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/vs_memory.h>
53 #include <asm/uaccess.h>
54 #include <asm/mmu_context.h>
57 #include <linux/kmod.h>
61 char core_pattern[65] = "core";
62 /* The maximal length of core_pattern is also specified in sysctl.c */
64 static struct linux_binfmt *formats;
65 static DEFINE_RWLOCK(binfmt_lock);
67 int register_binfmt(struct linux_binfmt * fmt)
69 struct linux_binfmt ** tmp = &formats;
75 write_lock(&binfmt_lock);
78 write_unlock(&binfmt_lock);
85 write_unlock(&binfmt_lock);
89 EXPORT_SYMBOL(register_binfmt);
91 int unregister_binfmt(struct linux_binfmt * fmt)
93 struct linux_binfmt ** tmp = &formats;
95 write_lock(&binfmt_lock);
99 write_unlock(&binfmt_lock);
104 write_unlock(&binfmt_lock);
108 EXPORT_SYMBOL(unregister_binfmt);
110 static inline void put_binfmt(struct linux_binfmt * fmt)
112 module_put(fmt->module);
116 * Note that a shared library must be both readable and executable due to
119 * Also note that we take the address to load from from the file itself.
121 asmlinkage long sys_uselib(const char __user * library)
127 nd.intent.open.flags = FMODE_READ;
128 error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
133 if (!S_ISREG(nd.dentry->d_inode->i_mode))
136 error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
140 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
141 error = PTR_ERR(file);
147 struct linux_binfmt * fmt;
149 read_lock(&binfmt_lock);
150 for (fmt = formats ; fmt ; fmt = fmt->next) {
151 if (!fmt->load_shlib)
153 if (!try_module_get(fmt->module))
155 read_unlock(&binfmt_lock);
156 error = fmt->load_shlib(file);
157 read_lock(&binfmt_lock);
159 if (error != -ENOEXEC)
162 read_unlock(&binfmt_lock);
173 * count() counts the number of strings in array ARGV.
175 static int count(char __user * __user * argv, int max)
183 if (get_user(p, argv))
197 * 'copy_strings()' copies argument/environment strings from user
198 * memory to free pages in kernel mem. These are in a format ready
199 * to be put directly into the top of new user memory.
201 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
203 struct page *kmapped_page = NULL;
212 if (get_user(str, argv+argc) ||
213 !(len = strnlen_user(str, bprm->p))) {
224 /* XXX: add architecture specific overflow check here. */
229 int offset, bytes_to_copy;
232 offset = pos % PAGE_SIZE;
234 page = bprm->page[i];
237 page = alloc_page(GFP_HIGHUSER);
238 bprm->page[i] = page;
246 if (page != kmapped_page) {
248 kunmap(kmapped_page);
250 kaddr = kmap(kmapped_page);
253 memset(kaddr, 0, offset);
254 bytes_to_copy = PAGE_SIZE - offset;
255 if (bytes_to_copy > len) {
258 memset(kaddr+offset+len, 0,
259 PAGE_SIZE-offset-len);
261 err = copy_from_user(kaddr+offset, str, bytes_to_copy);
267 pos += bytes_to_copy;
268 str += bytes_to_copy;
269 len -= bytes_to_copy;
275 kunmap(kmapped_page);
280 * Like copy_strings, but get argv and its values from kernel memory.
282 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
285 mm_segment_t oldfs = get_fs();
287 r = copy_strings(argc, (char __user * __user *)argv, bprm);
292 EXPORT_SYMBOL(copy_strings_kernel);
296 * This routine is used to map in a page into an address space: needed by
297 * execve() for the initial stack and environment pages.
299 * vma->vm_mm->mmap_sem is held for writing.
