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>
58 #include <asm/uaccess.h>
59 #include <asm/mmu_context.h>
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 unsigned int core_pipe_limit;
66 int suid_dumpable = 0;
68 /* The maximal length of core_pattern is also specified in sysctl.c */
70 static LIST_HEAD(formats);
71 static DEFINE_RWLOCK(binfmt_lock);
73 int __register_binfmt(struct linux_binfmt * fmt, int insert)
77 write_lock(&binfmt_lock);
78 insert ? list_add(&fmt->lh, &formats) :
79 list_add_tail(&fmt->lh, &formats);
80 write_unlock(&binfmt_lock);
84 EXPORT_SYMBOL(__register_binfmt);
86 void unregister_binfmt(struct linux_binfmt * fmt)
88 write_lock(&binfmt_lock);
90 write_unlock(&binfmt_lock);
93 EXPORT_SYMBOL(unregister_binfmt);
95 static inline void put_binfmt(struct linux_binfmt * fmt)
97 module_put(fmt->module);
101 * Note that a shared library must be both readable and executable due to
104 * Also note that we take the address to load from from the file itself.
106 SYSCALL_DEFINE1(uselib, const char __user *, library)
109 char *tmp = getname(library);
110 int error = PTR_ERR(tmp);
115 file = do_filp_open(AT_FDCWD, tmp,
116 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
117 MAY_READ | MAY_EXEC | MAY_OPEN);
119 error = PTR_ERR(file);
124 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
128 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
131 fsnotify_open(file->f_path.dentry);
135 struct linux_binfmt * fmt;
137 read_lock(&binfmt_lock);
138 list_for_each_entry(fmt, &formats, lh) {
139 if (!fmt->load_shlib)
141 if (!try_module_get(fmt->module))
143 read_unlock(&binfmt_lock);
144 error = fmt->load_shlib(file);
145 read_lock(&binfmt_lock);
147 if (error != -ENOEXEC)
150 read_unlock(&binfmt_lock);
160 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
166 #ifdef CONFIG_STACK_GROWSUP
168 ret = expand_stack_downwards(bprm->vma, pos);
173 ret = get_user_pages(current, bprm->mm, pos,
174 1, write, 1, &page, NULL);
179 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
183 * We've historically supported up to 32 pages (ARG_MAX)
184 * of argument strings even with small stacks
190 * Limit to 1/4-th the stack size for the argv+env strings.
192 * - the remaining binfmt code will not run out of stack space,
193 * - the program will have a reasonable amount of stack left
196 rlim = current->signal->rlim;
197 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
206 static void put_arg_page(struct page *page)
211 static void free_arg_page(struct linux_binprm *bprm, int i)
215 static void free_arg_pages(struct linux_binprm *bprm)
219 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
222 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
225 static int __bprm_mm_init(struct linux_binprm *bprm)
228 struct vm_area_struct *vma = NULL;
229 struct mm_struct *mm = bprm->mm;
231 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
235 down_write(&mm->mmap_sem);
239 * Place the stack at the largest stack address the architecture
240 * supports. Later, we'll move this to an appropriate place. We don't
241 * use STACK_TOP because that can depend on attributes which aren't
244 BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
245 vma->vm_end = STACK_TOP_MAX;
246 vma->vm_start = vma->vm_end - PAGE_SIZE;
247 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
248 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
249 INIT_LIST_HEAD(&vma->anon_vma_chain);
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))
386 * 'copy_strings()' copies argument/environment strings from the old
387 * processes's memory to the new process's stack. The call to get_user_pages()
388 * ensures the destination page is created and not swapped out.
