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)
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(const char __user * const __user * argv, int max)
370 const char __user * p;
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, const char __user *const __user *argv,
391 struct linux_binprm *bprm)
393 struct page *kmapped_page = NULL;
395 unsigned long kpos = 0;
399 const char __user *str;
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;
424 offset = pos % PAGE_SIZE;
428 bytes_to_copy = offset;
429 if (bytes_to_copy > len)
432 offset -= bytes_to_copy;
433 pos -= bytes_to_copy;
434 str -= bytes_to_copy;
435 len -= bytes_to_copy;
437 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
440 page = get_arg_page(bprm, pos, 1);
447 flush_kernel_dcache_page(kmapped_page);
448 kunmap(kmapped_page);
449 put_arg_page(kmapped_page);
452 kaddr = kmap(kmapped_page);
453 kpos = pos & PAGE_MASK;
454 flush_arg_page(bprm, kpos, kmapped_page);
456 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
465 flush_kernel_dcache_page(kmapped_page);
466 kunmap(kmapped_page);
467 put_arg_page(kmapped_page);
473 * Like copy_strings, but get argv and its values from kernel memory.
475 int copy_strings_kernel(int argc, const char *const *argv,
476 struct linux_binprm *bprm)
479 mm_segment_t oldfs = get_fs();
481 r = copy_strings(argc, (const char __user *const __user *)argv, bprm);
485 EXPORT_SYMBOL(copy_strings_kernel);
490 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
491 * the binfmt code determines where the new stack should reside, we shift it to
492 * its final location. The process proceeds as follows:
494 * 1) Use shift to calculate the new vma endpoints.
495 * 2) Extend vma to cover both the old and new ranges. This ensures the
496 * arguments passed to subsequent functions are consistent.
497 * 3) Move vma's page tables to the new range.
498 * 4) Free up any cleared pgd range.
499 * 5) Shrink the vma to cover only the new range.
501 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
503 struct mm_struct *mm = vma->vm_mm;
504 unsigned long old_start = vma->vm_start;
505 unsigned long old_end = vma->vm_end;
506 unsigned long length = old_end - old_start;
507 unsigned long new_start = old_start - shift;
508 unsigned long new_end = old_end - shift;
509 struct mmu_gather *tlb;
511 BUG_ON(new_start > new_end);
514 * ensure there are no vmas between where we want to go
517 if (vma != find_vma(mm, new_start))
521 * cover the whole range: [new_start, old_end)
523 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
527 * move the page tables downwards, on failure we rely on
528 * process cleanup to remove whatever mess we made.
530 if (length != move_page_tables(vma, old_start,
531 vma, new_start, length))
535 tlb = tlb_gather_mmu(mm, 0);
536 if (new_end > old_start) {
538 * when the old and new regions overlap clear from new_end.
540 free_pgd_range(tlb, new_end, old_end, new_end,
541 vma->vm_next ? vma->vm_next->vm_start : 0);
544 * otherwise, clean from old_start; this is done to not touch
545 * the address space in [new_end, old_start) some architectures
546 * have constraints on va-space that make this illegal (IA64) -
547 * for the others its just a little faster.
549 free_pgd_range(tlb, old_start, old_end, new_end,
550 vma->vm_next ? vma->vm_next->vm_start : 0);
552 tlb_finish_mmu(tlb, new_end, old_end);
555 * Shrink the vma to just the new range. Always succeeds.
557 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
563 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
564 * the stack is optionally relocated, and some extra space is added.
566 int setup_arg_pages(struct linux_binprm *bprm,
567 unsigned long stack_top,
568 int executable_stack)
571 unsigned long stack_shift;
572 struct mm_struct *mm = current->mm;
573 struct vm_area_struct *vma = bprm->vma;
574 struct vm_area_struct *prev = NULL;
575 unsigned long vm_flags;
576 unsigned long stack_base;
577 unsigned long stack_size;
578 unsigned long stack_expand;
579 unsigned long rlim_stack;
581 #ifdef CONFIG_STACK_GROWSUP
582 /* Limit stack size to 1GB */
583 stack_base = rlimit_max(RLIMIT_STACK);
584 if (stack_base > (1 << 30))
585 stack_base = 1 << 30;
587 /* Make sure we didn't let the argument array grow too large. */
588 if (vma->vm_end - vma->vm_start > stack_base)
591 stack_base = PAGE_ALIGN(stack_top - stack_base);
593 stack_shift = vma->vm_start - stack_base;
594 mm->arg_start = bprm->p - stack_shift;
595 bprm->p = vma->vm_end - stack_shift;
597 stack_top = arch_align_stack(stack_top);
598 stack_top = PAGE_ALIGN(stack_top);
600 if (unlikely(stack_top < mmap_min_addr) ||
601 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
604 stack_shift = vma->vm_end - stack_top;
606 bprm->p -= stack_shift;
607 mm->arg_start = bprm->p;
611 bprm->loader -= stack_shift;
612 bprm->exec -= stack_shift;
614 down_write(&mm->mmap_sem);
615 vm_flags = VM_STACK_FLAGS;
618 * Adjust stack execute permissions; explicitly enable for
619 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
620 * (arch default) otherwise.
