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/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/audit.h>
51 #include <linux/tracehook.h>
52 #include <linux/kmod.h>
53 #include <linux/fsnotify.h>
54 #include <linux/fs_struct.h>
55 #include <linux/pipe_fs_i.h>
56 #include <linux/oom.h>
57 #include <linux/compat.h>
59 #include <trace/events/fs.h>
61 #include <asm/uaccess.h>
62 #include <asm/mmu_context.h>
67 char core_pattern[CORENAME_MAX_SIZE] = "core";
68 unsigned int core_pipe_limit;
69 int suid_dumpable = 0;
75 static atomic_t call_count = ATOMIC_INIT(1);
77 /* The maximal length of core_pattern is also specified in sysctl.c */
79 static LIST_HEAD(formats);
80 static DEFINE_RWLOCK(binfmt_lock);
82 int __register_binfmt(struct linux_binfmt * fmt, int insert)
86 write_lock(&binfmt_lock);
87 insert ? list_add(&fmt->lh, &formats) :
88 list_add_tail(&fmt->lh, &formats);
89 write_unlock(&binfmt_lock);
93 EXPORT_SYMBOL(__register_binfmt);
95 void unregister_binfmt(struct linux_binfmt * fmt)
97 write_lock(&binfmt_lock);
99 write_unlock(&binfmt_lock);
102 EXPORT_SYMBOL(unregister_binfmt);
104 static inline void put_binfmt(struct linux_binfmt * fmt)
106 module_put(fmt->module);
110 * Note that a shared library must be both readable and executable due to
113 * Also note that we take the address to load from from the file itself.
115 SYSCALL_DEFINE1(uselib, const char __user *, library)
118 char *tmp = getname(library);
119 int error = PTR_ERR(tmp);
120 static const struct open_flags uselib_flags = {
121 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
122 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
123 .intent = LOOKUP_OPEN
129 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
131 error = PTR_ERR(file);
136 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
140 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
147 struct linux_binfmt * fmt;
149 read_lock(&binfmt_lock);
150 list_for_each_entry(fmt, &formats, lh) {
151 if (!fmt->load_shlib)
153 if (!try_module_get(fmt->module))
155 read_unlock(&binfmt_lock);
156 error = fmt->load_shlib(file);
157 read_lock(&binfmt_lock);
159 if (error != -ENOEXEC)
162 read_unlock(&binfmt_lock);
172 * The nascent bprm->mm is not visible until exec_mmap() but it can
173 * use a lot of memory, account these pages in current->mm temporary
174 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
175 * change the counter back via acct_arg_size(0).
177 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
179 struct mm_struct *mm = current->mm;
180 long diff = (long)(pages - bprm->vma_pages);
185 bprm->vma_pages = pages;
186 add_mm_counter(mm, MM_ANONPAGES, diff);
189 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
195 #ifdef CONFIG_STACK_GROWSUP
197 ret = expand_downwards(bprm->vma, pos);
202 ret = get_user_pages(current, bprm->mm, pos,
203 1, write, 1, &page, NULL);
208 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
211 acct_arg_size(bprm, size / PAGE_SIZE);
214 * We've historically supported up to 32 pages (ARG_MAX)
215 * of argument strings even with small stacks
221 * Limit to 1/4-th the stack size for the argv+env strings.
223 * - the remaining binfmt code will not run out of stack space,
224 * - the program will have a reasonable amount of stack left
227 rlim = current->signal->rlim;
228 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
237 static void put_arg_page(struct page *page)
242 static void free_arg_page(struct linux_binprm *bprm, int i)
246 static void free_arg_pages(struct linux_binprm *bprm)
250 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
253 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
256 static int __bprm_mm_init(struct linux_binprm *bprm)
259 struct vm_area_struct *vma = NULL;
260 struct mm_struct *mm = bprm->mm;
262 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
266 down_write(&mm->mmap_sem);
270 * Place the stack at the largest stack address the architecture
271 * supports. Later, we'll move this to an appropriate place. We don't
272 * use STACK_TOP because that can depend on attributes which aren't
275 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
276 vma->vm_end = STACK_TOP_MAX;
277 vma->vm_start = vma->vm_end - PAGE_SIZE;
278 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
279 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
280 INIT_LIST_HEAD(&vma->anon_vma_chain);
282 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
286 err = insert_vm_struct(mm, vma);
290 mm->stack_vm = mm->total_vm = 1;
291 up_write(&mm->mmap_sem);
292 bprm->p = vma->vm_end - sizeof(void *);
295 up_write(&mm->mmap_sem);
297 kmem_cache_free(vm_area_cachep, vma);
301 static bool valid_arg_len(struct linux_binprm *bprm, long len)
303 return len <= MAX_ARG_STRLEN;
308 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
312 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
317 page = bprm->page[pos / PAGE_SIZE];
318 if (!page && write) {
319 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
322 bprm->page[pos / PAGE_SIZE] = page;
328 static void put_arg_page(struct page *page)
332 static void free_arg_page(struct linux_binprm *bprm, int i)
335 __free_page(bprm->page[i]);
336 bprm->page[i] = NULL;
340 static void free_arg_pages(struct linux_binprm *bprm)
344 for (i = 0; i < MAX_ARG_PAGES; i++)
345 free_arg_page(bprm, i);
348 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
353 static int __bprm_mm_init(struct linux_binprm *bprm)
355 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
359 static bool valid_arg_len(struct linux_binprm *bprm, long len)
361 return len <= bprm->p;
364 #endif /* CONFIG_MMU */
367 * Create a new mm_struct and populate it with a temporary stack
368 * vm_area_struct. We don't have enough context at this point to set the stack
369 * flags, permissions, and offset, so we use temporary values. We'll update
370 * them later in setup_arg_pages().