390 static int copy_strings(int argc, char __user * __user * argv,
391 struct linux_binprm *bprm)
393 struct page *kmapped_page = NULL;
395 unsigned long kpos = 0;
403 if (get_user(str, argv+argc) ||
404 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
409 if (!valid_arg_len(bprm, len)) {
414 /* We're going to work our way backwords. */
420 int offset, bytes_to_copy;
422 offset = pos % PAGE_SIZE;
426 bytes_to_copy = offset;
427 if (bytes_to_copy > len)
430 offset -= bytes_to_copy;
431 pos -= bytes_to_copy;
432 str -= bytes_to_copy;
433 len -= bytes_to_copy;
435 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
438 page = get_arg_page(bprm, pos, 1);
445 flush_kernel_dcache_page(kmapped_page);
446 kunmap(kmapped_page);
447 put_arg_page(kmapped_page);
450 kaddr = kmap(kmapped_page);
451 kpos = pos & PAGE_MASK;
452 flush_arg_page(bprm, kpos, kmapped_page);
454 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
463 flush_kernel_dcache_page(kmapped_page);
464 kunmap(kmapped_page);
465 put_arg_page(kmapped_page);
471 * Like copy_strings, but get argv and its values from kernel memory.
473 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
476 mm_segment_t oldfs = get_fs();
478 r = copy_strings(argc, (char __user * __user *)argv, bprm);
482 EXPORT_SYMBOL(copy_strings_kernel);
487 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
488 * the binfmt code determines where the new stack should reside, we shift it to
489 * its final location. The process proceeds as follows:
491 * 1) Use shift to calculate the new vma endpoints.
492 * 2) Extend vma to cover both the old and new ranges. This ensures the
493 * arguments passed to subsequent functions are consistent.
494 * 3) Move vma's page tables to the new range.
495 * 4) Free up any cleared pgd range.
496 * 5) Shrink the vma to cover only the new range.
498 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
500 struct mm_struct *mm = vma->vm_mm;
501 unsigned long old_start = vma->vm_start;
502 unsigned long old_end = vma->vm_end;
503 unsigned long length = old_end - old_start;
504 unsigned long new_start = old_start - shift;
505 unsigned long new_end = old_end - shift;
506 struct mmu_gather *tlb;
508 BUG_ON(new_start > new_end);
511 * ensure there are no vmas between where we want to go
514 if (vma != find_vma(mm, new_start))
518 * cover the whole range: [new_start, old_end)
520 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
524 * move the page tables downwards, on failure we rely on
525 * process cleanup to remove whatever mess we made.
527 if (length != move_page_tables(vma, old_start,
528 vma, new_start, length))
532 tlb = tlb_gather_mmu(mm, 0);
533 if (new_end > old_start) {
535 * when the old and new regions overlap clear from new_end.
537 free_pgd_range(tlb, new_end, old_end, new_end,
538 vma->vm_next ? vma->vm_next->vm_start : 0);
541 * otherwise, clean from old_start; this is done to not touch
542 * the address space in [new_end, old_start) some architectures
543 * have constraints on va-space that make this illegal (IA64) -
544 * for the others its just a little faster.
546 free_pgd_range(tlb, old_start, old_end, new_end,
547 vma->vm_next ? vma->vm_next->vm_start : 0);
549 tlb_finish_mmu(tlb, new_end, old_end);
552 * Shrink the vma to just the new range. Always succeeds.
554 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
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 = rlimit_max(RLIMIT_STACK);
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;
619 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
621 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
627 /* Move stack pages down in memory. */
629 ret = shift_arg_pages(vma, stack_shift);
634 /* mprotect_fixup is overkill to remove the temporary stack flags */
635 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
637 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
638 stack_size = vma->vm_end - vma->vm_start;
640 * Align this down to a page boundary as expand_stack
643 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
644 #ifdef CONFIG_STACK_GROWSUP
645 if (stack_size + stack_expand > rlim_stack)
646 stack_base = vma->vm_start + rlim_stack;
648 stack_base = vma->vm_end + stack_expand;
650 if (stack_size + stack_expand > rlim_stack)
651 stack_base = vma->vm_end - rlim_stack;
653 stack_base = vma->vm_start - stack_expand;
655 ret = expand_stack(vma, stack_base);
660 up_write(&mm->mmap_sem);
663 EXPORT_SYMBOL(setup_arg_pages);
665 #endif /* CONFIG_MMU */
667 struct file *open_exec(const char *name)
672 file = do_filp_open(AT_FDCWD, name,
673 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
674 MAY_EXEC | MAY_OPEN);
679 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
682 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
685 fsnotify_open(file->f_path.dentry);
687 err = deny_write_access(file);
698 EXPORT_SYMBOL(open_exec);
700 int kernel_read(struct file *file, loff_t offset,
701 char *addr, unsigned long count)
709 /* The cast to a user pointer is valid due to the set_fs() */
710 result = vfs_read(file, (void __user *)addr, count, &pos);
715 EXPORT_SYMBOL(kernel_read);
717 static int exec_mmap(struct mm_struct *mm)
719 struct task_struct *tsk;
720 struct mm_struct * old_mm, *active_mm;
722 /* Notify parent that we're no longer interested in the old VM */
724 old_mm = current->mm;
725 sync_mm_rss(tsk, old_mm);
726 mm_release(tsk, old_mm);
730 * Make sure that if there is a core dump in progress
731 * for the old mm, we get out and die instead of going
732 * through with the exec. We must hold mmap_sem around
733 * checking core_state and changing tsk->mm.
