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 <asm/uaccess.h>
60 #include <asm/mmu_context.h>
65 char core_pattern[CORENAME_MAX_SIZE] = "core";
66 unsigned int core_pipe_limit;
67 int suid_dumpable = 0;
73 static atomic_t call_count = ATOMIC_INIT(1);
75 /* The maximal length of core_pattern is also specified in sysctl.c */
77 static LIST_HEAD(formats);
78 static DEFINE_RWLOCK(binfmt_lock);
80 int __register_binfmt(struct linux_binfmt * fmt, int insert)
84 write_lock(&binfmt_lock);
85 insert ? list_add(&fmt->lh, &formats) :
86 list_add_tail(&fmt->lh, &formats);
87 write_unlock(&binfmt_lock);
91 EXPORT_SYMBOL(__register_binfmt);
93 void unregister_binfmt(struct linux_binfmt * fmt)
95 write_lock(&binfmt_lock);
97 write_unlock(&binfmt_lock);
100 EXPORT_SYMBOL(unregister_binfmt);
102 static inline void put_binfmt(struct linux_binfmt * fmt)
104 module_put(fmt->module);
108 * Note that a shared library must be both readable and executable due to
111 * Also note that we take the address to load from from the file itself.
113 SYSCALL_DEFINE1(uselib, const char __user *, library)
116 char *tmp = getname(library);
117 int error = PTR_ERR(tmp);
118 static const struct open_flags uselib_flags = {
119 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
120 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
121 .intent = LOOKUP_OPEN
127 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
129 error = PTR_ERR(file);
134 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
138 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
145 struct linux_binfmt * fmt;
147 read_lock(&binfmt_lock);
148 list_for_each_entry(fmt, &formats, lh) {
149 if (!fmt->load_shlib)
151 if (!try_module_get(fmt->module))
153 read_unlock(&binfmt_lock);
154 error = fmt->load_shlib(file);
155 read_lock(&binfmt_lock);
157 if (error != -ENOEXEC)
160 read_unlock(&binfmt_lock);
170 * The nascent bprm->mm is not visible until exec_mmap() but it can
171 * use a lot of memory, account these pages in current->mm temporary
172 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
173 * change the counter back via acct_arg_size(0).
175 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
177 struct mm_struct *mm = current->mm;
178 long diff = (long)(pages - bprm->vma_pages);
183 bprm->vma_pages = pages;
184 add_mm_counter(mm, MM_ANONPAGES, diff);
187 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
193 #ifdef CONFIG_STACK_GROWSUP
195 ret = expand_downwards(bprm->vma, pos);
200 ret = get_user_pages(current, bprm->mm, pos,
201 1, write, 1, &page, NULL);
206 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
209 acct_arg_size(bprm, size / PAGE_SIZE);
212 * We've historically supported up to 32 pages (ARG_MAX)
213 * of argument strings even with small stacks
219 * Limit to 1/4-th the stack size for the argv+env strings.
221 * - the remaining binfmt code will not run out of stack space,
222 * - the program will have a reasonable amount of stack left
225 rlim = current->signal->rlim;
226 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
235 static void put_arg_page(struct page *page)
240 static void free_arg_page(struct linux_binprm *bprm, int i)
244 static void free_arg_pages(struct linux_binprm *bprm)
248 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
251 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
254 static int __bprm_mm_init(struct linux_binprm *bprm)
257 struct vm_area_struct *vma = NULL;
258 struct mm_struct *mm = bprm->mm;
260 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
264 down_write(&mm->mmap_sem);
268 * Place the stack at the largest stack address the architecture
269 * supports. Later, we'll move this to an appropriate place. We don't
270 * use STACK_TOP because that can depend on attributes which aren't
273 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
274 vma->vm_end = STACK_TOP_MAX;
275 vma->vm_start = vma->vm_end - PAGE_SIZE;
276 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
277 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
278 INIT_LIST_HEAD(&vma->anon_vma_chain);
280 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
284 err = insert_vm_struct(mm, vma);
288 mm->stack_vm = mm->total_vm = 1;
289 up_write(&mm->mmap_sem);
290 bprm->p = vma->vm_end - sizeof(void *);
293 up_write(&mm->mmap_sem);
295 kmem_cache_free(vm_area_cachep, vma);
299 static bool valid_arg_len(struct linux_binprm *bprm, long len)
301 return len <= MAX_ARG_STRLEN;
306 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
310 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
315 page = bprm->page[pos / PAGE_SIZE];
316 if (!page && write) {
317 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
320 bprm->page[pos / PAGE_SIZE] = page;
326 static void put_arg_page(struct page *page)
330 static void free_arg_page(struct linux_binprm *bprm, int i)
333 __free_page(bprm->page[i]);
334 bprm->page[i] = NULL;
338 static void free_arg_pages(struct linux_binprm *bprm)
342 for (i = 0; i < MAX_ARG_PAGES; i++)
343 free_arg_page(bprm, i);
346 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
351 static int __bprm_mm_init(struct linux_binprm *bprm)
353 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
357 static bool valid_arg_len(struct linux_binprm *bprm, long len)
359 return len <= bprm->p;
362 #endif /* CONFIG_MMU */
365 * Create a new mm_struct and populate it with a temporary stack
366 * vm_area_struct. We don't have enough context at this point to set the stack
367 * flags, permissions, and offset, so we use temporary values. We'll update
368 * them later in setup_arg_pages().