622 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
624 else if (executable_stack == EXSTACK_DISABLE_X)
625 vm_flags &= ~VM_EXEC;
626 vm_flags |= mm->def_flags;
627 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
629 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
635 /* Move stack pages down in memory. */
637 ret = shift_arg_pages(vma, stack_shift);
642 /* mprotect_fixup is overkill to remove the temporary stack flags */
643 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
645 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
646 stack_size = vma->vm_end - vma->vm_start;
648 * Align this down to a page boundary as expand_stack
651 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
652 #ifdef CONFIG_STACK_GROWSUP
653 if (stack_size + stack_expand > rlim_stack)
654 stack_base = vma->vm_start + rlim_stack;
656 stack_base = vma->vm_end + stack_expand;
658 if (stack_size + stack_expand > rlim_stack)
659 stack_base = vma->vm_end - rlim_stack;
661 stack_base = vma->vm_start - stack_expand;
663 current->mm->start_stack = bprm->p;
664 ret = expand_stack(vma, stack_base);
669 up_write(&mm->mmap_sem);
672 EXPORT_SYMBOL(setup_arg_pages);
674 #endif /* CONFIG_MMU */
676 struct file *open_exec(const char *name)
681 file = do_filp_open(AT_FDCWD, name,
682 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
683 MAY_EXEC | MAY_OPEN);
688 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
691 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
696 err = deny_write_access(file);
707 EXPORT_SYMBOL(open_exec);
709 int kernel_read(struct file *file, loff_t offset,
710 char *addr, unsigned long count)
718 /* The cast to a user pointer is valid due to the set_fs() */
719 result = vfs_read(file, (void __user *)addr, count, &pos);
724 EXPORT_SYMBOL(kernel_read);
726 static int exec_mmap(struct mm_struct *mm)
728 struct task_struct *tsk;
729 struct mm_struct * old_mm, *active_mm;
731 /* Notify parent that we're no longer interested in the old VM */
733 old_mm = current->mm;
734 sync_mm_rss(tsk, old_mm);
735 mm_release(tsk, old_mm);
739 * Make sure that if there is a core dump in progress
740 * for the old mm, we get out and die instead of going
741 * through with the exec. We must hold mmap_sem around
742 * checking core_state and changing tsk->mm.
744 down_read(&old_mm->mmap_sem);
745 if (unlikely(old_mm->core_state)) {
746 up_read(&old_mm->mmap_sem);
751 active_mm = tsk->active_mm;
754 activate_mm(active_mm, mm);
756 arch_pick_mmap_layout(mm);
758 up_read(&old_mm->mmap_sem);
759 BUG_ON(active_mm != old_mm);
760 mm_update_next_owner(old_mm);
769 * This function makes sure the current process has its own signal table,
770 * so that flush_signal_handlers can later reset the handlers without
771 * disturbing other processes. (Other processes might share the signal
772 * table via the CLONE_SIGHAND option to clone().)
774 static int de_thread(struct task_struct *tsk)
776 struct signal_struct *sig = tsk->signal;
777 struct sighand_struct *oldsighand = tsk->sighand;
778 spinlock_t *lock = &oldsighand->siglock;
780 if (thread_group_empty(tsk))
781 goto no_thread_group;
784 * Kill all other threads in the thread group.
787 if (signal_group_exit(sig)) {
789 * Another group action in progress, just
790 * return so that the signal is processed.
