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))
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 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
523 * move the page tables downwards, on failure we rely on
524 * process cleanup to remove whatever mess we made.
526 if (length != move_page_tables(vma, old_start,
527 vma, new_start, length))
531 tlb = tlb_gather_mmu(mm, 0);
532 if (new_end > old_start) {
534 * when the old and new regions overlap clear from new_end.
536 free_pgd_range(tlb, new_end, old_end, new_end,
537 vma->vm_next ? vma->vm_next->vm_start : 0);
540 * otherwise, clean from old_start; this is done to not touch
541 * the address space in [new_end, old_start) some architectures
542 * have constraints on va-space that make this illegal (IA64) -
543 * for the others its just a little faster.
545 free_pgd_range(tlb, old_start, old_end, new_end,
546 vma->vm_next ? vma->vm_next->vm_start : 0);
548 tlb_finish_mmu(tlb, new_end, old_end);
551 * shrink the vma to just the new range.
553 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
558 #define EXTRA_STACK_VM_PAGES 20 /* random */
561 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
562 * the stack is optionally relocated, and some extra space is added.
564 int setup_arg_pages(struct linux_binprm *bprm,
565 unsigned long stack_top,
566 int executable_stack)
569 unsigned long stack_shift;
570 struct mm_struct *mm = current->mm;
571 struct vm_area_struct *vma = bprm->vma;
572 struct vm_area_struct *prev = NULL;
573 unsigned long vm_flags;
574 unsigned long stack_base;
575 unsigned long stack_size;
576 unsigned long stack_expand;
577 unsigned long rlim_stack;
579 #ifdef CONFIG_STACK_GROWSUP
580 /* Limit stack size to 1GB */
581 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
582 if (stack_base > (1 << 30))
583 stack_base = 1 << 30;
585 /* Make sure we didn't let the argument array grow too large. */
586 if (vma->vm_end - vma->vm_start > stack_base)
589 stack_base = PAGE_ALIGN(stack_top - stack_base);
591 stack_shift = vma->vm_start - stack_base;
592 mm->arg_start = bprm->p - stack_shift;
593 bprm->p = vma->vm_end - stack_shift;
595 stack_top = arch_align_stack(stack_top);
596 stack_top = PAGE_ALIGN(stack_top);
597 stack_shift = vma->vm_end - stack_top;
599 bprm->p -= stack_shift;
600 mm->arg_start = bprm->p;
604 bprm->loader -= stack_shift;
605 bprm->exec -= stack_shift;
607 down_write(&mm->mmap_sem);
608 vm_flags = VM_STACK_FLAGS;
611 * Adjust stack execute permissions; explicitly enable for
612 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
613 * (arch default) otherwise.
615 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
617 else if (executable_stack == EXSTACK_DISABLE_X)
618 vm_flags &= ~VM_EXEC;
619 vm_flags |= mm->def_flags;
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 stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
635 stack_size = vma->vm_end - vma->vm_start;
637 * Align this down to a page boundary as expand_stack
640 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
641 #ifdef CONFIG_STACK_GROWSUP
642 if (stack_size + stack_expand > rlim_stack)
643 stack_base = vma->vm_start + rlim_stack;
645 stack_base = vma->vm_end + stack_expand;
647 if (stack_size + stack_expand > rlim_stack)
648 stack_base = vma->vm_end - rlim_stack;
650 stack_base = vma->vm_start - stack_expand;
652 ret = expand_stack(vma, stack_base);
657 up_write(&mm->mmap_sem);
660 EXPORT_SYMBOL(setup_arg_pages);
662 #endif /* CONFIG_MMU */
664 struct file *open_exec(const char *name)
669 file = do_filp_open(AT_FDCWD, name,
670 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
671 MAY_EXEC | MAY_OPEN);
676 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
679 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
682 fsnotify_open(file->f_path.dentry);
684 err = deny_write_access(file);
695 EXPORT_SYMBOL(open_exec);
697 int kernel_read(struct file *file, loff_t offset,
698 char *addr, unsigned long count)
706 /* The cast to a user pointer is valid due to the set_fs() */
707 result = vfs_read(file, (void __user *)addr, count, &pos);
712 EXPORT_SYMBOL(kernel_read);
714 static int exec_mmap(struct mm_struct *mm)
716 struct task_struct *tsk;
717 struct mm_struct * old_mm, *active_mm;
719 /* Notify parent that we're no longer interested in the old VM */
721 old_mm = current->mm;
722 mm_release(tsk, old_mm);
726 * Make sure that if there is a core dump in progress
727 * for the old mm, we get out and die instead of going
728 * through with the exec. We must hold mmap_sem around
729 * checking core_state and changing tsk->mm.
