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/ima.h>
50 #include <linux/syscalls.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54 #include <linux/tracehook.h>
55 #include <linux/kmod.h>
56 #include <linux/fsnotify.h>
57 #include <linux/fs_struct.h>
58 #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);
75 int __register_binfmt(struct linux_binfmt * fmt, int insert)
79 write_lock(&binfmt_lock);
80 insert ? list_add(&fmt->lh, &formats) :
81 list_add_tail(&fmt->lh, &formats);
82 write_unlock(&binfmt_lock);
86 EXPORT_SYMBOL(__register_binfmt);
88 void unregister_binfmt(struct linux_binfmt * fmt)
90 write_lock(&binfmt_lock);
92 write_unlock(&binfmt_lock);
95 EXPORT_SYMBOL(unregister_binfmt);
97 static inline void put_binfmt(struct linux_binfmt * fmt)
99 module_put(fmt->module);
103 * Note that a shared library must be both readable and executable due to
106 * Also note that we take the address to load from from the file itself.
108 SYSCALL_DEFINE1(uselib, const char __user *, library)
111 char *tmp = getname(library);
112 int error = PTR_ERR(tmp);
117 file = do_filp_open(AT_FDCWD, tmp,
118 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
119 MAY_READ | MAY_EXEC | MAY_OPEN);
121 error = PTR_ERR(file);
126 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
130 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
133 fsnotify_open(file->f_path.dentry);
137 struct linux_binfmt * fmt;
139 read_lock(&binfmt_lock);
140 list_for_each_entry(fmt, &formats, lh) {
141 if (!fmt->load_shlib)
143 if (!try_module_get(fmt->module))
145 read_unlock(&binfmt_lock);
146 error = fmt->load_shlib(file);
147 read_lock(&binfmt_lock);
149 if (error != -ENOEXEC)
152 read_unlock(&binfmt_lock);
162 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
168 #ifdef CONFIG_STACK_GROWSUP
170 ret = expand_stack_downwards(bprm->vma, pos);
175 ret = get_user_pages(current, bprm->mm, pos,
176 1, write, 1, &page, NULL);
181 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
185 * We've historically supported up to 32 pages (ARG_MAX)
186 * of argument strings even with small stacks
192 * Limit to 1/4-th the stack size for the argv+env strings.
194 * - the remaining binfmt code will not run out of stack space,
195 * - the program will have a reasonable amount of stack left
198 rlim = current->signal->rlim;
199 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
208 static void put_arg_page(struct page *page)
213 static void free_arg_page(struct linux_binprm *bprm, int i)
217 static void free_arg_pages(struct linux_binprm *bprm)
221 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
224 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
227 static int __bprm_mm_init(struct linux_binprm *bprm)
230 struct vm_area_struct *vma = NULL;
231 struct mm_struct *mm = bprm->mm;
233 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
237 down_write(&mm->mmap_sem);
241 * Place the stack at the largest stack address the architecture
242 * supports. Later, we'll move this to an appropriate place. We don't
243 * use STACK_TOP because that can depend on attributes which aren't
246 vma->vm_end = STACK_TOP_MAX;
247 vma->vm_start = vma->vm_end - PAGE_SIZE;
248 vma->vm_flags = VM_STACK_FLAGS;
249 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
250 err = insert_vm_struct(mm, vma);
254 mm->stack_vm = mm->total_vm = 1;
255 up_write(&mm->mmap_sem);
256 bprm->p = vma->vm_end - sizeof(void *);
259 up_write(&mm->mmap_sem);
261 kmem_cache_free(vm_area_cachep, vma);
265 static bool valid_arg_len(struct linux_binprm *bprm, long len)
267 return len <= MAX_ARG_STRLEN;
272 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
277 page = bprm->page[pos / PAGE_SIZE];
278 if (!page && write) {
279 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
282 bprm->page[pos / PAGE_SIZE] = page;
288 static void put_arg_page(struct page *page)
292 static void free_arg_page(struct linux_binprm *bprm, int i)
295 __free_page(bprm->page[i]);
296 bprm->page[i] = NULL;
300 static void free_arg_pages(struct linux_binprm *bprm)
304 for (i = 0; i < MAX_ARG_PAGES; i++)
305 free_arg_page(bprm, i);
308 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
313 static int __bprm_mm_init(struct linux_binprm *bprm)
315 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
319 static bool valid_arg_len(struct linux_binprm *bprm, long len)
321 return len <= bprm->p;
324 #endif /* CONFIG_MMU */
327 * Create a new mm_struct and populate it with a temporary stack
328 * vm_area_struct. We don't have enough context at this point to set the stack
329 * flags, permissions, and offset, so we use temporary values. We'll update
330 * them later in setup_arg_pages().
332 int bprm_mm_init(struct linux_binprm *bprm)
335 struct mm_struct *mm = NULL;
337 bprm->mm = mm = mm_alloc();
342 err = init_new_context(current, mm);
346 err = __bprm_mm_init(bprm);
362 * count() counts the number of strings in array ARGV.