792 spin_unlock_irq(lock);
796 sig->group_exit_task = tsk;
797 sig->notify_count = zap_other_threads(tsk);
798 if (!thread_group_leader(tsk))
801 while (sig->notify_count) {
802 __set_current_state(TASK_UNINTERRUPTIBLE);
803 spin_unlock_irq(lock);
807 spin_unlock_irq(lock);
810 * At this point all other threads have exited, all we have to
811 * do is to wait for the thread group leader to become inactive,
812 * and to assume its PID:
814 if (!thread_group_leader(tsk)) {
815 struct task_struct *leader = tsk->group_leader;
817 sig->notify_count = -1; /* for exit_notify() */
819 write_lock_irq(&tasklist_lock);
820 if (likely(leader->exit_state))
822 __set_current_state(TASK_UNINTERRUPTIBLE);
823 write_unlock_irq(&tasklist_lock);
828 * The only record we have of the real-time age of a
829 * process, regardless of execs it's done, is start_time.
830 * All the past CPU time is accumulated in signal_struct
831 * from sister threads now dead. But in this non-leader
832 * exec, nothing survives from the original leader thread,
833 * whose birth marks the true age of this process now.
834 * When we take on its identity by switching to its PID, we
835 * also take its birthdate (always earlier than our own).
837 tsk->start_time = leader->start_time;
839 BUG_ON(!same_thread_group(leader, tsk));
840 BUG_ON(has_group_leader_pid(tsk));
842 * An exec() starts a new thread group with the
843 * TGID of the previous thread group. Rehash the
844 * two threads with a switched PID, and release
845 * the former thread group leader:
848 /* Become a process group leader with the old leader's pid.
849 * The old leader becomes a thread of the this thread group.
850 * Note: The old leader also uses this pid until release_task
851 * is called. Odd but simple and correct.
853 detach_pid(tsk, PIDTYPE_PID);
854 tsk->pid = leader->pid;
855 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
856 transfer_pid(leader, tsk, PIDTYPE_PGID);
857 transfer_pid(leader, tsk, PIDTYPE_SID);
859 list_replace_rcu(&leader->tasks, &tsk->tasks);
860 list_replace_init(&leader->sibling, &tsk->sibling);
862 tsk->group_leader = tsk;
863 leader->group_leader = tsk;
865 tsk->exit_signal = SIGCHLD;
867 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
868 leader->exit_state = EXIT_DEAD;
869 write_unlock_irq(&tasklist_lock);
871 release_task(leader);
874 sig->group_exit_task = NULL;
875 sig->notify_count = 0;
879 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
882 flush_itimer_signals();
884 if (atomic_read(&oldsighand->count) != 1) {
885 struct sighand_struct *newsighand;
887 * This ->sighand is shared with the CLONE_SIGHAND
888 * but not CLONE_THREAD task, switch to the new one.
890 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
894 atomic_set(&newsighand->count, 1);
895 memcpy(newsighand->action, oldsighand->action,
896 sizeof(newsighand->action));
898 write_lock_irq(&tasklist_lock);
899 spin_lock(&oldsighand->siglock);
900 rcu_assign_pointer(tsk->sighand, newsighand);
901 spin_unlock(&oldsighand->siglock);
902 write_unlock_irq(&tasklist_lock);
904 __cleanup_sighand(oldsighand);
907 BUG_ON(!thread_group_leader(tsk));
912 * These functions flushes out all traces of the currently running executable
913 * so that a new one can be started
915 static void flush_old_files(struct files_struct * files)
920 spin_lock(&files->file_lock);
922 unsigned long set, i;
926 fdt = files_fdtable(files);
927 if (i >= fdt->max_fds)
929 set = fdt->close_on_exec->fds_bits[j];
932 fdt->close_on_exec->fds_bits[j] = 0;
933 spin_unlock(&files->file_lock);
934 for ( ; set ; i++,set >>= 1) {
939 spin_lock(&files->file_lock);
942 spin_unlock(&files->file_lock);
945 char *get_task_comm(char *buf, struct task_struct *tsk)
947 /* buf must be at least sizeof(tsk->comm) in size */
949 strncpy(buf, tsk->comm, sizeof(tsk->comm));
954 void set_task_comm(struct task_struct *tsk, char *buf)
959 * Threads may access current->comm without holding
960 * the task lock, so write the string carefully.
961 * Readers without a lock may see incomplete new
962 * names but are safe from non-terminating string reads.
964 memset(tsk->comm, 0, TASK_COMM_LEN);
966 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
968 perf_event_comm(tsk);
971 int flush_old_exec(struct linux_binprm * bprm)
976 * Make sure we have a private signal table and that
977 * we are unassociated from the previous thread group.