731 down_read(&old_mm->mmap_sem);
732 if (unlikely(old_mm->core_state)) {
733 up_read(&old_mm->mmap_sem);
738 active_mm = tsk->active_mm;
741 activate_mm(active_mm, mm);
743 arch_pick_mmap_layout(mm);
745 up_read(&old_mm->mmap_sem);
746 BUG_ON(active_mm != old_mm);
747 mm_update_next_owner(old_mm);
756 * This function makes sure the current process has its own signal table,
757 * so that flush_signal_handlers can later reset the handlers without
758 * disturbing other processes. (Other processes might share the signal
759 * table via the CLONE_SIGHAND option to clone().)
761 static int de_thread(struct task_struct *tsk)
763 struct signal_struct *sig = tsk->signal;
764 struct sighand_struct *oldsighand = tsk->sighand;
765 spinlock_t *lock = &oldsighand->siglock;
768 if (thread_group_empty(tsk))
769 goto no_thread_group;
772 * Kill all other threads in the thread group.
775 if (signal_group_exit(sig)) {
777 * Another group action in progress, just
778 * return so that the signal is processed.
780 spin_unlock_irq(lock);
783 sig->group_exit_task = tsk;
784 zap_other_threads(tsk);
786 /* Account for the thread group leader hanging around: */
787 count = thread_group_leader(tsk) ? 1 : 2;
788 sig->notify_count = count;
789 while (atomic_read(&sig->count) > count) {
790 __set_current_state(TASK_UNINTERRUPTIBLE);
791 spin_unlock_irq(lock);
795 spin_unlock_irq(lock);
798 * At this point all other threads have exited, all we have to
799 * do is to wait for the thread group leader to become inactive,
800 * and to assume its PID:
802 if (!thread_group_leader(tsk)) {
803 struct task_struct *leader = tsk->group_leader;
805 sig->notify_count = -1; /* for exit_notify() */
807 write_lock_irq(&tasklist_lock);
808 if (likely(leader->exit_state))
810 __set_current_state(TASK_UNINTERRUPTIBLE);
811 write_unlock_irq(&tasklist_lock);
816 * The only record we have of the real-time age of a
817 * process, regardless of execs it's done, is start_time.
818 * All the past CPU time is accumulated in signal_struct
819 * from sister threads now dead. But in this non-leader
820 * exec, nothing survives from the original leader thread,
821 * whose birth marks the true age of this process now.
822 * When we take on its identity by switching to its PID, we
823 * also take its birthdate (always earlier than our own).
825 tsk->start_time = leader->start_time;
827 BUG_ON(!same_thread_group(leader, tsk));
828 BUG_ON(has_group_leader_pid(tsk));
830 * An exec() starts a new thread group with the
831 * TGID of the previous thread group. Rehash the
832 * two threads with a switched PID, and release
833 * the former thread group leader:
836 /* Become a process group leader with the old leader's pid.
837 * The old leader becomes a thread of the this thread group.
838 * Note: The old leader also uses this pid until release_task
839 * is called. Odd but simple and correct.