372 int bprm_mm_init(struct linux_binprm *bprm)
375 struct mm_struct *mm = NULL;
377 bprm->mm = mm = mm_alloc();
382 err = init_new_context(current, mm);
386 err = __bprm_mm_init(bprm);
401 struct user_arg_ptr {
406 const char __user *const __user *native;
408 compat_uptr_t __user *compat;
413 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
415 const char __user *native;
418 if (unlikely(argv.is_compat)) {
419 compat_uptr_t compat;
421 if (get_user(compat, argv.ptr.compat + nr))
422 return ERR_PTR(-EFAULT);
424 return compat_ptr(compat);
428 if (get_user(native, argv.ptr.native + nr))
429 return ERR_PTR(-EFAULT);
435 * count() counts the number of strings in array ARGV.
437 static int count(struct user_arg_ptr argv, int max)
441 if (argv.ptr.native != NULL) {
443 const char __user *p = get_user_arg_ptr(argv, i);
454 if (fatal_signal_pending(current))
455 return -ERESTARTNOHAND;
463 * 'copy_strings()' copies argument/environment strings from the old
464 * processes's memory to the new process's stack. The call to get_user_pages()
465 * ensures the destination page is created and not swapped out.
467 static int copy_strings(int argc, struct user_arg_ptr argv,
468 struct linux_binprm *bprm)
470 struct page *kmapped_page = NULL;
472 unsigned long kpos = 0;
476 const char __user *str;
481 str = get_user_arg_ptr(argv, argc);
485 len = strnlen_user(str, MAX_ARG_STRLEN);
490 if (!valid_arg_len(bprm, len))
493 /* We're going to work our way backwords. */
499 int offset, bytes_to_copy;
501 if (fatal_signal_pending(current)) {
502 ret = -ERESTARTNOHAND;
507 offset = pos % PAGE_SIZE;
511 bytes_to_copy = offset;
512 if (bytes_to_copy > len)
515 offset -= bytes_to_copy;
516 pos -= bytes_to_copy;
517 str -= bytes_to_copy;
518 len -= bytes_to_copy;
520 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
523 page = get_arg_page(bprm, pos, 1);
530 flush_kernel_dcache_page(kmapped_page);
531 kunmap(kmapped_page);
532 put_arg_page(kmapped_page);
535 kaddr = kmap(kmapped_page);
536 kpos = pos & PAGE_MASK;
537 flush_arg_page(bprm, kpos, kmapped_page);
539 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
548 flush_kernel_dcache_page(kmapped_page);
549 kunmap(kmapped_page);
550 put_arg_page(kmapped_page);
556 * Like copy_strings, but get argv and its values from kernel memory.
558 int copy_strings_kernel(int argc, const char *const *__argv,
559 struct linux_binprm *bprm)
562 mm_segment_t oldfs = get_fs();
563 struct user_arg_ptr argv = {
564 .ptr.native = (const char __user *const __user *)__argv,
568 r = copy_strings(argc, argv, bprm);
573 EXPORT_SYMBOL(copy_strings_kernel);
578 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
579 * the binfmt code determines where the new stack should reside, we shift it to
580 * its final location. The process proceeds as follows:
582 * 1) Use shift to calculate the new vma endpoints.
583 * 2) Extend vma to cover both the old and new ranges. This ensures the
584 * arguments passed to subsequent functions are consistent.
585 * 3) Move vma's page tables to the new range.
586 * 4) Free up any cleared pgd range.
587 * 5) Shrink the vma to cover only the new range.
589 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
591 struct mm_struct *mm = vma->vm_mm;
592 unsigned long old_start = vma->vm_start;
593 unsigned long old_end = vma->vm_end;
594 unsigned long length = old_end - old_start;
595 unsigned long new_start = old_start - shift;
596 unsigned long new_end = old_end - shift;
597 struct mmu_gather tlb;
599 BUG_ON(new_start > new_end);
602 * ensure there are no vmas between where we want to go
605 if (vma != find_vma(mm, new_start))
609 * cover the whole range: [new_start, old_end)
611 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
615 * move the page tables downwards, on failure we rely on
616 * process cleanup to remove whatever mess we made.
618 if (length != move_page_tables(vma, old_start,
619 vma, new_start, length))
623 tlb_gather_mmu(&tlb, mm, 0);
624 if (new_end > old_start) {
626 * when the old and new regions overlap clear from new_end.
628 free_pgd_range(&tlb, new_end, old_end, new_end,
629 vma->vm_next ? vma->vm_next->vm_start : 0);
632 * otherwise, clean from old_start; this is done to not touch
633 * the address space in [new_end, old_start) some architectures
634 * have constraints on va-space that make this illegal (IA64) -
635 * for the others its just a little faster.
637 free_pgd_range(&tlb, old_start, old_end, new_end,
638 vma->vm_next ? vma->vm_next->vm_start : 0);
640 tlb_finish_mmu(&tlb, new_end, old_end);
643 * Shrink the vma to just the new range. Always succeeds.
645 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
651 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
652 * the stack is optionally relocated, and some extra space is added.
