 unsafe |= LSM_UNSAFE_PTRACE;
993 if (atomic_read(&p->fs->count) > 1 ||
994 atomic_read(&p->files->count) > 1 ||
995 atomic_read(&p->sighand->count) > 1)
996 unsafe |= LSM_UNSAFE_SHARE;
1001 void compute_creds(struct linux_binprm *bprm)
1005 if (bprm->e_uid != current->uid)
1010 unsafe = unsafe_exec(current);
1011 security_bprm_apply_creds(bprm, unsafe);
1012 task_unlock(current);
1013 security_bprm_post_apply_creds(bprm);
1016 EXPORT_SYMBOL(compute_creds);
1018 void remove_arg_zero(struct linux_binprm *bprm)
1021 unsigned long offset;
1025 offset = bprm->p % PAGE_SIZE;
1028 while (bprm->p++, *(kaddr+offset++)) {
1029 if (offset != PAGE_SIZE)
1032 kunmap_atomic(kaddr, KM_USER0);
1034 page = bprm->page[bprm->p/PAGE_SIZE];
1035 kaddr = kmap_atomic(page, KM_USER0);
1037 kunmap_atomic(kaddr, KM_USER0);
1042 EXPORT_SYMBOL(remove_arg_zero);
1045 * cycle the list of binary formats handler, until one recognizes the image
1047 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1050 struct linux_binfmt *fmt;
1052 /* handle /sbin/loader.. */
1054 struct exec * eh = (struct exec *) bprm->buf;
1056 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1057 (eh->fh.f_flags & 0x3000) == 0x3000)
1060 unsigned long loader;
1062 allow_write_access(bprm->file);
1066 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1068 file = open_exec("/sbin/loader");
1069 retval = PTR_ERR(file);
1073 /* Remember if the application is TASO. */
1074 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1077 bprm->loader = loader;
1078 retval = prepare_binprm(bprm);
1081 /* should call search_binary_handler recursively here,
1082 but it does not matter */
1086 retval = security_bprm_check(bprm);
1090 /* kernel module loader fixup */
1091 /* so we don't try to load run modprobe in kernel space. */
1094 for (try=0; try<2; try++) {
1095 read_lock(&binfmt_lock);
1096 for (fmt = formats ; fmt ; fmt = fmt->next) {
1097 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1100 if (!try_module_get(fmt->module))
1102 read_unlock(&binfmt_lock);
1103 retval = fn(bprm, regs);
1106 allow_write_access(bprm->file);
1110 current->did_exec = 1;
1111 proc_exec_connector(current);
1114 read_lock(&binfmt_lock);
1116 if (retval != -ENOEXEC || bprm->mm == NULL)
1119 read_unlock(&binfmt_lock);
1123 read_unlock(&binfmt_lock);
1124 if (retval != -ENOEXEC || bprm->mm == NULL) {
1128 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1129 if (printable(bprm->buf[0]) &&
1130 printable(bprm->buf[1]) &&
1131 printable(bprm->buf[2]) &&
1132 printable(bprm->buf[3]))
1133 break; /* -ENOEXEC */
1134 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1141 EXPORT_SYMBOL(search_binary_handler);
1144 * sys_execve() executes a new program.
1146 int do_execve(char * filename,
1147 char __user *__user *argv,
1148 char __user *__user *envp,
1149 struct pt_regs * regs)
1151 struct linux_binprm *bprm;
1157 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1161 file = open_exec(filename);
1162 retval = PTR_ERR(file);
1168 bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1171 bprm->filename = filename;
1172 bprm->interp = filename;
1173 bprm->mm = mm_alloc();
1178 retval = init_new_context(current, bprm->mm);
1182 bprm->argc = count(argv, bprm->p / sizeof(void *));
1183 if ((retval = bprm->argc) < 0)
1186 bprm->envc = count(envp, bprm->p / sizeof(void *));
1187 if ((retval = bprm->envc) < 0)
1190 retval = security_bprm_alloc(bprm);
1194 retval = prepare_binprm(bprm);
1198 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1202 bprm->exec = bprm->p;
1203 retval = copy_strings(bprm->envc, envp, bprm);
1207 retval = copy_strings(bprm->argc, argv, bprm);
1211 retval = search_binary_handler(bprm,regs);
1213 free_arg_pages(bprm);
1215 /* execve success */
1216 security_bprm_free(bprm);
1217 acct_update_integrals(current);
1223 /* Something went wrong, return the inode and free the argument pages*/
1224 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1225 struct page * page = bprm->page[i];
1231 security_bprm_free(bprm);
1239 allow_write_access(bprm->file);
1250 int set_binfmt(struct linux_binfmt *new)
1252 struct linux_binfmt *old = current->binfmt;
1255 if (!try_module_get(new->module))
1258 current->binfmt = new;
1260 module_put(old->module);
1264 EXPORT_SYMBOL(set_binfmt);
1266 #define CORENAME_MAX_SIZE 64
1268 /* format_corename will inspect the pattern parameter, and output a
1269 * name into corename, which must have space for at least
1270 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1272 static void format_corename(char *corename, const char *pattern, long signr)
1274 const char *pat_ptr = pattern;
1275 char *out_ptr = corename;
1276 char *const out_end = corename + CORENAME_MAX_SIZE;
1278 int pid_in_pattern = 0;
1280 /* Repeat as long as we have more pattern to process and more output
