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>
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
65 char core_pattern[CORENAME_MAX_SIZE] = "core";
66 unsigned int core_pipe_limit;
67 int suid_dumpable = 0;
69 /* The maximal length of core_pattern is also specified in sysctl.c */
71 static LIST_HEAD(formats);
72 static DEFINE_RWLOCK(binfmt_lock);
74 int __register_binfmt(struct linux_binfmt * fmt, int insert)
78 write_lock(&binfmt_lock);
79 insert ? list_add(&fmt->lh, &formats) :
80 list_add_tail(&fmt->lh, &formats);
81 write_unlock(&binfmt_lock);
85 EXPORT_SYMBOL(__register_binfmt);
87 void unregister_binfmt(struct linux_binfmt * fmt)
89 write_lock(&binfmt_lock);
91 write_unlock(&binfmt_lock);
94 EXPORT_SYMBOL(unregister_binfmt);
96 static inline void put_binfmt(struct linux_binfmt * fmt)
98 module_put(fmt->module);
102 * Note that a shared library must be both readable and executable due to
105 * Also note that we take the address to load from from the file itself.
107 SYSCALL_DEFINE1(uselib, const char __user *, library)
110 char *tmp = getname(library);
111 int error = PTR_ERR(tmp);
116 file = do_filp_open(AT_FDCWD, tmp,
117 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
118 MAY_READ | MAY_EXEC | MAY_OPEN);
120 error = PTR_ERR(file);
125 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
129 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
132 fsnotify_open(file->f_path.dentry);
136 struct linux_binfmt * fmt;
138 read_lock(&binfmt_lock);
139 list_for_each_entry(fmt, &formats, lh) {
140 if (!fmt->load_shlib)
142 if (!try_module_get(fmt->module))
144 read_unlock(&binfmt_lock);
145 error = fmt->load_shlib(file);
146 read_lock(&binfmt_lock);
148 if (error != -ENOEXEC)
151 read_unlock(&binfmt_lock);
161 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
167 #ifdef CONFIG_STACK_GROWSUP
169 ret = expand_stack_downwards(bprm->vma, pos);
174 ret = get_user_pages(current, bprm->mm, pos,
175 1, write, 1, &page, NULL);
180 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
184 * We've historically supported up to 32 pages (ARG_MAX)
185 * of argument strings even with small stacks
191 * Limit to 1/4-th the stack size for the argv+env strings.
193 * - the remaining binfmt code will not run out of stack space,
194 * - the program will have a reasonable amount of stack left
197 rlim = current->signal->rlim;
198 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
207 static void put_arg_page(struct page *page)
212 static void free_arg_page(struct linux_binprm *bprm, int i)
216 static void free_arg_pages(struct linux_binprm *bprm)
220 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
223 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
226 static int __bprm_mm_init(struct linux_binprm *bprm)
229 struct vm_area_struct *vma = NULL;
230 struct mm_struct *mm = bprm->mm;
232 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
236 down_write(&mm->mmap_sem);
240 * Place the stack at the largest stack address the architecture
241 * supports. Later, we'll move this to an appropriate place. We don't
242 * use STACK_TOP because that can depend on attributes which aren't
245 vma->vm_end = STACK_TOP_MAX;
246 vma->vm_start = vma->vm_end - PAGE_SIZE;
247 vma->vm_flags = VM_STACK_FLAGS;
248 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
249 err = insert_vm_struct(mm, vma);
253 mm->stack_vm = mm->total_vm = 1;
254 up_write(&mm->mmap_sem);
255 bprm->p = vma->vm_end - sizeof(void *);
258 up_write(&mm->mmap_sem);
260 kmem_cache_free(vm_area_cachep, vma);
264 static bool valid_arg_len(struct linux_binprm *bprm, long len)
266 return len <= MAX_ARG_STRLEN;
271 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
276 page = bprm->page[pos / PAGE_SIZE];
277 if (!page && write) {
278 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
281 bprm->page[pos / PAGE_SIZE] = page;
287 static void put_arg_page(struct page *page)
291 static void free_arg_page(struct linux_binprm *bprm, int i)
294 __free_page(bprm->page[i]);
295 bprm->page[i] = NULL;
299 static void free_arg_pages(struct linux_binprm *bprm)
303 for (i = 0; i < MAX_ARG_PAGES; i++)
304 free_arg_page(bprm, i);
307 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
312 static int __bprm_mm_init(struct linux_binprm *bprm)
314 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
318 static bool valid_arg_len(struct linux_binprm *bprm, long len)
320 return len <= bprm->p;
323 #endif /* CONFIG_MMU */
326 * Create a new mm_struct and populate it with a temporary stack
327 * vm_area_struct. We don't have enough context at this point to set the stack
328 * flags, permissions, and offset, so we use temporary values. We'll update
329 * them later in setup_arg_pages().
331 int bprm_mm_init(struct linux_binprm *bprm)
334 struct mm_struct *mm = NULL;
336 bprm->mm = mm = mm_alloc();
341 err = init_new_context(current, mm);
345 err = __bprm_mm_init(bprm);
361 * count() counts the number of strings in array ARGV.
