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;
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, const char *const *argv,
474 struct linux_binprm *bprm)
477 mm_segment_t oldfs = get_fs();
479 r = copy_strings(argc, (const char __user *const __user *)argv, bprm);
483 EXPORT_SYMBOL(copy_strings_kernel);
488 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
489 * the binfmt code determines where the new stack should reside, we shift it to
490 * its final location. The process proceeds as follows:
492 * 1) Use shift to calculate the new vma endpoints.
493 * 2) Extend vma to cover both the old and new ranges. This ensures the
494 * arguments passed to subsequent functions are consistent.
495 * 3) Move vma's page tables to the new range.
496 * 4) Free up any cleared pgd range.
497 * 5) Shrink the vma to cover only the new range.
499 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
501 struct mm_struct *mm = vma->vm_mm;
502 unsigned long old_start = vma->vm_start;
503 unsigned long old_end = vma->vm_end;
504 unsigned long length = old_end - old_start;
505 unsigned long new_start = old_start - shift;
506 unsigned long new_end = old_end - shift;
507 struct mmu_gather *tlb;
509 BUG_ON(new_start > new_end);
512 * ensure there are no vmas between where we want to go
515 if (vma != find_vma(mm, new_start))
519 * cover the whole range: [new_start, old_end)
521 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
525 * move the page tables downwards, on failure we rely on
526 * process cleanup to remove whatever mess we made.
528 if (length != move_page_tables(vma, old_start,
529 vma, new_start, length))
533 tlb = tlb_gather_mmu(mm, 0);
534 if (new_end > old_start) {
536 * when the old and new regions overlap clear from new_end.
538 free_pgd_range(tlb, new_end, old_end, new_end,
539 vma->vm_next ? vma->vm_next->vm_start : 0);
542 * otherwise, clean from old_start; this is done to not touch
543 * the address space in [new_end, old_start) some architectures
544 * have constraints on va-space that make this illegal (IA64) -
545 * for the others its just a little faster.
547 free_pgd_range(tlb, old_start, old_end, new_end,
548 vma->vm_next ? vma->vm_next->vm_start : 0);
550 tlb_finish_mmu(tlb, new_end, old_end);
553 * Shrink the vma to just the new range. Always succeeds.
555 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
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 = rlimit_max(RLIMIT_STACK);
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);
598 if (unlikely(stack_top < mmap_min_addr) ||
599 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
602 stack_shift = vma->vm_end - stack_top;
604 bprm->p -= stack_shift;
605 mm->arg_start = bprm->p;
609 bprm->loader -= stack_shift;
610 bprm->exec -= stack_shift;
612 down_write(&mm->mmap_sem);
613 vm_flags = VM_STACK_FLAGS;
616 * Adjust stack execute permissions; explicitly enable for
617 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
618 * (arch default) otherwise.
620 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
622 else if (executable_stack == EXSTACK_DISABLE_X)
623 vm_flags &= ~VM_EXEC;
624 vm_flags |= mm->def_flags;
625 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
627 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
633 /* Move stack pages down in memory. */
635 ret = shift_arg_pages(vma, stack_shift);
640 /* mprotect_fixup is overkill to remove the temporary stack flags */
641 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
643 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
644 stack_size = vma->vm_end - vma->vm_start;
646 * Align this down to a page boundary as expand_stack
649 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
650 #ifdef CONFIG_STACK_GROWSUP
651 if (stack_size + stack_expand > rlim_stack)
652 stack_base = vma->vm_start + rlim_stack;
654 stack_base = vma->vm_end + stack_expand;
656 if (stack_size + stack_expand > rlim_stack)
657 stack_base = vma->vm_end - rlim_stack;
659 stack_base = vma->vm_start - stack_expand;
661 current->mm->start_stack = bprm->p;
662 ret = expand_stack(vma, stack_base);
667 up_write(&mm->mmap_sem);
670 EXPORT_SYMBOL(setup_arg_pages);
672 #endif /* CONFIG_MMU */
674 struct file *open_exec(const char *name)
679 file = do_filp_open(AT_FDCWD, name,
680 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
681 MAY_EXEC | MAY_OPEN);
686 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
689 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
694 err = deny_write_access(file);
705 EXPORT_SYMBOL(open_exec);
707 int kernel_read(struct file *file, loff_t offset,
708 char *addr, unsigned long count)
716 /* The cast to a user pointer is valid due to the set_fs() */
717 result = vfs_read(file, (void __user *)addr, count, &pos);
722 EXPORT_SYMBOL(kernel_read);
724 static int exec_mmap(struct mm_struct *mm)
726 struct task_struct *tsk;
727 struct mm_struct * old_mm, *active_mm;
729 /* Notify parent that we're no longer interested in the old VM */
731 old_mm = current->mm;
732 sync_mm_rss(tsk, old_mm);
733 mm_release(tsk, old_mm);
737 * Make sure that if there is a core dump in progress
738 * for the old mm, we get out and die instead of going
739 * through with the exec. We must hold mmap_sem around
740 * checking core_state and changing tsk->mm.