364 static int count(char __user * __user * argv, int max)
372 if (get_user(p, argv))
380 if (fatal_signal_pending(current))
381 return -ERESTARTNOHAND;
389 * 'copy_strings()' copies argument/environment strings from the old
390 * processes's memory to the new process's stack. The call to get_user_pages()
391 * ensures the destination page is created and not swapped out.
393 static int copy_strings(int argc, char __user * __user * argv,
394 struct linux_binprm *bprm)
396 struct page *kmapped_page = NULL;
398 unsigned long kpos = 0;
406 if (get_user(str, argv+argc) ||
407 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
412 if (!valid_arg_len(bprm, len)) {
417 /* We're going to work our way backwords. */
423 int offset, bytes_to_copy;
425 if (fatal_signal_pending(current)) {
426 ret = -ERESTARTNOHAND;
431 offset = pos % PAGE_SIZE;
435 bytes_to_copy = offset;
436 if (bytes_to_copy > len)
439 offset -= bytes_to_copy;
440 pos -= bytes_to_copy;
441 str -= bytes_to_copy;
442 len -= bytes_to_copy;
444 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
447 page = get_arg_page(bprm, pos, 1);
454 flush_kernel_dcache_page(kmapped_page);
455 kunmap(kmapped_page);
456 put_arg_page(kmapped_page);
459 kaddr = kmap(kmapped_page);
460 kpos = pos & PAGE_MASK;
461 flush_arg_page(bprm, kpos, kmapped_page);
463 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
472 flush_kernel_dcache_page(kmapped_page);
473 kunmap(kmapped_page);
474 put_arg_page(kmapped_page);
480 * Like copy_strings, but get argv and its values from kernel memory.
482 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
485 mm_segment_t oldfs = get_fs();
487 r = copy_strings(argc, (char __user * __user *)argv, bprm);
491 EXPORT_SYMBOL(copy_strings_kernel);
496 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
497 * the binfmt code determines where the new stack should reside, we shift it to
498 * its final location. The process proceeds as follows:
500 * 1) Use shift to calculate the new vma endpoints.
501 * 2) Extend vma to cover both the old and new ranges. This ensures the
502 * arguments passed to subsequent functions are consistent.
503 * 3) Move vma's page tables to the new range.
504 * 4) Free up any cleared pgd range.
505 * 5) Shrink the vma to cover only the new range.
507 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
509 struct mm_struct *mm = vma->vm_mm;
510 unsigned long old_start = vma->vm_start;
511 unsigned long old_end = vma->vm_end;
512 unsigned long length = old_end - old_start;
513 unsigned long new_start = old_start - shift;
514 unsigned long new_end = old_end - shift;
515 struct mmu_gather *tlb;
517 BUG_ON(new_start > new_end);
520 * ensure there are no vmas between where we want to go
523 if (vma != find_vma(mm, new_start))
527 * cover the whole range: [new_start, old_end)
529 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
532 * move the page tables downwards, on failure we rely on
533 * process cleanup to remove whatever mess we made.
535 if (length != move_page_tables(vma, old_start,
536 vma, new_start, length))
540 tlb = tlb_gather_mmu(mm, 0);
541 if (new_end > old_start) {
543 * when the old and new regions overlap clear from new_end.
545 free_pgd_range(tlb, new_end, old_end, new_end,
546 vma->vm_next ? vma->vm_next->vm_start : 0);
549 * otherwise, clean from old_start; this is done to not touch
550 * the address space in [new_end, old_start) some architectures
551 * have constraints on va-space that make this illegal (IA64) -
552 * for the others its just a little faster.
554 free_pgd_range(tlb, old_start, old_end, new_end,
555 vma->vm_next ? vma->vm_next->vm_start : 0);
557 tlb_finish_mmu(tlb, new_end, old_end);
560 * shrink the vma to just the new range.
562 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
567 #define EXTRA_STACK_VM_PAGES 20 /* random */
570 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
571 * the stack is optionally relocated, and some extra space is added.
573 int setup_arg_pages(struct linux_binprm *bprm,
574 unsigned long stack_top,
575 int executable_stack)
578 unsigned long stack_shift;
579 struct mm_struct *mm = current->mm;
580 struct vm_area_struct *vma = bprm->vma;
581 struct vm_area_struct *prev = NULL;
582 unsigned long vm_flags;
583 unsigned long stack_base;
584 unsigned long stack_size;
585 unsigned long stack_expand;
586 unsigned long rlim_stack;
588 #ifdef CONFIG_STACK_GROWSUP
589 /* Limit stack size to 1GB */
590 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
591 if (stack_base > (1 << 30))
592 stack_base = 1 << 30;
594 /* Make sure we didn't let the argument array grow too large. */
595 if (vma->vm_end - vma->vm_start > stack_base)
598 stack_base = PAGE_ALIGN(stack_top - stack_base);
600 stack_shift = vma->vm_start - stack_base;
601 mm->arg_start = bprm->p - stack_shift;
602 bprm->p = vma->vm_end - stack_shift;
604 stack_top = arch_align_stack(stack_top);
605 stack_top = PAGE_ALIGN(stack_top);
607 if (unlikely(stack_top < mmap_min_addr) ||
608 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
611 stack_shift = vma->vm_end - stack_top;
613 bprm->p -= stack_shift;
614 mm->arg_start = bprm->p;
618 bprm->loader -= stack_shift;
619 bprm->exec -= stack_shift;
621 down_write(&mm->mmap_sem);
622 vm_flags = VM_STACK_FLAGS;
625 * Adjust stack execute permissions; explicitly enable for
626 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
627 * (arch default) otherwise.