735 down_read(&old_mm->mmap_sem);
736 if (unlikely(old_mm->core_state)) {
737 up_read(&old_mm->mmap_sem);
742 active_mm = tsk->active_mm;
745 activate_mm(active_mm, mm);
747 arch_pick_mmap_layout(mm);
749 up_read(&old_mm->mmap_sem);
750 BUG_ON(active_mm != old_mm);
751 mm_update_next_owner(old_mm);
760 * This function makes sure the current process has its own signal table,
761 * so that flush_signal_handlers can later reset the handlers without
762 * disturbing other processes. (Other processes might share the signal
763 * table via the CLONE_SIGHAND option to clone().)
765 static int de_thread(struct task_struct *tsk)
767 struct signal_struct *sig = tsk->signal;
768 struct sighand_struct *oldsighand = tsk->sighand;
769 spinlock_t *lock = &oldsighand->siglock;
771 if (thread_group_empty(tsk))
772 goto no_thread_group;
775 * Kill all other threads in the thread group.
778 if (signal_group_exit(sig)) {
780 * Another group action in progress, just
781 * return so that the signal is processed.
783 spin_unlock_irq(lock);
787 sig->group_exit_task = tsk;
788 sig->notify_count = zap_other_threads(tsk);
789 if (!thread_group_leader(tsk))
792 while (sig->notify_count) {
793 __set_current_state(TASK_UNINTERRUPTIBLE);
794 spin_unlock_irq(lock);
798 spin_unlock_irq(lock);
801 * At this point all other threads have exited, all we have to
802 * do is to wait for the thread group leader to become inactive,
803 * and to assume its PID:
805 if (!thread_group_leader(tsk)) {
806 struct task_struct *leader = tsk->group_leader;
808 sig->notify_count = -1; /* for exit_notify() */
810 write_lock_irq(&tasklist_lock);
811 if (likely(leader->exit_state))
813 __set_current_state(TASK_UNINTERRUPTIBLE);
814 write_unlock_irq(&tasklist_lock);
819 * The only record we have of the real-time age of a
820 * process, regardless of execs it's done, is start_time.
821 * All the past CPU time is accumulated in signal_struct
822 * from sister threads now dead. But in this non-leader
823 * exec, nothing survives from the original leader thread,
824 * whose birth marks the true age of this process now.
825 * When we take on its identity by switching to its PID, we
826 * also take its birthdate (always earlier than our own).
828 tsk->start_time = leader->start_time;
830 BUG_ON(!same_thread_group(leader, tsk));
831 BUG_ON(has_group_leader_pid(tsk));
833 * An exec() starts a new thread group with the
834 * TGID of the previous thread group. Rehash the
835 * two threads with a switched PID, and release
836 * the former thread group leader:
839 /* Become a process group leader with the old leader's pid.
840 * The old leader becomes a thread of the this thread group.
841 * Note: The old leader also uses this pid until release_task
842 * is called. Odd but simple and correct.
844 detach_pid(tsk, PIDTYPE_PID);
845 tsk->pid = leader->pid;
846 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
847 transfer_pid(leader, tsk, PIDTYPE_PGID);
848 transfer_pid(leader, tsk, PIDTYPE_SID);
850 list_replace_rcu(&leader->tasks, &tsk->tasks);
851 list_replace_init(&leader->sibling, &tsk->sibling);
853 tsk->group_leader = tsk;
854 leader->group_leader = tsk;
856 tsk->exit_signal = SIGCHLD;
858 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
859 leader->exit_state = EXIT_DEAD;
860 write_unlock_irq(&tasklist_lock);
862 release_task(leader);
865 sig->group_exit_task = NULL;
866 sig->notify_count = 0;
870 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
873 flush_itimer_signals();
875 if (atomic_read(&oldsighand->count) != 1) {
876 struct sighand_struct *newsighand;
878 * This ->sighand is shared with the CLONE_SIGHAND
879 * but not CLONE_THREAD task, switch to the new one.