742 down_read(&old_mm->mmap_sem);
743 if (unlikely(old_mm->core_state)) {
744 up_read(&old_mm->mmap_sem);
749 active_mm = tsk->active_mm;
752 activate_mm(active_mm, mm);
754 arch_pick_mmap_layout(mm);
756 up_read(&old_mm->mmap_sem);
757 BUG_ON(active_mm != old_mm);
758 mm_update_next_owner(old_mm);
767 * This function makes sure the current process has its own signal table,
768 * so that flush_signal_handlers can later reset the handlers without
769 * disturbing other processes. (Other processes might share the signal
770 * table via the CLONE_SIGHAND option to clone().)
772 static int de_thread(struct task_struct *tsk)
774 struct signal_struct *sig = tsk->signal;
775 struct sighand_struct *oldsighand = tsk->sighand;
776 spinlock_t *lock = &oldsighand->siglock;
778 if (thread_group_empty(tsk))
779 goto no_thread_group;
782 * Kill all other threads in the thread group.
785 if (signal_group_exit(sig)) {
787 * Another group action in progress, just
788 * return so that the signal is processed.
790 spin_unlock_irq(lock);
794 sig->group_exit_task = tsk;
795 sig->notify_count = zap_other_threads(tsk);
796 if (!thread_group_leader(tsk))
799 while (sig->notify_count) {
800 __set_current_state(TASK_UNINTERRUPTIBLE);
801 spin_unlock_irq(lock);
805 spin_unlock_irq(lock);
808 * At this point all other threads have exited, all we have to
809 * do is to wait for the thread group leader to become inactive,
810 * and to assume its PID:
812 if (!thread_group_leader(tsk)) {
813 struct task_struct *leader = tsk->group_leader;
815 sig->notify_count = -1; /* for exit_notify() */
817 write_lock_irq(&tasklist_lock);
818 if (likely(leader->exit_state))
820 __set_current_state(TASK_UNINTERRUPTIBLE);
821 write_unlock_irq(&tasklist_lock);
826 * The only record we have of the real-time age of a
827 * process, regardless of execs it's done, is start_time.
828 * All the past CPU time is accumulated in signal_struct
829 * from sister threads now dead. But in this non-leader
830 * exec, nothing survives from the original leader thread,
831 * whose birth marks the true age of this process now.
832 * When we take on its identity by switching to its PID, we
833 * also take its birthdate (always earlier than our own).
835 tsk->start_time = leader->start_time;
837 BUG_ON(!same_thread_group(leader, tsk));
838 BUG_ON(has_group_leader_pid(tsk));
840 * An exec() starts a new thread group with the
841 * TGID of the previous thread group. Rehash the
842 * two threads with a switched PID, and release
843 * the former thread group leader:
846 /* Become a process group leader with the old leader's pid.
847 * The old leader becomes a thread of the this thread group.
848 * Note: The old leader also uses this pid until release_task
849 * is called. Odd but simple and correct.