301 void install_arg_page(struct vm_area_struct *vma,
302 struct page *page, unsigned long address)
304 struct mm_struct *mm = vma->vm_mm;
310 if (unlikely(anon_vma_prepare(vma)))
313 flush_dcache_page(page);
314 pgd = pgd_offset(mm, address);
316 spin_lock(&mm->page_table_lock);
317 pud = pud_alloc(mm, pgd, address);
320 pmd = pmd_alloc(mm, pud, address);
323 pte = pte_alloc_map(mm, pmd, address);
326 if (!pte_none(*pte)) {
332 lru_cache_add_active(page);
333 set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(
334 page, vma->vm_page_prot))));
335 page_add_anon_rmap(page, vma, address);
337 spin_unlock(&mm->page_table_lock);
339 /* no need for flush_tlb */
342 spin_unlock(&mm->page_table_lock);
345 force_sig(SIGKILL, current);
348 #define EXTRA_STACK_VM_PAGES 20 /* random */
350 int setup_arg_pages(struct linux_binprm *bprm,
351 unsigned long stack_top,
352 int executable_stack)
354 unsigned long stack_base;
355 struct vm_area_struct *mpnt;
356 struct mm_struct *mm = current->mm;
360 #ifdef CONFIG_STACK_GROWSUP
361 /* Move the argument and environment strings to the bottom of the
367 /* Start by shifting all the pages down */
369 for (j = 0; j < MAX_ARG_PAGES; j++) {
370 struct page *page = bprm->page[j];
373 bprm->page[i++] = page;
376 /* Now move them within their pages */
377 offset = bprm->p % PAGE_SIZE;
378 to = kmap(bprm->page[0]);
379 for (j = 1; j < i; j++) {
380 memmove(to, to + offset, PAGE_SIZE - offset);
381 from = kmap(bprm->page[j]);
382 memcpy(to + PAGE_SIZE - offset, from, offset);
383 kunmap(bprm->page[j - 1]);
386 memmove(to, to + offset, PAGE_SIZE - offset);
387 kunmap(bprm->page[j - 1]);
389 /* Limit stack size to 1GB */
390 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
391 if (stack_base > (1 << 30))
392 stack_base = 1 << 30;
393 stack_base = PAGE_ALIGN(stack_top - stack_base);
395 /* Adjust bprm->p to point to the end of the strings. */
396 bprm->p = stack_base + PAGE_SIZE * i - offset;
398 mm->arg_start = stack_base;
399 arg_size = i << PAGE_SHIFT;
401 /* zero pages that were copied above */
402 while (i < MAX_ARG_PAGES)
403 bprm->page[i++] = NULL;
405 stack_base = stack_top - MAX_ARG_PAGES * PAGE_SIZE;
406 bprm->p += stack_base;
407 mm->arg_start = bprm->p;
408 arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
411 arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
414 bprm->loader += stack_base;
415 bprm->exec += stack_base;
417 mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
421 if (security_vm_enough_memory(arg_size >> PAGE_SHIFT) ||
422 !vx_vmpages_avail(mm, arg_size >> PAGE_SHIFT)) {
423 kmem_cache_free(vm_area_cachep, mpnt);
427 memset(mpnt, 0, sizeof(*mpnt));
429 down_write(&mm->mmap_sem);
432 #ifdef CONFIG_STACK_GROWSUP
433 mpnt->vm_start = stack_base;
434 mpnt->vm_end = stack_base + arg_size;
436 mpnt->vm_end = stack_top;
437 mpnt->vm_start = mpnt->vm_end - arg_size;
439 /* Adjust stack execute permissions; explicitly enable
440 * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
441 * and leave alone (arch default) otherwise. */
442 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
443 mpnt->vm_flags = VM_STACK_FLAGS | VM_EXEC;
444 else if (executable_stack == EXSTACK_DISABLE_X)
445 mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
447 mpnt->vm_flags = VM_STACK_FLAGS;
448 mpnt->vm_flags |= mm->def_flags;
449 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
450 if ((ret = insert_vm_struct(mm, mpnt))) {
451 up_write(&mm->mmap_sem);
452 kmem_cache_free(vm_area_cachep, mpnt);
455 // mm->stack_vm = mm->total_vm = vma_pages(mpnt);
456 vx_vmpages_sub(mm, mm->total_vm - vma_pages(mpnt));
457 mm->stack_vm = mm->total_vm;
460 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
461 struct page *page = bprm->page[i];
463 bprm->page[i] = NULL;
464 install_arg_page(mpnt, page, stack_base);
466 stack_base += PAGE_SIZE;
468 up_write(&mm->mmap_sem);
473 EXPORT_SYMBOL(setup_arg_pages);
