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/swap.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/perf_event.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53 #include <linux/kmod.h>
54 #include <linux/fsnotify.h>
55 #include <linux/fs_struct.h>
56 #include <linux/pipe_fs_i.h>
57 #include <linux/oom.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)
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 > ACCESS_ONCE(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 BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
246 vma->vm_end = STACK_TOP_MAX;
247 vma->vm_start = vma->vm_end - PAGE_SIZE;
248 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
249 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
250 INIT_LIST_HEAD(&vma->anon_vma_chain);
251 err = insert_vm_struct(mm, vma);
255 mm->stack_vm = mm->total_vm = 1;
256 up_write(&mm->mmap_sem);
257 bprm->p = vma->vm_end - sizeof(void *);
260 up_write(&mm->mmap_sem);
262 kmem_cache_free(vm_area_cachep, vma);
266 static bool valid_arg_len(struct linux_binprm *bprm, long len)
268 return len <= MAX_ARG_STRLEN;
273 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
278 page = bprm->page[pos / PAGE_SIZE];
279 if (!page && write) {
280 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
283 bprm->page[pos / PAGE_SIZE] = page;
289 static void put_arg_page(struct page *page)
293 static void free_arg_page(struct linux_binprm *bprm, int i)
296 __free_page(bprm->page[i]);
297 bprm->page[i] = NULL;
301 static void free_arg_pages(struct linux_binprm *bprm)
305 for (i = 0; i < MAX_ARG_PAGES; i++)
306 free_arg_page(bprm, i);
309 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
314 static int __bprm_mm_init(struct linux_binprm *bprm)
316 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
320 static bool valid_arg_len(struct linux_binprm *bprm, long len)
322 return len <= bprm->p;
325 #endif /* CONFIG_MMU */
328 * Create a new mm_struct and populate it with a temporary stack
329 * vm_area_struct. We don't have enough context at this point to set the stack
330 * flags, permissions, and offset, so we use temporary values. We'll update
331 * them later in setup_arg_pages().
333 int bprm_mm_init(struct linux_binprm *bprm)
336 struct mm_struct *mm = NULL;
338 bprm->mm = mm = mm_alloc();
343 err = init_new_context(current, mm);
347 err = __bprm_mm_init(bprm);
363 * count() counts the number of strings in array ARGV.
365 static int count(const char __user * const __user * argv, int max)
371 const char __user * p;
373 if (get_user(p, argv))
381 if (fatal_signal_pending(current))
382 return -ERESTARTNOHAND;
390 * 'copy_strings()' copies argument/environment strings from the old
391 * processes's memory to the new process's stack. The call to get_user_pages()
392 * ensures the destination page is created and not swapped out.
394 static int copy_strings(int argc, const char __user *const __user *argv,
395 struct linux_binprm *bprm)
397 struct page *kmapped_page = NULL;
399 unsigned long kpos = 0;
403 const char __user *str;
407 if (get_user(str, argv+argc) ||
408 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
413 if (!valid_arg_len(bprm, len)) {
418 /* We're going to work our way backwords. */
424 int offset, bytes_to_copy;
426 if (fatal_signal_pending(current)) {
427 ret = -ERESTARTNOHAND;
432 offset = pos % PAGE_SIZE;
436 bytes_to_copy = offset;
437 if (bytes_to_copy > len)
440 offset -= bytes_to_copy;
441 pos -= bytes_to_copy;
442 str -= bytes_to_copy;
443 len -= bytes_to_copy;
445 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
448 page = get_arg_page(bprm, pos, 1);
455 flush_kernel_dcache_page(kmapped_page);
456 kunmap(kmapped_page);
457 put_arg_page(kmapped_page);
460 kaddr = kmap(kmapped_page);
461 kpos = pos & PAGE_MASK;
462 flush_arg_page(bprm, kpos, kmapped_page);
464 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
473 flush_kernel_dcache_page(kmapped_page);
474 kunmap(kmapped_page);
475 put_arg_page(kmapped_page);
481 * Like copy_strings, but get argv and its values from kernel memory.
483 int copy_strings_kernel(int argc, const char *const *argv,
484 struct linux_binprm *bprm)
487 mm_segment_t oldfs = get_fs();
489 r = copy_strings(argc, (const char __user *const __user *)argv, bprm);
493 EXPORT_SYMBOL(copy_strings_kernel);
498 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
499 * the binfmt code determines where the new stack should reside, we shift it to
500 * its final location. The process proceeds as follows:
502 * 1) Use shift to calculate the new vma endpoints.
503 * 2) Extend vma to cover both the old and new ranges. This ensures the
504 * arguments passed to subsequent functions are consistent.
505 * 3) Move vma's page tables to the new range.
506 * 4) Free up any cleared pgd range.
507 * 5) Shrink the vma to cover only the new range.
