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/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/perf_event.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/proc_fs.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/tsacct_kern.h>
51 #include <linux/cn_proc.h>
52 #include <linux/audit.h>
53 #include <linux/tracehook.h>
54 #include <linux/kmod.h>
55 #include <linux/fsnotify.h>
56 #include <linux/fs_struct.h>
57 #include <linux/pipe_fs_i.h>
60 #include <asm/uaccess.h>
61 #include <asm/mmu_context.h>
66 char core_pattern[CORENAME_MAX_SIZE] = "core";
67 unsigned int core_pipe_limit;
68 int suid_dumpable = 0;
70 /* The maximal length of core_pattern is also specified in sysctl.c */
72 static LIST_HEAD(formats);
73 static DEFINE_RWLOCK(binfmt_lock);
76 * Also used in compat.c.
78 DEFINE_TRACE(fs_exec);
80 int __register_binfmt(struct linux_binfmt * fmt, int insert)
84 write_lock(&binfmt_lock);
85 insert ? list_add(&fmt->lh, &formats) :
86 list_add_tail(&fmt->lh, &formats);
87 write_unlock(&binfmt_lock);
91 EXPORT_SYMBOL(__register_binfmt);
93 void unregister_binfmt(struct linux_binfmt * fmt)
95 write_lock(&binfmt_lock);
97 write_unlock(&binfmt_lock);
100 EXPORT_SYMBOL(unregister_binfmt);
102 static inline void put_binfmt(struct linux_binfmt * fmt)
104 module_put(fmt->module);
108 * Note that a shared library must be both readable and executable due to
111 * Also note that we take the address to load from from the file itself.
113 SYSCALL_DEFINE1(uselib, const char __user *, library)
116 char *tmp = getname(library);
117 int error = PTR_ERR(tmp);
122 file = do_filp_open(AT_FDCWD, tmp,
123 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
124 MAY_READ | MAY_EXEC | MAY_OPEN);
126 error = PTR_ERR(file);
131 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
135 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
138 fsnotify_open(file->f_path.dentry);
142 struct linux_binfmt * fmt;
144 read_lock(&binfmt_lock);
145 list_for_each_entry(fmt, &formats, lh) {
146 if (!fmt->load_shlib)
148 if (!try_module_get(fmt->module))
150 read_unlock(&binfmt_lock);
151 error = fmt->load_shlib(file);
152 read_lock(&binfmt_lock);
154 if (error != -ENOEXEC)
157 read_unlock(&binfmt_lock);
167 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
173 #ifdef CONFIG_STACK_GROWSUP
175 ret = expand_stack_downwards(bprm->vma, pos);
180 ret = get_user_pages(current, bprm->mm, pos,
181 1, write, 1, &page, NULL);
186 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
190 * We've historically supported up to 32 pages (ARG_MAX)
191 * of argument strings even with small stacks
197 * Limit to 1/4-th the stack size for the argv+env strings.
199 * - the remaining binfmt code will not run out of stack space,
200 * - the program will have a reasonable amount of stack left
203 rlim = current->signal->rlim;
204 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
213 static void put_arg_page(struct page *page)
218 static void free_arg_page(struct linux_binprm *bprm, int i)
222 static void free_arg_pages(struct linux_binprm *bprm)
226 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
229 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
232 static int __bprm_mm_init(struct linux_binprm *bprm)
235 struct vm_area_struct *vma = NULL;
236 struct mm_struct *mm = bprm->mm;
238 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
242 down_write(&mm->mmap_sem);
246 * Place the stack at the largest stack address the architecture
247 * supports. Later, we'll move this to an appropriate place. We don't
248 * use STACK_TOP because that can depend on attributes which aren't
251 vma->vm_end = STACK_TOP_MAX;
252 vma->vm_start = vma->vm_end - PAGE_SIZE;
253 vma->vm_flags = VM_STACK_FLAGS;
254 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
255 err = insert_vm_struct(mm, vma);
259 mm->stack_vm = mm->total_vm = 1;
260 up_write(&mm->mmap_sem);
261 bprm->p = vma->vm_end - sizeof(void *);
264 up_write(&mm->mmap_sem);
266 kmem_cache_free(vm_area_cachep, vma);
270 static bool valid_arg_len(struct linux_binprm *bprm, long len)
272 return len <= MAX_ARG_STRLEN;
277 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
282 page = bprm->page[pos / PAGE_SIZE];
283 if (!page && write) {
284 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
287 bprm->page[pos / PAGE_SIZE] = page;
293 static void put_arg_page(struct page *page)
297 static void free_arg_page(struct linux_binprm *bprm, int i)
300 __free_page(bprm->page[i]);
301 bprm->page[i] = NULL;
305 static void free_arg_pages(struct linux_binprm *bprm)
309 for (i = 0; i < MAX_ARG_PAGES; i++)
310 free_arg_page(bprm, i);
313 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
318 static int __bprm_mm_init(struct linux_binprm *bprm)
320 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
324 static bool valid_arg_len(struct linux_binprm *bprm, long len)
326 return len <= bprm->p;
329 #endif /* CONFIG_MMU */
332 * Create a new mm_struct and populate it with a temporary stack
333 * vm_area_struct. We don't have enough context at this point to set the stack
334 * flags, permissions, and offset, so we use temporary values. We'll update
335 * them later in setup_arg_pages().