363 static int count(char __user * __user * argv, int max)
371 if (get_user(p, argv))
385 * 'copy_strings()' copies argument/environment strings from the old
386 * processes's memory to the new process's stack. The call to get_user_pages()
387 * ensures the destination page is created and not swapped out.
389 static int copy_strings(int argc, char __user * __user * argv,
390 struct linux_binprm *bprm)
392 struct page *kmapped_page = NULL;
394 unsigned long kpos = 0;
402 if (get_user(str, argv+argc) ||
403 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
408 if (!valid_arg_len(bprm, len)) {
413 /* We're going to work our way backwords. */
419 int offset, bytes_to_copy;
421 offset = pos % PAGE_SIZE;
425 bytes_to_copy = offset;
426 if (bytes_to_copy > len)
429 offset -= bytes_to_copy;
430 pos -= bytes_to_copy;
431 str -= bytes_to_copy;
432 len -= bytes_to_copy;
434 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
437 page = get_arg_page(bprm, pos, 1);
444 flush_kernel_dcache_page(kmapped_page);
445 kunmap(kmapped_page);
446 put_arg_page(kmapped_page);
449 kaddr = kmap(kmapped_page);
450 kpos = pos & PAGE_MASK;
451 flush_arg_page(bprm, kpos, kmapped_page);
453 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
462 flush_kernel_dcache_page(kmapped_page);
463 kunmap(kmapped_page);
464 put_arg_page(kmapped_page);
470 * Like copy_strings, but get argv and its values from kernel memory.
472 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
475 mm_segment_t oldfs = get_fs();
477 r = copy_strings(argc, (char __user * __user *)argv, bprm);
481 EXPORT_SYMBOL(copy_strings_kernel);
486 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
487 * the binfmt code determines where the new stack should reside, we shift it to
488 * its final location. The process proceeds as follows:
490 * 1) Use shift to calculate the new vma endpoints.
491 * 2) Extend vma to cover both the old and new ranges. This ensures the
492 * arguments passed to subsequent functions are consistent.
493 * 3) Move vma's page tables to the new range.
494 * 4) Free up any cleared pgd range.
495 * 5) Shrink the vma to cover only the new range.
497 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
499 struct mm_struct *mm = vma->vm_mm;
500 unsigned long old_start = vma->vm_start;
501 unsigned long old_end = vma->vm_end;
502 unsigned long length = old_end - old_start;
503 unsigned long new_start = old_start - shift;
504 unsigned long new_end = old_end - shift;
505 struct mmu_gather *tlb;
507 BUG_ON(new_start > new_end);
510 * ensure there are no vmas between where we want to go
513 if (vma != find_vma(mm, new_start))
517 * cover the whole range: [new_start, old_end)
519 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
522 * move the page tables downwards, on failure we rely on
523 * process cleanup to remove whatever mess we made.
525 if (length != move_page_tables(vma, old_start,
526 vma, new_start, length))
530 tlb = tlb_gather_mmu(mm, 0);
531 if (new_end > old_start) {
533 * when the old and new regions overlap clear from new_end.
535 free_pgd_range(tlb, new_end, old_end, new_end,
536 vma->vm_next ? vma->vm_next->vm_start : 0);
539 * otherwise, clean from old_start; this is done to not touch
540 * the address space in [new_end, old_start) some architectures
541 * have constraints on va-space that make this illegal (IA64) -
542 * for the others its just a little faster.
544 free_pgd_range(tlb, old_start, old_end, new_end,
545 vma->vm_next ? vma->vm_next->vm_start : 0);
547 tlb_finish_mmu(tlb, new_end, old_end);
550 * shrink the vma to just the new range.
552 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
557 #define EXTRA_STACK_VM_PAGES 20 /* random */
560 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
561 * the stack is optionally relocated, and some extra space is added.
563 int setup_arg_pages(struct linux_binprm *bprm,
564 unsigned long stack_top,
565 int executable_stack)
568 unsigned long stack_shift;
569 struct mm_struct *mm = current->mm;
570 struct vm_area_struct *vma = bprm->vma;
571 struct vm_area_struct *prev = NULL;
572 unsigned long vm_flags;
573 unsigned long stack_base;
574 unsigned long stack_size;
575 unsigned long stack_expand;
576 unsigned long rlim_stack;
578 #ifdef CONFIG_STACK_GROWSUP
579 /* Limit stack size to 1GB */
580 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
581 if (stack_base > (1 << 30))
582 stack_base = 1 << 30;
584 /* Make sure we didn't let the argument array grow too large. */
585 if (vma->vm_end - vma->vm_start > stack_base)
588 stack_base = PAGE_ALIGN(stack_top - stack_base);
590 stack_shift = vma->vm_start - stack_base;
591 mm->arg_start = bprm->p - stack_shift;
592 bprm->p = vma->vm_end - stack_shift;
594 stack_top = arch_align_stack(stack_top);
595 stack_top = PAGE_ALIGN(stack_top);
596 stack_shift = vma->vm_end - stack_top;
598 bprm->p -= stack_shift;
599 mm->arg_start = bprm->p;
603 bprm->loader -= stack_shift;
604 bprm->exec -= stack_shift;
606 down_write(&mm->mmap_sem);
607 vm_flags = VM_STACK_FLAGS;
610 * Adjust stack execute permissions; explicitly enable for
611 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
612 * (arch default) otherwise.