509 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
511 struct mm_struct *mm = vma->vm_mm;
512 unsigned long old_start = vma->vm_start;
513 unsigned long old_end = vma->vm_end;
514 unsigned long length = old_end - old_start;
515 unsigned long new_start = old_start - shift;
516 unsigned long new_end = old_end - shift;
517 struct mmu_gather *tlb;
519 BUG_ON(new_start > new_end);
522 * ensure there are no vmas between where we want to go
525 if (vma != find_vma(mm, new_start))
529 * cover the whole range: [new_start, old_end)
531 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
535 * move the page tables downwards, on failure we rely on
536 * process cleanup to remove whatever mess we made.
538 if (length != move_page_tables(vma, old_start,
539 vma, new_start, length))
543 tlb = tlb_gather_mmu(mm, 0);
544 if (new_end > old_start) {
546 * when the old and new regions overlap clear from new_end.
548 free_pgd_range(tlb, new_end, old_end, new_end,
549 vma->vm_next ? vma->vm_next->vm_start : 0);
552 * otherwise, clean from old_start; this is done to not touch
553 * the address space in [new_end, old_start) some architectures
554 * have constraints on va-space that make this illegal (IA64) -
555 * for the others its just a little faster.
557 free_pgd_range(tlb, old_start, old_end, new_end,
558 vma->vm_next ? vma->vm_next->vm_start : 0);
560 tlb_finish_mmu(tlb, new_end, old_end);
563 * Shrink the vma to just the new range. Always succeeds.
565 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
571 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
572 * the stack is optionally relocated, and some extra space is added.
574 int setup_arg_pages(struct linux_binprm *bprm,
575 unsigned long stack_top,
576 int executable_stack)
579 unsigned long stack_shift;
580 struct mm_struct *mm = current->mm;
581 struct vm_area_struct *vma = bprm->vma;
582 struct vm_area_struct *prev = NULL;
583 unsigned long vm_flags;
584 unsigned long stack_base;
585 unsigned long stack_size;
586 unsigned long stack_expand;
587 unsigned long rlim_stack;
589 #ifdef CONFIG_STACK_GROWSUP
590 /* Limit stack size to 1GB */
591 stack_base = rlimit_max(RLIMIT_STACK);
592 if (stack_base > (1 << 30))
593 stack_base = 1 << 30;
595 /* Make sure we didn't let the argument array grow too large. */
596 if (vma->vm_end - vma->vm_start > stack_base)
599 stack_base = PAGE_ALIGN(stack_top - stack_base);
601 stack_shift = vma->vm_start - stack_base;
602 mm->arg_start = bprm->p - stack_shift;
603 bprm->p = vma->vm_end - stack_shift;
605 stack_top = arch_align_stack(stack_top);
606 stack_top = PAGE_ALIGN(stack_top);
608 if (unlikely(stack_top < mmap_min_addr) ||
609 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
612 stack_shift = vma->vm_end - stack_top;
614 bprm->p -= stack_shift;
615 mm->arg_start = bprm->p;
619 bprm->loader -= stack_shift;
620 bprm->exec -= stack_shift;
622 down_write(&mm->mmap_sem);
623 vm_flags = VM_STACK_FLAGS;
626 * Adjust stack execute permissions; explicitly enable for
627 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
628 * (arch default) otherwise.
630 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
632 else if (executable_stack == EXSTACK_DISABLE_X)
633 vm_flags &= ~VM_EXEC;
634 vm_flags |= mm->def_flags;
635 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
637 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
643 /* Move stack pages down in memory. */
645 ret = shift_arg_pages(vma, stack_shift);
650 /* mprotect_fixup is overkill to remove the temporary stack flags */
651 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
653 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
654 stack_size = vma->vm_end - vma->vm_start;
656 * Align this down to a page boundary as expand_stack
659 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
660 #ifdef CONFIG_STACK_GROWSUP
661 if (stack_size + stack_expand > rlim_stack)
662 stack_base = vma->vm_start + rlim_stack;
664 stack_base = vma->vm_end + stack_expand;
666 if (stack_size + stack_expand > rlim_stack)
667 stack_base = vma->vm_end - rlim_stack;
669 stack_base = vma->vm_start - stack_expand;
671 current->mm->start_stack = bprm->p;
672 ret = expand_stack(vma, stack_base);
677 up_write(&mm->mmap_sem);
680 EXPORT_SYMBOL(setup_arg_pages);
682 #endif /* CONFIG_MMU */
684 struct file *open_exec(const char *name)
689 file = do_filp_open(AT_FDCWD, name,
690 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
691 MAY_EXEC | MAY_OPEN);
696 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
699 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
704 err = deny_write_access(file);
715 EXPORT_SYMBOL(open_exec);
717 int kernel_read(struct file *file, loff_t offset,
718 char *addr, unsigned long count)
726 /* The cast to a user pointer is valid due to the set_fs() */
727 result = vfs_read(file, (void __user *)addr, count, &pos);
732 EXPORT_SYMBOL(kernel_read);
734 static int exec_mmap(struct mm_struct *mm)
736 struct task_struct *tsk;
737 struct mm_struct * old_mm, *active_mm;
739 /* Notify parent that we're no longer interested in the old VM */
741 old_mm = current->mm;
742 sync_mm_rss(tsk, old_mm);
743 mm_release(tsk, old_mm);
747 * Make sure that if there is a core dump in progress
748 * for the old mm, we get out and die instead of going
749 * through with the exec. We must hold mmap_sem around
750 * checking core_state and changing tsk->mm.