979 retval = de_thread(current);
983 set_mm_exe_file(bprm->mm, bprm->file);
986 * Release all of the old mmap stuff
988 retval = exec_mmap(bprm->mm);
992 bprm->mm = NULL; /* We're using it now */
994 current->flags &= ~PF_RANDOMIZE;
996 current->personality &= ~bprm->per_clear;
1003 EXPORT_SYMBOL(flush_old_exec);
1005 void setup_new_exec(struct linux_binprm * bprm)
1009 char tcomm[sizeof(current->comm)];
1011 arch_pick_mmap_layout(current->mm);
1013 /* This is the point of no return */
1014 current->sas_ss_sp = current->sas_ss_size = 0;
1016 if (current_euid() == current_uid() && current_egid() == current_gid())
1017 set_dumpable(current->mm, 1);
1019 set_dumpable(current->mm, suid_dumpable);
1021 name = bprm->filename;
1023 /* Copies the binary name from after last slash */
1024 for (i=0; (ch = *(name++)) != '\0';) {
1026 i = 0; /* overwrite what we wrote */
1028 if (i < (sizeof(tcomm) - 1))
1032 set_task_comm(current, tcomm);
1034 /* Set the new mm task size. We have to do that late because it may
1035 * depend on TIF_32BIT which is only updated in flush_thread() on
1036 * some architectures like powerpc
1038 current->mm->task_size = TASK_SIZE;
1040 /* install the new credentials */
1041 if (bprm->cred->uid != current_euid() ||
1042 bprm->cred->gid != current_egid()) {
1043 current->pdeath_signal = 0;
1044 } else if (file_permission(bprm->file, MAY_READ) ||
1045 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1046 set_dumpable(current->mm, suid_dumpable);
1050 * Flush performance counters when crossing a
1053 if (!get_dumpable(current->mm))
1054 perf_event_exit_task(current);
1056 /* An exec changes our domain. We are no longer part of the thread
1059 current->self_exec_id++;
1061 flush_signal_handlers(current, 0);
1062 flush_old_files(current->files);
1064 EXPORT_SYMBOL(setup_new_exec);
1067 * Prepare credentials and lock ->cred_guard_mutex.
1068 * install_exec_creds() commits the new creds and drops the lock.
1069 * Or, if exec fails before, free_bprm() should release ->cred and
1072 int prepare_bprm_creds(struct linux_binprm *bprm)
1074 if (mutex_lock_interruptible(¤t->cred_guard_mutex))
1075 return -ERESTARTNOINTR;
1077 bprm->cred = prepare_exec_creds();
1078 if (likely(bprm->cred))
1081 mutex_unlock(¤t->cred_guard_mutex);
1085 void free_bprm(struct linux_binprm *bprm)
1087 free_arg_pages(bprm);
1089 mutex_unlock(¤t->cred_guard_mutex);
1090 abort_creds(bprm->cred);
1096 * install the new credentials for this executable
1098 void install_exec_creds(struct linux_binprm *bprm)
1100 security_bprm_committing_creds(bprm);
1102 commit_creds(bprm->cred);
1105 * cred_guard_mutex must be held at least to this point to prevent
1106 * ptrace_attach() from altering our determination of the task's
1107 * credentials; any time after this it may be unlocked.
1109 security_bprm_committed_creds(bprm);
1110 mutex_unlock(¤t->cred_guard_mutex);
1112 EXPORT_SYMBOL(install_exec_creds);
1115 * determine how safe it is to execute the proposed program
1116 * - the caller must hold current->cred_guard_mutex to protect against
1119 int check_unsafe_exec(struct linux_binprm *bprm)
1121 struct task_struct *p = current, *t;
1125 bprm->unsafe = tracehook_unsafe_exec(p);
1128 spin_lock(&p->fs->lock);
1130 for (t = next_thread(p); t != p; t = next_thread(t)) {
1136 if (p->fs->users > n_fs) {
1137 bprm->unsafe |= LSM_UNSAFE_SHARE;
1140 if (!p->fs->in_exec) {
1145 spin_unlock(&p->fs->lock);
1151 * Fill the binprm structure from the inode.
1152 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1154 * This may be called multiple times for binary chains (scripts for example).