881 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
885 atomic_set(&newsighand->count, 1);
886 memcpy(newsighand->action, oldsighand->action,
887 sizeof(newsighand->action));
889 write_lock_irq(&tasklist_lock);
890 spin_lock(&oldsighand->siglock);
891 rcu_assign_pointer(tsk->sighand, newsighand);
892 spin_unlock(&oldsighand->siglock);
893 write_unlock_irq(&tasklist_lock);
895 __cleanup_sighand(oldsighand);
898 BUG_ON(!thread_group_leader(tsk));
903 * These functions flushes out all traces of the currently running executable
904 * so that a new one can be started
906 static void flush_old_files(struct files_struct * files)
911 spin_lock(&files->file_lock);
913 unsigned long set, i;
917 fdt = files_fdtable(files);
918 if (i >= fdt->max_fds)
920 set = fdt->close_on_exec->fds_bits[j];
923 fdt->close_on_exec->fds_bits[j] = 0;
924 spin_unlock(&files->file_lock);
925 for ( ; set ; i++,set >>= 1) {
930 spin_lock(&files->file_lock);
933 spin_unlock(&files->file_lock);
936 char *get_task_comm(char *buf, struct task_struct *tsk)
938 /* buf must be at least sizeof(tsk->comm) in size */
940 strncpy(buf, tsk->comm, sizeof(tsk->comm));
945 void set_task_comm(struct task_struct *tsk, char *buf)
950 * Threads may access current->comm without holding
951 * the task lock, so write the string carefully.
952 * Readers without a lock may see incomplete new
953 * names but are safe from non-terminating string reads.
955 memset(tsk->comm, 0, TASK_COMM_LEN);
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);
1249 /* kernel module loader fixup */
1250 /* so we don't try to load run modprobe in kernel space. */
1253 retval = audit_bprm(bprm);
1258 for (try=0; try<2; try++) {
1259 read_lock(&binfmt_lock);
1260 list_for_each_entry(fmt, &formats, lh) {
1261 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1264 if (!try_module_get(fmt->module))
1266 read_unlock(&binfmt_lock);
1267 retval = fn(bprm, regs);
1269 * Restore the depth counter to its starting value
1270 * in this call, so we don't have to rely on every
1271 * load_binary function to restore it on return.
1273 bprm->recursion_depth = depth;
1276 tracehook_report_exec(fmt, bprm, regs);
1278 allow_write_access(bprm->file);
1282 current->did_exec = 1;
1283 proc_exec_connector(current);
1286 read_lock(&binfmt_lock);
1288 if (retval != -ENOEXEC || bprm->mm == NULL)
1291 read_unlock(&binfmt_lock);
1295 read_unlock(&binfmt_lock);
1296 if (retval != -ENOEXEC || bprm->mm == NULL) {
1298 #ifdef CONFIG_MODULES
1300 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1301 if (printable(bprm->buf[0]) &&
1302 printable(bprm->buf[1]) &&
1303 printable(bprm->buf[2]) &&
1304 printable(bprm->buf[3]))
1305 break; /* -ENOEXEC */
1306 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1313 EXPORT_SYMBOL(search_binary_handler);
1316 * sys_execve() executes a new program.