752 down_read(&old_mm->mmap_sem);
753 if (unlikely(old_mm->core_state)) {
754 up_read(&old_mm->mmap_sem);
759 active_mm = tsk->active_mm;
762 activate_mm(active_mm, mm);
763 if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
764 atomic_dec(&old_mm->oom_disable_count);
765 atomic_inc(&tsk->mm->oom_disable_count);
768 arch_pick_mmap_layout(mm);
770 up_read(&old_mm->mmap_sem);
771 BUG_ON(active_mm != old_mm);
772 mm_update_next_owner(old_mm);
781 * This function makes sure the current process has its own signal table,
782 * so that flush_signal_handlers can later reset the handlers without
783 * disturbing other processes. (Other processes might share the signal
784 * table via the CLONE_SIGHAND option to clone().)
786 static int de_thread(struct task_struct *tsk)
788 struct signal_struct *sig = tsk->signal;
789 struct sighand_struct *oldsighand = tsk->sighand;
790 spinlock_t *lock = &oldsighand->siglock;
792 if (thread_group_empty(tsk))
793 goto no_thread_group;
796 * Kill all other threads in the thread group.
799 if (signal_group_exit(sig)) {
801 * Another group action in progress, just
802 * return so that the signal is processed.
804 spin_unlock_irq(lock);
808 sig->group_exit_task = tsk;
809 sig->notify_count = zap_other_threads(tsk);
810 if (!thread_group_leader(tsk))
813 while (sig->notify_count) {
814 __set_current_state(TASK_UNINTERRUPTIBLE);
815 spin_unlock_irq(lock);
819 spin_unlock_irq(lock);
822 * At this point all other threads have exited, all we have to
823 * do is to wait for the thread group leader to become inactive,
824 * and to assume its PID:
826 if (!thread_group_leader(tsk)) {
827 struct task_struct *leader = tsk->group_leader;
829 sig->notify_count = -1; /* for exit_notify() */
831 write_lock_irq(&tasklist_lock);
832 if (likely(leader->exit_state))
834 __set_current_state(TASK_UNINTERRUPTIBLE);
835 write_unlock_irq(&tasklist_lock);
840 * The only record we have of the real-time age of a
841 * process, regardless of execs it's done, is start_time.
842 * All the past CPU time is accumulated in signal_struct
843 * from sister threads now dead. But in this non-leader
844 * exec, nothing survives from the original leader thread,
845 * whose birth marks the true age of this process now.
846 * When we take on its identity by switching to its PID, we
847 * also take its birthdate (always earlier than our own).
849 tsk->start_time = leader->start_time;
851 BUG_ON(!same_thread_group(leader, tsk));
852 BUG_ON(has_group_leader_pid(tsk));
854 * An exec() starts a new thread group with the
855 * TGID of the previous thread group. Rehash the
856 * two threads with a switched PID, and release
857 * the former thread group leader:
860 /* Become a process group leader with the old leader's pid.
861 * The old leader becomes a thread of the this thread group.
862 * Note: The old leader also uses this pid until release_task
863 * is called. Odd but simple and correct.
865 detach_pid(tsk, PIDTYPE_PID);
866 tsk->pid = leader->pid;
867 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
868 transfer_pid(leader, tsk, PIDTYPE_PGID);
869 transfer_pid(leader, tsk, PIDTYPE_SID);
871 list_replace_rcu(&leader->tasks, &tsk->tasks);
872 list_replace_init(&leader->sibling, &tsk->sibling);
874 tsk->group_leader = tsk;
875 leader->group_leader = tsk;
877 tsk->exit_signal = SIGCHLD;
879 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
880 leader->exit_state = EXIT_DEAD;
881 write_unlock_irq(&tasklist_lock);
883 release_task(leader);
886 sig->group_exit_task = NULL;
887 sig->notify_count = 0;
891 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
894 flush_itimer_signals();
896 if (atomic_read(&oldsighand->count) != 1) {
897 struct sighand_struct *newsighand;
899 * This ->sighand is shared with the CLONE_SIGHAND
900 * but not CLONE_THREAD task, switch to the new one.