337 int bprm_mm_init(struct linux_binprm *bprm)
340 struct mm_struct *mm = NULL;
342 bprm->mm = mm = mm_alloc();
347 err = init_new_context(current, mm);
351 err = __bprm_mm_init(bprm);
367 * count() counts the number of strings in array ARGV.
369 static int count(char __user * __user * argv, int max)
377 if (get_user(p, argv))
391 * 'copy_strings()' copies argument/environment strings from the old
392 * processes's memory to the new process's stack. The call to get_user_pages()
393 * ensures the destination page is created and not swapped out.
395 static int copy_strings(int argc, char __user * __user * argv,
396 struct linux_binprm *bprm)
398 struct page *kmapped_page = NULL;
400 unsigned long kpos = 0;
408 if (get_user(str, argv+argc) ||
409 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
414 if (!valid_arg_len(bprm, len)) {
419 /* We're going to work our way backwords. */
425 int offset, bytes_to_copy;
427 offset = pos % PAGE_SIZE;
431 bytes_to_copy = offset;
432 if (bytes_to_copy > len)
435 offset -= bytes_to_copy;
436 pos -= bytes_to_copy;
437 str -= bytes_to_copy;
438 len -= bytes_to_copy;
440 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
443 page = get_arg_page(bprm, pos, 1);
450 flush_kernel_dcache_page(kmapped_page);
451 kunmap(kmapped_page);
452 put_arg_page(kmapped_page);
455 kaddr = kmap(kmapped_page);
456 kpos = pos & PAGE_MASK;
457 flush_arg_page(bprm, kpos, kmapped_page);
459 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
468 flush_kernel_dcache_page(kmapped_page);
469 kunmap(kmapped_page);
470 put_arg_page(kmapped_page);
476 * Like copy_strings, but get argv and its values from kernel memory.
478 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
481 mm_segment_t oldfs = get_fs();
483 r = copy_strings(argc, (char __user * __user *)argv, bprm);
487 EXPORT_SYMBOL(copy_strings_kernel);
492 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
493 * the binfmt code determines where the new stack should reside, we shift it to
494 * its final location. The process proceeds as follows:
496 * 1) Use shift to calculate the new vma endpoints.
497 * 2) Extend vma to cover both the old and new ranges. This ensures the
498 * arguments passed to subsequent functions are consistent.
499 * 3) Move vma's page tables to the new range.
500 * 4) Free up any cleared pgd range.
501 * 5) Shrink the vma to cover only the new range.
503 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
505 struct mm_struct *mm = vma->vm_mm;
506 unsigned long old_start = vma->vm_start;
507 unsigned long old_end = vma->vm_end;
508 unsigned long length = old_end - old_start;
509 unsigned long new_start = old_start - shift;
510 unsigned long new_end = old_end - shift;
511 struct mmu_gather *tlb;
513 BUG_ON(new_start > new_end);
516 * ensure there are no vmas between where we want to go
519 if (vma != find_vma(mm, new_start))
523 * cover the whole range: [new_start, old_end)
525 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
528 * move the page tables downwards, on failure we rely on
529 * process cleanup to remove whatever mess we made.
531 if (length != move_page_tables(vma, old_start,
532 vma, new_start, length))
536 tlb = tlb_gather_mmu(mm, 0);
537 if (new_end > old_start) {
539 * when the old and new regions overlap clear from new_end.
541 free_pgd_range(tlb, new_end, old_end, new_end,
542 vma->vm_next ? vma->vm_next->vm_start : 0);
545 * otherwise, clean from old_start; this is done to not touch
546 * the address space in [new_end, old_start) some architectures
547 * have constraints on va-space that make this illegal (IA64) -
548 * for the others its just a little faster.
550 free_pgd_range(tlb, old_start, old_end, new_end,
551 vma->vm_next ? vma->vm_next->vm_start : 0);
553 tlb_finish_mmu(tlb, new_end, old_end);
556 * shrink the vma to just the new range.
558 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
563 #define EXTRA_STACK_VM_PAGES 20 /* random */
566 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
567 * the stack is optionally relocated, and some extra space is added.