1318 int do_execve(char * filename,
1319 char __user *__user *argv,
1320 char __user *__user *envp,
1321 struct pt_regs * regs)
1323 struct linux_binprm *bprm;
1325 struct files_struct *displaced;
1329 retval = unshare_files(&displaced);
1334 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1338 retval = prepare_bprm_creds(bprm);
1342 retval = check_unsafe_exec(bprm);
1345 clear_in_exec = retval;
1346 current->in_execve = 1;
1348 file = open_exec(filename);
1349 retval = PTR_ERR(file);
1356 bprm->filename = filename;
1357 bprm->interp = filename;
1359 retval = bprm_mm_init(bprm);
1363 bprm->argc = count(argv, MAX_ARG_STRINGS);
1364 if ((retval = bprm->argc) < 0)
1367 bprm->envc = count(envp, MAX_ARG_STRINGS);
1368 if ((retval = bprm->envc) < 0)
1371 retval = prepare_binprm(bprm);
1375 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1379 bprm->exec = bprm->p;
1380 retval = copy_strings(bprm->envc, envp, bprm);
1384 retval = copy_strings(bprm->argc, argv, bprm);
1388 current->flags &= ~PF_KTHREAD;
1389 retval = search_binary_handler(bprm,regs);
1393 /* execve succeeded */
1394 current->fs->in_exec = 0;
1395 current->in_execve = 0;
1396 acct_update_integrals(current);
1399 put_files_struct(displaced);
1408 allow_write_access(bprm->file);
1414 current->fs->in_exec = 0;
1415 current->in_execve = 0;
1422 reset_files_struct(displaced);
1427 void set_binfmt(struct linux_binfmt *new)
1429 struct mm_struct *mm = current->mm;
1432 module_put(mm->binfmt->module);
1436 __module_get(new->module);
1439 EXPORT_SYMBOL(set_binfmt);
1441 /* format_corename will inspect the pattern parameter, and output a
1442 * name into corename, which must have space for at least
1443 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1445 static int format_corename(char *corename, long signr)
1447 const struct cred *cred = current_cred();
1448 const char *pat_ptr = core_pattern;
1449 int ispipe = (*pat_ptr == '|');
1450 char *out_ptr = corename;
1451 char *const out_end = corename + CORENAME_MAX_SIZE;
1453 int pid_in_pattern = 0;
1455 /* Repeat as long as we have more pattern to process and more output
1458 if (*pat_ptr != '%') {
1459 if (out_ptr == out_end)
1461 *out_ptr++ = *pat_ptr++;
1463 switch (*++pat_ptr) {
1466 /* Double percent, output one percent */
1468 if (out_ptr == out_end)
1475 rc = snprintf(out_ptr, out_end - out_ptr,
1476 "%d", task_tgid_vnr(current));
1477 if (rc > out_end - out_ptr)
1483 rc = snprintf(out_ptr, out_end - out_ptr,
1485 if (rc > out_end - out_ptr)
1491 rc = snprintf(out_ptr, out_end - out_ptr,
1493 if (rc > out_end - out_ptr)
1497 /* signal that caused the coredump */
1499 rc = snprintf(out_ptr, out_end - out_ptr,
1501 if (rc > out_end - out_ptr)
1505 /* UNIX time of coredump */
1508 do_gettimeofday(&tv);
1509 rc = snprintf(out_ptr, out_end - out_ptr,
1511 if (rc > out_end - out_ptr)
1518 down_read(&uts_sem);
1519 rc = snprintf(out_ptr, out_end - out_ptr,
1520 "%s", utsname()->nodename);
1522 if (rc > out_end - out_ptr)
1528 rc = snprintf(out_ptr, out_end - out_ptr,
1529 "%s", current->comm);
1530 if (rc > out_end - out_ptr)
1534 /* core limit size */
1536 rc = snprintf(out_ptr, out_end - out_ptr,
1537 "%lu", rlimit(RLIMIT_CORE));
1538 if (rc > out_end - out_ptr)
1548 /* Backward compatibility with core_uses_pid:
1550 * If core_pattern does not include a %p (as is the default)
1551 * and core_uses_pid is set, then .%pid will be appended to
1552 * the filename. Do not do this for piped commands. */
1553 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1554 rc = snprintf(out_ptr, out_end - out_ptr,
1555 ".%d", task_tgid_vnr(current));
1556 if (rc > out_end - out_ptr)
1565 static int zap_process(struct task_struct *start, int exit_code)
1567 struct task_struct *t;
1570 start->signal->flags = SIGNAL_GROUP_EXIT;
1571 start->signal->group_exit_code = exit_code;
1572 start->signal->group_stop_count = 0;
1576 if (t != current && t->mm) {
1577 sigaddset(&t->pending.signal, SIGKILL);
1578 signal_wake_up(t, 1);
1581 } while_each_thread(start, t);
1586 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1587 struct core_state *core_state, int exit_code)
1589 struct task_struct *g, *p;
1590 unsigned long flags;
1593 spin_lock_irq(&tsk->sighand->siglock);
1594 if (!signal_group_exit(tsk->signal)) {
1595 mm->core_state = core_state;
1596 nr = zap_process(tsk, exit_code);
1598 spin_unlock_irq(&tsk->sighand->siglock);
1599 if (unlikely(nr < 0))
1602 if (atomic_read(&mm->mm_users) == nr + 1)
1605 * We should find and kill all tasks which use this mm, and we should
1606 * count them correctly into ->nr_threads. We don't take tasklist
1607 * lock, but this is safe wrt:
1610 * None of sub-threads can fork after zap_process(leader). All
1611 * processes which were created before this point should be
1612 * visible to zap_threads() because copy_process() adds the new
1613 * process to the tail of init_task.tasks list, and lock/unlock
1614 * of ->siglock provides a memory barrier.