902 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
906 atomic_set(&newsighand->count, 1);
907 memcpy(newsighand->action, oldsighand->action,
908 sizeof(newsighand->action));
910 write_lock_irq(&tasklist_lock);
911 spin_lock(&oldsighand->siglock);
912 rcu_assign_pointer(tsk->sighand, newsighand);
913 spin_unlock(&oldsighand->siglock);
914 write_unlock_irq(&tasklist_lock);
916 __cleanup_sighand(oldsighand);
919 BUG_ON(!thread_group_leader(tsk));
924 * These functions flushes out all traces of the currently running executable
925 * so that a new one can be started
927 static void flush_old_files(struct files_struct * files)
932 spin_lock(&files->file_lock);
934 unsigned long set, i;
938 fdt = files_fdtable(files);
939 if (i >= fdt->max_fds)
941 set = fdt->close_on_exec->fds_bits[j];
944 fdt->close_on_exec->fds_bits[j] = 0;
945 spin_unlock(&files->file_lock);
946 for ( ; set ; i++,set >>= 1) {
951 spin_lock(&files->file_lock);
954 spin_unlock(&files->file_lock);
957 char *get_task_comm(char *buf, struct task_struct *tsk)
959 /* buf must be at least sizeof(tsk->comm) in size */
961 strncpy(buf, tsk->comm, sizeof(tsk->comm));
966 void set_task_comm(struct task_struct *tsk, char *buf)
971 * Threads may access current->comm without holding
972 * the task lock, so write the string carefully.
973 * Readers without a lock may see incomplete new
974 * names but are safe from non-terminating string reads.
976 memset(tsk->comm, 0, TASK_COMM_LEN);
978 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
980 perf_event_comm(tsk);
983 int flush_old_exec(struct linux_binprm * bprm)
988 * Make sure we have a private signal table and that
989 * we are unassociated from the previous thread group.
991 retval = de_thread(current);
995 set_mm_exe_file(bprm->mm, bprm->file);
998 * Release all of the old mmap stuff
1000 retval = exec_mmap(bprm->mm);
1004 bprm->mm = NULL; /* We're using it now */
1006 current->flags &= ~PF_RANDOMIZE;
1008 current->personality &= ~bprm->per_clear;
1015 EXPORT_SYMBOL(flush_old_exec);
1017 void setup_new_exec(struct linux_binprm * bprm)
1021 char tcomm[sizeof(current->comm)];
1023 arch_pick_mmap_layout(current->mm);
1025 /* This is the point of no return */
1026 current->sas_ss_sp = current->sas_ss_size = 0;
1028 if (current_euid() == current_uid() && current_egid() == current_gid())
1029 set_dumpable(current->mm, 1);
1031 set_dumpable(current->mm, suid_dumpable);
1033 name = bprm->filename;
1035 /* Copies the binary name from after last slash */
1036 for (i=0; (ch = *(name++)) != '\0';) {
1038 i = 0; /* overwrite what we wrote */
1040 if (i < (sizeof(tcomm) - 1))
1044 set_task_comm(current, tcomm);
1046 /* Set the new mm task size. We have to do that late because it may
1047 * depend on TIF_32BIT which is only updated in flush_thread() on
1048 * some architectures like powerpc
1050 current->mm->task_size = TASK_SIZE;
1052 /* install the new credentials */
1053 if (bprm->cred->uid != current_euid() ||
1054 bprm->cred->gid != current_egid()) {
1055 current->pdeath_signal = 0;
1056 } else if (file_permission(bprm->file, MAY_READ) ||
1057 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1058 set_dumpable(current->mm, suid_dumpable);
1062 * Flush performance counters when crossing a
1065 if (!get_dumpable(current->mm))
1066 perf_event_exit_task(current);
1068 /* An exec changes our domain. We are no longer part of the thread
1071 current->self_exec_id++;
1073 flush_signal_handlers(current, 0);
1074 flush_old_files(current->files);
1076 EXPORT_SYMBOL(setup_new_exec);
1079 * Prepare credentials and lock ->cred_guard_mutex.
1080 * install_exec_creds() commits the new creds and drops the lock.
1081 * Or, if exec fails before, free_bprm() should release ->cred and
1084 int prepare_bprm_creds(struct linux_binprm *bprm)
1086 if (mutex_lock_interruptible(¤t->cred_guard_mutex))
1087 return -ERESTARTNOINTR;
1089 bprm->cred = prepare_exec_creds();
1090 if (likely(bprm->cred))
1093 mutex_unlock(¤t->cred_guard_mutex);
1097 void free_bprm(struct linux_binprm *bprm)
1099 free_arg_pages(bprm);
1101 mutex_unlock(¤t->cred_guard_mutex);
1102 abort_creds(bprm->cred);
1108 * install the new credentials for this executable
1110 void install_exec_creds(struct linux_binprm *bprm)
1112 security_bprm_committing_creds(bprm);
1114 commit_creds(bprm->cred);
1117 * cred_guard_mutex must be held at least to this point to prevent
1118 * ptrace_attach() from altering our determination of the task's
1119 * credentials; any time after this it may be unlocked.