629 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
631 else if (executable_stack == EXSTACK_DISABLE_X)
632 vm_flags &= ~VM_EXEC;
633 vm_flags |= mm->def_flags;
635 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
641 /* Move stack pages down in memory. */
643 ret = shift_arg_pages(vma, stack_shift);
648 stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
649 stack_size = vma->vm_end - vma->vm_start;
651 * Align this down to a page boundary as expand_stack
654 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
655 #ifdef CONFIG_STACK_GROWSUP
656 if (stack_size + stack_expand > rlim_stack)
657 stack_base = vma->vm_start + rlim_stack;
659 stack_base = vma->vm_end + stack_expand;
661 if (stack_size + stack_expand > rlim_stack)
662 stack_base = vma->vm_end - rlim_stack;
664 stack_base = vma->vm_start - stack_expand;
666 ret = expand_stack(vma, stack_base);
671 up_write(&mm->mmap_sem);
674 EXPORT_SYMBOL(setup_arg_pages);
676 #endif /* CONFIG_MMU */
678 struct file *open_exec(const char *name)
683 file = do_filp_open(AT_FDCWD, name,
684 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
685 MAY_EXEC | MAY_OPEN);
690 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
693 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
696 fsnotify_open(file->f_path.dentry);
698 err = deny_write_access(file);
709 EXPORT_SYMBOL(open_exec);
711 int kernel_read(struct file *file, loff_t offset,
712 char *addr, unsigned long count)
720 /* The cast to a user pointer is valid due to the set_fs() */
721 result = vfs_read(file, (void __user *)addr, count, &pos);
726 EXPORT_SYMBOL(kernel_read);
728 static int exec_mmap(struct mm_struct *mm)
730 struct task_struct *tsk;
731 struct mm_struct * old_mm, *active_mm;
733 /* Notify parent that we're no longer interested in the old VM */
735 old_mm = current->mm;
736 mm_release(tsk, old_mm);
740 * Make sure that if there is a core dump in progress
741 * for the old mm, we get out and die instead of going
742 * through with the exec. We must hold mmap_sem around
743 * checking core_state and changing tsk->mm.
745 down_read(&old_mm->mmap_sem);
746 if (unlikely(old_mm->core_state)) {
747 up_read(&old_mm->mmap_sem);
752 active_mm = tsk->active_mm;
755 activate_mm(active_mm, mm);
757 arch_pick_mmap_layout(mm);
759 up_read(&old_mm->mmap_sem);
760 BUG_ON(active_mm != old_mm);
761 mm_update_next_owner(old_mm);
770 * This function makes sure the current process has its own signal table,
771 * so that flush_signal_handlers can later reset the handlers without
772 * disturbing other processes. (Other processes might share the signal
773 * table via the CLONE_SIGHAND option to clone().)
775 static int de_thread(struct task_struct *tsk)
777 struct signal_struct *sig = tsk->signal;
778 struct sighand_struct *oldsighand = tsk->sighand;
779 spinlock_t *lock = &oldsighand->siglock;
782 if (thread_group_empty(tsk))
783 goto no_thread_group;
786 * Kill all other threads in the thread group.
789 if (signal_group_exit(sig)) {
791 * Another group action in progress, just
792 * return so that the signal is processed.
794 spin_unlock_irq(lock);
797 sig->group_exit_task = tsk;
798 zap_other_threads(tsk);
800 /* Account for the thread group leader hanging around: */
801 count = thread_group_leader(tsk) ? 1 : 2;
802 sig->notify_count = count;
803 while (atomic_read(&sig->count) > count) {
804 __set_current_state(TASK_UNINTERRUPTIBLE);
805 spin_unlock_irq(lock);
809 spin_unlock_irq(lock);
812 * At this point all other threads have exited, all we have to
813 * do is to wait for the thread group leader to become inactive,
814 * and to assume its PID:
816 if (!thread_group_leader(tsk)) {
817 struct task_struct *leader = tsk->group_leader;
819 sig->notify_count = -1; /* for exit_notify() */
821 write_lock_irq(&tasklist_lock);
822 if (likely(leader->exit_state))
824 __set_current_state(TASK_UNINTERRUPTIBLE);
825 write_unlock_irq(&tasklist_lock);
830 * The only record we have of the real-time age of a
831 * process, regardless of execs it's done, is start_time.