1617 * The caller holds mm->mmap_sem. This means that the task which
1618 * uses this mm can't pass exit_mm(), so it can't exit or clear
1622 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1623 * we must see either old or new leader, this does not matter.
1624 * However, it can change p->sighand, so lock_task_sighand(p)
1625 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1628 * Note also that "g" can be the old leader with ->mm == NULL
1629 * and already unhashed and thus removed from ->thread_group.
1630 * This is OK, __unhash_process()->list_del_rcu() does not
1631 * clear the ->next pointer, we will find the new leader via
1635 for_each_process(g) {
1636 if (g == tsk->group_leader)
1638 if (g->flags & PF_KTHREAD)
1643 if (unlikely(p->mm == mm)) {
1644 lock_task_sighand(p, &flags);
1645 nr += zap_process(p, exit_code);
1646 unlock_task_sighand(p, &flags);
1650 } while_each_thread(g, p);
1654 atomic_set(&core_state->nr_threads, nr);
1658 static int coredump_wait(int exit_code, struct core_state *core_state)
1660 struct task_struct *tsk = current;
1661 struct mm_struct *mm = tsk->mm;
1662 struct completion *vfork_done;
1663 int core_waiters = -EBUSY;
1665 init_completion(&core_state->startup);
1666 core_state->dumper.task = tsk;
1667 core_state->dumper.next = NULL;
1669 down_write(&mm->mmap_sem);
1670 if (!mm->core_state)
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 static int __get_dumpable(unsigned long mm_flags)
1759 ret = mm_flags & MMF_DUMPABLE_MASK;
1760 return (ret >= 2) ? 2 : ret;
1763 int get_dumpable(struct mm_struct *mm)
1765 return __get_dumpable(mm->flags);
1768 static void wait_for_dump_helpers(struct file *file)
1770 struct pipe_inode_info *pipe;
1772 pipe = file->f_path.dentry->d_inode->i_pipe;
1778 while ((pipe->readers > 1) && (!signal_pending(current))) {
1779 wake_up_interruptible_sync(&pipe->wait);
1780 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1793 * helper function to customize the process used
1794 * to collect the core in userspace. Specifically
1795 * it sets up a pipe and installs it as fd 0 (stdin)
1796 * for the process. Returns 0 on success, or
1797 * PTR_ERR on failure.