654 int setup_arg_pages(struct linux_binprm *bprm,
655 unsigned long stack_top,
656 int executable_stack)
659 unsigned long stack_shift;
660 struct mm_struct *mm = current->mm;
661 struct vm_area_struct *vma = bprm->vma;
662 struct vm_area_struct *prev = NULL;
663 unsigned long vm_flags;
664 unsigned long stack_base;
665 unsigned long stack_size;
666 unsigned long stack_expand;
667 unsigned long rlim_stack;
669 #ifdef CONFIG_STACK_GROWSUP
670 /* Limit stack size to 1GB */
671 stack_base = rlimit_max(RLIMIT_STACK);
672 if (stack_base > (1 << 30))
673 stack_base = 1 << 30;
675 /* Make sure we didn't let the argument array grow too large. */
676 if (vma->vm_end - vma->vm_start > stack_base)
679 stack_base = PAGE_ALIGN(stack_top - stack_base);
681 stack_shift = vma->vm_start - stack_base;
682 mm->arg_start = bprm->p - stack_shift;
683 bprm->p = vma->vm_end - stack_shift;
685 stack_top = arch_align_stack(stack_top);
686 stack_top = PAGE_ALIGN(stack_top);
688 if (unlikely(stack_top < mmap_min_addr) ||
689 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
692 stack_shift = vma->vm_end - stack_top;
694 bprm->p -= stack_shift;
695 mm->arg_start = bprm->p;
699 bprm->loader -= stack_shift;
700 bprm->exec -= stack_shift;
702 down_write(&mm->mmap_sem);
703 vm_flags = VM_STACK_FLAGS;
706 * Adjust stack execute permissions; explicitly enable for
707 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
708 * (arch default) otherwise.
710 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
712 else if (executable_stack == EXSTACK_DISABLE_X)
713 vm_flags &= ~VM_EXEC;
714 vm_flags |= mm->def_flags;
715 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
717 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
723 /* Move stack pages down in memory. */
725 ret = shift_arg_pages(vma, stack_shift);
730 /* mprotect_fixup is overkill to remove the temporary stack flags */
731 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
733 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
734 stack_size = vma->vm_end - vma->vm_start;
736 * Align this down to a page boundary as expand_stack
739 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
740 #ifdef CONFIG_STACK_GROWSUP
741 if (stack_size + stack_expand > rlim_stack)
742 stack_base = vma->vm_start + rlim_stack;
744 stack_base = vma->vm_end + stack_expand;
746 if (stack_size + stack_expand > rlim_stack)
747 stack_base = vma->vm_end - rlim_stack;
749 stack_base = vma->vm_start - stack_expand;
751 current->mm->start_stack = bprm->p;
752 ret = expand_stack(vma, stack_base);
757 up_write(&mm->mmap_sem);
760 EXPORT_SYMBOL(setup_arg_pages);
762 #endif /* CONFIG_MMU */
764 struct file *open_exec(const char *name)
768 static const struct open_flags open_exec_flags = {
769 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
770 .acc_mode = MAY_EXEC | MAY_OPEN,
771 .intent = LOOKUP_OPEN
774 file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
779 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
782 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
787 trace_open_exec(name);
789 err = deny_write_access(file);
800 EXPORT_SYMBOL(open_exec);
802 int kernel_read(struct file *file, loff_t offset,
803 char *addr, unsigned long count)
811 /* The cast to a user pointer is valid due to the set_fs() */
812 result = vfs_read(file, (void __user *)addr, count, &pos);
817 EXPORT_SYMBOL(kernel_read);
819 static int exec_mmap(struct mm_struct *mm)
821 struct task_struct *tsk;
822 struct mm_struct * old_mm, *active_mm;
824 /* Notify parent that we're no longer interested in the old VM */
826 old_mm = current->mm;
827 sync_mm_rss(tsk, old_mm);
828 mm_release(tsk, old_mm);
832 * Make sure that if there is a core dump in progress
833 * for the old mm, we get out and die instead of going
834 * through with the exec. We must hold mmap_sem around
835 * checking core_state and changing tsk->mm.
837 down_read(&old_mm->mmap_sem);
838 if (unlikely(old_mm->core_state)) {
839 up_read(&old_mm->mmap_sem);
844 active_mm = tsk->active_mm;
847 activate_mm(active_mm, mm);
849 arch_pick_mmap_layout(mm);
851 up_read(&old_mm->mmap_sem);
852 BUG_ON(active_mm != old_mm);
853 mm_update_next_owner(old_mm);
862 * This function makes sure the current process has its own signal table,
863 * so that flush_signal_handlers can later reset the handlers without
864 * disturbing other processes. (Other processes might share the signal
865 * table via the CLONE_SIGHAND option to clone().)
867 static int de_thread(struct task_struct *tsk)
869 struct signal_struct *sig = tsk->signal;
870 struct sighand_struct *oldsighand = tsk->sighand;
871 spinlock_t *lock = &oldsighand->siglock;
873 if (thread_group_empty(tsk))
874 goto no_thread_group;
877 * Kill all other threads in the thread group.
880 if (signal_group_exit(sig)) {
882 * Another group action in progress, just
883 * return so that the signal is processed.
885 spin_unlock_irq(lock);
889 sig->group_exit_task = tsk;
890 sig->notify_count = zap_other_threads(tsk);
891 if (!thread_group_leader(tsk))
894 while (sig->notify_count) {
895 __set_current_state(TASK_UNINTERRUPTIBLE);
896 spin_unlock_irq(lock);
900 spin_unlock_irq(lock);
903 * At this point all other threads have exited, all we have to
904 * do is to wait for the thread group leader to become inactive,
905 * and to assume its PID:
907 if (!thread_group_leader(tsk)) {
908 struct task_struct *leader = tsk->group_leader;
910 sig->notify_count = -1; /* for exit_notify() */
912 write_lock_irq(&tasklist_lock);
913 if (likely(leader->exit_state))
915 __set_current_state(TASK_UNINTERRUPTIBLE);
916 write_unlock_irq(&tasklist_lock);
921 * The only record we have of the real-time age of a
922 * process, regardless of execs it's done, is start_time.
923 * All the past CPU time is accumulated in signal_struct
924 * from sister threads now dead. But in this non-leader
925 * exec, nothing survives from the original leader thread,
926 * whose birth marks the true age of this process now.