832 * All the past CPU time is accumulated in signal_struct
833 * from sister threads now dead. But in this non-leader
834 * exec, nothing survives from the original leader thread,
835 * whose birth marks the true age of this process now.
836 * When we take on its identity by switching to its PID, we
837 * also take its birthdate (always earlier than our own).
839 tsk->start_time = leader->start_time;
841 BUG_ON(!same_thread_group(leader, tsk));
842 BUG_ON(has_group_leader_pid(tsk));
844 * An exec() starts a new thread group with the
845 * TGID of the previous thread group. Rehash the
846 * two threads with a switched PID, and release
847 * the former thread group leader:
850 /* Become a process group leader with the old leader's pid.
851 * The old leader becomes a thread of the this thread group.
852 * Note: The old leader also uses this pid until release_task
853 * is called. Odd but simple and correct.
855 detach_pid(tsk, PIDTYPE_PID);
856 tsk->pid = leader->pid;
857 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
858 transfer_pid(leader, tsk, PIDTYPE_PGID);
859 transfer_pid(leader, tsk, PIDTYPE_SID);
860 list_replace_rcu(&leader->tasks, &tsk->tasks);
862 tsk->group_leader = tsk;
863 leader->group_leader = tsk;
865 tsk->exit_signal = SIGCHLD;
867 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
868 leader->exit_state = EXIT_DEAD;
869 write_unlock_irq(&tasklist_lock);
871 release_task(leader);
874 sig->group_exit_task = NULL;
875 sig->notify_count = 0;
879 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
882 flush_itimer_signals();
884 if (atomic_read(&oldsighand->count) != 1) {
885 struct sighand_struct *newsighand;
887 * This ->sighand is shared with the CLONE_SIGHAND
888 * but not CLONE_THREAD task, switch to the new one.
890 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
894 atomic_set(&newsighand->count, 1);
895 memcpy(newsighand->action, oldsighand->action,
896 sizeof(newsighand->action));
898 write_lock_irq(&tasklist_lock);
899 spin_lock(&oldsighand->siglock);
900 rcu_assign_pointer(tsk->sighand, newsighand);
901 spin_unlock(&oldsighand->siglock);
902 write_unlock_irq(&tasklist_lock);
904 __cleanup_sighand(oldsighand);
907 BUG_ON(!thread_group_leader(tsk));
912 * These functions flushes out all traces of the currently running executable
913 * so that a new one can be started
915 static void flush_old_files(struct files_struct * files)
920 spin_lock(&files->file_lock);
922 unsigned long set, i;
926 fdt = files_fdtable(files);
927 if (i >= fdt->max_fds)
929 set = fdt->close_on_exec->fds_bits[j];
932 fdt->close_on_exec->fds_bits[j] = 0;
933 spin_unlock(&files->file_lock);
934 for ( ; set ; i++,set >>= 1) {
939 spin_lock(&files->file_lock);
942 spin_unlock(&files->file_lock);
945 char *get_task_comm(char *buf, struct task_struct *tsk)
947 /* buf must be at least sizeof(tsk->comm) in size */
949 strncpy(buf, tsk->comm, sizeof(tsk->comm));
954 void set_task_comm(struct task_struct *tsk, char *buf)
957 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
959 perf_event_comm(tsk);
962 int flush_old_exec(struct linux_binprm * bprm)
967 * Make sure we have a private signal table and that
968 * we are unassociated from the previous thread group.
970 retval = de_thread(current);
974 set_mm_exe_file(bprm->mm, bprm->file);
977 * Release all of the old mmap stuff
979 retval = exec_mmap(bprm->mm);
983 bprm->mm = NULL; /* We're using it now */
985 current->flags &= ~PF_RANDOMIZE;
987 current->personality &= ~bprm->per_clear;
994 EXPORT_SYMBOL(flush_old_exec);
996 void setup_new_exec(struct linux_binprm * bprm)
1000 char tcomm[sizeof(current->comm)];
1002 arch_pick_mmap_layout(current->mm);
1004 /* This is the point of no return */
1005 current->sas_ss_sp = current->sas_ss_size = 0;
1007 if (current_euid() == current_uid() && current_egid() == current_gid())
1008 set_dumpable(current->mm, 1);
1010 set_dumpable(current->mm, suid_dumpable);
1012 name = bprm->filename;
1014 /* Copies the binary name from after last slash */
1015 for (i=0; (ch = *(name++)) != '\0';) {
1017 i = 0; /* overwrite what we wrote */
1019 if (i < (sizeof(tcomm) - 1))
1023 set_task_comm(current, tcomm);
1025 /* Set the new mm task size. We have to do that late because it may
1026 * depend on TIF_32BIT which is only updated in flush_thread() on
1027 * some architectures like powerpc
1029 current->mm->task_size = TASK_SIZE;
1031 /* install the new credentials */
1032 if (bprm->cred->uid != current_euid() ||
1033 bprm->cred->gid != current_egid()) {
1034 current->pdeath_signal = 0;
1035 } else if (file_permission(bprm->file, MAY_READ) ||
1036 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1037 set_dumpable(current->mm, suid_dumpable);
1041 * Flush performance counters when crossing a
1044 if (!get_dumpable(current->mm))
1045 perf_event_exit_task(current);
1047 /* An exec changes our domain. We are no longer part of the thread
1050 current->self_exec_id++;
1052 flush_signal_handlers(current, 0);
1053 flush_old_files(current->files);
1055 EXPORT_SYMBOL(setup_new_exec);
1058 * Prepare credentials and lock ->cred_guard_mutex.