841 detach_pid(tsk, PIDTYPE_PID);
842 tsk->pid = leader->pid;
843 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
844 transfer_pid(leader, tsk, PIDTYPE_PGID);
845 transfer_pid(leader, tsk, PIDTYPE_SID);
846 list_replace_rcu(&leader->tasks, &tsk->tasks);
848 tsk->group_leader = tsk;
849 leader->group_leader = tsk;
851 tsk->exit_signal = SIGCHLD;
853 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
854 leader->exit_state = EXIT_DEAD;
855 write_unlock_irq(&tasklist_lock);
857 release_task(leader);
860 sig->group_exit_task = NULL;
861 sig->notify_count = 0;
865 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
868 flush_itimer_signals();
870 if (atomic_read(&oldsighand->count) != 1) {
871 struct sighand_struct *newsighand;
873 * This ->sighand is shared with the CLONE_SIGHAND
874 * but not CLONE_THREAD task, switch to the new one.
876 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
880 atomic_set(&newsighand->count, 1);
881 memcpy(newsighand->action, oldsighand->action,
882 sizeof(newsighand->action));
884 write_lock_irq(&tasklist_lock);
885 spin_lock(&oldsighand->siglock);
886 rcu_assign_pointer(tsk->sighand, newsighand);
887 spin_unlock(&oldsighand->siglock);
888 write_unlock_irq(&tasklist_lock);
890 __cleanup_sighand(oldsighand);
893 BUG_ON(!thread_group_leader(tsk));
898 * These functions flushes out all traces of the currently running executable
899 * so that a new one can be started
901 static void flush_old_files(struct files_struct * files)
906 spin_lock(&files->file_lock);
908 unsigned long set, i;
912 fdt = files_fdtable(files);
913 if (i >= fdt->max_fds)
915 set = fdt->close_on_exec->fds_bits[j];
918 fdt->close_on_exec->fds_bits[j] = 0;
919 spin_unlock(&files->file_lock);
920 for ( ; set ; i++,set >>= 1) {
925 spin_lock(&files->file_lock);
928 spin_unlock(&files->file_lock);
931 char *get_task_comm(char *buf, struct task_struct *tsk)
933 /* buf must be at least sizeof(tsk->comm) in size */
935 strncpy(buf, tsk->comm, sizeof(tsk->comm));
940 void set_task_comm(struct task_struct *tsk, char *buf)
943 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
945 perf_event_comm(tsk);
948 int flush_old_exec(struct linux_binprm * bprm)
953 * Make sure we have a private signal table and that
954 * we are unassociated from the previous thread group.
956 retval = de_thread(current);
960 set_mm_exe_file(bprm->mm, bprm->file);
963 * Release all of the old mmap stuff
965 retval = exec_mmap(bprm->mm);
969 bprm->mm = NULL; /* We're using it now */
971 current->flags &= ~PF_RANDOMIZE;
973 current->personality &= ~bprm->per_clear;
980 EXPORT_SYMBOL(flush_old_exec);
982 void setup_new_exec(struct linux_binprm * bprm)
986 char tcomm[sizeof(current->comm)];
988 arch_pick_mmap_layout(current->mm);
990 /* This is the point of no return */
991 current->sas_ss_sp = current->sas_ss_size = 0;
993 if (current_euid() == current_uid() && current_egid() == current_gid())
994 set_dumpable(current->mm, 1);
996 set_dumpable(current->mm, suid_dumpable);
998 name = bprm->filename;
1000 /* Copies the binary name from after last slash */
1001 for (i=0; (ch = *(name++)) != '\0';) {
1003 i = 0; /* overwrite what we wrote */
1005 if (i < (sizeof(tcomm) - 1))
1009 set_task_comm(current, tcomm);
1011 /* Set the new mm task size. We have to do that late because it may
1012 * depend on TIF_32BIT which is only updated in flush_thread() on
1013 * some architectures like powerpc
1015 current->mm->task_size = TASK_SIZE;
1017 /* install the new credentials */
1018 if (bprm->cred->uid != current_euid() ||
1019 bprm->cred->gid != current_egid()) {
1020 current->pdeath_signal = 0;
1021 } else if (file_permission(bprm->file, MAY_READ) ||
1022 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1023 set_dumpable(current->mm, suid_dumpable);
1027 * Flush performance counters when crossing a
1030 if (!get_dumpable(current->mm))
1031 perf_event_exit_task(current);
1033 /* An exec changes our domain. We are no longer part of the thread
1036 current->self_exec_id++;
1038 flush_signal_handlers(current, 0);
1039 flush_old_files(current->files);
1041 EXPORT_SYMBOL(setup_new_exec);
1044 * Prepare credentials and lock ->cred_guard_mutex.