475 #define free_arg_pages(bprm) do { } while (0)
479 static inline void free_arg_pages(struct linux_binprm *bprm)
483 for (i = 0; i < MAX_ARG_PAGES; i++) {
485 __free_page(bprm->page[i]);
486 bprm->page[i] = NULL;
490 #endif /* CONFIG_MMU */
492 struct file *open_exec(const char *name)
498 nd.intent.open.flags = FMODE_READ;
499 err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
503 struct inode *inode = nd.dentry->d_inode;
504 file = ERR_PTR(-EACCES);
505 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
506 S_ISREG(inode->i_mode)) {
507 int err = permission(inode, MAY_EXEC, &nd);
508 if (!err && !(inode->i_mode & 0111))
512 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
514 err = deny_write_access(file);
529 EXPORT_SYMBOL(open_exec);
531 int kernel_read(struct file *file, unsigned long offset,
532 char *addr, unsigned long count)
540 /* The cast to a user pointer is valid due to the set_fs() */
541 result = vfs_read(file, (void __user *)addr, count, &pos);
546 EXPORT_SYMBOL(kernel_read);
548 static int exec_mmap(struct mm_struct *mm)
550 struct task_struct *tsk;
551 struct mm_struct * old_mm, *active_mm;
553 /* Notify parent that we're no longer interested in the old VM */
555 old_mm = current->mm;
556 mm_release(tsk, old_mm);
560 * Make sure that if there is a core dump in progress
561 * for the old mm, we get out and die instead of going
562 * through with the exec. We must hold mmap_sem around
563 * checking core_waiters and changing tsk->mm. The
564 * core-inducing thread will increment core_waiters for
565 * each thread whose ->mm == old_mm.
567 down_read(&old_mm->mmap_sem);
568 if (unlikely(old_mm->core_waiters)) {
569 up_read(&old_mm->mmap_sem);
574 active_mm = tsk->active_mm;
577 activate_mm(active_mm, mm);
579 arch_pick_mmap_layout(mm);
581 up_read(&old_mm->mmap_sem);
582 if (active_mm != old_mm) BUG();
591 * This function makes sure the current process has its own signal table,
592 * so that flush_signal_handlers can later reset the handlers without
593 * disturbing other processes. (Other processes might share the signal
594 * table via the CLONE_SIGHAND option to clone().)
596 static inline int de_thread(struct task_struct *tsk)
598 struct signal_struct *sig = tsk->signal;
599 struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
600 spinlock_t *lock = &oldsighand->siglock;
604 * If we don't share sighandlers, then we aren't sharing anything
605 * and we can just re-use it all.
607 if (atomic_read(&oldsighand->count) <= 1) {
608 BUG_ON(atomic_read(&sig->count) != 1);
613 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
617 if (thread_group_empty(current))
618 goto no_thread_group;
621 * Kill all other threads in the thread group.
622 * We must hold tasklist_lock to call zap_other_threads.
624 read_lock(&tasklist_lock);
626 if (sig->flags & SIGNAL_GROUP_EXIT) {
628 * Another group action in progress, just
629 * return so that the signal is processed.
631 spin_unlock_irq(lock);
632 read_unlock(&tasklist_lock);
633 kmem_cache_free(sighand_cachep, newsighand);
636 zap_other_threads(current);
637 read_unlock(&tasklist_lock);
640 * Account for the thread group leader hanging around:
643 if (thread_group_leader(current))
645 while (atomic_read(&sig->count) > count) {
646 sig->group_exit_task = current;
647 sig->notify_count = count;
648 __set_current_state(TASK_UNINTERRUPTIBLE);
649 spin_unlock_irq(lock);
653 sig->group_exit_task = NULL;
654 sig->notify_count = 0;
655 spin_unlock_irq(lock);
658 * At this point all other threads have exited, all we have to
659 * do is to wait for the thread group leader to become inactive,
660 * and to assume its PID:
662 if (!thread_group_leader(current)) {
663 struct task_struct *leader = current->group_leader, *parent;
664 struct dentry *proc_dentry1, *proc_dentry2;
665 unsigned long exit_state, ptrace;
668 * Wait for the thread group leader to be a zombie.