1059 * install_exec_creds() commits the new creds and drops the lock.
1060 * Or, if exec fails before, free_bprm() should release ->cred and
1063 int prepare_bprm_creds(struct linux_binprm *bprm)
1065 if (mutex_lock_interruptible(¤t->cred_guard_mutex))
1066 return -ERESTARTNOINTR;
1068 bprm->cred = prepare_exec_creds();
1069 if (likely(bprm->cred))
1072 mutex_unlock(¤t->cred_guard_mutex);
1076 void free_bprm(struct linux_binprm *bprm)
1078 free_arg_pages(bprm);
1080 mutex_unlock(¤t->cred_guard_mutex);
1081 abort_creds(bprm->cred);
1087 * install the new credentials for this executable
1089 void install_exec_creds(struct linux_binprm *bprm)
1091 security_bprm_committing_creds(bprm);
1093 commit_creds(bprm->cred);
1096 * cred_guard_mutex must be held at least to this point to prevent
1097 * ptrace_attach() from altering our determination of the task's
1098 * credentials; any time after this it may be unlocked.
1100 security_bprm_committed_creds(bprm);
1101 mutex_unlock(¤t->cred_guard_mutex);
1103 EXPORT_SYMBOL(install_exec_creds);
1106 * determine how safe it is to execute the proposed program
1107 * - the caller must hold current->cred_guard_mutex to protect against
1110 int check_unsafe_exec(struct linux_binprm *bprm)
1112 struct task_struct *p = current, *t;
1116 bprm->unsafe = tracehook_unsafe_exec(p);
1119 write_lock(&p->fs->lock);
1121 for (t = next_thread(p); t != p; t = next_thread(t)) {
1127 if (p->fs->users > n_fs) {
1128 bprm->unsafe |= LSM_UNSAFE_SHARE;
1131 if (!p->fs->in_exec) {
1136 write_unlock(&p->fs->lock);
1142 * Fill the binprm structure from the inode.
1143 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1145 * This may be called multiple times for binary chains (scripts for example).
1147 int prepare_binprm(struct linux_binprm *bprm)
1150 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1153 mode = inode->i_mode;
1154 if (bprm->file->f_op == NULL)
1157 /* clear any previous set[ug]id data from a previous binary */
1158 bprm->cred->euid = current_euid();
1159 bprm->cred->egid = current_egid();
1161 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1163 if (mode & S_ISUID) {
1164 bprm->per_clear |= PER_CLEAR_ON_SETID;
1165 bprm->cred->euid = inode->i_uid;
1170 * If setgid is set but no group execute bit then this
1171 * is a candidate for mandatory locking, not a setgid
1174 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1175 bprm->per_clear |= PER_CLEAR_ON_SETID;
1176 bprm->cred->egid = inode->i_gid;
1180 /* fill in binprm security blob */
1181 retval = security_bprm_set_creds(bprm);
1184 bprm->cred_prepared = 1;
1186 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1187 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1190 EXPORT_SYMBOL(prepare_binprm);
1193 * Arguments are '\0' separated strings found at the location bprm->p
1194 * points to; chop off the first by relocating brpm->p to right after
1195 * the first '\0' encountered.
1197 int remove_arg_zero(struct linux_binprm *bprm)
1200 unsigned long offset;
1208 offset = bprm->p & ~PAGE_MASK;
1209 page = get_arg_page(bprm, bprm->p, 0);
1214 kaddr = kmap_atomic(page, KM_USER0);
1216 for (; offset < PAGE_SIZE && kaddr[offset];
1217 offset++, bprm->p++)
1220 kunmap_atomic(kaddr, KM_USER0);
1223 if (offset == PAGE_SIZE)
1224 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1225 } while (offset == PAGE_SIZE);
1234 EXPORT_SYMBOL(remove_arg_zero);
1237 * cycle the list of binary formats handler, until one recognizes the image
1239 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1241 unsigned int depth = bprm->recursion_depth;
1243 struct linux_binfmt *fmt;
1245 retval = security_bprm_check(bprm);
1248 retval = ima_bprm_check(bprm);
1252 /* kernel module loader fixup */
1253 /* so we don't try to load run modprobe in kernel space. */
1256 retval = audit_bprm(bprm);
1261 for (try=0; try<2; try++) {
1262 read_lock(&binfmt_lock);
1263 list_for_each_entry(fmt, &formats, lh) {
1264 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1267 if (!try_module_get(fmt->module))
1269 read_unlock(&binfmt_lock);
1270 retval = fn(bprm, regs);
1272 * Restore the depth counter to its starting value
1273 * in this call, so we don't have to rely on every
1274 * load_binary function to restore it on return.