1045 * install_exec_creds() commits the new creds and drops the lock.
1046 * Or, if exec fails before, free_bprm() should release ->cred and
1049 int prepare_bprm_creds(struct linux_binprm *bprm)
1051 if (mutex_lock_interruptible(¤t->cred_guard_mutex))
1052 return -ERESTARTNOINTR;
1054 bprm->cred = prepare_exec_creds();
1055 if (likely(bprm->cred))
1058 mutex_unlock(¤t->cred_guard_mutex);
1062 void free_bprm(struct linux_binprm *bprm)
1064 free_arg_pages(bprm);
1066 mutex_unlock(¤t->cred_guard_mutex);
1067 abort_creds(bprm->cred);
1073 * install the new credentials for this executable
1075 void install_exec_creds(struct linux_binprm *bprm)
1077 security_bprm_committing_creds(bprm);
1079 commit_creds(bprm->cred);
1082 * cred_guard_mutex must be held at least to this point to prevent
1083 * ptrace_attach() from altering our determination of the task's
1084 * credentials; any time after this it may be unlocked.
1086 security_bprm_committed_creds(bprm);
1087 mutex_unlock(¤t->cred_guard_mutex);
1089 EXPORT_SYMBOL(install_exec_creds);
1092 * determine how safe it is to execute the proposed program
1093 * - the caller must hold current->cred_guard_mutex to protect against
1096 int check_unsafe_exec(struct linux_binprm *bprm)
1098 struct task_struct *p = current, *t;
1102 bprm->unsafe = tracehook_unsafe_exec(p);
1105 write_lock(&p->fs->lock);
1107 for (t = next_thread(p); t != p; t = next_thread(t)) {
1113 if (p->fs->users > n_fs) {
1114 bprm->unsafe |= LSM_UNSAFE_SHARE;
1117 if (!p->fs->in_exec) {
1122 write_unlock(&p->fs->lock);
1128 * Fill the binprm structure from the inode.
1129 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1131 * This may be called multiple times for binary chains (scripts for example).
1133 int prepare_binprm(struct linux_binprm *bprm)
1136 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1139 mode = inode->i_mode;
1140 if (bprm->file->f_op == NULL)
1143 /* clear any previous set[ug]id data from a previous binary */
1144 bprm->cred->euid = current_euid();
1145 bprm->cred->egid = current_egid();
1147 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1149 if (mode & S_ISUID) {
1150 bprm->per_clear |= PER_CLEAR_ON_SETID;
1151 bprm->cred->euid = inode->i_uid;
1156 * If setgid is set but no group execute bit then this
1157 * is a candidate for mandatory locking, not a setgid
1160 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1161 bprm->per_clear |= PER_CLEAR_ON_SETID;
1162 bprm->cred->egid = inode->i_gid;
1166 /* fill in binprm security blob */
1167 retval = security_bprm_set_creds(bprm);
1170 bprm->cred_prepared = 1;
1172 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1173 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1176 EXPORT_SYMBOL(prepare_binprm);
1179 * Arguments are '\0' separated strings found at the location bprm->p
1180 * points to; chop off the first by relocating brpm->p to right after
1181 * the first '\0' encountered.