614 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
616 else if (executable_stack == EXSTACK_DISABLE_X)
617 vm_flags &= ~VM_EXEC;
618 vm_flags |= mm->def_flags;
620 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
626 /* Move stack pages down in memory. */
628 ret = shift_arg_pages(vma, stack_shift);
633 stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
634 stack_size = vma->vm_end - vma->vm_start;
636 * Align this down to a page boundary as expand_stack
639 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
640 #ifdef CONFIG_STACK_GROWSUP
641 if (stack_size + stack_expand > rlim_stack)
642 stack_base = vma->vm_start + rlim_stack;
644 stack_base = vma->vm_end + stack_expand;
646 if (stack_size + stack_expand > rlim_stack)
647 stack_base = vma->vm_end - rlim_stack;
649 stack_base = vma->vm_start - stack_expand;
651 ret = expand_stack(vma, stack_base);
656 up_write(&mm->mmap_sem);
659 EXPORT_SYMBOL(setup_arg_pages);
661 #endif /* CONFIG_MMU */
663 struct file *open_exec(const char *name)
668 file = do_filp_open(AT_FDCWD, name,
669 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
670 MAY_EXEC | MAY_OPEN);
675 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
678 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
681 fsnotify_open(file->f_path.dentry);
683 err = deny_write_access(file);
694 EXPORT_SYMBOL(open_exec);
696 int kernel_read(struct file *file, loff_t offset,
697 char *addr, unsigned long count)
705 /* The cast to a user pointer is valid due to the set_fs() */
706 result = vfs_read(file, (void __user *)addr, count, &pos);
711 EXPORT_SYMBOL(kernel_read);
713 static int exec_mmap(struct mm_struct *mm)
715 struct task_struct *tsk;
716 struct mm_struct * old_mm, *active_mm;
718 /* Notify parent that we're no longer interested in the old VM */
720 old_mm = current->mm;
721 mm_release(tsk, old_mm);
725 * Make sure that if there is a core dump in progress
726 * for the old mm, we get out and die instead of going
727 * through with the exec. We must hold mmap_sem around
728 * checking core_state and changing tsk->mm.
730 down_read(&old_mm->mmap_sem);
731 if (unlikely(old_mm->core_state)) {
732 up_read(&old_mm->mmap_sem);
737 active_mm = tsk->active_mm;
740 activate_mm(active_mm, mm);
742 arch_pick_mmap_layout(mm);
744 up_read(&old_mm->mmap_sem);
745 BUG_ON(active_mm != old_mm);
746 mm_update_next_owner(old_mm);
755 * This function makes sure the current process has its own signal table,
756 * so that flush_signal_handlers can later reset the handlers without
757 * disturbing other processes. (Other processes might share the signal
758 * table via the CLONE_SIGHAND option to clone().)
760 static int de_thread(struct task_struct *tsk)
762 struct signal_struct *sig = tsk->signal;
763 struct sighand_struct *oldsighand = tsk->sighand;
764 spinlock_t *lock = &oldsighand->siglock;
767 if (thread_group_empty(tsk))
768 goto no_thread_group;
771 * Kill all other threads in the thread group.
774 if (signal_group_exit(sig)) {
776 * Another group action in progress, just
777 * return so that the signal is processed.
779 spin_unlock_irq(lock);
782 sig->group_exit_task = tsk;
783 zap_other_threads(tsk);
785 /* Account for the thread group leader hanging around: */
786 count = thread_group_leader(tsk) ? 1 : 2;
787 sig->notify_count = count;
788 while (atomic_read(&sig->count) > count) {
789 __set_current_state(TASK_UNINTERRUPTIBLE);
790 spin_unlock_irq(lock);
794 spin_unlock_irq(lock);
797 * At this point all other threads have exited, all we have to
798 * do is to wait for the thread group leader to become inactive,
799 * and to assume its PID:
801 if (!thread_group_leader(tsk)) {
802 struct task_struct *leader = tsk->group_leader;
804 sig->notify_count = -1; /* for exit_notify() */
806 write_lock_irq(&tasklist_lock);
807 if (likely(leader->exit_state))
809 __set_current_state(TASK_UNINTERRUPTIBLE);
810 write_unlock_irq(&tasklist_lock);
815 * The only record we have of the real-time age of a
816 * process, regardless of execs it's done, is start_time.
817 * All the past CPU time is accumulated in signal_struct
818 * from sister threads now dead. But in this non-leader
819 * exec, nothing survives from the original leader thread,
820 * whose birth marks the true age of this process now.