1276 bprm->recursion_depth = depth;
1279 tracehook_report_exec(fmt, bprm, regs);
1281 allow_write_access(bprm->file);
1285 current->did_exec = 1;
1286 proc_exec_connector(current);
1289 read_lock(&binfmt_lock);
1291 if (retval != -ENOEXEC || bprm->mm == NULL)
1294 read_unlock(&binfmt_lock);
1298 read_unlock(&binfmt_lock);
1299 if (retval != -ENOEXEC || bprm->mm == NULL) {
1301 #ifdef CONFIG_MODULES
1303 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1304 if (printable(bprm->buf[0]) &&
1305 printable(bprm->buf[1]) &&
1306 printable(bprm->buf[2]) &&
1307 printable(bprm->buf[3]))
1308 break; /* -ENOEXEC */
1309 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1316 EXPORT_SYMBOL(search_binary_handler);
1319 * sys_execve() executes a new program.
1321 int do_execve(char * filename,
1322 char __user *__user *argv,
1323 char __user *__user *envp,
1324 struct pt_regs * regs)
1326 struct linux_binprm *bprm;
1328 struct files_struct *displaced;
1332 retval = unshare_files(&displaced);
1337 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1341 retval = prepare_bprm_creds(bprm);
1345 retval = check_unsafe_exec(bprm);
1348 clear_in_exec = retval;
1349 current->in_execve = 1;
1351 file = open_exec(filename);
1352 retval = PTR_ERR(file);
1359 bprm->filename = filename;
1360 bprm->interp = filename;
1362 retval = bprm_mm_init(bprm);
1366 bprm->argc = count(argv, MAX_ARG_STRINGS);
1367 if ((retval = bprm->argc) < 0)
1370 bprm->envc = count(envp, MAX_ARG_STRINGS);
1371 if ((retval = bprm->envc) < 0)
1374 retval = prepare_binprm(bprm);
1378 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1382 bprm->exec = bprm->p;
1383 retval = copy_strings(bprm->envc, envp, bprm);
1387 retval = copy_strings(bprm->argc, argv, bprm);
1391 current->flags &= ~PF_KTHREAD;
1392 retval = search_binary_handler(bprm,regs);
1396 /* execve succeeded */
1397 current->fs->in_exec = 0;
1398 current->in_execve = 0;
1399 acct_update_integrals(current);
1402 put_files_struct(displaced);
1411 allow_write_access(bprm->file);
1417 current->fs->in_exec = 0;
1418 current->in_execve = 0;
1425 reset_files_struct(displaced);
1430 void set_binfmt(struct linux_binfmt *new)
1432 struct mm_struct *mm = current->mm;
1435 module_put(mm->binfmt->module);
1439 __module_get(new->module);
1442 EXPORT_SYMBOL(set_binfmt);
1444 /* format_corename will inspect the pattern parameter, and output a
1445 * name into corename, which must have space for at least
1446 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1448 static int format_corename(char *corename, long signr)
1450 const struct cred *cred = current_cred();
1451 const char *pat_ptr = core_pattern;
1452 int ispipe = (*pat_ptr == '|');
1453 char *out_ptr = corename;
1454 char *const out_end = corename + CORENAME_MAX_SIZE;
1456 int pid_in_pattern = 0;
1458 /* Repeat as long as we have more pattern to process and more output
1461 if (*pat_ptr != '%') {
1462 if (out_ptr == out_end)
1464 *out_ptr++ = *pat_ptr++;
1466 switch (*++pat_ptr) {
1469 /* Double percent, output one percent */
1471 if (out_ptr == out_end)
1478 rc = snprintf(out_ptr, out_end - out_ptr,
1479 "%d", task_tgid_vnr(current));
1480 if (rc > out_end - out_ptr)
1486 rc = snprintf(out_ptr, out_end - out_ptr,
1488 if (rc > out_end - out_ptr)
1494 rc = snprintf(out_ptr, out_end - out_ptr,
1496 if (rc > out_end - out_ptr)
1500 /* signal that caused the coredump */
1502 rc = snprintf(out_ptr, out_end - out_ptr,
1504 if (rc > out_end - out_ptr)
1508 /* UNIX time of coredump */
1511 do_gettimeofday(&tv);
1512 rc = snprintf(out_ptr, out_end - out_ptr,
1514 if (rc > out_end - out_ptr)
1521 down_read(&uts_sem);
1522 rc = snprintf(out_ptr, out_end - out_ptr,
1523 "%s", utsname()->nodename);
1525 if (rc > out_end - out_ptr)
1531 rc = snprintf(out_ptr, out_end - out_ptr,
1532 "%s", current->comm);
1533 if (rc > out_end - out_ptr)
1537 /* core limit size */
1539 rc = snprintf(out_ptr, out_end - out_ptr,
1540 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1541 if (rc > out_end - out_ptr)
1551 /* Backward compatibility with core_uses_pid:
1553 * If core_pattern does not include a %p (as is the default)