370 int bprm_mm_init(struct linux_binprm *bprm)
373 struct mm_struct *mm = NULL;
375 bprm->mm = mm = mm_alloc();
380 err = init_new_context(current, mm);
384 err = __bprm_mm_init(bprm);
399 struct user_arg_ptr {
404 const char __user *const __user *native;
406 compat_uptr_t __user *compat;
411 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
413 const char __user *native;
416 if (unlikely(argv.is_compat)) {
417 compat_uptr_t compat;
419 if (get_user(compat, argv.ptr.compat + nr))
420 return ERR_PTR(-EFAULT);
422 return compat_ptr(compat);
426 if (get_user(native, argv.ptr.native + nr))
427 return ERR_PTR(-EFAULT);
433 * count() counts the number of strings in array ARGV.
435 static int count(struct user_arg_ptr argv, int max)
439 if (argv.ptr.native != NULL) {
441 const char __user *p = get_user_arg_ptr(argv, i);
452 if (fatal_signal_pending(current))
453 return -ERESTARTNOHAND;
461 * 'copy_strings()' copies argument/environment strings from the old
462 * processes's memory to the new process's stack. The call to get_user_pages()
463 * ensures the destination page is created and not swapped out.
465 static int copy_strings(int argc, struct user_arg_ptr argv,
466 struct linux_binprm *bprm)
468 struct page *kmapped_page = NULL;
470 unsigned long kpos = 0;
474 const char __user *str;
479 str = get_user_arg_ptr(argv, argc);
483 len = strnlen_user(str, MAX_ARG_STRLEN);
488 if (!valid_arg_len(bprm, len))
491 /* We're going to work our way backwords. */
497 int offset, bytes_to_copy;
499 if (fatal_signal_pending(current)) {
500 ret = -ERESTARTNOHAND;
505 offset = pos % PAGE_SIZE;
509 bytes_to_copy = offset;
510 if (bytes_to_copy > len)
513 offset -= bytes_to_copy;
514 pos -= bytes_to_copy;
515 str -= bytes_to_copy;
516 len -= bytes_to_copy;
518 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
521 page = get_arg_page(bprm, pos, 1);
528 flush_kernel_dcache_page(kmapped_page);
529 kunmap(kmapped_page);
530 put_arg_page(kmapped_page);
533 kaddr = kmap(kmapped_page);
534 kpos = pos & PAGE_MASK;
535 flush_arg_page(bprm, kpos, kmapped_page);
537 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
546 flush_kernel_dcache_page(kmapped_page);
547 kunmap(kmapped_page);
548 put_arg_page(kmapped_page);
554 * Like copy_strings, but get argv and its values from kernel memory.
556 int copy_strings_kernel(int argc, const char *const *__argv,
557 struct linux_binprm *bprm)
560 mm_segment_t oldfs = get_fs();
561 struct user_arg_ptr argv = {
562 .ptr.native = (const char __user *const __user *)__argv,
566 r = copy_strings(argc, argv, bprm);
571 EXPORT_SYMBOL(copy_strings_kernel);
576 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
577 * the binfmt code determines where the new stack should reside, we shift it to
578 * its final location. The process proceeds as follows:
580 * 1) Use shift to calculate the new vma endpoints.
581 * 2) Extend vma to cover both the old and new ranges. This ensures the
582 * arguments passed to subsequent functions are consistent.
583 * 3) Move vma's page tables to the new range.
584 * 4) Free up any cleared pgd range.
585 * 5) Shrink the vma to cover only the new range.
587 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
589 struct mm_struct *mm = vma->vm_mm;
590 unsigned long old_start = vma->vm_start;
591 unsigned long old_end = vma->vm_end;
592 unsigned long length = old_end - old_start;
593 unsigned long new_start = old_start - shift;
594 unsigned long new_end = old_end - shift;
595 struct mmu_gather tlb;
597 BUG_ON(new_start > new_end);
600 * ensure there are no vmas between where we want to go
603 if (vma != find_vma(mm, new_start))
607 * cover the whole range: [new_start, old_end)
609 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
613 * move the page tables downwards, on failure we rely on
614 * process cleanup to remove whatever mess we made.
616 if (length != move_page_tables(vma, old_start,
617 vma, new_start, length))
621 tlb_gather_mmu(&tlb, mm, 0);
622 if (new_end > old_start) {
624 * when the old and new regions overlap clear from new_end.
626 free_pgd_range(&tlb, new_end, old_end, new_end,
627 vma->vm_next ? vma->vm_next->vm_start : 0);
630 * otherwise, clean from old_start; this is done to not touch
631 * the address space in [new_end, old_start) some architectures
632 * have constraints on va-space that make this illegal (IA64) -
633 * for the others its just a little faster.
635 free_pgd_range(&tlb, old_start, old_end, new_end,
636 vma->vm_next ? vma->vm_next->vm_start : 0);
638 tlb_finish_mmu(&tlb, new_end, old_end);
641 * Shrink the vma to just the new range. Always succeeds.