1121 security_bprm_committed_creds(bprm);
1122 mutex_unlock(¤t->cred_guard_mutex);
1124 EXPORT_SYMBOL(install_exec_creds);
1127 * determine how safe it is to execute the proposed program
1128 * - the caller must hold current->cred_guard_mutex to protect against
1131 int check_unsafe_exec(struct linux_binprm *bprm)
1133 struct task_struct *p = current, *t;
1137 bprm->unsafe = tracehook_unsafe_exec(p);
1140 spin_lock(&p->fs->lock);
1142 for (t = next_thread(p); t != p; t = next_thread(t)) {
1148 if (p->fs->users > n_fs) {
1149 bprm->unsafe |= LSM_UNSAFE_SHARE;
1152 if (!p->fs->in_exec) {
1157 spin_unlock(&p->fs->lock);
1163 * Fill the binprm structure from the inode.
1164 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1166 * This may be called multiple times for binary chains (scripts for example).
1168 int prepare_binprm(struct linux_binprm *bprm)
1171 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1174 mode = inode->i_mode;
1175 if (bprm->file->f_op == NULL)
1178 /* clear any previous set[ug]id data from a previous binary */
1179 bprm->cred->euid = current_euid();
1180 bprm->cred->egid = current_egid();
1182 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1184 if (mode & S_ISUID) {
1185 bprm->per_clear |= PER_CLEAR_ON_SETID;
1186 bprm->cred->euid = inode->i_uid;
1191 * If setgid is set but no group execute bit then this
1192 * is a candidate for mandatory locking, not a setgid
1195 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1196 bprm->per_clear |= PER_CLEAR_ON_SETID;
1197 bprm->cred->egid = inode->i_gid;
1201 /* fill in binprm security blob */
1202 retval = security_bprm_set_creds(bprm);
1205 bprm->cred_prepared = 1;
1207 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1208 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1211 EXPORT_SYMBOL(prepare_binprm);
1214 * Arguments are '\0' separated strings found at the location bprm->p
1215 * points to; chop off the first by relocating brpm->p to right after
1216 * the first '\0' encountered.
1218 int remove_arg_zero(struct linux_binprm *bprm)
1221 unsigned long offset;
1229 offset = bprm->p & ~PAGE_MASK;
1230 page = get_arg_page(bprm, bprm->p, 0);
1235 kaddr = kmap_atomic(page, KM_USER0);
1237 for (; offset < PAGE_SIZE && kaddr[offset];
1238 offset++, bprm->p++)
1241 kunmap_atomic(kaddr, KM_USER0);
1244 if (offset == PAGE_SIZE)
1245 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1246 } while (offset == PAGE_SIZE);
1255 EXPORT_SYMBOL(remove_arg_zero);
1258 * cycle the list of binary formats handler, until one recognizes the image
1260 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1262 unsigned int depth = bprm->recursion_depth;
1264 struct linux_binfmt *fmt;
1266 retval = security_bprm_check(bprm);
1270 /* kernel module loader fixup */
1271 /* so we don't try to load run modprobe in kernel space. */
1274 retval = audit_bprm(bprm);
1279 for (try=0; try<2; try++) {
1280 read_lock(&binfmt_lock);
1281 list_for_each_entry(fmt, &formats, lh) {
1282 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1285 if (!try_module_get(fmt->module))
1287 read_unlock(&binfmt_lock);
1288 retval = fn(bprm, regs);
1290 * Restore the depth counter to its starting value
1291 * in this call, so we don't have to rely on every
1292 * load_binary function to restore it on return.
1294 bprm->recursion_depth = depth;
1297 tracehook_report_exec(fmt, bprm, regs);
1299 allow_write_access(bprm->file);
1303 current->did_exec = 1;
1304 proc_exec_connector(current);
1307 read_lock(&binfmt_lock);
1309 if (retval != -ENOEXEC || bprm->mm == NULL)
1312 read_unlock(&binfmt_lock);
1316 read_unlock(&binfmt_lock);
1317 if (retval != -ENOEXEC || bprm->mm == NULL) {
1319 #ifdef CONFIG_MODULES
1321 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1322 if (printable(bprm->buf[0]) &&
1323 printable(bprm->buf[1]) &&
1324 printable(bprm->buf[2]) &&
1325 printable(bprm->buf[3]))
1326 break; /* -ENOEXEC */
1327 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1334 EXPORT_SYMBOL(search_binary_handler);
1337 * sys_execve() executes a new program.