569 int setup_arg_pages(struct linux_binprm *bprm,
570 unsigned long stack_top,
571 int executable_stack)
574 unsigned long stack_shift;
575 struct mm_struct *mm = current->mm;
576 struct vm_area_struct *vma = bprm->vma;
577 struct vm_area_struct *prev = NULL;
578 unsigned long vm_flags;
579 unsigned long stack_base;
580 unsigned long stack_size;
581 unsigned long stack_expand;
582 unsigned long rlim_stack;
584 #ifdef CONFIG_STACK_GROWSUP
585 /* Limit stack size to 1GB */
586 stack_base = rlimit_max(RLIMIT_STACK);
587 if (stack_base > (1 << 30))
588 stack_base = 1 << 30;
590 /* Make sure we didn't let the argument array grow too large. */
591 if (vma->vm_end - vma->vm_start > stack_base)
594 stack_base = PAGE_ALIGN(stack_top - stack_base);
596 stack_shift = vma->vm_start - stack_base;
597 mm->arg_start = bprm->p - stack_shift;
598 bprm->p = vma->vm_end - stack_shift;
600 stack_top = arch_align_stack(stack_top);
601 stack_top = PAGE_ALIGN(stack_top);
602 stack_shift = vma->vm_end - stack_top;
604 bprm->p -= stack_shift;
605 mm->arg_start = bprm->p;
609 bprm->loader -= stack_shift;
610 bprm->exec -= stack_shift;
612 down_write(&mm->mmap_sem);
613 vm_flags = VM_STACK_FLAGS;
616 * Adjust stack execute permissions; explicitly enable for
617 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
618 * (arch default) otherwise.
620 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
622 else if (executable_stack == EXSTACK_DISABLE_X)
623 vm_flags &= ~VM_EXEC;
624 vm_flags |= mm->def_flags;
626 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
632 /* Move stack pages down in memory. */
634 ret = shift_arg_pages(vma, stack_shift);
639 stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
640 stack_size = vma->vm_end - vma->vm_start;
642 * Align this down to a page boundary as expand_stack
645 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
646 #ifdef CONFIG_STACK_GROWSUP
647 if (stack_size + stack_expand > rlim_stack)
648 stack_base = vma->vm_start + rlim_stack;
650 stack_base = vma->vm_end + stack_expand;
652 if (stack_size + stack_expand > rlim_stack)
653 stack_base = vma->vm_end - rlim_stack;
655 stack_base = vma->vm_start - stack_expand;
657 ret = expand_stack(vma, stack_base);
662 up_write(&mm->mmap_sem);
665 EXPORT_SYMBOL(setup_arg_pages);
667 #endif /* CONFIG_MMU */
669 struct file *open_exec(const char *name)
674 file = do_filp_open(AT_FDCWD, name,
675 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
676 MAY_EXEC | MAY_OPEN);
681 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
684 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
687 fsnotify_open(file->f_path.dentry);
689 if (file->f_op && file->f_op->open_exec) {
690 err = file->f_op->open_exec(file->f_path.dentry->d_inode);
695 err = deny_write_access(file);
706 EXPORT_SYMBOL(open_exec);
708 int kernel_read(struct file *file, loff_t offset,
709 char *addr, unsigned long count)
717 /* The cast to a user pointer is valid due to the set_fs() */
718 result = vfs_read(file, (void __user *)addr, count, &pos);
723 EXPORT_SYMBOL(kernel_read);
725 static int exec_mmap(struct mm_struct *mm)
727 struct task_struct *tsk;
728 struct mm_struct * old_mm, *active_mm;
730 /* Notify parent that we're no longer interested in the old VM */
732 old_mm = current->mm;
733 mm_release(tsk, old_mm);
737 * Make sure that if there is a core dump in progress
738 * for the old mm, we get out and die instead of going
739 * through with the exec. We must hold mmap_sem around
740 * checking core_state and changing tsk->mm.
742 down_read(&old_mm->mmap_sem);
743 if (unlikely(old_mm->core_state)) {
744 up_read(&old_mm->mmap_sem);
749 active_mm = tsk->active_mm;
752 activate_mm(active_mm, mm);
754 arch_pick_mmap_layout(mm);
756 up_read(&old_mm->mmap_sem);
757 BUG_ON(active_mm != old_mm);
758 mm_update_next_owner(old_mm);
767 * This function makes sure the current process has its own signal table,
768 * so that flush_signal_handlers can later reset the handlers without
769 * disturbing other processes. (Other processes might share the signal
770 * table via the CLONE_SIGHAND option to clone().)
772 static int de_thread(struct task_struct *tsk)
774 struct signal_struct *sig = tsk->signal;
775 struct sighand_struct *oldsighand = tsk->sighand;
776 spinlock_t *lock = &oldsighand->siglock;
779 if (thread_group_empty(tsk))
780 goto no_thread_group;
783 * Kill all other threads in the thread group.
786 if (signal_group_exit(sig)) {
788 * Another group action in progress, just
789 * return so that the signal is processed.
791 spin_unlock_irq(lock);
794 sig->group_exit_task = tsk;
795 zap_other_threads(tsk);
797 /* Account for the thread group leader hanging around: */
798 count = thread_group_leader(tsk) ? 1 : 2;
799 sig->notify_count = count;
800 while (atomic_read(&sig->count) > count) {
801 __set_current_state(TASK_UNINTERRUPTIBLE);
802 spin_unlock_irq(lock);
806 spin_unlock_irq(lock);
809 * At this point all other threads have exited, all we have to
810 * do is to wait for the thread group leader to become inactive,
811 * and to assume its PID:
813 if (!thread_group_leader(tsk)) {
814 struct task_struct *leader = tsk->group_leader;
816 sig->notify_count = -1; /* for exit_notify() */
818 write_lock_irq(&tasklist_lock);
819 if (likely(leader->exit_state))
821 __set_current_state(TASK_UNINTERRUPTIBLE);
822 write_unlock_irq(&tasklist_lock);
827 * The only record we have of the real-time age of a
828 * process, regardless of execs it's done, is start_time.