851 detach_pid(tsk, PIDTYPE_PID);
852 tsk->pid = leader->pid;
853 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
854 transfer_pid(leader, tsk, PIDTYPE_PGID);
855 transfer_pid(leader, tsk, PIDTYPE_SID);
857 list_replace_rcu(&leader->tasks, &tsk->tasks);
858 list_replace_init(&leader->sibling, &tsk->sibling);
860 tsk->group_leader = tsk;
861 leader->group_leader = tsk;
863 tsk->exit_signal = SIGCHLD;
865 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
866 leader->exit_state = EXIT_DEAD;
867 write_unlock_irq(&tasklist_lock);
869 release_task(leader);
872 sig->group_exit_task = NULL;
873 sig->notify_count = 0;
877 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
880 flush_itimer_signals();
882 if (atomic_read(&oldsighand->count) != 1) {
883 struct sighand_struct *newsighand;
885 * This ->sighand is shared with the CLONE_SIGHAND
886 * but not CLONE_THREAD task, switch to the new one.
888 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
892 atomic_set(&newsighand->count, 1);
893 memcpy(newsighand->action, oldsighand->action,
894 sizeof(newsighand->action));
896 write_lock_irq(&tasklist_lock);
897 spin_lock(&oldsighand->siglock);
898 rcu_assign_pointer(tsk->sighand, newsighand);
899 spin_unlock(&oldsighand->siglock);
900 write_unlock_irq(&tasklist_lock);
902 __cleanup_sighand(oldsighand);
905 BUG_ON(!thread_group_leader(tsk));
910 * These functions flushes out all traces of the currently running executable
911 * so that a new one can be started
913 static void flush_old_files(struct files_struct * files)
918 spin_lock(&files->file_lock);
920 unsigned long set, i;
924 fdt = files_fdtable(files);
925 if (i >= fdt->max_fds)
927 set = fdt->close_on_exec->fds_bits[j];
930 fdt->close_on_exec->fds_bits[j] = 0;
931 spin_unlock(&files->file_lock);
932 for ( ; set ; i++,set >>= 1) {
937 spin_lock(&files->file_lock);
940 spin_unlock(&files->file_lock);
943 char *get_task_comm(char *buf, struct task_struct *tsk)
945 /* buf must be at least sizeof(tsk->comm) in size */
947 strncpy(buf, tsk->comm, sizeof(tsk->comm));
952 void set_task_comm(struct task_struct *tsk, char *buf)
957 * Threads may access current->comm without holding
958 * the task lock, so write the string carefully.
959 * Readers without a lock may see incomplete new
960 * names but are safe from non-terminating string reads.
962 memset(tsk->comm, 0, TASK_COMM_LEN);
964 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
966 perf_event_comm(tsk);
969 int flush_old_exec(struct linux_binprm * bprm)
974 * Make sure we have a private signal table and that
975 * we are unassociated from the previous thread group.
977 retval = de_thread(current);
981 set_mm_exe_file(bprm->mm, bprm->file);
984 * Release all of the old mmap stuff
986 retval = exec_mmap(bprm->mm);
990 bprm->mm = NULL; /* We're using it now */
992 current->flags &= ~PF_RANDOMIZE;
994 current->personality &= ~bprm->per_clear;
1001 EXPORT_SYMBOL(flush_old_exec);
1003 void setup_new_exec(struct linux_binprm * bprm)
1007 char tcomm[sizeof(current->comm)];
1009 arch_pick_mmap_layout(current->mm);
1011 /* This is the point of no return */
1012 current->sas_ss_sp = current->sas_ss_size = 0;
1014 if (current_euid() == current_uid() && current_egid() == current_gid())
1015 set_dumpable(current->mm, 1);
1017 set_dumpable(current->mm, suid_dumpable);
1019 name = bprm->filename;
1021 /* Copies the binary name from after last slash */
1022 for (i=0; (ch = *(name++)) != '\0';) {
1024 i = 0; /* overwrite what we wrote */
1026 if (i < (sizeof(tcomm) - 1))
1030 set_task_comm(current, tcomm);
1032 /* Set the new mm task size. We have to do that late because it may
1033 * depend on TIF_32BIT which is only updated in flush_thread() on
1034 * some architectures like powerpc
1036 current->mm->task_size = TASK_SIZE;
1038 /* install the new credentials */
1039 if (bprm->cred->uid != current_euid() ||
1040 bprm->cred->gid != current_egid()) {
1041 current->pdeath_signal = 0;
1042 } else if (file_permission(bprm->file, MAY_READ) ||
1043 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1044 set_dumpable(current->mm, suid_dumpable);
1048 * Flush performance counters when crossing a
1051 if (!get_dumpable(current->mm))
1052 perf_event_exit_task(current);
1054 /* An exec changes our domain. We are no longer part of the thread
1057 current->self_exec_id++;
1059 flush_signal_handlers(current, 0);
1060 flush_old_files(current->files);
1062 EXPORT_SYMBOL(setup_new_exec);
1065 * Prepare credentials and lock ->cred_guard_mutex.