821 * When we take on its identity by switching to its PID, we
822 * also take its birthdate (always earlier than our own).
824 tsk->start_time = leader->start_time;
826 BUG_ON(!same_thread_group(leader, tsk));
827 BUG_ON(has_group_leader_pid(tsk));
829 * An exec() starts a new thread group with the
830 * TGID of the previous thread group. Rehash the
831 * two threads with a switched PID, and release
832 * the former thread group leader:
835 /* Become a process group leader with the old leader's pid.
836 * The old leader becomes a thread of the this thread group.
837 * Note: The old leader also uses this pid until release_task
838 * is called. Odd but simple and correct.
840 detach_pid(tsk, PIDTYPE_PID);
841 tsk->pid = leader->pid;
842 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
843 transfer_pid(leader, tsk, PIDTYPE_PGID);
844 transfer_pid(leader, tsk, PIDTYPE_SID);
846 list_replace_rcu(&leader->tasks, &tsk->tasks);
847 list_replace_init(&leader->sibling, &tsk->sibling);
849 tsk->group_leader = tsk;
850 leader->group_leader = tsk;
852 tsk->exit_signal = SIGCHLD;
854 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
855 leader->exit_state = EXIT_DEAD;
856 write_unlock_irq(&tasklist_lock);
858 release_task(leader);
861 sig->group_exit_task = NULL;
862 sig->notify_count = 0;
866 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
869 flush_itimer_signals();
871 if (atomic_read(&oldsighand->count) != 1) {
872 struct sighand_struct *newsighand;
874 * This ->sighand is shared with the CLONE_SIGHAND
875 * but not CLONE_THREAD task, switch to the new one.
877 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
881 atomic_set(&newsighand->count, 1);
882 memcpy(newsighand->action, oldsighand->action,
883 sizeof(newsighand->action));
885 write_lock_irq(&tasklist_lock);
886 spin_lock(&oldsighand->siglock);
887 rcu_assign_pointer(tsk->sighand, newsighand);
888 spin_unlock(&oldsighand->siglock);
889 write_unlock_irq(&tasklist_lock);
891 __cleanup_sighand(oldsighand);
894 BUG_ON(!thread_group_leader(tsk));
899 * These functions flushes out all traces of the currently running executable
900 * so that a new one can be started
902 static void flush_old_files(struct files_struct * files)
907 spin_lock(&files->file_lock);
909 unsigned long set, i;
913 fdt = files_fdtable(files);
914 if (i >= fdt->max_fds)
916 set = fdt->close_on_exec->fds_bits[j];
919 fdt->close_on_exec->fds_bits[j] = 0;
920 spin_unlock(&files->file_lock);
921 for ( ; set ; i++,set >>= 1) {
926 spin_lock(&files->file_lock);
929 spin_unlock(&files->file_lock);
932 char *get_task_comm(char *buf, struct task_struct *tsk)
934 /* buf must be at least sizeof(tsk->comm) in size */
936 strncpy(buf, tsk->comm, sizeof(tsk->comm));
941 void set_task_comm(struct task_struct *tsk, char *buf)
946 * Threads may access current->comm without holding
947 * the task lock, so write the string carefully.
948 * Readers without a lock may see incomplete new
949 * names but are safe from non-terminating string reads.
951 memset(tsk->comm, 0, TASK_COMM_LEN);
953 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
955 perf_event_comm(tsk);
958 int flush_old_exec(struct linux_binprm * bprm)
963 * Make sure we have a private signal table and that
964 * we are unassociated from the previous thread group.
966 retval = de_thread(current);
970 set_mm_exe_file(bprm->mm, bprm->file);
973 * Release all of the old mmap stuff
975 retval = exec_mmap(bprm->mm);
979 bprm->mm = NULL; /* We're using it now */
981 current->flags &= ~PF_RANDOMIZE;
983 current->personality &= ~bprm->per_clear;
990 EXPORT_SYMBOL(flush_old_exec);
992 void setup_new_exec(struct linux_binprm * bprm)
996 char tcomm[sizeof(current->comm)];
998 arch_pick_mmap_layout(current->mm);
1000 /* This is the point of no return */
1001 current->sas_ss_sp = current->sas_ss_size = 0;
1003 if (current_euid() == current_uid() && current_egid() == current_gid())
1004 set_dumpable(current->mm, 1);
1006 set_dumpable(current->mm, suid_dumpable);
1008 name = bprm->filename;
1010 /* Copies the binary name from after last slash */
1011 for (i=0; (ch = *(name++)) != '\0';) {
1013 i = 0; /* overwrite what we wrote */
1015 if (i < (sizeof(tcomm) - 1))
1019 set_task_comm(current, tcomm);
1021 /* Set the new mm task size. We have to do that late because it may
1022 * depend on TIF_32BIT which is only updated in flush_thread() on
1023 * some architectures like powerpc
1025 current->mm->task_size = TASK_SIZE;
1027 /* install the new credentials */
1028 if (bprm->cred->uid != current_euid() ||
1029 bprm->cred->gid != current_egid()) {
1030 current->pdeath_signal = 0;
1031 } else if (file_permission(bprm->file, MAY_READ) ||
1032 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1033 set_dumpable(current->mm, suid_dumpable);
1037 * Flush performance counters when crossing a
1040 if (!get_dumpable(current->mm))
1041 perf_event_exit_task(current);
1043 /* An exec changes our domain. We are no longer part of the thread
1046 current->self_exec_id++;
1048 flush_signal_handlers(current, 0);
1049 flush_old_files(current->files);
1051 EXPORT_SYMBOL(setup_new_exec);
1054 * Prepare credentials and lock ->cred_guard_mutex.