1339 int do_execve(const char * filename,
1340 const char __user *const __user *argv,
1341 const char __user *const __user *envp,
1342 struct pt_regs * regs)
1344 struct linux_binprm *bprm;
1346 struct files_struct *displaced;
1350 retval = unshare_files(&displaced);
1355 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1359 retval = prepare_bprm_creds(bprm);
1363 retval = check_unsafe_exec(bprm);
1366 clear_in_exec = retval;
1367 current->in_execve = 1;
1369 file = open_exec(filename);
1370 retval = PTR_ERR(file);
1377 bprm->filename = filename;
1378 bprm->interp = filename;
1380 retval = bprm_mm_init(bprm);
1384 bprm->argc = count(argv, MAX_ARG_STRINGS);
1385 if ((retval = bprm->argc) < 0)
1388 bprm->envc = count(envp, MAX_ARG_STRINGS);
1389 if ((retval = bprm->envc) < 0)
1392 retval = prepare_binprm(bprm);
1396 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1400 bprm->exec = bprm->p;
1401 retval = copy_strings(bprm->envc, envp, bprm);
1405 retval = copy_strings(bprm->argc, argv, bprm);
1409 current->flags &= ~PF_KTHREAD;
1410 retval = search_binary_handler(bprm,regs);
1414 /* execve succeeded */
1415 current->fs->in_exec = 0;
1416 current->in_execve = 0;
1417 acct_update_integrals(current);
1420 put_files_struct(displaced);
1429 allow_write_access(bprm->file);
1435 current->fs->in_exec = 0;
1436 current->in_execve = 0;
1443 reset_files_struct(displaced);
1448 void set_binfmt(struct linux_binfmt *new)
1450 struct mm_struct *mm = current->mm;
1453 module_put(mm->binfmt->module);
1457 __module_get(new->module);
1460 EXPORT_SYMBOL(set_binfmt);
1462 /* format_corename will inspect the pattern parameter, and output a
1463 * name into corename, which must have space for at least
1464 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1466 static int format_corename(char *corename, long signr)
1468 const struct cred *cred = current_cred();
1469 const char *pat_ptr = core_pattern;
1470 int ispipe = (*pat_ptr == '|');
1471 char *out_ptr = corename;
1472 char *const out_end = corename + CORENAME_MAX_SIZE;
1474 int pid_in_pattern = 0;
1476 /* Repeat as long as we have more pattern to process and more output
1479 if (*pat_ptr != '%') {
1480 if (out_ptr == out_end)
1482 *out_ptr++ = *pat_ptr++;
1484 switch (*++pat_ptr) {
1487 /* Double percent, output one percent */
1489 if (out_ptr == out_end)
1496 rc = snprintf(out_ptr, out_end - out_ptr,
1497 "%d", task_tgid_vnr(current));
1498 if (rc > out_end - out_ptr)
1504 rc = snprintf(out_ptr, out_end - out_ptr,
1506 if (rc > out_end - out_ptr)
1512 rc = snprintf(out_ptr, out_end - out_ptr,
1514 if (rc > out_end - out_ptr)
1518 /* signal that caused the coredump */
1520 rc = snprintf(out_ptr, out_end - out_ptr,
1522 if (rc > out_end - out_ptr)
1526 /* UNIX time of coredump */
1529 do_gettimeofday(&tv);
1530 rc = snprintf(out_ptr, out_end - out_ptr,
1532 if (rc > out_end - out_ptr)
1539 down_read(&uts_sem);
1540 rc = snprintf(out_ptr, out_end - out_ptr,
1541 "%s", utsname()->nodename);
1543 if (rc > out_end - out_ptr)
1549 rc = snprintf(out_ptr, out_end - out_ptr,
1550 "%s", current->comm);
1551 if (rc > out_end - out_ptr)
1555 /* core limit size */
1557 rc = snprintf(out_ptr, out_end - out_ptr,
1558 "%lu", rlimit(RLIMIT_CORE));
1559 if (rc > out_end - out_ptr)
1569 /* Backward compatibility with core_uses_pid:
1571 * If core_pattern does not include a %p (as is the default)
1572 * and core_uses_pid is set, then .%pid will be appended to
1573 * the filename. Do not do this for piped commands. */
1574 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1575 rc = snprintf(out_ptr, out_end - out_ptr,
1576 ".%d", task_tgid_vnr(current));
1577 if (rc > out_end - out_ptr)
1586 static int zap_process(struct task_struct *start, int exit_code)
1588 struct task_struct *t;
1591 start->signal->flags = SIGNAL_GROUP_EXIT;
1592 start->signal->group_exit_code = exit_code;
1593 start->signal->group_stop_count = 0;
1597 if (t != current && t->mm) {
1598 sigaddset(&t->pending.signal, SIGKILL);
1599 signal_wake_up(t, 1);
1602 } while_each_thread(start, t);
1607 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1608 struct core_state *core_state, int exit_code)
1610 struct task_struct *g, *p;
1611 unsigned long flags;
1614 spin_lock_irq(&tsk->sighand->siglock);
1615 if (!signal_group_exit(tsk->signal)) {
1616 mm->core_state = core_state;
1617 nr = zap_process(tsk, exit_code);
1619 spin_unlock_irq(&tsk->sighand->siglock);
1620 if (unlikely(nr < 0))
1623 if (atomic_read(&mm->mm_users) == nr + 1)
1626 * We should find and kill all tasks which use this mm, and we should
1627 * count them correctly into ->nr_threads. We don't take tasklist
1628 * lock, but this is safe wrt:
1631 * None of sub-threads can fork after zap_process(leader). All
1632 * processes which were created before this point should be
1633 * visible to zap_threads() because copy_process() adds the new
1634 * process to the tail of init_task.tasks list, and lock/unlock
1635 * of ->siglock provides a memory barrier.