1066 * install_exec_creds() commits the new creds and drops the lock.
1067 * Or, if exec fails before, free_bprm() should release ->cred and
1070 int prepare_bprm_creds(struct linux_binprm *bprm)
1072 if (mutex_lock_interruptible(¤t->cred_guard_mutex))
1073 return -ERESTARTNOINTR;
1075 bprm->cred = prepare_exec_creds();
1076 if (likely(bprm->cred))
1079 mutex_unlock(¤t->cred_guard_mutex);
1083 void free_bprm(struct linux_binprm *bprm)
1085 free_arg_pages(bprm);
1087 mutex_unlock(¤t->cred_guard_mutex);
1088 abort_creds(bprm->cred);
1094 * install the new credentials for this executable
1096 void install_exec_creds(struct linux_binprm *bprm)
1098 security_bprm_committing_creds(bprm);
1100 commit_creds(bprm->cred);
1103 * cred_guard_mutex must be held at least to this point to prevent
1104 * ptrace_attach() from altering our determination of the task's
1105 * credentials; any time after this it may be unlocked.
1107 security_bprm_committed_creds(bprm);
1108 mutex_unlock(¤t->cred_guard_mutex);
1110 EXPORT_SYMBOL(install_exec_creds);
1113 * determine how safe it is to execute the proposed program
1114 * - the caller must hold current->cred_guard_mutex to protect against
1117 int check_unsafe_exec(struct linux_binprm *bprm)
1119 struct task_struct *p = current, *t;
1123 bprm->unsafe = tracehook_unsafe_exec(p);
1126 spin_lock(&p->fs->lock);
1128 for (t = next_thread(p); t != p; t = next_thread(t)) {
1134 if (p->fs->users > n_fs) {
1135 bprm->unsafe |= LSM_UNSAFE_SHARE;
1138 if (!p->fs->in_exec) {
1143 spin_unlock(&p->fs->lock);
1149 * Fill the binprm structure from the inode.
1150 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1152 * This may be called multiple times for binary chains (scripts for example).
1154 int prepare_binprm(struct linux_binprm *bprm)
1157 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1160 mode = inode->i_mode;
1161 if (bprm->file->f_op == NULL)
1164 /* clear any previous set[ug]id data from a previous binary */
1165 bprm->cred->euid = current_euid();
1166 bprm->cred->egid = current_egid();
1168 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1170 if (mode & S_ISUID) {
1171 bprm->per_clear |= PER_CLEAR_ON_SETID;
1172 bprm->cred->euid = inode->i_uid;
1177 * If setgid is set but no group execute bit then this
1178 * is a candidate for mandatory locking, not a setgid
1181 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1182 bprm->per_clear |= PER_CLEAR_ON_SETID;
1183 bprm->cred->egid = inode->i_gid;
1187 /* fill in binprm security blob */
1188 retval = security_bprm_set_creds(bprm);
1191 bprm->cred_prepared = 1;
1193 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1194 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1197 EXPORT_SYMBOL(prepare_binprm);
1200 * Arguments are '\0' separated strings found at the location bprm->p
1201 * points to; chop off the first by relocating brpm->p to right after
1202 * the first '\0' encountered.