1055 * install_exec_creds() commits the new creds and drops the lock.
1056 * Or, if exec fails before, free_bprm() should release ->cred and
1059 int prepare_bprm_creds(struct linux_binprm *bprm)
1061 if (mutex_lock_interruptible(¤t->cred_guard_mutex))
1062 return -ERESTARTNOINTR;
1064 bprm->cred = prepare_exec_creds();
1065 if (likely(bprm->cred))
1068 mutex_unlock(¤t->cred_guard_mutex);
1072 void free_bprm(struct linux_binprm *bprm)
1074 free_arg_pages(bprm);
1076 mutex_unlock(¤t->cred_guard_mutex);
1077 abort_creds(bprm->cred);
1083 * install the new credentials for this executable
1085 void install_exec_creds(struct linux_binprm *bprm)
1087 security_bprm_committing_creds(bprm);
1089 commit_creds(bprm->cred);
1092 * cred_guard_mutex must be held at least to this point to prevent
1093 * ptrace_attach() from altering our determination of the task's
1094 * credentials; any time after this it may be unlocked.
1096 security_bprm_committed_creds(bprm);
1097 mutex_unlock(¤t->cred_guard_mutex);
1099 EXPORT_SYMBOL(install_exec_creds);
1102 * determine how safe it is to execute the proposed program
1103 * - the caller must hold current->cred_guard_mutex to protect against
1106 int check_unsafe_exec(struct linux_binprm *bprm)
1108 struct task_struct *p = current, *t;
1112 bprm->unsafe = tracehook_unsafe_exec(p);
1115 write_lock(&p->fs->lock);
1117 for (t = next_thread(p); t != p; t = next_thread(t)) {
1123 if (p->fs->users > n_fs) {
1124 bprm->unsafe |= LSM_UNSAFE_SHARE;
1127 if (!p->fs->in_exec) {
1132 write_unlock(&p->fs->lock);
1138 * Fill the binprm structure from the inode.
1139 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1141 * This may be called multiple times for binary chains (scripts for example).
1143 int prepare_binprm(struct linux_binprm *bprm)
1146 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1149 mode = inode->i_mode;
1150 if (bprm->file->f_op == NULL)
1153 /* clear any previous set[ug]id data from a previous binary */
1154 bprm->cred->euid = current_euid();
1155 bprm->cred->egid = current_egid();
1157 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1159 if (mode & S_ISUID) {
1160 bprm->per_clear |= PER_CLEAR_ON_SETID;
1161 bprm->cred->euid = inode->i_uid;
1166 * If setgid is set but no group execute bit then this
1167 * is a candidate for mandatory locking, not a setgid
1170 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1171 bprm->per_clear |= PER_CLEAR_ON_SETID;
1172 bprm->cred->egid = inode->i_gid;
1176 /* fill in binprm security blob */
1177 retval = security_bprm_set_creds(bprm);
1180 bprm->cred_prepared = 1;
1182 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1183 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1186 EXPORT_SYMBOL(prepare_binprm);
1189 * Arguments are '\0' separated strings found at the location bprm->p
1190 * points to; chop off the first by relocating brpm->p to right after
1191 * the first '\0' encountered.
1193 int remove_arg_zero(struct linux_binprm *bprm)
1196 unsigned long offset;
1204 offset = bprm->p & ~PAGE_MASK;
1205 page = get_arg_page(bprm, bprm->p, 0);
1210 kaddr = kmap_atomic(page, KM_USER0);
1212 for (; offset < PAGE_SIZE && kaddr[offset];
1213 offset++, bprm->p++)
1216 kunmap_atomic(kaddr, KM_USER0);
1219 if (offset == PAGE_SIZE)
1220 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1221 } while (offset == PAGE_SIZE);
1230 EXPORT_SYMBOL(remove_arg_zero);
1233 * cycle the list of binary formats handler, until one recognizes the image
1235 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1237 unsigned int depth = bprm->recursion_depth;
1239 struct linux_binfmt *fmt;
1241 retval = security_bprm_check(bprm);
1245 /* kernel module loader fixup */
1246 /* so we don't try to load run modprobe in kernel space. */
1249 retval = audit_bprm(bprm);
1254 for (try=0; try<2; try++) {
1255 read_lock(&binfmt_lock);
1256 list_for_each_entry(fmt, &formats, lh) {
1257 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1260 if (!try_module_get(fmt->module))
1262 read_unlock(&binfmt_lock);
1263 retval = fn(bprm, regs);
1265 * Restore the depth counter to its starting value
1266 * in this call, so we don't have to rely on every
1267 * load_binary function to restore it on return.