669 * It should already be zombie at this point, most
672 while (leader->exit_state != EXIT_ZOMBIE)
675 spin_lock(&leader->proc_lock);
676 spin_lock(¤t->proc_lock);
677 proc_dentry1 = proc_pid_unhash(current);
678 proc_dentry2 = proc_pid_unhash(leader);
679 write_lock_irq(&tasklist_lock);
681 if (leader->tgid != current->tgid)
683 if (current->pid == current->tgid)
686 * An exec() starts a new thread group with the
687 * TGID of the previous thread group. Rehash the
688 * two threads with a switched PID, and release
689 * the former thread group leader:
691 ptrace = leader->ptrace;
692 parent = leader->parent;
693 if (unlikely(ptrace) && unlikely(parent == current)) {
695 * Joker was ptracing his own group leader,
696 * and now he wants to be his own parent!
697 * We can't have that.
702 ptrace_unlink(current);
703 ptrace_unlink(leader);
704 remove_parent(current);
705 remove_parent(leader);
707 switch_exec_pids(leader, current);
709 current->parent = current->real_parent = leader->real_parent;
710 leader->parent = leader->real_parent = child_reaper;
711 current->group_leader = current;
712 leader->group_leader = leader;
714 add_parent(current, current->parent);
715 add_parent(leader, leader->parent);
717 current->ptrace = ptrace;
718 __ptrace_link(current, parent);
721 list_del(¤t->tasks);
722 list_add_tail(¤t->tasks, &init_task.tasks);
723 current->exit_signal = SIGCHLD;
724 exit_state = leader->exit_state;
726 write_unlock_irq(&tasklist_lock);
727 spin_unlock(&leader->proc_lock);
728 spin_unlock(¤t->proc_lock);
729 proc_pid_flush(proc_dentry1);
730 proc_pid_flush(proc_dentry2);
732 if (exit_state != EXIT_ZOMBIE)
734 release_task(leader);
738 * Now there are really no other threads at all,
739 * so it's safe to stop telling them to kill themselves.
744 BUG_ON(atomic_read(&sig->count) != 1);
747 if (atomic_read(&oldsighand->count) == 1) {
749 * Now that we nuked the rest of the thread group,
750 * it turns out we are not sharing sighand any more either.
751 * So we can just keep it.
753 kmem_cache_free(sighand_cachep, newsighand);
756 * Move our state over to newsighand and switch it in.
758 spin_lock_init(&newsighand->siglock);
759 atomic_set(&newsighand->count, 1);
760 memcpy(newsighand->action, oldsighand->action,
761 sizeof(newsighand->action));
763 write_lock_irq(&tasklist_lock);
764 spin_lock(&oldsighand->siglock);
765 spin_lock(&newsighand->siglock);
767 current->sighand = newsighand;
770 spin_unlock(&newsighand->siglock);
771 spin_unlock(&oldsighand->siglock);
772 write_unlock_irq(&tasklist_lock);
774 if (atomic_dec_and_test(&oldsighand->count))
775 kmem_cache_free(sighand_cachep, oldsighand);
778 if (!thread_group_empty(current))
780 if (!thread_group_leader(current))
786 * These functions flushes out all traces of the currently running executable
787 * so that a new one can be started
790 static inline void flush_old_files(struct files_struct * files)
794 spin_lock(&files->file_lock);
796 unsigned long set, i;
800 if (i >= files->max_fds || i >= files->max_fdset)
802 set = files->close_on_exec->fds_bits[j];
805 files->close_on_exec->fds_bits[j] = 0;
806 spin_unlock(&files->file_lock);
807 for ( ; set ; i++,set >>= 1) {
812 spin_lock(&files->file_lock);
815 spin_unlock(&files->file_lock);
818 void get_task_comm(char *buf, struct task_struct *tsk)
820 /* buf must be at least sizeof(tsk->comm) in size */
822 strncpy(buf, tsk->comm, sizeof(tsk->comm));
826 void set_task_comm(struct task_struct *tsk, char *buf)
829 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
833 int flush_old_exec(struct linux_binprm * bprm)
837 struct files_struct *files;
838 char tcomm[sizeof(current->comm)];
841 * Make sure we have a private signal table and that
842 * we are unassociated from the previous thread group.
844 retval = de_thread(current);
849 * Make sure we have private file handles. Ask the
850 * fork helper to do the work for us and the exit
851 * helper to do the cleanup of the old one.