1204 int remove_arg_zero(struct linux_binprm *bprm)
1207 unsigned long offset;
1215 offset = bprm->p & ~PAGE_MASK;
1216 page = get_arg_page(bprm, bprm->p, 0);
1221 kaddr = kmap_atomic(page, KM_USER0);
1223 for (; offset < PAGE_SIZE && kaddr[offset];
1224 offset++, bprm->p++)
1227 kunmap_atomic(kaddr, KM_USER0);
1230 if (offset == PAGE_SIZE)
1231 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1232 } while (offset == PAGE_SIZE);
1241 EXPORT_SYMBOL(remove_arg_zero);
1244 * cycle the list of binary formats handler, until one recognizes the image
1246 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1248 unsigned int depth = bprm->recursion_depth;
1250 struct linux_binfmt *fmt;
1252 retval = security_bprm_check(bprm);
1256 /* kernel module loader fixup */
1257 /* so we don't try to load run modprobe in kernel space. */
1260 retval = audit_bprm(bprm);
1265 for (try=0; try<2; try++) {
1266 read_lock(&binfmt_lock);
1267 list_for_each_entry(fmt, &formats, lh) {
1268 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1271 if (!try_module_get(fmt->module))
1273 read_unlock(&binfmt_lock);
1274 retval = fn(bprm, regs);
1276 * Restore the depth counter to its starting value
1277 * in this call, so we don't have to rely on every
1278 * load_binary function to restore it on return.
1280 bprm->recursion_depth = depth;
1283 tracehook_report_exec(fmt, bprm, regs);
1285 allow_write_access(bprm->file);
1289 current->did_exec = 1;
1290 proc_exec_connector(current);
1293 read_lock(&binfmt_lock);
1295 if (retval != -ENOEXEC || bprm->mm == NULL)
1298 read_unlock(&binfmt_lock);
1302 read_unlock(&binfmt_lock);
1303 if (retval != -ENOEXEC || bprm->mm == NULL) {
1305 #ifdef CONFIG_MODULES
1307 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1308 if (printable(bprm->buf[0]) &&
1309 printable(bprm->buf[1]) &&
1310 printable(bprm->buf[2]) &&
1311 printable(bprm->buf[3]))
1312 break; /* -ENOEXEC */
1313 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1320 EXPORT_SYMBOL(search_binary_handler);
1323 * sys_execve() executes a new program.
1325 int do_execve(const char * filename,
1326 const char __user *const __user *argv,
1327 const char __user *const __user *envp,
1328 struct pt_regs * regs)
1330 struct linux_binprm *bprm;
1332 struct files_struct *displaced;
1336 retval = unshare_files(&displaced);
1341 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1345 retval = prepare_bprm_creds(bprm);
1349 retval = check_unsafe_exec(bprm);
1352 clear_in_exec = retval;
1353 current->in_execve = 1;
1355 file = open_exec(filename);
1356 retval = PTR_ERR(file);
1363 bprm->filename = filename;
1364 bprm->interp = filename;
1366 retval = bprm_mm_init(bprm);
1370 bprm->argc = count(argv, MAX_ARG_STRINGS);
1371 if ((retval = bprm->argc) < 0)
1374 bprm->envc = count(envp, MAX_ARG_STRINGS);
1375 if ((retval = bprm->envc) < 0)
1378 retval = prepare_binprm(bprm);
1382 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1386 bprm->exec = bprm->p;
1387 retval = copy_strings(bprm->envc, envp, bprm);
1391 retval = copy_strings(bprm->argc, argv, bprm);
1395 current->flags &= ~PF_KTHREAD;
1396 retval = search_binary_handler(bprm,regs);
1400 /* execve succeeded */
1401 current->fs->in_exec = 0;
1402 current->in_execve = 0;
1403 acct_update_integrals(current);
1406 put_files_struct(displaced);
1415 allow_write_access(bprm->file);
1421 current->fs->in_exec = 0;
1422 current->in_execve = 0;
1429 reset_files_struct(displaced);
1434 void set_binfmt(struct linux_binfmt *new)
1436 struct mm_struct *mm = current->mm;
1439 module_put(mm->binfmt->module);
1443 __module_get(new->module);
1446 EXPORT_SYMBOL(set_binfmt);
1448 /* format_corename will inspect the pattern parameter, and output a
1449 * name into corename, which must have space for at least