1269 bprm->recursion_depth = depth;
1272 tracehook_report_exec(fmt, bprm, regs);
1274 allow_write_access(bprm->file);
1278 current->did_exec = 1;
1279 proc_exec_connector(current);
1282 read_lock(&binfmt_lock);
1284 if (retval != -ENOEXEC || bprm->mm == NULL)
1287 read_unlock(&binfmt_lock);
1291 read_unlock(&binfmt_lock);
1292 if (retval != -ENOEXEC || bprm->mm == NULL) {
1294 #ifdef CONFIG_MODULES
1296 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1297 if (printable(bprm->buf[0]) &&
1298 printable(bprm->buf[1]) &&
1299 printable(bprm->buf[2]) &&
1300 printable(bprm->buf[3]))
1301 break; /* -ENOEXEC */
1302 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1309 EXPORT_SYMBOL(search_binary_handler);
1312 * sys_execve() executes a new program.
1314 int do_execve(char * filename,
1315 char __user *__user *argv,
1316 char __user *__user *envp,
1317 struct pt_regs * regs)
1319 struct linux_binprm *bprm;
1321 struct files_struct *displaced;
1325 retval = unshare_files(&displaced);
1330 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1334 retval = prepare_bprm_creds(bprm);
1338 retval = check_unsafe_exec(bprm);
1341 clear_in_exec = retval;
1342 current->in_execve = 1;
1344 file = open_exec(filename);
1345 retval = PTR_ERR(file);
1352 bprm->filename = filename;
1353 bprm->interp = filename;
1355 retval = bprm_mm_init(bprm);
1359 bprm->argc = count(argv, MAX_ARG_STRINGS);
1360 if ((retval = bprm->argc) < 0)
1363 bprm->envc = count(envp, MAX_ARG_STRINGS);
1364 if ((retval = bprm->envc) < 0)
1367 retval = prepare_binprm(bprm);
1371 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1375 bprm->exec = bprm->p;
1376 retval = copy_strings(bprm->envc, envp, bprm);
1380 retval = copy_strings(bprm->argc, argv, bprm);
1384 current->flags &= ~PF_KTHREAD;
1385 retval = search_binary_handler(bprm,regs);
1389 current->stack_start = current->mm->start_stack;
1391 /* execve succeeded */
1392 current->fs->in_exec = 0;
1393 current->in_execve = 0;
1394 acct_update_integrals(current);
1397 put_files_struct(displaced);
1406 allow_write_access(bprm->file);
1412 current->fs->in_exec = 0;
1413 current->in_execve = 0;
1420 reset_files_struct(displaced);
1425 void set_binfmt(struct linux_binfmt *new)
1427 struct mm_struct *mm = current->mm;
1430 module_put(mm->binfmt->module);
1434 __module_get(new->module);
1437 EXPORT_SYMBOL(set_binfmt);
1439 /* format_corename will inspect the pattern parameter, and output a
1440 * name into corename, which must have space for at least
1441 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1443 static int format_corename(char *corename, long signr)
1445 const struct cred *cred = current_cred();
1446 const char *pat_ptr = core_pattern;
1447 int ispipe = (*pat_ptr == '|');
1448 char *out_ptr = corename;
1449 char *const out_end = corename + CORENAME_MAX_SIZE;
1451 int pid_in_pattern = 0;
1453 /* Repeat as long as we have more pattern to process and more output
1456 if (*pat_ptr != '%') {
1457 if (out_ptr == out_end)
1459 *out_ptr++ = *pat_ptr++;
1461 switch (*++pat_ptr) {
1464 /* Double percent, output one percent */
1466 if (out_ptr == out_end)
1473 rc = snprintf(out_ptr, out_end - out_ptr,
1474 "%d", task_tgid_vnr(current));
1475 if (rc > out_end - out_ptr)
1481 rc = snprintf(out_ptr, out_end - out_ptr,
1483 if (rc > out_end - out_ptr)
1489 rc = snprintf(out_ptr, out_end - out_ptr,
1491 if (rc > out_end - out_ptr)
1495 /* signal that caused the coredump */
1497 rc = snprintf(out_ptr, out_end - out_ptr,
1499 if (rc > out_end - out_ptr)
1503 /* UNIX time of coredump */
1506 do_gettimeofday(&tv);
1507 rc = snprintf(out_ptr, out_end - out_ptr,
1509 if (rc > out_end - out_ptr)
1516 down_read(&uts_sem);
1517 rc = snprintf(out_ptr, out_end - out_ptr,
1518 "%s", utsname()->nodename);
1520 if (rc > out_end - out_ptr)
1526 rc = snprintf(out_ptr, out_end - out_ptr,
1527 "%s", current->comm);
1528 if (rc > out_end - out_ptr)
1532 /* core limit size */
1534 rc = snprintf(out_ptr, out_end - out_ptr,
1535 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1536 if (rc > out_end - out_ptr)
1546 /* Backward compatibility with core_uses_pid:
1548 * If core_pattern does not include a %p (as is the default)