1554 * and core_uses_pid is set, then .%pid will be appended to
1555 * the filename. Do not do this for piped commands. */
1556 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1557 rc = snprintf(out_ptr, out_end - out_ptr,
1558 ".%d", task_tgid_vnr(current));
1559 if (rc > out_end - out_ptr)
1568 static int zap_process(struct task_struct *start)
1570 struct task_struct *t;
1573 start->signal->flags = SIGNAL_GROUP_EXIT;
1574 start->signal->group_stop_count = 0;
1578 if (t != current && t->mm) {
1579 sigaddset(&t->pending.signal, SIGKILL);
1580 signal_wake_up(t, 1);
1583 } while_each_thread(start, t);
1588 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1589 struct core_state *core_state, int exit_code)
1591 struct task_struct *g, *p;
1592 unsigned long flags;
1595 spin_lock_irq(&tsk->sighand->siglock);
1596 if (!signal_group_exit(tsk->signal)) {
1597 mm->core_state = core_state;
1598 tsk->signal->group_exit_code = exit_code;
1599 nr = zap_process(tsk);
1601 spin_unlock_irq(&tsk->sighand->siglock);
1602 if (unlikely(nr < 0))
1605 if (atomic_read(&mm->mm_users) == nr + 1)
1608 * We should find and kill all tasks which use this mm, and we should
1609 * count them correctly into ->nr_threads. We don't take tasklist
1610 * lock, but this is safe wrt:
1613 * None of sub-threads can fork after zap_process(leader). All
1614 * processes which were created before this point should be
1615 * visible to zap_threads() because copy_process() adds the new
1616 * process to the tail of init_task.tasks list, and lock/unlock
1617 * of ->siglock provides a memory barrier.
1620 * The caller holds mm->mmap_sem. This means that the task which
1621 * uses this mm can't pass exit_mm(), so it can't exit or clear
1625 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1626 * we must see either old or new leader, this does not matter.
1627 * However, it can change p->sighand, so lock_task_sighand(p)
1628 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1631 * Note also that "g" can be the old leader with ->mm == NULL
1632 * and already unhashed and thus removed from ->thread_group.
1633 * This is OK, __unhash_process()->list_del_rcu() does not
1634 * clear the ->next pointer, we will find the new leader via
1638 for_each_process(g) {
1639 if (g == tsk->group_leader)
1641 if (g->flags & PF_KTHREAD)
1646 if (unlikely(p->mm == mm)) {
1647 lock_task_sighand(p, &flags);
1648 nr += zap_process(p);
1649 unlock_task_sighand(p, &flags);
1653 } while_each_thread(g, p);
1657 atomic_set(&core_state->nr_threads, nr);
1661 static int coredump_wait(int exit_code, struct core_state *core_state)
1663 struct task_struct *tsk = current;
1664 struct mm_struct *mm = tsk->mm;
1665 struct completion *vfork_done;
1668 init_completion(&core_state->startup);
1669 core_state->dumper.task = tsk;
1670 core_state->dumper.next = NULL;
1671 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1672 up_write(&mm->mmap_sem);
1674 if (unlikely(core_waiters < 0))
1678 * Make sure nobody is waiting for us to release the VM,
1679 * otherwise we can deadlock when we wait on each other
1681 vfork_done = tsk->vfork_done;
1683 tsk->vfork_done = NULL;
1684 complete(vfork_done);
1688 wait_for_completion(&core_state->startup);
1690 return core_waiters;
1693 static void coredump_finish(struct mm_struct *mm)
1695 struct core_thread *curr, *next;
1696 struct task_struct *task;
1698 next = mm->core_state->dumper.next;
1699 while ((curr = next) != NULL) {
1703 * see exit_mm(), curr->task must not see
1704 * ->task == NULL before we read ->next.
1708 wake_up_process(task);
1711 mm->core_state = NULL;
1715 * set_dumpable converts traditional three-value dumpable to two flags and
1716 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1717 * these bits are not changed atomically. So get_dumpable can observe the
1718 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1719 * return either old dumpable or new one by paying attention to the order of
1720 * modifying the bits.
1722 * dumpable | mm->flags (binary)
1723 * old new | initial interim final
1724 * ---------+-----------------------
1732 * (*) get_dumpable regards interim value of 10 as 11.