927 * When we take on its identity by switching to its PID, we
928 * also take its birthdate (always earlier than our own).
930 tsk->start_time = leader->start_time;
932 BUG_ON(!same_thread_group(leader, tsk));
933 BUG_ON(has_group_leader_pid(tsk));
935 * An exec() starts a new thread group with the
936 * TGID of the previous thread group. Rehash the
937 * two threads with a switched PID, and release
938 * the former thread group leader:
941 /* Become a process group leader with the old leader's pid.
942 * The old leader becomes a thread of the this thread group.
943 * Note: The old leader also uses this pid until release_task
944 * is called. Odd but simple and correct.
946 detach_pid(tsk, PIDTYPE_PID);
947 tsk->pid = leader->pid;
948 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
949 transfer_pid(leader, tsk, PIDTYPE_PGID);
950 transfer_pid(leader, tsk, PIDTYPE_SID);
952 list_replace_rcu(&leader->tasks, &tsk->tasks);
953 list_replace_init(&leader->sibling, &tsk->sibling);
955 tsk->group_leader = tsk;
956 leader->group_leader = tsk;
958 tsk->exit_signal = SIGCHLD;
959 leader->exit_signal = -1;
961 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
962 leader->exit_state = EXIT_DEAD;
965 * We are going to release_task()->ptrace_unlink() silently,
966 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
967 * the tracer wont't block again waiting for this thread.
969 if (unlikely(leader->ptrace))
970 __wake_up_parent(leader, leader->parent);
971 write_unlock_irq(&tasklist_lock);
973 release_task(leader);
976 sig->group_exit_task = NULL;
977 sig->notify_count = 0;
981 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
984 flush_itimer_signals();
986 if (atomic_read(&oldsighand->count) != 1) {
987 struct sighand_struct *newsighand;
989 * This ->sighand is shared with the CLONE_SIGHAND
990 * but not CLONE_THREAD task, switch to the new one.
992 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
996 atomic_set(&newsighand->count, 1);
997 memcpy(newsighand->action, oldsighand->action,
998 sizeof(newsighand->action));
1000 write_lock_irq(&tasklist_lock);
1001 spin_lock(&oldsighand->siglock);
1002 rcu_assign_pointer(tsk->sighand, newsighand);
1003 spin_unlock(&oldsighand->siglock);
1004 write_unlock_irq(&tasklist_lock);
1006 __cleanup_sighand(oldsighand);
1009 BUG_ON(!thread_group_leader(tsk));
1014 * These functions flushes out all traces of the currently running executable
1015 * so that a new one can be started
1017 static void flush_old_files(struct files_struct * files)
1020 struct fdtable *fdt;
1022 spin_lock(&files->file_lock);
1024 unsigned long set, i;
1028 fdt = files_fdtable(files);
1029 if (i >= fdt->max_fds)
1031 set = fdt->close_on_exec->fds_bits[j];
1034 fdt->close_on_exec->fds_bits[j] = 0;
1035 spin_unlock(&files->file_lock);
1036 for ( ; set ; i++,set >>= 1) {
1041 spin_lock(&files->file_lock);
1044 spin_unlock(&files->file_lock);
1047 char *get_task_comm(char *buf, struct task_struct *tsk)
1049 /* buf must be at least sizeof(tsk->comm) in size */
1051 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1055 EXPORT_SYMBOL_GPL(get_task_comm);
1057 void set_task_comm(struct task_struct *tsk, char *buf)
1062 * Threads may access current->comm without holding
1063 * the task lock, so write the string carefully.
1064 * Readers without a lock may see incomplete new
1065 * names but are safe from non-terminating string reads.
1067 memset(tsk->comm, 0, TASK_COMM_LEN);
1069 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1071 perf_event_comm(tsk);
1074 int flush_old_exec(struct linux_binprm * bprm)
1079 * Make sure we have a private signal table and that
1080 * we are unassociated from the previous thread group.
1082 retval = de_thread(current);
1086 set_mm_exe_file(bprm->mm, bprm->file);
1089 * Release all of the old mmap stuff
1091 acct_arg_size(bprm, 0);
1092 retval = exec_mmap(bprm->mm);
1096 bprm->mm = NULL; /* We're using it now */
1099 current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1101 current->personality &= ~bprm->per_clear;
1108 EXPORT_SYMBOL(flush_old_exec);
1110 void would_dump(struct linux_binprm *bprm, struct file *file)
1112 if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1113 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1115 EXPORT_SYMBOL(would_dump);
1117 void setup_new_exec(struct linux_binprm * bprm)
1121 char tcomm[sizeof(current->comm)];
1123 arch_pick_mmap_layout(current->mm);
1125 /* This is the point of no return */
1126 current->sas_ss_sp = current->sas_ss_size = 0;
1128 if (current_euid() == current_uid() && current_egid() == current_gid())
1129 set_dumpable(current->mm, 1);
1131 set_dumpable(current->mm, suid_dumpable);
1133 name = bprm->filename;
1135 /* Copies the binary name from after last slash */
1136 for (i=0; (ch = *(name++)) != '\0';) {
1138 i = 0; /* overwrite what we wrote */
1140 if (i < (sizeof(tcomm) - 1))
1144 set_task_comm(current, tcomm);
1146 /* Set the new mm task size. We have to do that late because it may
1147 * depend on TIF_32BIT which is only updated in flush_thread() on
1148 * some architectures like powerpc
1150 current->mm->task_size = TASK_SIZE;
1152 /* install the new credentials */
1153 if (bprm->cred->uid != current_euid() ||
1154 bprm->cred->gid != current_egid()) {
1155 current->pdeath_signal = 0;
1157 would_dump(bprm, bprm->file);
1158 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1159 set_dumpable(current->mm, suid_dumpable);
1163 * Flush performance counters when crossing a
1166 if (!get_dumpable(current->mm))
1167 perf_event_exit_task(current);
1169 /* An exec changes our domain. We are no longer part of the thread
1172 current->self_exec_id++;
1174 flush_signal_handlers(current, 0);
1175 flush_old_files(current->files);
1177 EXPORT_SYMBOL(setup_new_exec);
1180 * Prepare credentials and lock ->cred_guard_mutex.