1549 * and core_uses_pid is set, then .%pid will be appended to
1550 * the filename. Do not do this for piped commands. */
1551 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1552 rc = snprintf(out_ptr, out_end - out_ptr,
1553 ".%d", task_tgid_vnr(current));
1554 if (rc > out_end - out_ptr)
1563 static int zap_process(struct task_struct *start)
1565 struct task_struct *t;
1568 start->signal->flags = SIGNAL_GROUP_EXIT;
1569 start->signal->group_stop_count = 0;
1573 if (t != current && t->mm) {
1574 sigaddset(&t->pending.signal, SIGKILL);
1575 signal_wake_up(t, 1);
1578 } while_each_thread(start, t);
1583 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1584 struct core_state *core_state, int exit_code)
1586 struct task_struct *g, *p;
1587 unsigned long flags;
1590 spin_lock_irq(&tsk->sighand->siglock);
1591 if (!signal_group_exit(tsk->signal)) {
1592 mm->core_state = core_state;
1593 tsk->signal->group_exit_code = exit_code;
1594 nr = zap_process(tsk);
1596 spin_unlock_irq(&tsk->sighand->siglock);
1597 if (unlikely(nr < 0))
1600 if (atomic_read(&mm->mm_users) == nr + 1)
1603 * We should find and kill all tasks which use this mm, and we should
1604 * count them correctly into ->nr_threads. We don't take tasklist
1605 * lock, but this is safe wrt:
1608 * None of sub-threads can fork after zap_process(leader). All
1609 * processes which were created before this point should be
1610 * visible to zap_threads() because copy_process() adds the new
1611 * process to the tail of init_task.tasks list, and lock/unlock
1612 * of ->siglock provides a memory barrier.
1615 * The caller holds mm->mmap_sem. This means that the task which
1616 * uses this mm can't pass exit_mm(), so it can't exit or clear
1620 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1621 * we must see either old or new leader, this does not matter.
1622 * However, it can change p->sighand, so lock_task_sighand(p)
1623 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1626 * Note also that "g" can be the old leader with ->mm == NULL
1627 * and already unhashed and thus removed from ->thread_group.
1628 * This is OK, __unhash_process()->list_del_rcu() does not
1629 * clear the ->next pointer, we will find the new leader via
1633 for_each_process(g) {
1634 if (g == tsk->group_leader)
1636 if (g->flags & PF_KTHREAD)
1641 if (unlikely(p->mm == mm)) {
1642 lock_task_sighand(p, &flags);
1643 nr += zap_process(p);
1644 unlock_task_sighand(p, &flags);
1648 } while_each_thread(g, p);
1652 atomic_set(&core_state->nr_threads, nr);
1656 static int coredump_wait(int exit_code, struct core_state *core_state)
1658 struct task_struct *tsk = current;
1659 struct mm_struct *mm = tsk->mm;
1660 struct completion *vfork_done;
1663 init_completion(&core_state->startup);
1664 core_state->dumper.task = tsk;
1665 core_state->dumper.next = NULL;
1666 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1667 up_write(&mm->mmap_sem);
1669 if (unlikely(core_waiters < 0))
1673 * Make sure nobody is waiting for us to release the VM,
1674 * otherwise we can deadlock when we wait on each other
1676 vfork_done = tsk->vfork_done;
1678 tsk->vfork_done = NULL;
1679 complete(vfork_done);
1683 wait_for_completion(&core_state->startup);
1685 return core_waiters;
1688 static void coredump_finish(struct mm_struct *mm)
1690 struct core_thread *curr, *next;
1691 struct task_struct *task;
1693 next = mm->core_state->dumper.next;
1694 while ((curr = next) != NULL) {
1698 * see exit_mm(), curr->task must not see
1699 * ->task == NULL before we read ->next.
1703 wake_up_process(task);
1706 mm->core_state = NULL;
1710 * set_dumpable converts traditional three-value dumpable to two flags and
1711 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1712 * these bits are not changed atomically. So get_dumpable can observe the
1713 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1714 * return either old dumpable or new one by paying attention to the order of
1715 * modifying the bits.
1717 * dumpable | mm->flags (binary)
1718 * old new | initial interim final
1719 * ---------+-----------------------
1727 * (*) get_dumpable regards interim value of 10 as 11.