1183 int remove_arg_zero(struct linux_binprm *bprm)
1186 unsigned long offset;
1194 offset = bprm->p & ~PAGE_MASK;
1195 page = get_arg_page(bprm, bprm->p, 0);
1200 kaddr = kmap_atomic(page, KM_USER0);
1202 for (; offset < PAGE_SIZE && kaddr[offset];
1203 offset++, bprm->p++)
1206 kunmap_atomic(kaddr, KM_USER0);
1209 if (offset == PAGE_SIZE)
1210 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1211 } while (offset == PAGE_SIZE);
1220 EXPORT_SYMBOL(remove_arg_zero);
1223 * cycle the list of binary formats handler, until one recognizes the image
1225 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1227 unsigned int depth = bprm->recursion_depth;
1229 struct linux_binfmt *fmt;
1231 retval = security_bprm_check(bprm);
1234 retval = ima_bprm_check(bprm);
1238 /* kernel module loader fixup */
1239 /* so we don't try to load run modprobe in kernel space. */
1242 retval = audit_bprm(bprm);
1247 for (try=0; try<2; try++) {
1248 read_lock(&binfmt_lock);
1249 list_for_each_entry(fmt, &formats, lh) {
1250 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1253 if (!try_module_get(fmt->module))
1255 read_unlock(&binfmt_lock);
1256 retval = fn(bprm, regs);
1258 * Restore the depth counter to its starting value
1259 * in this call, so we don't have to rely on every
1260 * load_binary function to restore it on return.
1262 bprm->recursion_depth = depth;
1265 tracehook_report_exec(fmt, bprm, regs);
1267 allow_write_access(bprm->file);
1271 current->did_exec = 1;
1272 proc_exec_connector(current);
1275 read_lock(&binfmt_lock);
1277 if (retval != -ENOEXEC || bprm->mm == NULL)
1280 read_unlock(&binfmt_lock);
1284 read_unlock(&binfmt_lock);
1285 if (retval != -ENOEXEC || bprm->mm == NULL) {
1287 #ifdef CONFIG_MODULES
1289 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1290 if (printable(bprm->buf[0]) &&
1291 printable(bprm->buf[1]) &&
1292 printable(bprm->buf[2]) &&
1293 printable(bprm->buf[3]))
1294 break; /* -ENOEXEC */
1295 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1302 EXPORT_SYMBOL(search_binary_handler);
1305 * sys_execve() executes a new program.
1307 int do_execve(char * filename,
1308 char __user *__user *argv,
1309 char __user *__user *envp,
1310 struct pt_regs * regs)
1312 struct linux_binprm *bprm;
1314 struct files_struct *displaced;
1318 retval = unshare_files(&displaced);
1323 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1327 retval = prepare_bprm_creds(bprm);
1331 retval = check_unsafe_exec(bprm);
1334 clear_in_exec = retval;
1335 current->in_execve = 1;
1337 file = open_exec(filename);
1338 retval = PTR_ERR(file);
1345 bprm->filename = filename;
1346 bprm->interp = filename;
1348 retval = bprm_mm_init(bprm);
1352 bprm->argc = count(argv, MAX_ARG_STRINGS);
1353 if ((retval = bprm->argc) < 0)
1356 bprm->envc = count(envp, MAX_ARG_STRINGS);
1357 if ((retval = bprm->envc) < 0)
1360 retval = prepare_binprm(bprm);
1364 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1368 bprm->exec = bprm->p;
1369 retval = copy_strings(bprm->envc, envp, bprm);
1373 retval = copy_strings(bprm->argc, argv, bprm);
1377 current->flags &= ~PF_KTHREAD;
1378 retval = search_binary_handler(bprm,regs);
1382 current->stack_start = current->mm->start_stack;
1384 /* execve succeeded */
1385 current->fs->in_exec = 0;
1386 current->in_execve = 0;
1387 acct_update_integrals(current);
1390 put_files_struct(displaced);
1399 allow_write_access(bprm->file);
1405 current->fs->in_exec = 0;
1406 current->in_execve = 0;
1413 reset_files_struct(displaced);
1418 void set_binfmt(struct linux_binfmt *new)