1450 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1452 static int format_corename(char *corename, long signr)
1454 const struct cred *cred = current_cred();
1455 const char *pat_ptr = core_pattern;
1456 int ispipe = (*pat_ptr == '|');
1457 char *out_ptr = corename;
1458 char *const out_end = corename + CORENAME_MAX_SIZE;
1460 int pid_in_pattern = 0;
1462 /* Repeat as long as we have more pattern to process and more output
1465 if (*pat_ptr != '%') {
1466 if (out_ptr == out_end)
1468 *out_ptr++ = *pat_ptr++;
1470 switch (*++pat_ptr) {
1473 /* Double percent, output one percent */
1475 if (out_ptr == out_end)
1482 rc = snprintf(out_ptr, out_end - out_ptr,
1483 "%d", task_tgid_vnr(current));
1484 if (rc > out_end - out_ptr)
1490 rc = snprintf(out_ptr, out_end - out_ptr,
1492 if (rc > out_end - out_ptr)
1498 rc = snprintf(out_ptr, out_end - out_ptr,
1500 if (rc > out_end - out_ptr)
1504 /* signal that caused the coredump */
1506 rc = snprintf(out_ptr, out_end - out_ptr,
1508 if (rc > out_end - out_ptr)
1512 /* UNIX time of coredump */
1515 do_gettimeofday(&tv);
1516 rc = snprintf(out_ptr, out_end - out_ptr,
1518 if (rc > out_end - out_ptr)
1525 down_read(&uts_sem);
1526 rc = snprintf(out_ptr, out_end - out_ptr,
1527 "%s", utsname()->nodename);
1529 if (rc > out_end - out_ptr)
1535 rc = snprintf(out_ptr, out_end - out_ptr,
1536 "%s", current->comm);
1537 if (rc > out_end - out_ptr)
1541 /* core limit size */
1543 rc = snprintf(out_ptr, out_end - out_ptr,
1544 "%lu", rlimit(RLIMIT_CORE));
1545 if (rc > out_end - out_ptr)
1555 /* Backward compatibility with core_uses_pid:
1557 * If core_pattern does not include a %p (as is the default)
1558 * and core_uses_pid is set, then .%pid will be appended to
1559 * the filename. Do not do this for piped commands. */
1560 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1561 rc = snprintf(out_ptr, out_end - out_ptr,
1562 ".%d", task_tgid_vnr(current));
1563 if (rc > out_end - out_ptr)
1572 static int zap_process(struct task_struct *start, int exit_code)
1574 struct task_struct *t;
1577 start->signal->flags = SIGNAL_GROUP_EXIT;
1578 start->signal->group_exit_code = exit_code;
1579 start->signal->group_stop_count = 0;
1583 if (t != current && t->mm) {
1584 sigaddset(&t->pending.signal, SIGKILL);
1585 signal_wake_up(t, 1);
1588 } while_each_thread(start, t);
1593 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1594 struct core_state *core_state, int exit_code)
1596 struct task_struct *g, *p;
1597 unsigned long flags;
1600 spin_lock_irq(&tsk->sighand->siglock);
1601 if (!signal_group_exit(tsk->signal)) {
1602 mm->core_state = core_state;
1603 nr = zap_process(tsk, exit_code);
1605 spin_unlock_irq(&tsk->sighand->siglock);
1606 if (unlikely(nr < 0))
1609 if (atomic_read(&mm->mm_users) == nr + 1)
1612 * We should find and kill all tasks which use this mm, and we should
1613 * count them correctly into ->nr_threads. We don't take tasklist
1614 * lock, but this is safe wrt:
1617 * None of sub-threads can fork after zap_process(leader). All
1618 * processes which were created before this point should be
1619 * visible to zap_threads() because copy_process() adds the new
1620 * process to the tail of init_task.tasks list, and lock/unlock
1621 * of ->siglock provides a memory barrier.
1624 * The caller holds mm->mmap_sem. This means that the task which
1625 * uses this mm can't pass exit_mm(), so it can't exit or clear
1629 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1630 * we must see either old or new leader, this does not matter.
1631 * However, it can change p->sighand, so lock_task_sighand(p)
1632 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1635 * Note also that "g" can be the old leader with ->mm == NULL
1636 * and already unhashed and thus removed from ->thread_group.