1156 int prepare_binprm(struct linux_binprm *bprm)
1159 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1162 mode = inode->i_mode;
1163 if (bprm->file->f_op == NULL)
1166 /* clear any previous set[ug]id data from a previous binary */
1167 bprm->cred->euid = current_euid();
1168 bprm->cred->egid = current_egid();
1170 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1172 if (mode & S_ISUID) {
1173 bprm->per_clear |= PER_CLEAR_ON_SETID;
1174 bprm->cred->euid = inode->i_uid;
1179 * If setgid is set but no group execute bit then this
1180 * is a candidate for mandatory locking, not a setgid
1183 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1184 bprm->per_clear |= PER_CLEAR_ON_SETID;
1185 bprm->cred->egid = inode->i_gid;
1189 /* fill in binprm security blob */
1190 retval = security_bprm_set_creds(bprm);
1193 bprm->cred_prepared = 1;
1195 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1196 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1199 EXPORT_SYMBOL(prepare_binprm);
1202 * Arguments are '\0' separated strings found at the location bprm->p
1203 * points to; chop off the first by relocating brpm->p to right after
1204 * the first '\0' encountered.
1206 int remove_arg_zero(struct linux_binprm *bprm)
1209 unsigned long offset;
1217 offset = bprm->p & ~PAGE_MASK;
1218 page = get_arg_page(bprm, bprm->p, 0);
1223 kaddr = kmap_atomic(page, KM_USER0);
1225 for (; offset < PAGE_SIZE && kaddr[offset];
1226 offset++, bprm->p++)
1229 kunmap_atomic(kaddr, KM_USER0);
1232 if (offset == PAGE_SIZE)
1233 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1234 } while (offset == PAGE_SIZE);
1243 EXPORT_SYMBOL(remove_arg_zero);
1246 * cycle the list of binary formats handler, until one recognizes the image
1248 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1250 unsigned int depth = bprm->recursion_depth;
1252 struct linux_binfmt *fmt;
1254 retval = security_bprm_check(bprm);
1258 /* kernel module loader fixup */
1259 /* so we don't try to load run modprobe in kernel space. */
1262 retval = audit_bprm(bprm);
1267 for (try=0; try<2; try++) {
1268 read_lock(&binfmt_lock);
1269 list_for_each_entry(fmt, &formats, lh) {
1270 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1273 if (!try_module_get(fmt->module))
1275 read_unlock(&binfmt_lock);
1276 retval = fn(bprm, regs);
1278 * Restore the depth counter to its starting value
1279 * in this call, so we don't have to rely on every
1280 * load_binary function to restore it on return.
1282 bprm->recursion_depth = depth;
1285 tracehook_report_exec(fmt, bprm, regs);
1287 allow_write_access(bprm->file);
1291 current->did_exec = 1;
1292 proc_exec_connector(current);
1295 read_lock(&binfmt_lock);
1297 if (retval != -ENOEXEC || bprm->mm == NULL)
1300 read_unlock(&binfmt_lock);
1304 read_unlock(&binfmt_lock);
1305 if (retval != -ENOEXEC || bprm->mm == NULL) {
1307 #ifdef CONFIG_MODULES
1309 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1310 if (printable(bprm->buf[0]) &&
1311 printable(bprm->buf[1]) &&
1312 printable(bprm->buf[2]) &&
1313 printable(bprm->buf[3]))
1314 break; /* -ENOEXEC */
1315 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1322 EXPORT_SYMBOL(search_binary_handler);
1325 * sys_execve() executes a new program.