829 * All the past CPU time is accumulated in signal_struct
830 * from sister threads now dead. But in this non-leader
831 * exec, nothing survives from the original leader thread,
832 * whose birth marks the true age of this process now.
833 * When we take on its identity by switching to its PID, we
834 * also take its birthdate (always earlier than our own).
836 tsk->start_time = leader->start_time;
838 BUG_ON(!same_thread_group(leader, tsk));
839 BUG_ON(has_group_leader_pid(tsk));
841 * An exec() starts a new thread group with the
842 * TGID of the previous thread group. Rehash the
843 * two threads with a switched PID, and release
844 * the former thread group leader:
847 /* Become a process group leader with the old leader's pid.
848 * The old leader becomes a thread of the this thread group.
849 * Note: The old leader also uses this pid until release_task
850 * is called. Odd but simple and correct.
852 detach_pid(tsk, PIDTYPE_PID);
853 tsk->pid = leader->pid;
854 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
855 transfer_pid(leader, tsk, PIDTYPE_PGID);
856 transfer_pid(leader, tsk, PIDTYPE_SID);
858 list_replace_rcu(&leader->tasks, &tsk->tasks);
859 list_replace_init(&leader->sibling, &tsk->sibling);
861 tsk->group_leader = tsk;
862 leader->group_leader = tsk;
864 tsk->exit_signal = SIGCHLD;
866 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
867 leader->exit_state = EXIT_DEAD;
868 write_unlock_irq(&tasklist_lock);
870 release_task(leader);
873 sig->group_exit_task = NULL;
874 sig->notify_count = 0;
878 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
881 flush_itimer_signals();
883 if (atomic_read(&oldsighand->count) != 1) {
884 struct sighand_struct *newsighand;
886 * This ->sighand is shared with the CLONE_SIGHAND
887 * but not CLONE_THREAD task, switch to the new one.
889 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
893 atomic_set(&newsighand->count, 1);
894 memcpy(newsighand->action, oldsighand->action,
895 sizeof(newsighand->action));
897 write_lock_irq(&tasklist_lock);
898 spin_lock(&oldsighand->siglock);
899 rcu_assign_pointer(tsk->sighand, newsighand);
900 spin_unlock(&oldsighand->siglock);
901 write_unlock_irq(&tasklist_lock);
903 __cleanup_sighand(oldsighand);
906 BUG_ON(!thread_group_leader(tsk));
911 * These functions flushes out all traces of the currently running executable
912 * so that a new one can be started
914 static void flush_old_files(struct files_struct * files)
919 spin_lock(&files->file_lock);
921 unsigned long set, i;
925 fdt = files_fdtable(files);
926 if (i >= fdt->max_fds)
928 set = fdt->close_on_exec->fds_bits[j];
931 fdt->close_on_exec->fds_bits[j] = 0;
932 spin_unlock(&files->file_lock);
933 for ( ; set ; i++,set >>= 1) {
938 spin_lock(&files->file_lock);
941 spin_unlock(&files->file_lock);
944 char *get_task_comm(char *buf, struct task_struct *tsk)
946 /* buf must be at least sizeof(tsk->comm) in size */
948 strncpy(buf, tsk->comm, sizeof(tsk->comm));
953 void set_task_comm(struct task_struct *tsk, char *buf)
958 * Threads may access current->comm without holding
959 * the task lock, so write the string carefully.
960 * Readers without a lock may see incomplete new
961 * names but are safe from non-terminating string reads.
963 memset(tsk->comm, 0, TASK_COMM_LEN);
965 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
967 perf_event_comm(tsk);
970 int flush_old_exec(struct linux_binprm * bprm)
975 * Make sure we have a private signal table and that
976 * we are unassociated from the previous thread group.