1637 * This is OK, __unhash_process()->list_del_rcu() does not
1638 * clear the ->next pointer, we will find the new leader via
1642 for_each_process(g) {
1643 if (g == tsk->group_leader)
1645 if (g->flags & PF_KTHREAD)
1650 if (unlikely(p->mm == mm)) {
1651 lock_task_sighand(p, &flags);
1652 nr += zap_process(p, exit_code);
1653 unlock_task_sighand(p, &flags);
1657 } while_each_thread(g, p);
1661 atomic_set(&core_state->nr_threads, nr);
1665 static int coredump_wait(int exit_code, struct core_state *core_state)
1667 struct task_struct *tsk = current;
1668 struct mm_struct *mm = tsk->mm;
1669 struct completion *vfork_done;
1670 int core_waiters = -EBUSY;
1672 init_completion(&core_state->startup);
1673 core_state->dumper.task = tsk;
1674 core_state->dumper.next = NULL;
1676 down_write(&mm->mmap_sem);
1677 if (!mm->core_state)
1678 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1679 up_write(&mm->mmap_sem);
1681 if (unlikely(core_waiters < 0))
1685 * Make sure nobody is waiting for us to release the VM,
1686 * otherwise we can deadlock when we wait on each other
1688 vfork_done = tsk->vfork_done;
1690 tsk->vfork_done = NULL;
1691 complete(vfork_done);
1695 wait_for_completion(&core_state->startup);
1697 return core_waiters;
1700 static void coredump_finish(struct mm_struct *mm)
1702 struct core_thread *curr, *next;
1703 struct task_struct *task;
1705 next = mm->core_state->dumper.next;
1706 while ((curr = next) != NULL) {
1710 * see exit_mm(), curr->task must not see
1711 * ->task == NULL before we read ->next.
1715 wake_up_process(task);
1718 mm->core_state = NULL;
1722 * set_dumpable converts traditional three-value dumpable to two flags and
1723 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1724 * these bits are not changed atomically. So get_dumpable can observe the
1725 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1726 * return either old dumpable or new one by paying attention to the order of
1727 * modifying the bits.
1729 * dumpable | mm->flags (binary)
1730 * old new | initial interim final
1731 * ---------+-----------------------
1739 * (*) get_dumpable regards interim value of 10 as 11.
1741 void set_dumpable(struct mm_struct *mm, int value)
1745 clear_bit(MMF_DUMPABLE, &mm->flags);
1747 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1750 set_bit(MMF_DUMPABLE, &mm->flags);
1752 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1755 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1757 set_bit(MMF_DUMPABLE, &mm->flags);
1762 static int __get_dumpable(unsigned long mm_flags)
1766 ret = mm_flags & MMF_DUMPABLE_MASK;
1767 return (ret >= 2) ? 2 : ret;
1770 int get_dumpable(struct mm_struct *mm)
1772 return __get_dumpable(mm->flags);
1775 static void wait_for_dump_helpers(struct file *file)
1777 struct pipe_inode_info *pipe;
1779 pipe = file->f_path.dentry->d_inode->i_pipe;
1785 while ((pipe->readers > 1) && (!signal_pending(current))) {
1786 wake_up_interruptible_sync(&pipe->wait);
1787 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1800 * helper function to customize the process used
1801 * to collect the core in userspace. Specifically
1802 * it sets up a pipe and installs it as fd 0 (stdin)
1803 * for the process. Returns 0 on success, or
1804 * PTR_ERR on failure.