1181 * install_exec_creds() commits the new creds and drops the lock.
1182 * Or, if exec fails before, free_bprm() should release ->cred and
1185 int prepare_bprm_creds(struct linux_binprm *bprm)
1187 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1188 return -ERESTARTNOINTR;
1190 bprm->cred = prepare_exec_creds();
1191 if (likely(bprm->cred))
1194 mutex_unlock(¤t->signal->cred_guard_mutex);
1198 void free_bprm(struct linux_binprm *bprm)
1200 free_arg_pages(bprm);
1202 mutex_unlock(¤t->signal->cred_guard_mutex);
1203 abort_creds(bprm->cred);
1209 * install the new credentials for this executable
1211 void install_exec_creds(struct linux_binprm *bprm)
1213 security_bprm_committing_creds(bprm);
1215 commit_creds(bprm->cred);
1218 * cred_guard_mutex must be held at least to this point to prevent
1219 * ptrace_attach() from altering our determination of the task's
1220 * credentials; any time after this it may be unlocked.
1222 security_bprm_committed_creds(bprm);
1223 mutex_unlock(¤t->signal->cred_guard_mutex);
1225 EXPORT_SYMBOL(install_exec_creds);
1228 * determine how safe it is to execute the proposed program
1229 * - the caller must hold ->cred_guard_mutex to protect against
1232 int check_unsafe_exec(struct linux_binprm *bprm)
1234 struct task_struct *p = current, *t;
1239 if (p->ptrace & PT_PTRACE_CAP)
1240 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1242 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1246 * This isn't strictly necessary, but it makes it harder for LSMs to
1249 if (current->no_new_privs)
1250 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1253 spin_lock(&p->fs->lock);
1255 for (t = next_thread(p); t != p; t = next_thread(t)) {
1261 if (p->fs->users > n_fs) {
1262 bprm->unsafe |= LSM_UNSAFE_SHARE;
1265 if (!p->fs->in_exec) {
1270 spin_unlock(&p->fs->lock);
1276 * Fill the binprm structure from the inode.
1277 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1279 * This may be called multiple times for binary chains (scripts for example).
1281 int prepare_binprm(struct linux_binprm *bprm)
1284 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1287 mode = inode->i_mode;
1288 if (bprm->file->f_op == NULL)
1291 /* clear any previous set[ug]id data from a previous binary */
1292 bprm->cred->euid = current_euid();
1293 bprm->cred->egid = current_egid();
1295 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1296 !current->no_new_privs) {
1298 if (mode & S_ISUID) {
1299 bprm->per_clear |= PER_CLEAR_ON_SETID;
1300 bprm->cred->euid = inode->i_uid;
1305 * If setgid is set but no group execute bit then this
1306 * is a candidate for mandatory locking, not a setgid
1309 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1310 bprm->per_clear |= PER_CLEAR_ON_SETID;
1311 bprm->cred->egid = inode->i_gid;
1315 /* fill in binprm security blob */
1316 retval = security_bprm_set_creds(bprm);
1319 bprm->cred_prepared = 1;
1321 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1322 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1325 EXPORT_SYMBOL(prepare_binprm);
1328 * Arguments are '\0' separated strings found at the location bprm->p
1329 * points to; chop off the first by relocating brpm->p to right after
1330 * the first '\0' encountered.
1332 int remove_arg_zero(struct linux_binprm *bprm)
1335 unsigned long offset;
1343 offset = bprm->p & ~PAGE_MASK;
1344 page = get_arg_page(bprm, bprm->p, 0);
1349 kaddr = kmap_atomic(page, KM_USER0);
1351 for (; offset < PAGE_SIZE && kaddr[offset];
1352 offset++, bprm->p++)
1355 kunmap_atomic(kaddr, KM_USER0);
1358 if (offset == PAGE_SIZE)
1359 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1360 } while (offset == PAGE_SIZE);
1369 EXPORT_SYMBOL(remove_arg_zero);
1372 * cycle the list of binary formats handler, until one recognizes the image
1374 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1376 unsigned int depth = bprm->recursion_depth;
1378 struct linux_binfmt *fmt;
1381 retval = security_bprm_check(bprm);
1385 retval = audit_bprm(bprm);
1389 /* Need to fetch pid before load_binary changes it */
1391 old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1395 for (try=0; try<2; try++) {
1396 read_lock(&binfmt_lock);
1397 list_for_each_entry(fmt, &formats, lh) {
1398 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1401 if (!try_module_get(fmt->module))
1403 read_unlock(&binfmt_lock);
1404 retval = fn(bprm, regs);
1406 * Restore the depth counter to its starting value
1407 * in this call, so we don't have to rely on every
1408 * load_binary function to restore it on return.
1410 bprm->recursion_depth = depth;
1413 ptrace_event(PTRACE_EVENT_EXEC,
1416 allow_write_access(bprm->file);
1420 current->did_exec = 1;
1421 proc_exec_connector(current);
1424 read_lock(&binfmt_lock);
1426 if (retval != -ENOEXEC || bprm->mm == NULL)
1429 read_unlock(&binfmt_lock);
1433 read_unlock(&binfmt_lock);
1434 #ifdef CONFIG_MODULES
1435 if (retval != -ENOEXEC || bprm->mm == NULL) {
1438 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1439 if (printable(bprm->buf[0]) &&
1440 printable(bprm->buf[1]) &&
1441 printable(bprm->buf[2]) &&
1442 printable(bprm->buf[3]))
1443 break; /* -ENOEXEC */
1445 break; /* -ENOEXEC */
1446 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1455 EXPORT_SYMBOL(search_binary_handler);
1458 * sys_execve() executes a new program.