1327 int do_execve(const char * filename,
1328 const char __user *const __user *argv,
1329 const char __user *const __user *envp,
1330 struct pt_regs * regs)
1332 struct linux_binprm *bprm;
1334 struct files_struct *displaced;
1338 retval = unshare_files(&displaced);
1343 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1347 retval = prepare_bprm_creds(bprm);
1351 retval = check_unsafe_exec(bprm);
1354 clear_in_exec = retval;
1355 current->in_execve = 1;
1357 file = open_exec(filename);
1358 retval = PTR_ERR(file);
1365 bprm->filename = filename;
1366 bprm->interp = filename;
1368 retval = bprm_mm_init(bprm);
1372 bprm->argc = count(argv, MAX_ARG_STRINGS);
1373 if ((retval = bprm->argc) < 0)
1376 bprm->envc = count(envp, MAX_ARG_STRINGS);
1377 if ((retval = bprm->envc) < 0)
1380 retval = prepare_binprm(bprm);
1384 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1388 bprm->exec = bprm->p;
1389 retval = copy_strings(bprm->envc, envp, bprm);
1393 retval = copy_strings(bprm->argc, argv, bprm);
1397 current->flags &= ~PF_KTHREAD;
1398 retval = search_binary_handler(bprm,regs);
1402 /* execve succeeded */
1403 current->fs->in_exec = 0;
1404 current->in_execve = 0;
1405 acct_update_integrals(current);
1408 put_files_struct(displaced);
1417 allow_write_access(bprm->file);
1423 current->fs->in_exec = 0;
1424 current->in_execve = 0;
1431 reset_files_struct(displaced);
1436 void set_binfmt(struct linux_binfmt *new)
1438 struct mm_struct *mm = current->mm;
1441 module_put(mm->binfmt->module);
1445 __module_get(new->module);
1448 EXPORT_SYMBOL(set_binfmt);
1450 /* format_corename will inspect the pattern parameter, and output a
1451 * name into corename, which must have space for at least
1452 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1454 static int format_corename(char *corename, long signr)
1456 const struct cred *cred = current_cred();
1457 const char *pat_ptr = core_pattern;
1458 int ispipe = (*pat_ptr == '|');
1459 char *out_ptr = corename;
1460 char *const out_end = corename + CORENAME_MAX_SIZE;
1462 int pid_in_pattern = 0;
1464 /* Repeat as long as we have more pattern to process and more output
1467 if (*pat_ptr != '%') {
1468 if (out_ptr == out_end)
1470 *out_ptr++ = *pat_ptr++;
1472 switch (*++pat_ptr) {
1475 /* Double percent, output one percent */
1477 if (out_ptr == out_end)
1484 rc = snprintf(out_ptr, out_end - out_ptr,
1485 "%d", task_tgid_vnr(current));
1486 if (rc > out_end - out_ptr)
1492 rc = snprintf(out_ptr, out_end - out_ptr,
1494 if (rc > out_end - out_ptr)
1500 rc = snprintf(out_ptr, out_end - out_ptr,
1502 if (rc > out_end - out_ptr)
1506 /* signal that caused the coredump */
1508 rc = snprintf(out_ptr, out_end - out_ptr,
1510 if (rc > out_end - out_ptr)
1514 /* UNIX time of coredump */
1517 do_gettimeofday(&tv);
1518 rc = snprintf(out_ptr, out_end - out_ptr,
1520 if (rc > out_end - out_ptr)
1527 down_read(&uts_sem);
1528 rc = snprintf(out_ptr, out_end - out_ptr,
1529 "%s", utsname()->nodename);
1531 if (rc > out_end - out_ptr)
1537 rc = snprintf(out_ptr, out_end - out_ptr,
1538 "%s", current->comm);
1539 if (rc > out_end - out_ptr)
1543 /* core limit size */
1545 rc = snprintf(out_ptr, out_end - out_ptr,
1546 "%lu", rlimit(RLIMIT_CORE));
1547 if (rc > out_end - out_ptr)
1557 /* Backward compatibility with core_uses_pid:
1559 * If core_pattern does not include a %p (as is the default)
1560 * and core_uses_pid is set, then .%pid will be appended to
1561 * the filename. Do not do this for piped commands. */
1562 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1563 rc = snprintf(out_ptr, out_end - out_ptr,
1564 ".%d", task_tgid_vnr(current));
1565 if (rc > out_end - out_ptr)
1574 static int zap_process(struct task_struct *start, int exit_code)
1576 struct task_struct *t;
1579 start->signal->flags = SIGNAL_GROUP_EXIT;
1580 start->signal->group_exit_code = exit_code;
1581 start->signal->group_stop_count = 0;
1585 if (t != current && t->mm) {
1586 sigaddset(&t->pending.signal, SIGKILL);
1587 signal_wake_up(t, 1);
1590 } while_each_thread(start, t);
1595 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1596 struct core_state *core_state, int exit_code)
1598 struct task_struct *g, *p;
1599 unsigned long flags;
1602 spin_lock_irq(&tsk->sighand->siglock);
1603 if (!signal_group_exit(tsk->signal)) {
1604 mm->core_state = core_state;
1605 nr = zap_process(tsk, exit_code);
1607 spin_unlock_irq(&tsk->sighand->siglock);
1608 if (unlikely(nr < 0))
1611 if (atomic_read(&mm->mm_users) == nr + 1)
1614 * We should find and kill all tasks which use this mm, and we should
1615 * count them correctly into ->nr_threads. We don't take tasklist
1616 * lock, but this is safe wrt:
1619 * None of sub-threads can fork after zap_process(leader). All
1620 * processes which were created before this point should be
1621 * visible to zap_threads() because copy_process() adds the new
1622 * process to the tail of init_task.tasks list, and lock/unlock
1623 * of ->siglock provides a memory barrier.