1420 struct mm_struct *mm = current->mm;
1423 module_put(mm->binfmt->module);
1427 __module_get(new->module);
1430 EXPORT_SYMBOL(set_binfmt);
1432 /* format_corename will inspect the pattern parameter, and output a
1433 * name into corename, which must have space for at least
1434 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1436 static int format_corename(char *corename, long signr)
1438 const struct cred *cred = current_cred();
1439 const char *pat_ptr = core_pattern;
1440 int ispipe = (*pat_ptr == '|');
1441 char *out_ptr = corename;
1442 char *const out_end = corename + CORENAME_MAX_SIZE;
1444 int pid_in_pattern = 0;
1446 /* Repeat as long as we have more pattern to process and more output
1449 if (*pat_ptr != '%') {
1450 if (out_ptr == out_end)
1452 *out_ptr++ = *pat_ptr++;
1454 switch (*++pat_ptr) {
1457 /* Double percent, output one percent */
1459 if (out_ptr == out_end)
1466 rc = snprintf(out_ptr, out_end - out_ptr,
1467 "%d", task_tgid_vnr(current));
1468 if (rc > out_end - out_ptr)
1474 rc = snprintf(out_ptr, out_end - out_ptr,
1476 if (rc > out_end - out_ptr)
1482 rc = snprintf(out_ptr, out_end - out_ptr,
1484 if (rc > out_end - out_ptr)
1488 /* signal that caused the coredump */
1490 rc = snprintf(out_ptr, out_end - out_ptr,
1492 if (rc > out_end - out_ptr)
1496 /* UNIX time of coredump */
1499 do_gettimeofday(&tv);
1500 rc = snprintf(out_ptr, out_end - out_ptr,
1502 if (rc > out_end - out_ptr)
1509 down_read(&uts_sem);
1510 rc = snprintf(out_ptr, out_end - out_ptr,
1511 "%s", utsname()->nodename);
1513 if (rc > out_end - out_ptr)
1519 rc = snprintf(out_ptr, out_end - out_ptr,
1520 "%s", current->comm);
1521 if (rc > out_end - out_ptr)
1525 /* core limit size */
1527 rc = snprintf(out_ptr, out_end - out_ptr,
1528 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1529 if (rc > out_end - out_ptr)
1539 /* Backward compatibility with core_uses_pid:
1541 * If core_pattern does not include a %p (as is the default)
1542 * and core_uses_pid is set, then .%pid will be appended to
1543 * the filename. Do not do this for piped commands. */
1544 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1545 rc = snprintf(out_ptr, out_end - out_ptr,
1546 ".%d", task_tgid_vnr(current));
1547 if (rc > out_end - out_ptr)
1556 static int zap_process(struct task_struct *start)
1558 struct task_struct *t;
1561 start->signal->flags = SIGNAL_GROUP_EXIT;
1562 start->signal->group_stop_count = 0;
1566 if (t != current && t->mm) {
1567 sigaddset(&t->pending.signal, SIGKILL);
1568 signal_wake_up(t, 1);
1571 } while_each_thread(start, t);
1576 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1577 struct core_state *core_state, int exit_code)
1579 struct task_struct *g, *p;
1580 unsigned long flags;
1583 spin_lock_irq(&tsk->sighand->siglock);
1584 if (!signal_group_exit(tsk->signal)) {
1585 mm->core_state = core_state;
1586 tsk->signal->group_exit_code = exit_code;
1587 nr = zap_process(tsk);
1589 spin_unlock_irq(&tsk->sighand->siglock);
1590 if (unlikely(nr < 0))
1593 if (atomic_read(&mm->mm_users) == nr + 1)
1596 * We should find and kill all tasks which use this mm, and we should
1597 * count them correctly into ->nr_threads. We don't take tasklist
1598 * lock, but this is safe wrt:
1601 * None of sub-threads can fork after zap_process(leader). All
1602 * processes which were created before this point should be
1603 * visible to zap_threads() because copy_process() adds the new
1604 * process to the tail of init_task.tasks list, and lock/unlock
1605 * of ->siglock provides a memory barrier.