643 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
649 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
650 * the stack is optionally relocated, and some extra space is added.
652 int setup_arg_pages(struct linux_binprm *bprm,
653 unsigned long stack_top,
654 int executable_stack)
657 unsigned long stack_shift;
658 struct mm_struct *mm = current->mm;
659 struct vm_area_struct *vma = bprm->vma;
660 struct vm_area_struct *prev = NULL;
661 unsigned long vm_flags;
662 unsigned long stack_base;
663 unsigned long stack_size;
664 unsigned long stack_expand;
665 unsigned long rlim_stack;
667 #ifdef CONFIG_STACK_GROWSUP
668 /* Limit stack size to 1GB */
669 stack_base = rlimit_max(RLIMIT_STACK);
670 if (stack_base > (1 << 30))
671 stack_base = 1 << 30;
673 /* Make sure we didn't let the argument array grow too large. */
674 if (vma->vm_end - vma->vm_start > stack_base)
677 stack_base = PAGE_ALIGN(stack_top - stack_base);
679 stack_shift = vma->vm_start - stack_base;
680 mm->arg_start = bprm->p - stack_shift;
681 bprm->p = vma->vm_end - stack_shift;
683 stack_top = arch_align_stack(stack_top);
684 stack_top = PAGE_ALIGN(stack_top);
686 if (unlikely(stack_top < mmap_min_addr) ||
687 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
690 stack_shift = vma->vm_end - stack_top;
692 bprm->p -= stack_shift;
693 mm->arg_start = bprm->p;
697 bprm->loader -= stack_shift;
698 bprm->exec -= stack_shift;
700 down_write(&mm->mmap_sem);
701 vm_flags = VM_STACK_FLAGS;
704 * Adjust stack execute permissions; explicitly enable for
705 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
706 * (arch default) otherwise.
708 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
710 else if (executable_stack == EXSTACK_DISABLE_X)
711 vm_flags &= ~VM_EXEC;
712 vm_flags |= mm->def_flags;
713 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
715 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
721 /* Move stack pages down in memory. */
723 ret = shift_arg_pages(vma, stack_shift);
728 /* mprotect_fixup is overkill to remove the temporary stack flags */
729 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
731 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
732 stack_size = vma->vm_end - vma->vm_start;
734 * Align this down to a page boundary as expand_stack
737 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
738 #ifdef CONFIG_STACK_GROWSUP
739 if (stack_size + stack_expand > rlim_stack)
740 stack_base = vma->vm_start + rlim_stack;
742 stack_base = vma->vm_end + stack_expand;
744 if (stack_size + stack_expand > rlim_stack)
745 stack_base = vma->vm_end - rlim_stack;
747 stack_base = vma->vm_start - stack_expand;
749 current->mm->start_stack = bprm->p;
750 ret = expand_stack(vma, stack_base);
755 up_write(&mm->mmap_sem);
758 EXPORT_SYMBOL(setup_arg_pages);
760 #endif /* CONFIG_MMU */
762 struct file *open_exec(const char *name)
766 static const struct open_flags open_exec_flags = {
767 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
768 .acc_mode = MAY_EXEC | MAY_OPEN,
769 .intent = LOOKUP_OPEN
772 file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
777 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
780 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
785 err = deny_write_access(file);
796 EXPORT_SYMBOL(open_exec);
798 int kernel_read(struct file *file, loff_t offset,
799 char *addr, unsigned long count)
807 /* The cast to a user pointer is valid due to the set_fs() */
808 result = vfs_read(file, (void __user *)addr, count, &pos);
813 EXPORT_SYMBOL(kernel_read);
815 static int exec_mmap(struct mm_struct *mm)
817 struct task_struct *tsk;
818 struct mm_struct * old_mm, *active_mm;
820 /* Notify parent that we're no longer interested in the old VM */
822 old_mm = current->mm;
823 sync_mm_rss(tsk, old_mm);
824 mm_release(tsk, old_mm);
828 * Make sure that if there is a core dump in progress
829 * for the old mm, we get out and die instead of going
830 * through with the exec. We must hold mmap_sem around
831 * checking core_state and changing tsk->mm.
833 down_read(&old_mm->mmap_sem);
834 if (unlikely(old_mm->core_state)) {
835 up_read(&old_mm->mmap_sem);
840 active_mm = tsk->active_mm;
843 activate_mm(active_mm, mm);
845 arch_pick_mmap_layout(mm);
847 up_read(&old_mm->mmap_sem);
848 BUG_ON(active_mm != old_mm);
849 mm_update_next_owner(old_mm);
858 * This function makes sure the current process has its own signal table,
859 * so that flush_signal_handlers can later reset the handlers without
860 * disturbing other processes. (Other processes might share the signal
861 * table via the CLONE_SIGHAND option to clone().)
863 static int de_thread(struct task_struct *tsk)
865 struct signal_struct *sig = tsk->signal;
866 struct sighand_struct *oldsighand = tsk->sighand;
867 spinlock_t *lock = &oldsighand->siglock;
869 if (thread_group_empty(tsk))
870 goto no_thread_group;
873 * Kill all other threads in the thread group.
876 if (signal_group_exit(sig)) {
878 * Another group action in progress, just
879 * return so that the signal is processed.