1734 void set_dumpable(struct mm_struct *mm, int value)
1738 clear_bit(MMF_DUMPABLE, &mm->flags);
1740 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1743 set_bit(MMF_DUMPABLE, &mm->flags);
1745 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1748 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1750 set_bit(MMF_DUMPABLE, &mm->flags);
1755 int get_dumpable(struct mm_struct *mm)
1759 ret = mm->flags & 0x3;
1760 return (ret >= 2) ? 2 : ret;
1763 static void wait_for_dump_helpers(struct file *file)
1765 struct pipe_inode_info *pipe;
1767 pipe = file->f_path.dentry->d_inode->i_pipe;
1773 while ((pipe->readers > 1) && (!signal_pending(current))) {
1774 wake_up_interruptible_sync(&pipe->wait);
1775 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1786 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1788 struct core_state core_state;
1789 char corename[CORENAME_MAX_SIZE + 1];
1790 struct mm_struct *mm = current->mm;
1791 struct linux_binfmt * binfmt;
1792 struct inode * inode;
1794 const struct cred *old_cred;
1799 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1800 char **helper_argv = NULL;
1801 int helper_argc = 0;
1803 static atomic_t core_dump_count = ATOMIC_INIT(0);
1805 audit_core_dumps(signr);
1807 binfmt = mm->binfmt;
1808 if (!binfmt || !binfmt->core_dump)
1811 cred = prepare_creds();
1817 down_write(&mm->mmap_sem);
1819 * If another thread got here first, or we are not dumpable, bail out.
1821 if (mm->core_state || !get_dumpable(mm)) {
1822 up_write(&mm->mmap_sem);
1828 * We cannot trust fsuid as being the "true" uid of the
1829 * process nor do we know its entire history. We only know it
1830 * was tainted so we dump it as root in mode 2.
1832 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1833 flag = O_EXCL; /* Stop rewrite attacks */
1834 cred->fsuid = 0; /* Dump root private */
1837 retval = coredump_wait(exit_code, &core_state);
1843 old_cred = override_creds(cred);
1846 * Clear any false indication of pending signals that might
1847 * be seen by the filesystem code called to write the core file.
1849 clear_thread_flag(TIF_SIGPENDING);
1852 * lock_kernel() because format_corename() is controlled by sysctl, which
1853 * uses lock_kernel()
1856 ispipe = format_corename(corename, signr);
1859 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1863 if (core_limit == 0) {
1865 * Normally core limits are irrelevant to pipes, since
1866 * we're not writing to the file system, but we use
1867 * core_limit of 0 here as a speacial value. Any
1868 * non-zero limit gets set to RLIM_INFINITY below, but
1869 * a limit of 0 skips the dump. This is a consistent
1870 * way to catch recursive crashes. We can still crash
1871 * if the core_pattern binary sets RLIM_CORE = !0
1872 * but it runs as root, and can do lots of stupid things
1873 * Note that we use task_tgid_vnr here to grab the pid
1874 * of the process group leader. That way we get the
1875 * right pid if a thread in a multi-threaded
1876 * core_pattern process dies.
1879 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1880 task_tgid_vnr(current), current->comm);
1881 printk(KERN_WARNING "Aborting core\n");
1885 dump_count = atomic_inc_return(&core_dump_count);
1886 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1887 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1888 task_tgid_vnr(current), current->comm);
1889 printk(KERN_WARNING "Skipping core dump\n");
1890 goto fail_dropcount;
1893 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1895 printk(KERN_WARNING "%s failed to allocate memory\n",
1897 goto fail_dropcount;
1900 core_limit = RLIM_INFINITY;
1902 /* SIGPIPE can happen, but it's just never processed */
1903 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1905 printk(KERN_INFO "Core dump to %s pipe failed\n",
1907 goto fail_dropcount;
1910 file = filp_open(corename,
1911 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1914 goto fail_dropcount;
1915 inode = file->f_path.dentry->d_inode;
1916 if (inode->i_nlink > 1)
1917 goto close_fail; /* multiple links - don't dump */
1918 if (!ispipe && d_unhashed(file->f_path.dentry))
1921 /* AK: actually i see no reason to not allow this for named pipes etc.,
1922 but keep the previous behaviour for now. */
1923 if (!ispipe && !S_ISREG(inode->i_mode))
1926 * Dont allow local users get cute and trick others to coredump
1927 * into their pre-created files:
1928 * Note, this is not relevant for pipes
1930 if (!ispipe && (inode->i_uid != current_fsuid()))
1934 if (!file->f_op->write)
1936 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1939 retval = binfmt->core_dump(signr, regs, file, core_limit);
1942 current->signal->group_exit_code |= 0x80;
1944 if (ispipe && core_pipe_limit)
1945 wait_for_dump_helpers(file);
1946 filp_close(file, NULL);
1949 atomic_dec(&core_dump_count);
1952 argv_free(helper_argv);
1954 revert_creds(old_cred);
1956 coredump_finish(mm);