1638 * The caller holds mm->mmap_sem. This means that the task which
1639 * uses this mm can't pass exit_mm(), so it can't exit or clear
1643 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1644 * we must see either old or new leader, this does not matter.
1645 * However, it can change p->sighand, so lock_task_sighand(p)
1646 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1649 * Note also that "g" can be the old leader with ->mm == NULL
1650 * and already unhashed and thus removed from ->thread_group.
1651 * This is OK, __unhash_process()->list_del_rcu() does not
1652 * clear the ->next pointer, we will find the new leader via
1656 for_each_process(g) {
1657 if (g == tsk->group_leader)
1659 if (g->flags & PF_KTHREAD)
1664 if (unlikely(p->mm == mm)) {
1665 lock_task_sighand(p, &flags);
1666 nr += zap_process(p, exit_code);
1667 unlock_task_sighand(p, &flags);
1671 } while_each_thread(g, p);
1675 atomic_set(&core_state->nr_threads, nr);
1679 static int coredump_wait(int exit_code, struct core_state *core_state)
1681 struct task_struct *tsk = current;
1682 struct mm_struct *mm = tsk->mm;
1683 struct completion *vfork_done;
1684 int core_waiters = -EBUSY;
1686 init_completion(&core_state->startup);
1687 core_state->dumper.task = tsk;
1688 core_state->dumper.next = NULL;
1690 down_write(&mm->mmap_sem);
1691 if (!mm->core_state)
1692 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1693 up_write(&mm->mmap_sem);
1695 if (unlikely(core_waiters < 0))
1699 * Make sure nobody is waiting for us to release the VM,
1700 * otherwise we can deadlock when we wait on each other
1702 vfork_done = tsk->vfork_done;
1704 tsk->vfork_done = NULL;
1705 complete(vfork_done);
1709 wait_for_completion(&core_state->startup);
1711 return core_waiters;
1714 static void coredump_finish(struct mm_struct *mm)
1716 struct core_thread *curr, *next;
1717 struct task_struct *task;
1719 next = mm->core_state->dumper.next;
1720 while ((curr = next) != NULL) {
1724 * see exit_mm(), curr->task must not see
1725 * ->task == NULL before we read ->next.
1729 wake_up_process(task);
1732 mm->core_state = NULL;
1736 * set_dumpable converts traditional three-value dumpable to two flags and
1737 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1738 * these bits are not changed atomically. So get_dumpable can observe the
1739 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1740 * return either old dumpable or new one by paying attention to the order of
1741 * modifying the bits.
1743 * dumpable | mm->flags (binary)
1744 * old new | initial interim final
1745 * ---------+-----------------------
1753 * (*) get_dumpable regards interim value of 10 as 11.
1755 void set_dumpable(struct mm_struct *mm, int value)
1759 clear_bit(MMF_DUMPABLE, &mm->flags);
1761 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1764 set_bit(MMF_DUMPABLE, &mm->flags);
1766 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1769 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1771 set_bit(MMF_DUMPABLE, &mm->flags);
1776 static int __get_dumpable(unsigned long mm_flags)
1780 ret = mm_flags & MMF_DUMPABLE_MASK;
1781 return (ret >= 2) ? 2 : ret;
1784 int get_dumpable(struct mm_struct *mm)
1786 return __get_dumpable(mm->flags);
1789 static void wait_for_dump_helpers(struct file *file)
1791 struct pipe_inode_info *pipe;
1793 pipe = file->f_path.dentry->d_inode->i_pipe;
1799 while ((pipe->readers > 1) && (!signal_pending(current))) {
1800 wake_up_interruptible_sync(&pipe->wait);
1801 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1814 * helper function to customize the process used
1815 * to collect the core in userspace. Specifically
1816 * it sets up a pipe and installs it as fd 0 (stdin)
1817 * for the process. Returns 0 on success, or
1818 * PTR_ERR on failure.