1460 static int do_execve_common(const char *filename,
1461 struct user_arg_ptr argv,
1462 struct user_arg_ptr envp,
1463 struct pt_regs *regs)
1465 struct linux_binprm *bprm;
1467 struct files_struct *displaced;
1470 const struct cred *cred = current_cred();
1473 * We move the actual failure in case of RLIMIT_NPROC excess from
1474 * set*uid() to execve() because too many poorly written programs
1475 * don't check setuid() return code. Here we additionally recheck
1476 * whether NPROC limit is still exceeded.
1478 if ((current->flags & PF_NPROC_EXCEEDED) &&
1479 atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1484 /* We're below the limit (still or again), so we don't want to make
1485 * further execve() calls fail. */
1486 current->flags &= ~PF_NPROC_EXCEEDED;
1488 retval = unshare_files(&displaced);
1493 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1497 retval = prepare_bprm_creds(bprm);
1501 retval = check_unsafe_exec(bprm);
1504 clear_in_exec = retval;
1505 current->in_execve = 1;
1507 file = open_exec(filename);
1508 retval = PTR_ERR(file);
1515 bprm->filename = filename;
1516 bprm->interp = filename;
1518 retval = bprm_mm_init(bprm);
1522 bprm->argc = count(argv, MAX_ARG_STRINGS);
1523 if ((retval = bprm->argc) < 0)
1526 bprm->envc = count(envp, MAX_ARG_STRINGS);
1527 if ((retval = bprm->envc) < 0)
1530 retval = prepare_binprm(bprm);
1534 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1538 bprm->exec = bprm->p;
1539 retval = copy_strings(bprm->envc, envp, bprm);
1543 retval = copy_strings(bprm->argc, argv, bprm);
1547 retval = search_binary_handler(bprm,regs);
1551 /* execve succeeded */
1552 current->fs->in_exec = 0;
1553 current->in_execve = 0;
1554 acct_update_integrals(current);
1557 put_files_struct(displaced);
1562 acct_arg_size(bprm, 0);
1568 allow_write_access(bprm->file);
1574 current->fs->in_exec = 0;
1575 current->in_execve = 0;
1582 reset_files_struct(displaced);
1587 int do_execve(const char *filename,
1588 const char __user *const __user *__argv,
1589 const char __user *const __user *__envp,
1590 struct pt_regs *regs)
1592 struct user_arg_ptr argv = { .ptr.native = __argv };
1593 struct user_arg_ptr envp = { .ptr.native = __envp };
1594 return do_execve_common(filename, argv, envp, regs);
1597 #ifdef CONFIG_COMPAT
1598 int compat_do_execve(char *filename,
1599 compat_uptr_t __user *__argv,
1600 compat_uptr_t __user *__envp,
1601 struct pt_regs *regs)
1603 struct user_arg_ptr argv = {
1605 .ptr.compat = __argv,
1607 struct user_arg_ptr envp = {
1609 .ptr.compat = __envp,
1611 return do_execve_common(filename, argv, envp, regs);
1615 void set_binfmt(struct linux_binfmt *new)
1617 struct mm_struct *mm = current->mm;
1620 module_put(mm->binfmt->module);
1624 __module_get(new->module);
1627 EXPORT_SYMBOL(set_binfmt);
1629 static int expand_corename(struct core_name *cn)
1631 char *old_corename = cn->corename;
1633 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1634 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1636 if (!cn->corename) {
1637 kfree(old_corename);
1644 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1652 need = vsnprintf(NULL, 0, fmt, arg);
1655 if (likely(need < cn->size - cn->used - 1))
1658 ret = expand_corename(cn);
1663 cur = cn->corename + cn->used;
1665 vsnprintf(cur, need + 1, fmt, arg);
1674 static void cn_escape(char *str)
1681 static int cn_print_exe_file(struct core_name *cn)
1683 struct file *exe_file;
1684 char *pathbuf, *path;
1687 exe_file = get_mm_exe_file(current->mm);
1689 char *commstart = cn->corename + cn->used;
1690 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1691 cn_escape(commstart);
1695 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1701 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1703 ret = PTR_ERR(path);
1709 ret = cn_printf(cn, "%s", path);
1718 /* format_corename will inspect the pattern parameter, and output a
1719 * name into corename, which must have space for at least
1720 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1722 static int format_corename(struct core_name *cn, long signr)
1724 const struct cred *cred = current_cred();
1725 const char *pat_ptr = core_pattern;
1726 int ispipe = (*pat_ptr == '|');
1727 int pid_in_pattern = 0;
1730 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1731 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1737 /* Repeat as long as we have more pattern to process and more output
1740 if (*pat_ptr != '%') {
1743 err = cn_printf(cn, "%c", *pat_ptr++);
1745 switch (*++pat_ptr) {
1746 /* single % at the end, drop that */
1749 /* Double percent, output one percent */
1751 err = cn_printf(cn, "%c", '%');
1756 err = cn_printf(cn, "%d",
1757 task_tgid_vnr(current));
1761 err = cn_printf(cn, "%d", cred->uid);
1765 err = cn_printf(cn, "%d", cred->gid);
1767 /* signal that caused the coredump */
1769 err = cn_printf(cn, "%ld", signr);
1771 /* UNIX time of coredump */
1774 do_gettimeofday(&tv);
1775 err = cn_printf(cn, "%lu", tv.tv_sec);
1780 char *namestart = cn->corename + cn->used;
1781 down_read(&uts_sem);
1782 err = cn_printf(cn, "%s",
1783 utsname()->nodename);
1785 cn_escape(namestart);
1790 char *commstart = cn->corename + cn->used;
1791 err = cn_printf(cn, "%s", current->comm);
1792 cn_escape(commstart);
1796 err = cn_print_exe_file(cn);
1798 /* core limit size */
1800 err = cn_printf(cn, "%lu",
1801 rlimit(RLIMIT_CORE));
1813 /* Backward compatibility with core_uses_pid:
1815 * If core_pattern does not include a %p (as is the default)
1816 * and core_uses_pid is set, then .%pid will be appended to
1817 * the filename. Do not do this for piped commands. */
1818 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1819 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1827 static int zap_process(struct task_struct *start, int exit_code)
1829 struct task_struct *t;
1832 start->signal->flags = SIGNAL_GROUP_EXIT;
1833 start->signal->group_exit_code = exit_code;
1834 start->signal->group_stop_count = 0;
1838 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1839 if (t != current && t->mm) {
1840 sigaddset(&t->pending.signal, SIGKILL);
1841 signal_wake_up(t, 1);
1844 } while_each_thread(start, t);
1849 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1850 struct core_state *core_state, int exit_code)
1852 struct task_struct *g, *p;
1853 unsigned long flags;
1856 spin_lock_irq(&tsk->sighand->siglock);
1857 if (!signal_group_exit(tsk->signal)) {
1858 mm->core_state = core_state;
1859 nr = zap_process(tsk, exit_code);
1861 spin_unlock_irq(&tsk->sighand->siglock);
1862 if (unlikely(nr < 0))
1865 if (atomic_read(&mm->mm_users) == nr + 1)
1868 * We should find and kill all tasks which use this mm, and we should
1869 * count them correctly into ->nr_threads. We don't take tasklist
1870 * lock, but this is safe wrt:
1873 * None of sub-threads can fork after zap_process(leader). All
1874 * processes which were created before this point should be
1875 * visible to zap_threads() because copy_process() adds the new
1876 * process to the tail of init_task.tasks list, and lock/unlock
1877 * of ->siglock provides a memory barrier.
1880 * The caller holds mm->mmap_sem. This means that the task which
1881 * uses this mm can't pass exit_mm(), so it can't exit or clear
1885 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1886 * we must see either old or new leader, this does not matter.
1887 * However, it can change p->sighand, so lock_task_sighand(p)
1888 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1891 * Note also that "g" can be the old leader with ->mm == NULL
1892 * and already unhashed and thus removed from ->thread_group.
1893 * This is OK, __unhash_process()->list_del_rcu() does not
1894 * clear the ->next pointer, we will find the new leader via
1898 for_each_process(g) {
1899 if (g == tsk->group_leader)
1901 if (g->flags & PF_KTHREAD)
1906 if (unlikely(p->mm == mm)) {
1907 lock_task_sighand(p, &flags);
1908 nr += zap_process(p, exit_code);
1909 unlock_task_sighand(p, &flags);
1913 } while_each_thread(g, p);
1917 atomic_set(&core_state->nr_threads, nr);
1921 static int coredump_wait(int exit_code, struct core_state *core_state)
1923 struct task_struct *tsk = current;
1924 struct mm_struct *mm = tsk->mm;
1925 struct completion *vfork_done;
1926 int core_waiters = -EBUSY;
1928 init_completion(&core_state->startup);
1929 core_state->dumper.task = tsk;
1930 core_state->dumper.next = NULL;
1932 down_write(&mm->mmap_sem);
1933 if (!mm->core_state)
1934 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1935 up_write(&mm->mmap_sem);
1937 if (unlikely(core_waiters < 0))
1941 * Make sure nobody is waiting for us to release the VM,
1942 * otherwise we can deadlock when we wait on each other
1944 vfork_done = tsk->vfork_done;
1946 tsk->vfork_done = NULL;
1947 complete(vfork_done);
1951 wait_for_completion(&core_state->startup);
1953 return core_waiters;
1956 static void coredump_finish(struct mm_struct *mm)
1958 struct core_thread *curr, *next;
1959 struct task_struct *task;
1961 next = mm->core_state->dumper.next;
1962 while ((curr = next) != NULL) {
1966 * see exit_mm(), curr->task must not see
1967 * ->task == NULL before we read ->next.
1971 wake_up_process(task);
1974 mm->core_state = NULL;
1978 * set_dumpable converts traditional three-value dumpable to two flags and
1979 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1980 * these bits are not changed atomically. So get_dumpable can observe the
1981 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1982 * return either old dumpable or new one by paying attention to the order of
1983 * modifying the bits.
1985 * dumpable | mm->flags (binary)
1986 * old new | initial interim final
1987 * ---------+-----------------------
1995 * (*) get_dumpable regards interim value of 10 as 11.
1997 void set_dumpable(struct mm_struct *mm, int value)
2001 clear_bit(MMF_DUMPABLE, &mm->flags);
2003 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2006 set_bit(MMF_DUMPABLE, &mm->flags);
2008 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2011 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2013 set_bit(MMF_DUMPABLE, &mm->flags);
2018 static int __get_dumpable(unsigned long mm_flags)
2022 ret = mm_flags & MMF_DUMPABLE_MASK;
2023 return (ret >= 2) ? 2 : ret;
2026 int get_dumpable(struct mm_struct *mm)
2028 return __get_dumpable(mm->flags);
2031 static void wait_for_dump_helpers(struct file *file)
2033 struct pipe_inode_info *pipe;
2035 pipe = file->f_path.dentry->d_inode->i_pipe;
2041 while ((pipe->readers > 1) && (!signal_pending(current))) {
2042 wake_up_interruptible_sync(&pipe->wait);
2043 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2056 * helper function to customize the process used
2057 * to collect the core in userspace. Specifically
2058 * it sets up a pipe and installs it as fd 0 (stdin)
2059 * for the process. Returns 0 on success, or
2060 * PTR_ERR on failure.