881 spin_unlock_irq(lock);
885 sig->group_exit_task = tsk;
886 sig->notify_count = zap_other_threads(tsk);
887 if (!thread_group_leader(tsk))
890 while (sig->notify_count) {
891 __set_current_state(TASK_UNINTERRUPTIBLE);
892 spin_unlock_irq(lock);
896 spin_unlock_irq(lock);
899 * At this point all other threads have exited, all we have to
900 * do is to wait for the thread group leader to become inactive,
901 * and to assume its PID:
903 if (!thread_group_leader(tsk)) {
904 struct task_struct *leader = tsk->group_leader;
906 sig->notify_count = -1; /* for exit_notify() */
908 write_lock_irq(&tasklist_lock);
909 if (likely(leader->exit_state))
911 __set_current_state(TASK_UNINTERRUPTIBLE);
912 write_unlock_irq(&tasklist_lock);
917 * The only record we have of the real-time age of a
918 * process, regardless of execs it's done, is start_time.
919 * All the past CPU time is accumulated in signal_struct
920 * from sister threads now dead. But in this non-leader
921 * exec, nothing survives from the original leader thread,
922 * whose birth marks the true age of this process now.
923 * When we take on its identity by switching to its PID, we
924 * also take its birthdate (always earlier than our own).
926 tsk->start_time = leader->start_time;
928 BUG_ON(!same_thread_group(leader, tsk));
929 BUG_ON(has_group_leader_pid(tsk));
931 * An exec() starts a new thread group with the
932 * TGID of the previous thread group. Rehash the
933 * two threads with a switched PID, and release
934 * the former thread group leader:
937 /* Become a process group leader with the old leader's pid.
938 * The old leader becomes a thread of the this thread group.
939 * Note: The old leader also uses this pid until release_task
940 * is called. Odd but simple and correct.
942 detach_pid(tsk, PIDTYPE_PID);
943 tsk->pid = leader->pid;
944 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
945 transfer_pid(leader, tsk, PIDTYPE_PGID);
946 transfer_pid(leader, tsk, PIDTYPE_SID);
948 list_replace_rcu(&leader->tasks, &tsk->tasks);
949 list_replace_init(&leader->sibling, &tsk->sibling);
951 tsk->group_leader = tsk;
952 leader->group_leader = tsk;
954 tsk->exit_signal = SIGCHLD;
955 leader->exit_signal = -1;
957 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
958 leader->exit_state = EXIT_DEAD;
961 * We are going to release_task()->ptrace_unlink() silently,
962 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
963 * the tracer wont't block again waiting for this thread.
965 if (unlikely(leader->ptrace))
966 __wake_up_parent(leader, leader->parent);
967 write_unlock_irq(&tasklist_lock);
969 release_task(leader);
972 sig->group_exit_task = NULL;
973 sig->notify_count = 0;
976 /* we have changed execution domain */
977 tsk->exit_signal = SIGCHLD;
980 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
983 flush_itimer_signals();
985 if (atomic_read(&oldsighand->count) != 1) {
986 struct sighand_struct *newsighand;
988 * This ->sighand is shared with the CLONE_SIGHAND
989 * but not CLONE_THREAD task, switch to the new one.
991 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
995 atomic_set(&newsighand->count, 1);
996 memcpy(newsighand->action, oldsighand->action,
997 sizeof(newsighand->action));
999 write_lock_irq(&tasklist_lock);
1000 spin_lock(&oldsighand->siglock);
1001 rcu_assign_pointer(tsk->sighand, newsighand);
1002 spin_unlock(&oldsighand->siglock);
1003 write_unlock_irq(&tasklist_lock);
1005 __cleanup_sighand(oldsighand);
1008 BUG_ON(!thread_group_leader(tsk));
1013 * These functions flushes out all traces of the currently running executable
1014 * so that a new one can be started
1016 static void flush_old_files(struct files_struct * files)
1019 struct fdtable *fdt;
1021 spin_lock(&files->file_lock);
1023 unsigned long set, i;
1027 fdt = files_fdtable(files);
1028 if (i >= fdt->max_fds)
1030 set = fdt->close_on_exec->fds_bits[j];
1033 fdt->close_on_exec->fds_bits[j] = 0;
1034 spin_unlock(&files->file_lock);
1035 for ( ; set ; i++,set >>= 1) {
1040 spin_lock(&files->file_lock);
1043 spin_unlock(&files->file_lock);
1046 char *get_task_comm(char *buf, struct task_struct *tsk)
1048 /* buf must be at least sizeof(tsk->comm) in size */
1050 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1054 EXPORT_SYMBOL_GPL(get_task_comm);
1056 void set_task_comm(struct task_struct *tsk, char *buf)
1061 * Threads may access current->comm without holding
1062 * the task lock, so write the string carefully.
1063 * Readers without a lock may see incomplete new
1064 * names but are safe from non-terminating string reads.
1066 memset(tsk->comm, 0, TASK_COMM_LEN);
1068 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1070 perf_event_comm(tsk);
1073 int flush_old_exec(struct linux_binprm * bprm)
1078 * Make sure we have a private signal table and that
1079 * we are unassociated from the previous thread group.