1805 * Note that it also sets the core limit to 1. This
1806 * is a special value that we use to trap recursive
1809 static int umh_pipe_setup(struct subprocess_info *info)
1811 struct file *rp, *wp;
1812 struct fdtable *fdt;
1813 struct coredump_params *cp = (struct coredump_params *)info->data;
1814 struct files_struct *cf = current->files;
1816 wp = create_write_pipe(0);
1820 rp = create_read_pipe(wp, 0);
1822 free_write_pipe(wp);
1830 spin_lock(&cf->file_lock);
1831 fdt = files_fdtable(cf);
1832 FD_SET(0, fdt->open_fds);
1833 FD_CLR(0, fdt->close_on_exec);
1834 spin_unlock(&cf->file_lock);
1836 /* and disallow core files too */
1837 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1842 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1844 struct core_state core_state;
1845 char corename[CORENAME_MAX_SIZE + 1];
1846 struct mm_struct *mm = current->mm;
1847 struct linux_binfmt * binfmt;
1848 const struct cred *old_cred;
1853 static atomic_t core_dump_count = ATOMIC_INIT(0);
1854 struct coredump_params cprm = {
1857 .limit = rlimit(RLIMIT_CORE),
1859 * We must use the same mm->flags while dumping core to avoid
1860 * inconsistency of bit flags, since this flag is not protected
1863 .mm_flags = mm->flags,
1866 audit_core_dumps(signr);
1868 binfmt = mm->binfmt;
1869 if (!binfmt || !binfmt->core_dump)
1871 if (!__get_dumpable(cprm.mm_flags))
1874 cred = prepare_creds();
1878 * We cannot trust fsuid as being the "true" uid of the
1879 * process nor do we know its entire history. We only know it
1880 * was tainted so we dump it as root in mode 2.
1882 if (__get_dumpable(cprm.mm_flags) == 2) {
1883 /* Setuid core dump mode */
1884 flag = O_EXCL; /* Stop rewrite attacks */
1885 cred->fsuid = 0; /* Dump root private */
1888 retval = coredump_wait(exit_code, &core_state);
1892 old_cred = override_creds(cred);
1895 * Clear any false indication of pending signals that might
1896 * be seen by the filesystem code called to write the core file.
1898 clear_thread_flag(TIF_SIGPENDING);
1900 ispipe = format_corename(corename, signr);
1906 if (cprm.limit == 1) {
1908 * Normally core limits are irrelevant to pipes, since
1909 * we're not writing to the file system, but we use
1910 * cprm.limit of 1 here as a speacial value. Any
1911 * non-1 limit gets set to RLIM_INFINITY below, but
1912 * a limit of 0 skips the dump. This is a consistent
1913 * way to catch recursive crashes. We can still crash
1914 * if the core_pattern binary sets RLIM_CORE = !1
1915 * but it runs as root, and can do lots of stupid things
1916 * Note that we use task_tgid_vnr here to grab the pid
1917 * of the process group leader. That way we get the
1918 * right pid if a thread in a multi-threaded
1919 * core_pattern process dies.
1922 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1923 task_tgid_vnr(current), current->comm);
1924 printk(KERN_WARNING "Aborting core\n");
1927 cprm.limit = RLIM_INFINITY;
1929 dump_count = atomic_inc_return(&core_dump_count);
1930 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1931 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1932 task_tgid_vnr(current), current->comm);
1933 printk(KERN_WARNING "Skipping core dump\n");
1934 goto fail_dropcount;
1937 helper_argv = argv_split(GFP_KERNEL, corename+1, NULL);
1939 printk(KERN_WARNING "%s failed to allocate memory\n",
1941 goto fail_dropcount;
1944 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1945 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1947 argv_free(helper_argv);
1949 printk(KERN_INFO "Core dump to %s pipe failed\n",
1954 struct inode *inode;
1956 if (cprm.limit < binfmt->min_coredump)
1959 cprm.file = filp_open(corename,
1960 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1962 if (IS_ERR(cprm.file))
1965 inode = cprm.file->f_path.dentry->d_inode;
1966 if (inode->i_nlink > 1)
1968 if (d_unhashed(cprm.file->f_path.dentry))
1971 * AK: actually i see no reason to not allow this for named
1972 * pipes etc, but keep the previous behaviour for now.
1974 if (!S_ISREG(inode->i_mode))
1977 * Dont allow local users get cute and trick others to coredump
1978 * into their pre-created files.
1980 if (inode->i_uid != current_fsuid())
1982 if (!cprm.file->f_op || !cprm.file->f_op->write)
1984 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
1988 retval = binfmt->core_dump(&cprm);
1990 current->signal->group_exit_code |= 0x80;
1992 if (ispipe && core_pipe_limit)
1993 wait_for_dump_helpers(cprm.file);
1996 filp_close(cprm.file, NULL);
1999 atomic_dec(&core_dump_count);
2001 coredump_finish(mm);
2002 revert_creds(old_cred);