853 files = current->files; /* refcounted so safe to hold */
854 retval = unshare_files();
858 * Release all of the old mmap stuff
860 retval = exec_mmap(bprm->mm);
864 bprm->mm = NULL; /* We're using it now */
866 /* This is the point of no return */
868 put_files_struct(files);
870 current->sas_ss_sp = current->sas_ss_size = 0;
872 if (current->euid == current->uid && current->egid == current->gid)
873 current->mm->dumpable = 1;
874 name = bprm->filename;
875 for (i=0; (ch = *(name++)) != '\0';) {
879 if (i < (sizeof(tcomm) - 1))
883 set_task_comm(current, tcomm);
887 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
888 permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
889 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
891 current->mm->dumpable = 0;
894 /* An exec changes our domain. We are no longer part of the thread
897 current->self_exec_id++;
899 flush_signal_handlers(current, 0);
900 flush_old_files(current->files);
905 put_files_struct(current->files);
906 current->files = files;
911 EXPORT_SYMBOL(flush_old_exec);
914 * Fill the binprm structure from the inode.
915 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
917 int prepare_binprm(struct linux_binprm *bprm)
920 struct inode * inode = bprm->file->f_dentry->d_inode;
923 mode = inode->i_mode;
925 * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
926 * generic_permission lets a non-executable through
928 if (!(mode & 0111)) /* with at least _one_ execute bit set */
930 if (bprm->file->f_op == NULL)
933 bprm->e_uid = current->euid;
934 bprm->e_gid = current->egid;
936 if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
938 if (mode & S_ISUID) {
939 current->personality &= ~PER_CLEAR_ON_SETID;
940 bprm->e_uid = inode->i_uid;
945 * If setgid is set but no group execute bit then this
946 * is a candidate for mandatory locking, not a setgid
949 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
950 current->personality &= ~PER_CLEAR_ON_SETID;
951 bprm->e_gid = inode->i_gid;
955 /* fill in binprm security blob */
956 retval = security_bprm_set(bprm);
960 memset(bprm->buf,0,BINPRM_BUF_SIZE);
961 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
964 EXPORT_SYMBOL(prepare_binprm);
966 static inline int unsafe_exec(struct task_struct *p)
969 if (p->ptrace & PT_PTRACED) {
970 if (p->ptrace & PT_PTRACE_CAP)
971 unsafe |= LSM_UNSAFE_PTRACE_CAP;
973 unsafe |= LSM_UNSAFE_PTRACE;
975 if (atomic_read(&p->fs->count) > 1 ||
976 atomic_read(&p->files->count) > 1 ||
977 atomic_read(&p->sighand->count) > 1)
978 unsafe |= LSM_UNSAFE_SHARE;
983 void compute_creds(struct linux_binprm *bprm)
987 if (bprm->e_uid != current->uid)
992 unsafe = unsafe_exec(current);
993 security_bprm_apply_creds(bprm, unsafe);
994 task_unlock(current);
995 security_bprm_post_apply_creds(bprm);
998 EXPORT_SYMBOL(compute_creds);
1000 void remove_arg_zero(struct linux_binprm *bprm)
1003 unsigned long offset;
1007 offset = bprm->p % PAGE_SIZE;
1010 while (bprm->p++, *(kaddr+offset++)) {
1011 if (offset != PAGE_SIZE)
1014 kunmap_atomic(kaddr, KM_USER0);
1016 page = bprm->page[bprm->p/PAGE_SIZE];
1017 kaddr = kmap_atomic(page, KM_USER0);
1019 kunmap_atomic(kaddr, KM_USER0);
1024 EXPORT_SYMBOL(remove_arg_zero);
1027 * cycle the list of binary formats handler, until one recognizes the image
1029 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1032 struct linux_binfmt *fmt;
1034 /* handle /sbin/loader.. */
1036 struct exec * eh = (struct exec *) bprm->buf;
1038 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1039 (eh->fh.f_flags & 0x3000) == 0x3000)
1042 unsigned long loader;
1044 allow_write_access(bprm->file);
1048 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1050 file = open_exec("/sbin/loader");
1051 retval = PTR_ERR(file);
1055 /* Remember if the application is TASO. */
1056 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1059 bprm->loader = loader;
1060 retval = prepare_binprm(bprm);
1063 /* should call search_binary_handler recursively here,
1064 but it does not matter */
1068 retval = security_bprm_check(bprm);
1072 /* kernel module loader fixup */
1073 /* so we don't try to load run modprobe in kernel space. */
1076 for (try=0; try<2; try++) {
1077 read_lock(&binfmt_lock);
1078 for (fmt = formats ; fmt ; fmt = fmt->next) {
1079 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1082 if (!try_module_get(fmt->module))
1084 read_unlock(&binfmt_lock);
1085 retval = fn(bprm, regs);
1088 allow_write_access(bprm->file);
1092 current->did_exec = 1;
1095 read_lock(&binfmt_lock);
1097 if (retval != -ENOEXEC || bprm->mm == NULL)
1100 read_unlock(&binfmt_lock);
1104 read_unlock(&binfmt_lock);
1105 if (retval != -ENOEXEC || bprm->mm == NULL) {
1109 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1110 if (printable(bprm->buf[0]) &&
1111 printable(bprm->buf[1]) &&
1112 printable(bprm->buf[2]) &&
1113 printable(bprm->buf[3]))
1114 break; /* -ENOEXEC */
1115 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1122 EXPORT_SYMBOL(search_binary_handler);
1125 * sys_execve() executes a new program.