1081 retval = de_thread(current);
1085 set_mm_exe_file(bprm->mm, bprm->file);
1088 * Release all of the old mmap stuff
1090 acct_arg_size(bprm, 0);
1091 retval = exec_mmap(bprm->mm);
1095 bprm->mm = NULL; /* We're using it now */
1099 ~(PF_RANDOMIZE | PF_KTHREAD | PF_NOFREEZE | PF_FREEZER_NOSIG);
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);
1162 /* An exec changes our domain. We are no longer part of the thread
1165 current->self_exec_id++;
1167 flush_signal_handlers(current, 0);
1168 flush_old_files(current->files);
1170 EXPORT_SYMBOL(setup_new_exec);
1173 * Prepare credentials and lock ->cred_guard_mutex.
1174 * install_exec_creds() commits the new creds and drops the lock.
1175 * Or, if exec fails before, free_bprm() should release ->cred and
1178 int prepare_bprm_creds(struct linux_binprm *bprm)
1180 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1181 return -ERESTARTNOINTR;
1183 bprm->cred = prepare_exec_creds();
1184 if (likely(bprm->cred))
1187 mutex_unlock(¤t->signal->cred_guard_mutex);
1191 void free_bprm(struct linux_binprm *bprm)
1193 free_arg_pages(bprm);
1195 mutex_unlock(¤t->signal->cred_guard_mutex);
1196 abort_creds(bprm->cred);
1198 /* If a binfmt changed the interp, free it. */
1199 if (bprm->interp != bprm->filename)
1200 kfree(bprm->interp);
1204 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1206 /* If a binfmt changed the interp, free it first. */
1207 if (bprm->interp != bprm->filename)
1208 kfree(bprm->interp);
1209 bprm->interp = kstrdup(interp, GFP_KERNEL);
1214 EXPORT_SYMBOL(bprm_change_interp);
1217 * install the new credentials for this executable
1219 void install_exec_creds(struct linux_binprm *bprm)
1221 security_bprm_committing_creds(bprm);
1223 commit_creds(bprm->cred);
1227 * Disable monitoring for regular users
1228 * when executing setuid binaries. Must
1229 * wait until new credentials are committed
1230 * by commit_creds() above
1232 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1233 perf_event_exit_task(current);
1235 * cred_guard_mutex must be held at least to this point to prevent
1236 * ptrace_attach() from altering our determination of the task's
1237 * credentials; any time after this it may be unlocked.
1239 security_bprm_committed_creds(bprm);
1240 mutex_unlock(¤t->signal->cred_guard_mutex);
1242 EXPORT_SYMBOL(install_exec_creds);
1245 * determine how safe it is to execute the proposed program
1246 * - the caller must hold ->cred_guard_mutex to protect against
1249 int check_unsafe_exec(struct linux_binprm *bprm)
1251 struct task_struct *p = current, *t;
1256 if (p->ptrace & PT_PTRACE_CAP)
1257 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1259 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1263 spin_lock(&p->fs->lock);
1265 for (t = next_thread(p); t != p; t = next_thread(t)) {
1271 if (p->fs->users > n_fs) {
1272 bprm->unsafe |= LSM_UNSAFE_SHARE;
1275 if (!p->fs->in_exec) {
1280 spin_unlock(&p->fs->lock);
1285 static void bprm_fill_uid(struct linux_binprm *bprm)
1287 struct inode *inode;
1292 /* clear any previous set[ug]id data from a previous binary */
1293 bprm->cred->euid = current_euid();
1294 bprm->cred->egid = current_egid();
1296 if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1299 inode = bprm->file->f_path.dentry->d_inode;
1300 mode = ACCESS_ONCE(inode->i_mode);
1301 if (!(mode & (S_ISUID|S_ISGID)))
1304 /* Be careful if suid/sgid is set */
1305 mutex_lock(&inode->i_mutex);
1307 /* reload atomically mode/uid/gid now that lock held */
1308 mode = inode->i_mode;
1311 mutex_unlock(&inode->i_mutex);
1313 if (mode & S_ISUID) {
1314 bprm->per_clear |= PER_CLEAR_ON_SETID;
1315 bprm->cred->euid = uid;
1318 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1319 bprm->per_clear |= PER_CLEAR_ON_SETID;
1320 bprm->cred->egid = gid;
1325 * Fill the binprm structure from the inode.
1326 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1328 * This may be called multiple times for binary chains (scripts for example).
1330 int prepare_binprm(struct linux_binprm *bprm)
1334 if (bprm->file->f_op == NULL)
1337 bprm_fill_uid(bprm);
1339 /* fill in binprm security blob */
1340 retval = security_bprm_set_creds(bprm);
1343 bprm->cred_prepared = 1;
1345 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1346 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1349 EXPORT_SYMBOL(prepare_binprm);
1352 * Arguments are '\0' separated strings found at the location bprm->p
1353 * points to; chop off the first by relocating brpm->p to right after
1354 * the first '\0' encountered.