978 retval = de_thread(current);
982 set_mm_exe_file(bprm->mm, bprm->file);
985 * Release all of the old mmap stuff
987 retval = exec_mmap(bprm->mm);
991 bprm->mm = NULL; /* We're using it now */
993 current->flags &= ~PF_RANDOMIZE;
995 current->personality &= ~bprm->per_clear;
1002 EXPORT_SYMBOL(flush_old_exec);
1004 void setup_new_exec(struct linux_binprm * bprm)
1008 char tcomm[sizeof(current->comm)];
1010 arch_pick_mmap_layout(current->mm);
1012 /* This is the point of no return */
1013 current->sas_ss_sp = current->sas_ss_size = 0;
1015 if (current_euid() == current_uid() && current_egid() == current_gid())
1016 set_dumpable(current->mm, 1);
1018 set_dumpable(current->mm, suid_dumpable);
1020 name = bprm->filename;
1022 /* Copies the binary name from after last slash */
1023 for (i=0; (ch = *(name++)) != '\0';) {
1025 i = 0; /* overwrite what we wrote */
1027 if (i < (sizeof(tcomm) - 1))
1031 set_task_comm(current, tcomm);
1033 /* Set the new mm task size. We have to do that late because it may
1034 * depend on TIF_32BIT which is only updated in flush_thread() on
1035 * some architectures like powerpc
1037 current->mm->task_size = TASK_SIZE;
1039 /* install the new credentials */
1040 if (bprm->cred->uid != current_euid() ||
1041 bprm->cred->gid != current_egid()) {
1042 current->pdeath_signal = 0;
1043 } else if (file_permission(bprm->file, MAY_READ) ||
1044 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1045 set_dumpable(current->mm, suid_dumpable);
1049 * Flush performance counters when crossing a
1052 if (!get_dumpable(current->mm))
1053 perf_event_exit_task(current);
1055 /* An exec changes our domain. We are no longer part of the thread
1058 current->self_exec_id++;
1060 flush_signal_handlers(current, 0);
1061 flush_old_files(current->files);
1063 EXPORT_SYMBOL(setup_new_exec);
1066 * Prepare credentials and lock ->cred_guard_mutex.
1067 * install_exec_creds() commits the new creds and drops the lock.
1068 * Or, if exec fails before, free_bprm() should release ->cred and
1071 int prepare_bprm_creds(struct linux_binprm *bprm)
1073 if (mutex_lock_interruptible(¤t->cred_guard_mutex))
1074 return -ERESTARTNOINTR;
1076 bprm->cred = prepare_exec_creds();
1077 if (likely(bprm->cred))
1080 mutex_unlock(¤t->cred_guard_mutex);
1084 void free_bprm(struct linux_binprm *bprm)
1086 free_arg_pages(bprm);
1088 mutex_unlock(¤t->cred_guard_mutex);
1089 abort_creds(bprm->cred);
1095 * install the new credentials for this executable
1097 void install_exec_creds(struct linux_binprm *bprm)
1099 security_bprm_committing_creds(bprm);
1101 commit_creds(bprm->cred);
1104 * cred_guard_mutex must be held at least to this point to prevent
1105 * ptrace_attach() from altering our determination of the task's
1106 * credentials; any time after this it may be unlocked.
1108 security_bprm_committed_creds(bprm);
1109 mutex_unlock(¤t->cred_guard_mutex);
1111 EXPORT_SYMBOL(install_exec_creds);
1114 * determine how safe it is to execute the proposed program
1115 * - the caller must hold current->cred_guard_mutex to protect against
1118 int check_unsafe_exec(struct linux_binprm *bprm)
1120 struct task_struct *p = current, *t;
1124 bprm->unsafe = tracehook_unsafe_exec(p);
1127 write_lock(&p->fs->lock);
1129 for (t = next_thread(p); t != p; t = next_thread(t)) {
1135 if (p->fs->users > n_fs) {
1136 bprm->unsafe |= LSM_UNSAFE_SHARE;
1139 if (!p->fs->in_exec) {
1144 write_unlock(&p->fs->lock);
1150 * Fill the binprm structure from the inode.
1151 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1153 * This may be called multiple times for binary chains (scripts for example).
1155 int prepare_binprm(struct linux_binprm *bprm)
1158 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1161 mode = inode->i_mode;
1162 if (bprm->file->f_op == NULL)
1165 /* clear any previous set[ug]id data from a previous binary */
1166 bprm->cred->euid = current_euid();
1167 bprm->cred->egid = current_egid();
1169 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1171 if (mode & S_ISUID) {
1172 bprm->per_clear |= PER_CLEAR_ON_SETID;
1173 bprm->cred->euid = inode->i_uid;
1178 * If setgid is set but no group execute bit then this
1179 * is a candidate for mandatory locking, not a setgid
1182 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1183 bprm->per_clear |= PER_CLEAR_ON_SETID;
1184 bprm->cred->egid = inode->i_gid;
1188 /* fill in binprm security blob */
1189 retval = security_bprm_set_creds(bprm);
1192 bprm->cred_prepared = 1;
1194 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1195 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1198 EXPORT_SYMBOL(prepare_binprm);
1201 * Arguments are '\0' separated strings found at the location bprm->p
1202 * points to; chop off the first by relocating brpm->p to right after
1203 * the first '\0' encountered.
1205 int remove_arg_zero(struct linux_binprm *bprm)
1208 unsigned long offset;
1216 offset = bprm->p & ~PAGE_MASK;
1217 page = get_arg_page(bprm, bprm->p, 0);
1222 kaddr = kmap_atomic(page, KM_USER0);
1224 for (; offset < PAGE_SIZE && kaddr[offset];
1225 offset++, bprm->p++)
1228 kunmap_atomic(kaddr, KM_USER0);
1231 if (offset == PAGE_SIZE)
1232 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1233 } while (offset == PAGE_SIZE);
1242 EXPORT_SYMBOL(remove_arg_zero);
1245 * cycle the list of binary formats handler, until one recognizes the image
1247 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1249 unsigned int depth = bprm->recursion_depth;
1251 struct linux_binfmt *fmt;
1253 retval = security_bprm_check(bprm);
1257 /* kernel module loader fixup */
1258 /* so we don't try to load run modprobe in kernel space. */
1261 retval = audit_bprm(bprm);
1266 for (try=0; try<2; try++) {
1267 read_lock(&binfmt_lock);
1268 list_for_each_entry(fmt, &formats, lh) {
1269 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1272 if (!try_module_get(fmt->module))
1274 read_unlock(&binfmt_lock);
1275 retval = fn(bprm, regs);
1277 * Restore the depth counter to its starting value
1278 * in this call, so we don't have to rely on every
1279 * load_binary function to restore it on return.