2061 * Note that it also sets the core limit to 1. This
2062 * is a special value that we use to trap recursive
2065 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2067 struct file *rp, *wp;
2068 struct fdtable *fdt;
2069 struct coredump_params *cp = (struct coredump_params *)info->data;
2070 struct files_struct *cf = current->files;
2072 wp = create_write_pipe(0);
2076 rp = create_read_pipe(wp, 0);
2078 free_write_pipe(wp);
2086 spin_lock(&cf->file_lock);
2087 fdt = files_fdtable(cf);
2088 FD_SET(0, fdt->open_fds);
2089 FD_CLR(0, fdt->close_on_exec);
2090 spin_unlock(&cf->file_lock);
2092 /* and disallow core files too */
2093 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2098 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2100 struct core_state core_state;
2101 struct core_name cn;
2102 struct mm_struct *mm = current->mm;
2103 struct linux_binfmt * binfmt;
2104 const struct cred *old_cred;
2109 static atomic_t core_dump_count = ATOMIC_INIT(0);
2110 struct coredump_params cprm = {
2113 .limit = rlimit(RLIMIT_CORE),
2115 * We must use the same mm->flags while dumping core to avoid
2116 * inconsistency of bit flags, since this flag is not protected
2119 .mm_flags = mm->flags,
2122 audit_core_dumps(signr);
2124 binfmt = mm->binfmt;
2125 if (!binfmt || !binfmt->core_dump)
2127 if (!__get_dumpable(cprm.mm_flags))
2130 cred = prepare_creds();
2134 * We cannot trust fsuid as being the "true" uid of the
2135 * process nor do we know its entire history. We only know it
2136 * was tainted so we dump it as root in mode 2.
2138 if (__get_dumpable(cprm.mm_flags) == 2) {
2139 /* Setuid core dump mode */
2140 flag = O_EXCL; /* Stop rewrite attacks */
2141 cred->fsuid = 0; /* Dump root private */
2144 retval = coredump_wait(exit_code, &core_state);
2148 old_cred = override_creds(cred);
2151 * Clear any false indication of pending signals that might
2152 * be seen by the filesystem code called to write the core file.
2154 clear_thread_flag(TIF_SIGPENDING);
2156 ispipe = format_corename(&cn, signr);
2163 printk(KERN_WARNING "format_corename failed\n");
2164 printk(KERN_WARNING "Aborting core\n");
2168 if (cprm.limit == 1) {
2170 * Normally core limits are irrelevant to pipes, since
2171 * we're not writing to the file system, but we use
2172 * cprm.limit of 1 here as a speacial value. Any
2173 * non-1 limit gets set to RLIM_INFINITY below, but
2174 * a limit of 0 skips the dump. This is a consistent
2175 * way to catch recursive crashes. We can still crash
2176 * if the core_pattern binary sets RLIM_CORE = !1
2177 * but it runs as root, and can do lots of stupid things
2178 * Note that we use task_tgid_vnr here to grab the pid
2179 * of the process group leader. That way we get the
2180 * right pid if a thread in a multi-threaded
2181 * core_pattern process dies.
2184 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2185 task_tgid_vnr(current), current->comm);
2186 printk(KERN_WARNING "Aborting core\n");
2189 cprm.limit = RLIM_INFINITY;
2191 dump_count = atomic_inc_return(&core_dump_count);
2192 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2193 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2194 task_tgid_vnr(current), current->comm);
2195 printk(KERN_WARNING "Skipping core dump\n");
2196 goto fail_dropcount;
2199 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2201 printk(KERN_WARNING "%s failed to allocate memory\n",
2203 goto fail_dropcount;
2206 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2207 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2209 argv_free(helper_argv);
2211 printk(KERN_INFO "Core dump to %s pipe failed\n",
2216 struct inode *inode;
2218 if (cprm.limit < binfmt->min_coredump)
2221 cprm.file = filp_open(cn.corename,
2222 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2224 if (IS_ERR(cprm.file))
2227 inode = cprm.file->f_path.dentry->d_inode;
2228 if (inode->i_nlink > 1)
2230 if (d_unhashed(cprm.file->f_path.dentry))
2233 * AK: actually i see no reason to not allow this for named
2234 * pipes etc, but keep the previous behaviour for now.
2236 if (!S_ISREG(inode->i_mode))
2239 * Dont allow local users get cute and trick others to coredump
2240 * into their pre-created files.
2242 if (inode->i_uid != current_fsuid())
2244 if (!cprm.file->f_op || !cprm.file->f_op->write)
2246 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2250 retval = binfmt->core_dump(&cprm);
2252 current->signal->group_exit_code |= 0x80;
2254 if (ispipe && core_pipe_limit)
2255 wait_for_dump_helpers(cprm.file);
2258 filp_close(cprm.file, NULL);
2261 atomic_dec(&core_dump_count);
2265 coredump_finish(mm);
2266 revert_creds(old_cred);
2274 * Core dumping helper functions. These are the only things you should
2275 * do on a core-file: use only these functions to write out all the
2278 int dump_write(struct file *file, const void *addr, int nr)
2280 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2282 EXPORT_SYMBOL(dump_write);
2284 int dump_seek(struct file *file, loff_t off)
2288 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2289 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2292 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2297 unsigned long n = off;
2301 if (!dump_write(file, buf, n)) {
2307 free_page((unsigned long)buf);
2311 EXPORT_SYMBOL(dump_seek);