1819 * Note that it also sets the core limit to 1. This
1820 * is a special value that we use to trap recursive
1823 static int umh_pipe_setup(struct subprocess_info *info)
1825 struct file *rp, *wp;
1826 struct fdtable *fdt;
1827 struct coredump_params *cp = (struct coredump_params *)info->data;
1828 struct files_struct *cf = current->files;
1830 wp = create_write_pipe(0);
1834 rp = create_read_pipe(wp, 0);
1836 free_write_pipe(wp);
1844 spin_lock(&cf->file_lock);
1845 fdt = files_fdtable(cf);
1846 FD_SET(0, fdt->open_fds);
1847 FD_CLR(0, fdt->close_on_exec);
1848 spin_unlock(&cf->file_lock);
1850 /* and disallow core files too */
1851 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1856 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1858 struct core_state core_state;
1859 char corename[CORENAME_MAX_SIZE + 1];
1860 struct mm_struct *mm = current->mm;
1861 struct linux_binfmt * binfmt;
1862 const struct cred *old_cred;
1867 static atomic_t core_dump_count = ATOMIC_INIT(0);
1868 struct coredump_params cprm = {
1871 .limit = rlimit(RLIMIT_CORE),
1873 * We must use the same mm->flags while dumping core to avoid
1874 * inconsistency of bit flags, since this flag is not protected
1877 .mm_flags = mm->flags,
1880 audit_core_dumps(signr);
1882 binfmt = mm->binfmt;
1883 if (!binfmt || !binfmt->core_dump)
1885 if (!__get_dumpable(cprm.mm_flags))
1888 cred = prepare_creds();
1892 * We cannot trust fsuid as being the "true" uid of the
1893 * process nor do we know its entire history. We only know it
1894 * was tainted so we dump it as root in mode 2.
1896 if (__get_dumpable(cprm.mm_flags) == 2) {
1897 /* Setuid core dump mode */
1898 flag = O_EXCL; /* Stop rewrite attacks */
1899 cred->fsuid = 0; /* Dump root private */
1902 retval = coredump_wait(exit_code, &core_state);
1906 old_cred = override_creds(cred);
1909 * Clear any false indication of pending signals that might
1910 * be seen by the filesystem code called to write the core file.
1912 clear_thread_flag(TIF_SIGPENDING);
1914 ispipe = format_corename(corename, signr);
1920 if (cprm.limit == 1) {
1922 * Normally core limits are irrelevant to pipes, since
1923 * we're not writing to the file system, but we use
1924 * cprm.limit of 1 here as a speacial value. Any
1925 * non-1 limit gets set to RLIM_INFINITY below, but
1926 * a limit of 0 skips the dump. This is a consistent
1927 * way to catch recursive crashes. We can still crash
1928 * if the core_pattern binary sets RLIM_CORE = !1
1929 * but it runs as root, and can do lots of stupid things
1930 * Note that we use task_tgid_vnr here to grab the pid
1931 * of the process group leader. That way we get the
1932 * right pid if a thread in a multi-threaded
1933 * core_pattern process dies.
1936 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1937 task_tgid_vnr(current), current->comm);
1938 printk(KERN_WARNING "Aborting core\n");
1941 cprm.limit = RLIM_INFINITY;
1943 dump_count = atomic_inc_return(&core_dump_count);
1944 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1945 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1946 task_tgid_vnr(current), current->comm);
1947 printk(KERN_WARNING "Skipping core dump\n");
1948 goto fail_dropcount;
1951 helper_argv = argv_split(GFP_KERNEL, corename+1, NULL);
1953 printk(KERN_WARNING "%s failed to allocate memory\n",
1955 goto fail_dropcount;
1958 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1959 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1961 argv_free(helper_argv);
1963 printk(KERN_INFO "Core dump to %s pipe failed\n",
1968 struct inode *inode;
1970 if (cprm.limit < binfmt->min_coredump)
1973 cprm.file = filp_open(corename,
1974 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1976 if (IS_ERR(cprm.file))
1979 inode = cprm.file->f_path.dentry->d_inode;
1980 if (inode->i_nlink > 1)
1982 if (d_unhashed(cprm.file->f_path.dentry))
1985 * AK: actually i see no reason to not allow this for named
1986 * pipes etc, but keep the previous behaviour for now.
1988 if (!S_ISREG(inode->i_mode))
1991 * Dont allow local users get cute and trick others to coredump
1992 * into their pre-created files.
1994 if (inode->i_uid != current_fsuid())
1996 if (!cprm.file->f_op || !cprm.file->f_op->write)
1998 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2002 retval = binfmt->core_dump(&cprm);
2004 current->signal->group_exit_code |= 0x80;
2006 if (ispipe && core_pipe_limit)
2007 wait_for_dump_helpers(cprm.file);
2010 filp_close(cprm.file, NULL);
2013 atomic_dec(&core_dump_count);
2015 coredump_finish(mm);
2016 revert_creds(old_cred);
2024 * Core dumping helper functions. These are the only things you should
2025 * do on a core-file: use only these functions to write out all the
2028 int dump_write(struct file *file, const void *addr, int nr)
2030 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2032 EXPORT_SYMBOL(dump_write);
2034 int dump_seek(struct file *file, loff_t off)
2038 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2039 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2042 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2047 unsigned long n = off;
2051 if (!dump_write(file, buf, n)) {
2057 free_page((unsigned long)buf);
2061 EXPORT_SYMBOL(dump_seek);