1729 void set_dumpable(struct mm_struct *mm, int value)
1733 clear_bit(MMF_DUMPABLE, &mm->flags);
1735 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1738 set_bit(MMF_DUMPABLE, &mm->flags);
1740 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1743 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1745 set_bit(MMF_DUMPABLE, &mm->flags);
1750 int get_dumpable(struct mm_struct *mm)
1754 ret = mm->flags & 0x3;
1755 return (ret >= 2) ? 2 : ret;
1758 static void wait_for_dump_helpers(struct file *file)
1760 struct pipe_inode_info *pipe;
1762 pipe = file->f_path.dentry->d_inode->i_pipe;
1768 while ((pipe->readers > 1) && (!signal_pending(current))) {
1769 wake_up_interruptible_sync(&pipe->wait);
1770 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1781 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1783 struct core_state core_state;
1784 char corename[CORENAME_MAX_SIZE + 1];
1785 struct mm_struct *mm = current->mm;
1786 struct linux_binfmt * binfmt;
1787 struct inode * inode;
1788 const struct cred *old_cred;
1793 char **helper_argv = NULL;
1794 int helper_argc = 0;
1796 static atomic_t core_dump_count = ATOMIC_INIT(0);
1797 struct coredump_params cprm = {
1800 .limit = current->signal->rlim[RLIMIT_CORE].rlim_cur,
1803 audit_core_dumps(signr);
1805 binfmt = mm->binfmt;
1806 if (!binfmt || !binfmt->core_dump)
1809 cred = prepare_creds();
1815 down_write(&mm->mmap_sem);
1817 * If another thread got here first, or we are not dumpable, bail out.
1819 if (mm->core_state || !get_dumpable(mm)) {
1820 up_write(&mm->mmap_sem);
1826 * We cannot trust fsuid as being the "true" uid of the
1827 * process nor do we know its entire history. We only know it
1828 * was tainted so we dump it as root in mode 2.
1830 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1831 flag = O_EXCL; /* Stop rewrite attacks */
1832 cred->fsuid = 0; /* Dump root private */
1835 retval = coredump_wait(exit_code, &core_state);
1841 old_cred = override_creds(cred);
1844 * Clear any false indication of pending signals that might
1845 * be seen by the filesystem code called to write the core file.
1847 clear_thread_flag(TIF_SIGPENDING);
1850 * lock_kernel() because format_corename() is controlled by sysctl, which
1851 * uses lock_kernel()
1854 ispipe = format_corename(corename, signr);
1857 if ((!ispipe) && (cprm.limit < binfmt->min_coredump))
1861 if (cprm.limit == 0) {
1863 * Normally core limits are irrelevant to pipes, since
1864 * we're not writing to the file system, but we use
1865 * cprm.limit of 0 here as a speacial value. Any
1866 * non-zero limit gets set to RLIM_INFINITY below, but
1867 * a limit of 0 skips the dump. This is a consistent
1868 * way to catch recursive crashes. We can still crash
1869 * if the core_pattern binary sets RLIM_CORE = !0
1870 * but it runs as root, and can do lots of stupid things
1871 * Note that we use task_tgid_vnr here to grab the pid
1872 * of the process group leader. That way we get the
1873 * right pid if a thread in a multi-threaded
1874 * core_pattern process dies.
1877 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1878 task_tgid_vnr(current), current->comm);
1879 printk(KERN_WARNING "Aborting core\n");
1883 dump_count = atomic_inc_return(&core_dump_count);
1884 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1885 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1886 task_tgid_vnr(current), current->comm);
1887 printk(KERN_WARNING "Skipping core dump\n");
1888 goto fail_dropcount;
1891 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1893 printk(KERN_WARNING "%s failed to allocate memory\n",
1895 goto fail_dropcount;
1898 cprm.limit = RLIM_INFINITY;
1900 /* SIGPIPE can happen, but it's just never processed */
1901 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1903 printk(KERN_INFO "Core dump to %s pipe failed\n",
1905 goto fail_dropcount;
1908 cprm.file = filp_open(corename,
1909 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1911 if (IS_ERR(cprm.file))
1912 goto fail_dropcount;
1913 inode = cprm.file->f_path.dentry->d_inode;
1914 if (inode->i_nlink > 1)
1915 goto close_fail; /* multiple links - don't dump */
1916 if (!ispipe && d_unhashed(cprm.file->f_path.dentry))
1919 /* AK: actually i see no reason to not allow this for named pipes etc.,
1920 but keep the previous behaviour for now. */
1921 if (!ispipe && !S_ISREG(inode->i_mode))
1924 * Dont allow local users get cute and trick others to coredump
1925 * into their pre-created files:
1927 if (inode->i_uid != current_fsuid())
1929 if (!cprm.file->f_op)
1931 if (!cprm.file->f_op->write)
1934 do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file) != 0)
1937 retval = binfmt->core_dump(&cprm);
1940 current->signal->group_exit_code |= 0x80;
1942 if (ispipe && core_pipe_limit)
1943 wait_for_dump_helpers(cprm.file);
1944 filp_close(cprm.file, NULL);
1947 atomic_dec(&core_dump_count);
1950 argv_free(helper_argv);
1952 revert_creds(old_cred);
1954 coredump_finish(mm);