1356 int remove_arg_zero(struct linux_binprm *bprm)
1359 unsigned long offset;
1367 offset = bprm->p & ~PAGE_MASK;
1368 page = get_arg_page(bprm, bprm->p, 0);
1373 kaddr = kmap_atomic(page, KM_USER0);
1375 for (; offset < PAGE_SIZE && kaddr[offset];
1376 offset++, bprm->p++)
1379 kunmap_atomic(kaddr, KM_USER0);
1382 if (offset == PAGE_SIZE)
1383 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1384 } while (offset == PAGE_SIZE);
1393 EXPORT_SYMBOL(remove_arg_zero);
1396 * cycle the list of binary formats handler, until one recognizes the image
1398 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1400 unsigned int depth = bprm->recursion_depth;
1402 struct linux_binfmt *fmt;
1405 /* This allows 4 levels of binfmt rewrites before failing hard. */
1409 retval = security_bprm_check(bprm);
1413 retval = audit_bprm(bprm);
1417 /* Need to fetch pid before load_binary changes it */
1419 old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1423 for (try=0; try<2; try++) {
1424 read_lock(&binfmt_lock);
1425 list_for_each_entry(fmt, &formats, lh) {
1426 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1429 if (!try_module_get(fmt->module))
1431 read_unlock(&binfmt_lock);
1432 bprm->recursion_depth = depth + 1;
1433 retval = fn(bprm, regs);
1434 bprm->recursion_depth = depth;
1437 ptrace_event(PTRACE_EVENT_EXEC,
1440 allow_write_access(bprm->file);
1444 current->did_exec = 1;
1445 proc_exec_connector(current);
1448 read_lock(&binfmt_lock);
1450 if (retval != -ENOEXEC || bprm->mm == NULL)
1453 read_unlock(&binfmt_lock);
1457 read_unlock(&binfmt_lock);
1458 #ifdef CONFIG_MODULES
1459 if (retval != -ENOEXEC || bprm->mm == NULL) {
1462 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1463 if (printable(bprm->buf[0]) &&
1464 printable(bprm->buf[1]) &&
1465 printable(bprm->buf[2]) &&
1466 printable(bprm->buf[3]))
1467 break; /* -ENOEXEC */
1469 break; /* -ENOEXEC */
1470 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1479 EXPORT_SYMBOL(search_binary_handler);
1482 * sys_execve() executes a new program.
1484 static int do_execve_common(const char *filename,
1485 struct user_arg_ptr argv,
1486 struct user_arg_ptr envp,
1487 struct pt_regs *regs)
1489 struct linux_binprm *bprm;
1491 struct files_struct *displaced;
1494 const struct cred *cred = current_cred();
1497 * We move the actual failure in case of RLIMIT_NPROC excess from
1498 * set*uid() to execve() because too many poorly written programs
1499 * don't check setuid() return code. Here we additionally recheck
1500 * whether NPROC limit is still exceeded.
1502 if ((current->flags & PF_NPROC_EXCEEDED) &&
1503 atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1508 /* We're below the limit (still or again), so we don't want to make
1509 * further execve() calls fail. */
1510 current->flags &= ~PF_NPROC_EXCEEDED;
1512 retval = unshare_files(&displaced);
1517 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1521 retval = prepare_bprm_creds(bprm);
1525 retval = check_unsafe_exec(bprm);
1528 clear_in_exec = retval;
1529 current->in_execve = 1;
1531 file = open_exec(filename);
1532 retval = PTR_ERR(file);
1539 bprm->filename = filename;
1540 bprm->interp = filename;
1542 retval = bprm_mm_init(bprm);
1546 bprm->argc = count(argv, MAX_ARG_STRINGS);
1547 if ((retval = bprm->argc) < 0)
1550 bprm->envc = count(envp, MAX_ARG_STRINGS);
1551 if ((retval = bprm->envc) < 0)
1554 retval = prepare_binprm(bprm);
1558 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1562 bprm->exec = bprm->p;
1563 retval = copy_strings(bprm->envc, envp, bprm);
1567 retval = copy_strings(bprm->argc, argv, bprm);
1571 retval = search_binary_handler(bprm,regs);
1575 /* execve succeeded */
1576 current->fs->in_exec = 0;
1577 current->in_execve = 0;
1578 acct_update_integrals(current);
1581 put_files_struct(displaced);
1586 acct_arg_size(bprm, 0);
1592 allow_write_access(bprm->file);
1598 current->fs->in_exec = 0;
1599 current->in_execve = 0;
1606 reset_files_struct(displaced);
1611 int do_execve(const char *filename,
1612 const char __user *const __user *__argv,
1613 const char __user *const __user *__envp,
1614 struct pt_regs *regs)
1616 struct user_arg_ptr argv = { .ptr.native = __argv };
1617 struct user_arg_ptr envp = { .ptr.native = __envp };
1618 return do_execve_common(filename, argv, envp, regs);
1621 #ifdef CONFIG_COMPAT
1622 int compat_do_execve(char *filename,
1623 compat_uptr_t __user *__argv,
1624 compat_uptr_t __user *__envp,
1625 struct pt_regs *regs)
1627 struct user_arg_ptr argv = {
1629 .ptr.compat = __argv,
1631 struct user_arg_ptr envp = {
1633 .ptr.compat = __envp,
1635 return do_execve_common(filename, argv, envp, regs);
1639 void set_binfmt(struct linux_binfmt *new)
1641 struct mm_struct *mm = current->mm;
1644 module_put(mm->binfmt->module);
1648 __module_get(new->module);
1651 EXPORT_SYMBOL(set_binfmt);