1281 bprm->recursion_depth = depth;
1284 tracehook_report_exec(fmt, bprm, regs);
1286 allow_write_access(bprm->file);
1290 current->did_exec = 1;
1291 proc_exec_connector(current);
1294 read_lock(&binfmt_lock);
1296 if (retval != -ENOEXEC || bprm->mm == NULL)
1299 read_unlock(&binfmt_lock);
1303 read_unlock(&binfmt_lock);
1304 if (retval != -ENOEXEC || bprm->mm == NULL) {
1306 #ifdef CONFIG_MODULES
1308 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1309 if (printable(bprm->buf[0]) &&
1310 printable(bprm->buf[1]) &&
1311 printable(bprm->buf[2]) &&
1312 printable(bprm->buf[3]))
1313 break; /* -ENOEXEC */
1314 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1321 EXPORT_SYMBOL(search_binary_handler);
1324 * sys_execve() executes a new program.
1326 int do_execve(char * filename,
1327 char __user *__user *argv,
1328 char __user *__user *envp,
1329 struct pt_regs * regs)
1331 struct linux_binprm *bprm;
1333 struct files_struct *displaced;
1337 retval = unshare_files(&displaced);
1342 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1346 retval = prepare_bprm_creds(bprm);
1350 retval = check_unsafe_exec(bprm);
1353 clear_in_exec = retval;
1354 current->in_execve = 1;
1356 file = open_exec(filename);
1357 retval = PTR_ERR(file);
1364 bprm->filename = filename;
1365 bprm->interp = filename;
1367 retval = bprm_mm_init(bprm);
1371 bprm->argc = count(argv, MAX_ARG_STRINGS);
1372 if ((retval = bprm->argc) < 0)
1375 bprm->envc = count(envp, MAX_ARG_STRINGS);
1376 if ((retval = bprm->envc) < 0)
1379 retval = prepare_binprm(bprm);
1383 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1387 bprm->exec = bprm->p;
1388 retval = copy_strings(bprm->envc, envp, bprm);
1392 retval = copy_strings(bprm->argc, argv, bprm);
1396 current->flags &= ~PF_KTHREAD;
1397 retval = search_binary_handler(bprm,regs);
1401 current->stack_start = current->mm->start_stack;
1403 trace_fs_exec(filename);
1404 /* execve succeeded */
1405 current->fs->in_exec = 0;
1406 current->in_execve = 0;
1407 acct_update_integrals(current);
1410 put_files_struct(displaced);
1419 allow_write_access(bprm->file);
1425 current->fs->in_exec = 0;
1426 current->in_execve = 0;
1433 reset_files_struct(displaced);
1438 void set_binfmt(struct linux_binfmt *new)
1440 struct mm_struct *mm = current->mm;
1443 module_put(mm->binfmt->module);
1447 __module_get(new->module);
1450 EXPORT_SYMBOL(set_binfmt);
1452 /* format_corename will inspect the pattern parameter, and output a
1453 * name into corename, which must have space for at least
1454 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1456 static int format_corename(char *corename, long signr)
1458 const struct cred *cred = current_cred();
1459 const char *pat_ptr = core_pattern;
1460 int ispipe = (*pat_ptr == '|');
1461 char *out_ptr = corename;
1462 char *const out_end = corename + CORENAME_MAX_SIZE;
1464 int pid_in_pattern = 0;
1466 /* Repeat as long as we have more pattern to process and more output
1469 if (*pat_ptr != '%') {
1470 if (out_ptr == out_end)
1472 *out_ptr++ = *pat_ptr++;
1474 switch (*++pat_ptr) {
1477 /* Double percent, output one percent */
1479 if (out_ptr == out_end)
1486 rc = snprintf(out_ptr, out_end - out_ptr,
1487 "%d", task_tgid_vnr(current));
1488 if (rc > out_end - out_ptr)
1494 rc = snprintf(out_ptr, out_end - out_ptr,
1496 if (rc > out_end - out_ptr)
1502 rc = snprintf(out_ptr, out_end - out_ptr,
1504 if (rc > out_end - out_ptr)
1508 /* signal that caused the coredump */
1510 rc = snprintf(out_ptr, out_end - out_ptr,
1512 if (rc > out_end - out_ptr)
1516 /* UNIX time of coredump */
1519 do_gettimeofday(&tv);
1520 rc = snprintf(out_ptr, out_end - out_ptr,
1522 if (rc > out_end - out_ptr)
1529 down_read(&uts_sem);
1530 rc = snprintf(out_ptr, out_end - out_ptr,
1531 "%s", utsname()->nodename);
1533 if (rc > out_end - out_ptr)
1539 rc = snprintf(out_ptr, out_end - out_ptr,
1540 "%s", current->comm);
1541 if (rc > out_end - out_ptr)
1545 /* core limit size */
1547 rc = snprintf(out_ptr, out_end - out_ptr,
1548 "%lu", rlimit(RLIMIT_CORE));
1549 if (rc > out_end - out_ptr)
1559 /* Backward compatibility with core_uses_pid:
1561 * If core_pattern does not include a %p (as is the default)