1798 * Note that it also sets the core limit to 1. This
1799 * is a special value that we use to trap recursive
1802 static int umh_pipe_setup(struct subprocess_info *info)
1804 struct file *rp, *wp;
1805 struct fdtable *fdt;
1806 struct coredump_params *cp = (struct coredump_params *)info->data;
1807 struct files_struct *cf = current->files;
1809 wp = create_write_pipe(0);
1813 rp = create_read_pipe(wp, 0);
1815 free_write_pipe(wp);
1823 spin_lock(&cf->file_lock);
1824 fdt = files_fdtable(cf);
1825 FD_SET(0, fdt->open_fds);
1826 FD_CLR(0, fdt->close_on_exec);
1827 spin_unlock(&cf->file_lock);
1829 /* and disallow core files too */
1830 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1835 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1837 struct core_state core_state;
1838 char corename[CORENAME_MAX_SIZE + 1];
1839 struct mm_struct *mm = current->mm;
1840 struct linux_binfmt * binfmt;
1841 const struct cred *old_cred;
1846 static atomic_t core_dump_count = ATOMIC_INIT(0);
1847 struct coredump_params cprm = {
1850 .limit = rlimit(RLIMIT_CORE),
1852 * We must use the same mm->flags while dumping core to avoid
1853 * inconsistency of bit flags, since this flag is not protected
1856 .mm_flags = mm->flags,
1859 audit_core_dumps(signr);
1861 binfmt = mm->binfmt;
1862 if (!binfmt || !binfmt->core_dump)
1864 if (!__get_dumpable(cprm.mm_flags))
1867 cred = prepare_creds();
1871 * We cannot trust fsuid as being the "true" uid of the
1872 * process nor do we know its entire history. We only know it
1873 * was tainted so we dump it as root in mode 2.
1875 if (__get_dumpable(cprm.mm_flags) == 2) {
1876 /* Setuid core dump mode */
1877 flag = O_EXCL; /* Stop rewrite attacks */
1878 cred->fsuid = 0; /* Dump root private */
1881 retval = coredump_wait(exit_code, &core_state);
1885 old_cred = override_creds(cred);
1888 * Clear any false indication of pending signals that might
1889 * be seen by the filesystem code called to write the core file.
1891 clear_thread_flag(TIF_SIGPENDING);
1893 ispipe = format_corename(corename, signr);
1899 if (cprm.limit == 1) {
1901 * Normally core limits are irrelevant to pipes, since
1902 * we're not writing to the file system, but we use
1903 * cprm.limit of 1 here as a speacial value. Any
1904 * non-1 limit gets set to RLIM_INFINITY below, but
1905 * a limit of 0 skips the dump. This is a consistent
1906 * way to catch recursive crashes. We can still crash
1907 * if the core_pattern binary sets RLIM_CORE = !1
1908 * but it runs as root, and can do lots of stupid things
1909 * Note that we use task_tgid_vnr here to grab the pid
1910 * of the process group leader. That way we get the
1911 * right pid if a thread in a multi-threaded
1912 * core_pattern process dies.
1915 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1916 task_tgid_vnr(current), current->comm);
1917 printk(KERN_WARNING "Aborting core\n");
1920 cprm.limit = RLIM_INFINITY;
1922 dump_count = atomic_inc_return(&core_dump_count);
1923 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1924 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1925 task_tgid_vnr(current), current->comm);
1926 printk(KERN_WARNING "Skipping core dump\n");
1927 goto fail_dropcount;
1930 helper_argv = argv_split(GFP_KERNEL, corename+1, NULL);
1932 printk(KERN_WARNING "%s failed to allocate memory\n",
1934 goto fail_dropcount;
1937 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1938 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1940 argv_free(helper_argv);
1942 printk(KERN_INFO "Core dump to %s pipe failed\n",
1947 struct inode *inode;
1949 if (cprm.limit < binfmt->min_coredump)
1952 cprm.file = filp_open(corename,
1953 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1955 if (IS_ERR(cprm.file))
1958 inode = cprm.file->f_path.dentry->d_inode;
1959 if (inode->i_nlink > 1)
1961 if (d_unhashed(cprm.file->f_path.dentry))
1964 * AK: actually i see no reason to not allow this for named
1965 * pipes etc, but keep the previous behaviour for now.
1967 if (!S_ISREG(inode->i_mode))
1970 * Dont allow local users get cute and trick others to coredump
1971 * into their pre-created files.
1973 if (inode->i_uid != current_fsuid())
1975 if (!cprm.file->f_op || !cprm.file->f_op->write)
1977 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
1981 retval = binfmt->core_dump(&cprm);
1983 current->signal->group_exit_code |= 0x80;
1985 if (ispipe && core_pipe_limit)
1986 wait_for_dump_helpers(cprm.file);
1989 filp_close(cprm.file, NULL);
1992 atomic_dec(&core_dump_count);
1994 coredump_finish(mm);
1995 revert_creds(old_cred);