1608 * The caller holds mm->mmap_sem. This means that the task which
1609 * uses this mm can't pass exit_mm(), so it can't exit or clear
1613 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1614 * we must see either old or new leader, this does not matter.
1615 * However, it can change p->sighand, so lock_task_sighand(p)
1616 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1619 * Note also that "g" can be the old leader with ->mm == NULL
1620 * and already unhashed and thus removed from ->thread_group.
1621 * This is OK, __unhash_process()->list_del_rcu() does not
1622 * clear the ->next pointer, we will find the new leader via
1626 for_each_process(g) {
1627 if (g == tsk->group_leader)
1629 if (g->flags & PF_KTHREAD)
1634 if (unlikely(p->mm == mm)) {
1635 lock_task_sighand(p, &flags);
1636 nr += zap_process(p);
1637 unlock_task_sighand(p, &flags);
1641 } while_each_thread(g, p);
1645 atomic_set(&core_state->nr_threads, nr);
1649 static int coredump_wait(int exit_code, struct core_state *core_state)
1651 struct task_struct *tsk = current;
1652 struct mm_struct *mm = tsk->mm;
1653 struct completion *vfork_done;
1656 init_completion(&core_state->startup);
1657 core_state->dumper.task = tsk;
1658 core_state->dumper.next = NULL;
1659 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1660 up_write(&mm->mmap_sem);
1662 if (unlikely(core_waiters < 0))
1666 * Make sure nobody is waiting for us to release the VM,
1667 * otherwise we can deadlock when we wait on each other
1669 vfork_done = tsk->vfork_done;
1671 tsk->vfork_done = NULL;
1672 complete(vfork_done);
1676 wait_for_completion(&core_state->startup);
1678 return core_waiters;
1681 static void coredump_finish(struct mm_struct *mm)
1683 struct core_thread *curr, *next;
1684 struct task_struct *task;
1686 next = mm->core_state->dumper.next;
1687 while ((curr = next) != NULL) {
1691 * see exit_mm(), curr->task must not see
1692 * ->task == NULL before we read ->next.
1696 wake_up_process(task);
1699 mm->core_state = NULL;
1703 * set_dumpable converts traditional three-value dumpable to two flags and
1704 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1705 * these bits are not changed atomically. So get_dumpable can observe the
1706 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1707 * return either old dumpable or new one by paying attention to the order of
1708 * modifying the bits.
1710 * dumpable | mm->flags (binary)
1711 * old new | initial interim final
1712 * ---------+-----------------------
1720 * (*) get_dumpable regards interim value of 10 as 11.
1722 void set_dumpable(struct mm_struct *mm, int value)
1726 clear_bit(MMF_DUMPABLE, &mm->flags);
1728 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1731 set_bit(MMF_DUMPABLE, &mm->flags);
1733 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1736 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1738 set_bit(MMF_DUMPABLE, &mm->flags);
1743 int get_dumpable(struct mm_struct *mm)
1747 ret = mm->flags & 0x3;
1748 return (ret >= 2) ? 2 : ret;
1751 static void wait_for_dump_helpers(struct file *file)
1753 struct pipe_inode_info *pipe;
1755 pipe = file->f_path.dentry->d_inode->i_pipe;
1761 while ((pipe->readers > 1) && (!signal_pending(current))) {
1762 wake_up_interruptible_sync(&pipe->wait);
1763 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1774 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1776 struct core_state core_state;
1777 char corename[CORENAME_MAX_SIZE + 1];
1778 struct mm_struct *mm = current->mm;
1779 struct linux_binfmt * binfmt;
1780 struct inode * inode;
1782 const struct cred *old_cred;
1787 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1788 char **helper_argv = NULL;
1789 int helper_argc = 0;
1791 static atomic_t core_dump_count = ATOMIC_INIT(0);
1793 audit_core_dumps(signr);
1795 binfmt = mm->binfmt;
1796 if (!binfmt || !binfmt->core_dump)
1799 cred = prepare_creds();
1805 down_write(&mm->mmap_sem);
1807 * If another thread got here first, or we are not dumpable, bail out.