1653 static int expand_corename(struct core_name *cn)
1655 char *old_corename = cn->corename;
1657 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1658 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1660 if (!cn->corename) {
1661 kfree(old_corename);
1668 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1676 need = vsnprintf(NULL, 0, fmt, arg);
1679 if (likely(need < cn->size - cn->used - 1))
1682 ret = expand_corename(cn);
1687 cur = cn->corename + cn->used;
1689 vsnprintf(cur, need + 1, fmt, arg);
1698 static void cn_escape(char *str)
1705 static int cn_print_exe_file(struct core_name *cn)
1707 struct file *exe_file;
1708 char *pathbuf, *path;
1711 exe_file = get_mm_exe_file(current->mm);
1713 char *commstart = cn->corename + cn->used;
1714 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1715 cn_escape(commstart);
1719 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1725 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1727 ret = PTR_ERR(path);
1733 ret = cn_printf(cn, "%s", path);
1742 /* format_corename will inspect the pattern parameter, and output a
1743 * name into corename, which must have space for at least
1744 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1746 static int format_corename(struct core_name *cn, long signr)
1748 const struct cred *cred = current_cred();
1749 const char *pat_ptr = core_pattern;
1750 int ispipe = (*pat_ptr == '|');
1751 int pid_in_pattern = 0;
1754 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1755 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1761 /* Repeat as long as we have more pattern to process and more output
1764 if (*pat_ptr != '%') {
1767 err = cn_printf(cn, "%c", *pat_ptr++);
1769 switch (*++pat_ptr) {
1770 /* single % at the end, drop that */
1773 /* Double percent, output one percent */
1775 err = cn_printf(cn, "%c", '%');
1780 err = cn_printf(cn, "%d",
1781 task_tgid_vnr(current));
1785 err = cn_printf(cn, "%d", cred->uid);
1789 err = cn_printf(cn, "%d", cred->gid);
1791 /* signal that caused the coredump */
1793 err = cn_printf(cn, "%ld", signr);
1795 /* UNIX time of coredump */
1798 do_gettimeofday(&tv);
1799 err = cn_printf(cn, "%lu", tv.tv_sec);
1804 char *namestart = cn->corename + cn->used;
1805 down_read(&uts_sem);
1806 err = cn_printf(cn, "%s",
1807 utsname()->nodename);
1809 cn_escape(namestart);
1814 char *commstart = cn->corename + cn->used;
1815 err = cn_printf(cn, "%s", current->comm);
1816 cn_escape(commstart);
1820 err = cn_print_exe_file(cn);
1822 /* core limit size */
1824 err = cn_printf(cn, "%lu",
1825 rlimit(RLIMIT_CORE));
1837 /* Backward compatibility with core_uses_pid:
1839 * If core_pattern does not include a %p (as is the default)
1840 * and core_uses_pid is set, then .%pid will be appended to
1841 * the filename. Do not do this for piped commands. */
1842 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1843 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1851 static int zap_process(struct task_struct *start, int exit_code)
1853 struct task_struct *t;
1856 start->signal->flags = SIGNAL_GROUP_EXIT;
1857 start->signal->group_exit_code = exit_code;
1858 start->signal->group_stop_count = 0;
1862 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1863 if (t != current && t->mm) {
1864 sigaddset(&t->pending.signal, SIGKILL);
1865 signal_wake_up(t, 1);
1868 } while_each_thread(start, t);
1873 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1874 struct core_state *core_state, int exit_code)
1876 struct task_struct *g, *p;
1877 unsigned long flags;
1880 spin_lock_irq(&tsk->sighand->siglock);
1881 if (!signal_group_exit(tsk->signal)) {
1882 mm->core_state = core_state;
1883 nr = zap_process(tsk, exit_code);
1885 spin_unlock_irq(&tsk->sighand->siglock);
1886 if (unlikely(nr < 0))
1889 if (atomic_read(&mm->mm_users) == nr + 1)
1892 * We should find and kill all tasks which use this mm, and we should
1893 * count them correctly into ->nr_threads. We don't take tasklist
1894 * lock, but this is safe wrt:
1897 * None of sub-threads can fork after zap_process(leader). All
1898 * processes which were created before this point should be
1899 * visible to zap_threads() because copy_process() adds the new
1900 * process to the tail of init_task.tasks list, and lock/unlock
1901 * of ->siglock provides a memory barrier.
1904 * The caller holds mm->mmap_sem. This means that the task which
1905 * uses this mm can't pass exit_mm(), so it can't exit or clear
1909 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1910 * we must see either old or new leader, this does not matter.
1911 * However, it can change p->sighand, so lock_task_sighand(p)
1912 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1915 * Note also that "g" can be the old leader with ->mm == NULL
1916 * and already unhashed and thus removed from ->thread_group.