1127 int do_execve(char * filename,
1128 char __user *__user *argv,
1129 char __user *__user *envp,
1130 struct pt_regs * regs)
1132 struct linux_binprm *bprm;
1138 bprm = kmalloc(sizeof(*bprm), GFP_KERNEL);
1141 memset(bprm, 0, sizeof(*bprm));
1143 file = open_exec(filename);
1144 retval = PTR_ERR(file);
1150 bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1153 bprm->filename = filename;
1154 bprm->interp = filename;
1155 bprm->mm = mm_alloc();
1160 retval = init_new_context(current, bprm->mm);
1164 bprm->argc = count(argv, bprm->p / sizeof(void *));
1165 if ((retval = bprm->argc) < 0)
1168 bprm->envc = count(envp, bprm->p / sizeof(void *));
1169 if ((retval = bprm->envc) < 0)
1172 retval = security_bprm_alloc(bprm);
1176 retval = prepare_binprm(bprm);
1180 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1184 bprm->exec = bprm->p;
1185 retval = copy_strings(bprm->envc, envp, bprm);
1189 retval = copy_strings(bprm->argc, argv, bprm);
1193 retval = search_binary_handler(bprm,regs);
1195 free_arg_pages(bprm);
1197 /* execve success */
1198 security_bprm_free(bprm);
1199 acct_update_integrals();
1200 update_mem_hiwater();
1206 /* Something went wrong, return the inode and free the argument pages*/
1207 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1208 struct page * page = bprm->page[i];
1214 security_bprm_free(bprm);
1222 allow_write_access(bprm->file);
1233 int set_binfmt(struct linux_binfmt *new)
1235 struct linux_binfmt *old = current->binfmt;
1238 if (!try_module_get(new->module))
1241 current->binfmt = new;
1243 module_put(old->module);
1247 EXPORT_SYMBOL(set_binfmt);
1249 #define CORENAME_MAX_SIZE 64
1251 /* format_corename will inspect the pattern parameter, and output a
1252 * name into corename, which must have space for at least
1253 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1255 static void format_corename(char *corename, const char *pattern, long signr)
1257 const char *pat_ptr = pattern;
1258 char *out_ptr = corename;
1259 char *const out_end = corename + CORENAME_MAX_SIZE;
1261 int pid_in_pattern = 0;
1263 /* Repeat as long as we have more pattern to process and more output
1266 if (*pat_ptr != '%') {
1267 if (out_ptr == out_end)
1269 *out_ptr++ = *pat_ptr++;
1271 switch (*++pat_ptr) {
1274 /* Double percent, output one percent */
1276 if (out_ptr == out_end)
1283 rc = snprintf(out_ptr, out_end - out_ptr,
1284 "%d", current->tgid);
1285 if (rc > out_end - out_ptr)
1291 rc = snprintf(out_ptr, out_end - out_ptr,
1292 "%d", current->uid);
1293 if (rc > out_end - out_ptr)
1299 rc = snprintf(out_ptr, out_end - out_ptr,
1300 "%d", current->gid);
1301 if (rc > out_end - out_ptr)
1305 /* signal that caused the coredump */
1307 rc = snprintf(out_ptr, out_end - out_ptr,
1309 if (rc > out_end - out_ptr)
1313 /* UNIX time of coredump */
1316 do_gettimeofday(&tv);
1317 rc = snprintf(out_ptr, out_end - out_ptr,
1319 if (rc > out_end - out_ptr)
1326 down_read(&uts_sem);
1327 rc = snprintf(out_ptr, out_end - out_ptr,
1328 "%s", system_utsname.nodename);
1330 if (rc > out_end - out_ptr)
1336 rc = snprintf(out_ptr, out_end - out_ptr,
1337 "%s", current->comm);
1338 if (rc > out_end - out_ptr)
1348 /* Backward compatibility with core_uses_pid:
1350 * If core_pattern does not include a %p (as is the default)