1626 * The caller holds mm->mmap_sem. This means that the task which
1627 * uses this mm can't pass exit_mm(), so it can't exit or clear
1631 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1632 * we must see either old or new leader, this does not matter.
1633 * However, it can change p->sighand, so lock_task_sighand(p)
1634 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1637 * Note also that "g" can be the old leader with ->mm == NULL
1638 * and already unhashed and thus removed from ->thread_group.
1639 * This is OK, __unhash_process()->list_del_rcu() does not
1640 * clear the ->next pointer, we will find the new leader via
1644 for_each_process(g) {
1645 if (g == tsk->group_leader)
1647 if (g->flags & PF_KTHREAD)
1652 if (unlikely(p->mm == mm)) {
1653 lock_task_sighand(p, &flags);
1654 nr += zap_process(p, exit_code);
1655 unlock_task_sighand(p, &flags);
1659 } while_each_thread(g, p);
1663 atomic_set(&core_state->nr_threads, nr);
1667 static int coredump_wait(int exit_code, struct core_state *core_state)
1669 struct task_struct *tsk = current;
1670 struct mm_struct *mm = tsk->mm;
1671 struct completion *vfork_done;
1672 int core_waiters = -EBUSY;
1674 init_completion(&core_state->startup);
1675 core_state->dumper.task = tsk;
1676 core_state->dumper.next = NULL;
1678 down_write(&mm->mmap_sem);
1679 if (!mm->core_state)
1680 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1681 up_write(&mm->mmap_sem);
1683 if (unlikely(core_waiters < 0))
1687 * Make sure nobody is waiting for us to release the VM,
1688 * otherwise we can deadlock when we wait on each other
1690 vfork_done = tsk->vfork_done;
1692 tsk->vfork_done = NULL;
1693 complete(vfork_done);
1697 wait_for_completion(&core_state->startup);
1699 return core_waiters;
1702 static void coredump_finish(struct mm_struct *mm)
1704 struct core_thread *curr, *next;
1705 struct task_struct *task;
1707 next = mm->core_state->dumper.next;
1708 while ((curr = next) != NULL) {
1712 * see exit_mm(), curr->task must not see
1713 * ->task == NULL before we read ->next.
1717 wake_up_process(task);
1720 mm->core_state = NULL;
1724 * set_dumpable converts traditional three-value dumpable to two flags and
1725 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1726 * these bits are not changed atomically. So get_dumpable can observe the
1727 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1728 * return either old dumpable or new one by paying attention to the order of
1729 * modifying the bits.
1731 * dumpable | mm->flags (binary)
1732 * old new | initial interim final
1733 * ---------+-----------------------
1741 * (*) get_dumpable regards interim value of 10 as 11.
1743 void set_dumpable(struct mm_struct *mm, int value)
1747 clear_bit(MMF_DUMPABLE, &mm->flags);
1749 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1752 set_bit(MMF_DUMPABLE, &mm->flags);
1754 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1757 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1759 set_bit(MMF_DUMPABLE, &mm->flags);
1764 static int __get_dumpable(unsigned long mm_flags)
1768 ret = mm_flags & MMF_DUMPABLE_MASK;
1769 return (ret >= 2) ? 2 : ret;
1772 int get_dumpable(struct mm_struct *mm)
1774 return __get_dumpable(mm->flags);
1777 static void wait_for_dump_helpers(struct file *file)
1779 struct pipe_inode_info *pipe;
1781 pipe = file->f_path.dentry->d_inode->i_pipe;
1787 while ((pipe->readers > 1) && (!signal_pending(current))) {
1788 wake_up_interruptible_sync(&pipe->wait);
1789 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1802 * helper function to customize the process used
1803 * to collect the core in userspace. Specifically
1804 * it sets up a pipe and installs it as fd 0 (stdin)
1805 * for the process. Returns 0 on success, or
1806 * PTR_ERR on failure.