1283 if (*pat_ptr != '%') {
1284 if (out_ptr == out_end)
1286 *out_ptr++ = *pat_ptr++;
1288 switch (*++pat_ptr) {
1291 /* Double percent, output one percent */
1293 if (out_ptr == out_end)
1300 rc = snprintf(out_ptr, out_end - out_ptr,
1301 "%d", current->tgid);
1302 if (rc > out_end - out_ptr)
1308 rc = snprintf(out_ptr, out_end - out_ptr,
1309 "%d", current->uid);
1310 if (rc > out_end - out_ptr)
1316 rc = snprintf(out_ptr, out_end - out_ptr,
1317 "%d", current->gid);
1318 if (rc > out_end - out_ptr)
1322 /* signal that caused the coredump */
1324 rc = snprintf(out_ptr, out_end - out_ptr,
1326 if (rc > out_end - out_ptr)
1330 /* UNIX time of coredump */
1333 do_gettimeofday(&tv);
1334 rc = snprintf(out_ptr, out_end - out_ptr,
1336 if (rc > out_end - out_ptr)
1343 down_read(&uts_sem);
1344 rc = snprintf(out_ptr, out_end - out_ptr,
1345 "%s", system_utsname.nodename);
1347 if (rc > out_end - out_ptr)
1353 rc = snprintf(out_ptr, out_end - out_ptr,
1354 "%s", current->comm);
1355 if (rc > out_end - out_ptr)
1365 /* Backward compatibility with core_uses_pid:
1367 * If core_pattern does not include a %p (as is the default)
1368 * and core_uses_pid is set, then .%pid will be appended to
1371 && (core_uses_pid || atomic_read(¤t->mm->mm_users) != 1)) {
1372 rc = snprintf(out_ptr, out_end - out_ptr,
1373 ".%d", current->tgid);
1374 if (rc > out_end - out_ptr)
1382 static void zap_threads (struct mm_struct *mm)
1384 struct task_struct *g, *p;
1385 struct task_struct *tsk = current;
1386 struct completion *vfork_done = tsk->vfork_done;
1390 * Make sure nobody is waiting for us to release the VM,
1391 * otherwise we can deadlock when we wait on each other
1394 tsk->vfork_done = NULL;
1395 complete(vfork_done);
1398 read_lock(&tasklist_lock);
1400 if (mm == p->mm && p != tsk) {
1401 force_sig_specific(SIGKILL, p);
1403 if (unlikely(p->ptrace) &&
1404 unlikely(p->parent->mm == mm))
1407 while_each_thread(g,p);
1409 read_unlock(&tasklist_lock);
1411 if (unlikely(traced)) {
1413 * We are zapping a thread and the thread it ptraces.
1414 * If the tracee went into a ptrace stop for exit tracing,
1415 * we could deadlock since the tracer is waiting for this
1416 * coredump to finish. Detach them so they can both die.
1418 write_lock_irq(&tasklist_lock);
1419 do_each_thread(g,p) {
1420 if (mm == p->mm && p != tsk &&
1421 p->ptrace && p->parent->mm == mm) {
1422 __ptrace_detach(p, 0);
1424 } while_each_thread(g,p);
1425 write_unlock_irq(&tasklist_lock);
1429 static void coredump_wait(struct mm_struct *mm)
1431 DECLARE_COMPLETION(startup_done);
1434 mm->core_startup_done = &startup_done;
1437 core_waiters = mm->core_waiters;
1438 up_write(&mm->mmap_sem);
1441 wait_for_completion(&startup_done);
1442 BUG_ON(mm->core_waiters);
1445 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1447 char corename[CORENAME_MAX_SIZE + 1];
1448 struct mm_struct *mm = current->mm;
1449 struct linux_binfmt * binfmt;
1450 struct inode * inode;
1453 int fsuid = current->fsuid;
1456 binfmt = current->binfmt;
1457 if (!binfmt || !binfmt->core_dump)
1459 down_write(&mm->mmap_sem);
1460 if (!mm->dumpable) {
1461 up_write(&mm->mmap_sem);
1466 * We cannot trust fsuid as being the "true" uid of the
1467 * process nor do we know its entire history. We only know it
1468 * was tainted so we dump it as root in mode 2.
1470 if (mm->dumpable == 2) { /* Setuid core dump mode */
1471 flag = O_EXCL; /* Stop rewrite attacks */
1472 current->fsuid = 0; /* Dump root private */
1477 spin_lock_irq(¤t->sighand->siglock);
1478 if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
1479 current->signal->flags = SIGNAL_GROUP_EXIT;
1480 current->signal->group_exit_code = exit_code;
1481 current->signal->group_stop_count = 0;
1484 spin_unlock_irq(¤t->sighand->siglock);
1486 up_write(&mm->mmap_sem);
1490 init_completion(&mm->core_done);
1494 * Clear any false indication of pending signals that might
1495 * be seen by the filesystem code called to write the core file.
1497 clear_thread_flag(TIF_SIGPENDING);
1499 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1503 * lock_kernel() because format_corename() is controlled by sysctl, which
1504 * uses lock_kernel()
1507 format_corename(corename, core_pattern, signr);
1509 file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1512 inode = file->f_dentry->d_inode;
1513 if (inode->i_nlink > 1)
1514 goto close_fail; /* multiple links - don't dump */
1515 if (d_unhashed(file->f_dentry))
1518 if (!S_ISREG(inode->i_mode))
1522 if (!file->f_op->write)
1524 if (do_truncate(file->f_dentry, 0, 0, file) != 0)
1527 retval = binfmt->core_dump(signr, regs, file);
1530 current->signal->group_exit_code |= 0x80;
1532 filp_close(file, NULL);
1534 current->fsuid = fsuid;
1535 complete_all(&mm->core_done);