1351 * and core_uses_pid is set, then .%pid will be appended to
1354 && (core_uses_pid || atomic_read(¤t->mm->mm_users) != 1)) {
1355 rc = snprintf(out_ptr, out_end - out_ptr,
1356 ".%d", current->tgid);
1357 if (rc > out_end - out_ptr)
1365 static void zap_threads (struct mm_struct *mm)
1367 struct task_struct *g, *p;
1368 struct task_struct *tsk = current;
1369 struct completion *vfork_done = tsk->vfork_done;
1373 * Make sure nobody is waiting for us to release the VM,
1374 * otherwise we can deadlock when we wait on each other
1377 tsk->vfork_done = NULL;
1378 complete(vfork_done);
1381 read_lock(&tasklist_lock);
1383 if (mm == p->mm && p != tsk) {
1384 force_sig_specific(SIGKILL, p);
1386 if (unlikely(p->ptrace) &&
1387 unlikely(p->parent->mm == mm))
1390 while_each_thread(g,p);
1392 read_unlock(&tasklist_lock);
1394 if (unlikely(traced)) {
1396 * We are zapping a thread and the thread it ptraces.
1397 * If the tracee went into a ptrace stop for exit tracing,
1398 * we could deadlock since the tracer is waiting for this
1399 * coredump to finish. Detach them so they can both die.
1401 write_lock_irq(&tasklist_lock);
1402 do_each_thread(g,p) {
1403 if (mm == p->mm && p != tsk &&
1404 p->ptrace && p->parent->mm == mm) {
1407 } while_each_thread(g,p);
1408 write_unlock_irq(&tasklist_lock);
1412 static void coredump_wait(struct mm_struct *mm)
1414 DECLARE_COMPLETION(startup_done);
1416 mm->core_waiters++; /* let other threads block */
1417 mm->core_startup_done = &startup_done;
1419 /* give other threads a chance to run: */
1423 if (--mm->core_waiters) {
1424 up_write(&mm->mmap_sem);
1425 wait_for_completion(&startup_done);
1427 up_write(&mm->mmap_sem);
1428 BUG_ON(mm->core_waiters);
1431 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1433 char corename[CORENAME_MAX_SIZE + 1];
1434 struct mm_struct *mm = current->mm;
1435 struct linux_binfmt * binfmt;
1436 struct inode * inode;
1440 binfmt = current->binfmt;
1441 if (!binfmt || !binfmt->core_dump)
1443 down_write(&mm->mmap_sem);
1444 if (!mm->dumpable) {
1445 up_write(&mm->mmap_sem);
1449 init_completion(&mm->core_done);
1450 spin_lock_irq(¤t->sighand->siglock);
1451 current->signal->flags = SIGNAL_GROUP_EXIT;
1452 current->signal->group_exit_code = exit_code;
1453 spin_unlock_irq(¤t->sighand->siglock);
1457 * Clear any false indication of pending signals that might
1458 * be seen by the filesystem code called to write the core file.
1460 current->signal->group_stop_count = 0;
1461 clear_thread_flag(TIF_SIGPENDING);
1463 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1467 * lock_kernel() because format_corename() is controlled by sysctl, which
1468 * uses lock_kernel()
1471 format_corename(corename, core_pattern, signr);
1473 file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
1476 inode = file->f_dentry->d_inode;
1477 if (inode->i_nlink > 1)
1478 goto close_fail; /* multiple links - don't dump */
1479 if (d_unhashed(file->f_dentry))
1482 if (!S_ISREG(inode->i_mode))
1486 if (!file->f_op->write)
1488 if (do_truncate(file->f_dentry, 0) != 0)
1491 retval = binfmt->core_dump(signr, regs, file);
1494 current->signal->group_exit_code |= 0x80;
1496 filp_close(file, NULL);
1498 complete_all(&mm->core_done);