1807 * Note that it also sets the core limit to 1. This
1808 * is a special value that we use to trap recursive
1811 static int umh_pipe_setup(struct subprocess_info *info)
1813 struct file *rp, *wp;
1814 struct fdtable *fdt;
1815 struct coredump_params *cp = (struct coredump_params *)info->data;
1816 struct files_struct *cf = current->files;
1818 wp = create_write_pipe(0);
1822 rp = create_read_pipe(wp, 0);
1824 free_write_pipe(wp);
1832 spin_lock(&cf->file_lock);
1833 fdt = files_fdtable(cf);
1834 FD_SET(0, fdt->open_fds);
1835 FD_CLR(0, fdt->close_on_exec);
1836 spin_unlock(&cf->file_lock);
1838 /* and disallow core files too */
1839 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1844 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1846 struct core_state core_state;
1847 char corename[CORENAME_MAX_SIZE + 1];
1848 struct mm_struct *mm = current->mm;
1849 struct linux_binfmt * binfmt;
1850 const struct cred *old_cred;
1855 static atomic_t core_dump_count = ATOMIC_INIT(0);
1856 struct coredump_params cprm = {
1859 .limit = rlimit(RLIMIT_CORE),
1861 * We must use the same mm->flags while dumping core to avoid
1862 * inconsistency of bit flags, since this flag is not protected
1865 .mm_flags = mm->flags,
1868 audit_core_dumps(signr);
1870 binfmt = mm->binfmt;
1871 if (!binfmt || !binfmt->core_dump)
1873 if (!__get_dumpable(cprm.mm_flags))
1876 cred = prepare_creds();
1880 * We cannot trust fsuid as being the "true" uid of the
1881 * process nor do we know its entire history. We only know it
1882 * was tainted so we dump it as root in mode 2.
1884 if (__get_dumpable(cprm.mm_flags) == 2) {
1885 /* Setuid core dump mode */
1886 flag = O_EXCL; /* Stop rewrite attacks */
1887 cred->fsuid = 0; /* Dump root private */
1890 retval = coredump_wait(exit_code, &core_state);
1894 old_cred = override_creds(cred);
1897 * Clear any false indication of pending signals that might
1898 * be seen by the filesystem code called to write the core file.
1900 clear_thread_flag(TIF_SIGPENDING);
1902 ispipe = format_corename(corename, signr);
1908 if (cprm.limit == 1) {
1910 * Normally core limits are irrelevant to pipes, since
1911 * we're not writing to the file system, but we use
1912 * cprm.limit of 1 here as a speacial value. Any
1913 * non-1 limit gets set to RLIM_INFINITY below, but
1914 * a limit of 0 skips the dump. This is a consistent
1915 * way to catch recursive crashes. We can still crash
1916 * if the core_pattern binary sets RLIM_CORE = !1
1917 * but it runs as root, and can do lots of stupid things
1918 * Note that we use task_tgid_vnr here to grab the pid
1919 * of the process group leader. That way we get the
1920 * right pid if a thread in a multi-threaded
1921 * core_pattern process dies.
1924 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1925 task_tgid_vnr(current), current->comm);
1926 printk(KERN_WARNING "Aborting core\n");
1929 cprm.limit = RLIM_INFINITY;
1931 dump_count = atomic_inc_return(&core_dump_count);
1932 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1933 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1934 task_tgid_vnr(current), current->comm);
1935 printk(KERN_WARNING "Skipping core dump\n");
1936 goto fail_dropcount;
1939 helper_argv = argv_split(GFP_KERNEL, corename+1, NULL);
1941 printk(KERN_WARNING "%s failed to allocate memory\n",
1943 goto fail_dropcount;
1946 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1947 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1949 argv_free(helper_argv);
1951 printk(KERN_INFO "Core dump to %s pipe failed\n",
1956 struct inode *inode;
1958 if (cprm.limit < binfmt->min_coredump)
1961 cprm.file = filp_open(corename,
1962 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1964 if (IS_ERR(cprm.file))
1967 inode = cprm.file->f_path.dentry->d_inode;
1968 if (inode->i_nlink > 1)
1970 if (d_unhashed(cprm.file->f_path.dentry))
1973 * AK: actually i see no reason to not allow this for named
1974 * pipes etc, but keep the previous behaviour for now.
1976 if (!S_ISREG(inode->i_mode))
1979 * Dont allow local users get cute and trick others to coredump
1980 * into their pre-created files.
1982 if (inode->i_uid != current_fsuid())
1984 if (!cprm.file->f_op || !cprm.file->f_op->write)
1986 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
1990 retval = binfmt->core_dump(&cprm);
1992 current->signal->group_exit_code |= 0x80;
1994 if (ispipe && core_pipe_limit)
1995 wait_for_dump_helpers(cprm.file);
1998 filp_close(cprm.file, NULL);
2001 atomic_dec(&core_dump_count);
2003 coredump_finish(mm);
2004 revert_creds(old_cred);