1809 if (mm->core_state || !get_dumpable(mm)) {
1810 up_write(&mm->mmap_sem);
1816 * We cannot trust fsuid as being the "true" uid of the
1817 * process nor do we know its entire history. We only know it
1818 * was tainted so we dump it as root in mode 2.
1820 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1821 flag = O_EXCL; /* Stop rewrite attacks */
1822 cred->fsuid = 0; /* Dump root private */
1825 retval = coredump_wait(exit_code, &core_state);
1831 old_cred = override_creds(cred);
1834 * Clear any false indication of pending signals that might
1835 * be seen by the filesystem code called to write the core file.
1837 clear_thread_flag(TIF_SIGPENDING);
1840 * lock_kernel() because format_corename() is controlled by sysctl, which
1841 * uses lock_kernel()
1844 ispipe = format_corename(corename, signr);
1847 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1851 if (core_limit == 0) {
1853 * Normally core limits are irrelevant to pipes, since
1854 * we're not writing to the file system, but we use
1855 * core_limit of 0 here as a speacial value. Any
1856 * non-zero limit gets set to RLIM_INFINITY below, but
1857 * a limit of 0 skips the dump. This is a consistent
1858 * way to catch recursive crashes. We can still crash
1859 * if the core_pattern binary sets RLIM_CORE = !0
1860 * but it runs as root, and can do lots of stupid things
1861 * Note that we use task_tgid_vnr here to grab the pid
1862 * of the process group leader. That way we get the
1863 * right pid if a thread in a multi-threaded
1864 * core_pattern process dies.
1867 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1868 task_tgid_vnr(current), current->comm);
1869 printk(KERN_WARNING "Aborting core\n");
1873 dump_count = atomic_inc_return(&core_dump_count);
1874 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1875 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1876 task_tgid_vnr(current), current->comm);
1877 printk(KERN_WARNING "Skipping core dump\n");
1878 goto fail_dropcount;
1881 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1883 printk(KERN_WARNING "%s failed to allocate memory\n",
1885 goto fail_dropcount;
1888 core_limit = RLIM_INFINITY;
1890 /* SIGPIPE can happen, but it's just never processed */
1891 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1893 printk(KERN_INFO "Core dump to %s pipe failed\n",
1895 goto fail_dropcount;
1898 file = filp_open(corename,
1899 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1902 goto fail_dropcount;
1903 inode = file->f_path.dentry->d_inode;
1904 if (inode->i_nlink > 1)
1905 goto close_fail; /* multiple links - don't dump */
1906 if (!ispipe && d_unhashed(file->f_path.dentry))
1909 /* AK: actually i see no reason to not allow this for named pipes etc.,
1910 but keep the previous behaviour for now. */
1911 if (!ispipe && !S_ISREG(inode->i_mode))
1914 * Dont allow local users get cute and trick others to coredump
1915 * into their pre-created files:
1917 if (inode->i_uid != current_fsuid())
1921 if (!file->f_op->write)
1923 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1926 retval = binfmt->core_dump(signr, regs, file, core_limit);
1929 current->signal->group_exit_code |= 0x80;
1931 if (ispipe && core_pipe_limit)
1932 wait_for_dump_helpers(file);
1933 filp_close(file, NULL);
1936 atomic_dec(&core_dump_count);
1939 argv_free(helper_argv);
1941 revert_creds(old_cred);
1943 coredump_finish(mm);