1562 * and core_uses_pid is set, then .%pid will be appended to
1563 * the filename. Do not do this for piped commands. */
1564 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1565 rc = snprintf(out_ptr, out_end - out_ptr,
1566 ".%d", task_tgid_vnr(current));
1567 if (rc > out_end - out_ptr)
1576 static int zap_process(struct task_struct *start)
1578 struct task_struct *t;
1581 start->signal->flags = SIGNAL_GROUP_EXIT;
1582 start->signal->group_stop_count = 0;
1586 if (t != current && t->mm) {
1587 sigaddset(&t->pending.signal, SIGKILL);
1588 signal_wake_up(t, 1);
1591 } while_each_thread(start, t);
1596 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1597 struct core_state *core_state, int exit_code)
1599 struct task_struct *g, *p;
1600 unsigned long flags;
1603 spin_lock_irq(&tsk->sighand->siglock);
1604 if (!signal_group_exit(tsk->signal)) {
1605 mm->core_state = core_state;
1606 tsk->signal->group_exit_code = exit_code;
1607 nr = zap_process(tsk);
1609 spin_unlock_irq(&tsk->sighand->siglock);
1610 if (unlikely(nr < 0))
1613 if (atomic_read(&mm->mm_users) == nr + 1)
1616 * We should find and kill all tasks which use this mm, and we should
1617 * count them correctly into ->nr_threads. We don't take tasklist
1618 * lock, but this is safe wrt:
1621 * None of sub-threads can fork after zap_process(leader). All
1622 * processes which were created before this point should be
1623 * visible to zap_threads() because copy_process() adds the new
1624 * process to the tail of init_task.tasks list, and lock/unlock
1625 * of ->siglock provides a memory barrier.
1628 * The caller holds mm->mmap_sem. This means that the task which
1629 * uses this mm can't pass exit_mm(), so it can't exit or clear
1633 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1634 * we must see either old or new leader, this does not matter.
1635 * However, it can change p->sighand, so lock_task_sighand(p)
1636 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1639 * Note also that "g" can be the old leader with ->mm == NULL
1640 * and already unhashed and thus removed from ->thread_group.
1641 * This is OK, __unhash_process()->list_del_rcu() does not
1642 * clear the ->next pointer, we will find the new leader via
1646 for_each_process(g) {
1647 if (g == tsk->group_leader)
1649 if (g->flags & PF_KTHREAD)
1654 if (unlikely(p->mm == mm)) {
1655 lock_task_sighand(p, &flags);
1656 nr += zap_process(p);
1657 unlock_task_sighand(p, &flags);
1661 } while_each_thread(g, p);
1665 atomic_set(&core_state->nr_threads, nr);
1669 static int coredump_wait(int exit_code, struct core_state *core_state)
1671 struct task_struct *tsk = current;
1672 struct mm_struct *mm = tsk->mm;
1673 struct completion *vfork_done;
1676 init_completion(&core_state->startup);
1677 core_state->dumper.task = tsk;
1678 core_state->dumper.next = NULL;
1679 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1680 up_write(&mm->mmap_sem);
1682 if (unlikely(core_waiters < 0))
1686 * Make sure nobody is waiting for us to release the VM,
1687 * otherwise we can deadlock when we wait on each other
1689 vfork_done = tsk->vfork_done;
1691 tsk->vfork_done = NULL;
1692 complete(vfork_done);
1696 wait_for_completion(&core_state->startup);
1698 return core_waiters;
1701 static void coredump_finish(struct mm_struct *mm)
1703 struct core_thread *curr, *next;
1704 struct task_struct *task;
1706 next = mm->core_state->dumper.next;
1707 while ((curr = next) != NULL) {
1711 * see exit_mm(), curr->task must not see
1712 * ->task == NULL before we read ->next.
1716 wake_up_process(task);
1719 mm->core_state = NULL;
1723 * set_dumpable converts traditional three-value dumpable to two flags and
1724 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1725 * these bits are not changed atomically. So get_dumpable can observe the
1726 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1727 * return either old dumpable or new one by paying attention to the order of
1728 * modifying the bits.
1730 * dumpable | mm->flags (binary)
1731 * old new | initial interim final
1732 * ---------+-----------------------
1740 * (*) get_dumpable regards interim value of 10 as 11.
1742 void set_dumpable(struct mm_struct *mm, int value)
1746 clear_bit(MMF_DUMPABLE, &mm->flags);
1748 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1751 set_bit(MMF_DUMPABLE, &mm->flags);
1753 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1756 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1758 set_bit(MMF_DUMPABLE, &mm->flags);
1763 int get_dumpable(struct mm_struct *mm)
1767 ret = mm->flags & 0x3;
1768 return (ret >= 2) ? 2 : ret;
1771 static void wait_for_dump_helpers(struct file *file)
1773 struct pipe_inode_info *pipe;
1775 pipe = file->f_path.dentry->d_inode->i_pipe;
1781 while ((pipe->readers > 1) && (!signal_pending(current))) {
1782 wake_up_interruptible_sync(&pipe->wait);
1783 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1794 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1796 struct core_state core_state;
1797 char corename[CORENAME_MAX_SIZE + 1];
1798 struct mm_struct *mm = current->mm;
1799 struct linux_binfmt * binfmt;
1800 struct inode * inode;
1801 const struct cred *old_cred;
1806 char **helper_argv = NULL;
1807 int helper_argc = 0;
1809 static atomic_t core_dump_count = ATOMIC_INIT(0);
1810 struct coredump_params cprm = {
1813 .limit = rlimit(RLIMIT_CORE),
1816 audit_core_dumps(signr);
1818 binfmt = mm->binfmt;
1819 if (!binfmt || !binfmt->core_dump)
1822 cred = prepare_creds();
1828 down_write(&mm->mmap_sem);
1830 * If another thread got here first, or we are not dumpable, bail out.
1832 if (mm->core_state || !get_dumpable(mm)) {
1833 up_write(&mm->mmap_sem);
1839 * We cannot trust fsuid as being the "true" uid of the
1840 * process nor do we know its entire history. We only know it
1841 * was tainted so we dump it as root in mode 2.
1843 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1844 flag = O_EXCL; /* Stop rewrite attacks */
1845 cred->fsuid = 0; /* Dump root private */
1848 retval = coredump_wait(exit_code, &core_state);
1854 old_cred = override_creds(cred);
1857 * Clear any false indication of pending signals that might
1858 * be seen by the filesystem code called to write the core file.
1860 clear_thread_flag(TIF_SIGPENDING);
1863 * lock_kernel() because format_corename() is controlled by sysctl, which
1864 * uses lock_kernel()
1867 ispipe = format_corename(corename, signr);
1870 if ((!ispipe) && (cprm.limit < binfmt->min_coredump))
1874 if (cprm.limit == 0) {
1876 * Normally core limits are irrelevant to pipes, since
1877 * we're not writing to the file system, but we use
1878 * cprm.limit of 0 here as a speacial value. Any
1879 * non-zero limit gets set to RLIM_INFINITY below, but
1880 * a limit of 0 skips the dump. This is a consistent
1881 * way to catch recursive crashes. We can still crash
1882 * if the core_pattern binary sets RLIM_CORE = !0
1883 * but it runs as root, and can do lots of stupid things
1884 * Note that we use task_tgid_vnr here to grab the pid
1885 * of the process group leader. That way we get the
1886 * right pid if a thread in a multi-threaded
1887 * core_pattern process dies.
1890 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1891 task_tgid_vnr(current), current->comm);
1892 printk(KERN_WARNING "Aborting core\n");
1896 dump_count = atomic_inc_return(&core_dump_count);
1897 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1898 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1899 task_tgid_vnr(current), current->comm);
1900 printk(KERN_WARNING "Skipping core dump\n");
1901 goto fail_dropcount;
1904 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1906 printk(KERN_WARNING "%s failed to allocate memory\n",
1908 goto fail_dropcount;
1911 cprm.limit = RLIM_INFINITY;
1913 /* SIGPIPE can happen, but it's just never processed */
1914 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1916 printk(KERN_INFO "Core dump to %s pipe failed\n",
1918 goto fail_dropcount;
1921 cprm.file = filp_open(corename,
1922 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1924 if (IS_ERR(cprm.file))
1925 goto fail_dropcount;
1926 inode = cprm.file->f_path.dentry->d_inode;
1927 if (inode->i_nlink > 1)
1928 goto close_fail; /* multiple links - don't dump */
1929 if (!ispipe && d_unhashed(cprm.file->f_path.dentry))
1932 /* AK: actually i see no reason to not allow this for named pipes etc.,
1933 but keep the previous behaviour for now. */
1934 if (!ispipe && !S_ISREG(inode->i_mode))
1937 * Dont allow local users get cute and trick others to coredump
1938 * into their pre-created files:
1940 if (inode->i_uid != current_fsuid())
1942 if (!cprm.file->f_op)
1944 if (!cprm.file->f_op->write)
1947 do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file) != 0)
1950 retval = binfmt->core_dump(&cprm);
1953 current->signal->group_exit_code |= 0x80;
1955 if (ispipe && core_pipe_limit)
1956 wait_for_dump_helpers(cprm.file);
1957 filp_close(cprm.file, NULL);
1960 atomic_dec(&core_dump_count);
1963 argv_free(helper_argv);
1965 revert_creds(old_cred);
1967 coredump_finish(mm);