1917 * This is OK, __unhash_process()->list_del_rcu() does not
1918 * clear the ->next pointer, we will find the new leader via
1922 for_each_process(g) {
1923 if (g == tsk->group_leader)
1925 if (g->flags & PF_KTHREAD)
1930 if (unlikely(p->mm == mm)) {
1931 lock_task_sighand(p, &flags);
1932 nr += zap_process(p, exit_code);
1933 unlock_task_sighand(p, &flags);
1937 } while_each_thread(g, p);
1941 atomic_set(&core_state->nr_threads, nr);
1945 static int coredump_wait(int exit_code, struct core_state *core_state)
1947 struct task_struct *tsk = current;
1948 struct mm_struct *mm = tsk->mm;
1949 struct completion *vfork_done;
1950 int core_waiters = -EBUSY;
1952 init_completion(&core_state->startup);
1953 core_state->dumper.task = tsk;
1954 core_state->dumper.next = NULL;
1956 down_write(&mm->mmap_sem);
1957 if (!mm->core_state)
1958 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1959 up_write(&mm->mmap_sem);
1961 if (unlikely(core_waiters < 0))
1965 * Make sure nobody is waiting for us to release the VM,
1966 * otherwise we can deadlock when we wait on each other
1968 vfork_done = tsk->vfork_done;
1970 tsk->vfork_done = NULL;
1971 complete(vfork_done);
1975 wait_for_completion(&core_state->startup);
1977 return core_waiters;
1980 static void coredump_finish(struct mm_struct *mm)
1982 struct core_thread *curr, *next;
1983 struct task_struct *task;
1985 next = mm->core_state->dumper.next;
1986 while ((curr = next) != NULL) {
1990 * see exit_mm(), curr->task must not see
1991 * ->task == NULL before we read ->next.
1995 wake_up_process(task);
1998 mm->core_state = NULL;
2002 * set_dumpable converts traditional three-value dumpable to two flags and
2003 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
2004 * these bits are not changed atomically. So get_dumpable can observe the
2005 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
2006 * return either old dumpable or new one by paying attention to the order of
2007 * modifying the bits.
2009 * dumpable | mm->flags (binary)
2010 * old new | initial interim final
2011 * ---------+-----------------------
2019 * (*) get_dumpable regards interim value of 10 as 11.
2021 void set_dumpable(struct mm_struct *mm, int value)
2025 clear_bit(MMF_DUMPABLE, &mm->flags);
2027 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2030 set_bit(MMF_DUMPABLE, &mm->flags);
2032 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2035 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2037 set_bit(MMF_DUMPABLE, &mm->flags);
2042 static int __get_dumpable(unsigned long mm_flags)
2046 ret = mm_flags & MMF_DUMPABLE_MASK;
2047 return (ret >= 2) ? 2 : ret;
2051 * This returns the actual value of the suid_dumpable flag. For things
2052 * that are using this for checking for privilege transitions, it must
2053 * test against SUID_DUMP_USER rather than treating it as a boolean
2056 int get_dumpable(struct mm_struct *mm)
2058 return __get_dumpable(mm->flags);
2061 static void wait_for_dump_helpers(struct file *file)
2063 struct pipe_inode_info *pipe;
2065 pipe = file->f_path.dentry->d_inode->i_pipe;
2071 while ((pipe->readers > 1) && (!signal_pending(current))) {
2072 wake_up_interruptible_sync(&pipe->wait);
2073 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2086 * helper function to customize the process used
2087 * to collect the core in userspace. Specifically
2088 * it sets up a pipe and installs it as fd 0 (stdin)
2089 * for the process. Returns 0 on success, or
2090 * PTR_ERR on failure.
2091 * Note that it also sets the core limit to 1. This
2092 * is a special value that we use to trap recursive
2095 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2097 struct file *rp, *wp;
2098 struct fdtable *fdt;
2099 struct coredump_params *cp = (struct coredump_params *)info->data;
2100 struct files_struct *cf = current->files;
2102 wp = create_write_pipe(0);
2106 rp = create_read_pipe(wp, 0);
2108 free_write_pipe(wp);
2116 spin_lock(&cf->file_lock);
2117 fdt = files_fdtable(cf);
2118 __set_open_fd(0, fdt);
2119 __clear_close_on_exec(0, fdt);
2120 spin_unlock(&cf->file_lock);
2122 /* and disallow core files too */
2123 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2128 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2130 struct core_state core_state;
2131 struct core_name cn;
2132 struct mm_struct *mm = current->mm;
2133 struct linux_binfmt * binfmt;
2134 const struct cred *old_cred;
2138 /* require nonrelative corefile path and be extra careful */
2139 bool need_suid_safe = false;
2140 static atomic_t core_dump_count = ATOMIC_INIT(0);
2141 struct coredump_params cprm = {
2144 .limit = rlimit(RLIMIT_CORE),
2146 * We must use the same mm->flags while dumping core to avoid
2147 * inconsistency of bit flags, since this flag is not protected
2150 .mm_flags = mm->flags,
2153 audit_core_dumps(signr);
2155 binfmt = mm->binfmt;
2156 if (!binfmt || !binfmt->core_dump)
2158 if (!__get_dumpable(cprm.mm_flags))
2161 cred = prepare_creds();
2165 * We cannot trust fsuid as being the "true" uid of the process
2166 * nor do we know its entire history. We only know it was tainted
2167 * so we dump it as root in mode 2, and only into a controlled
2168 * environment (pipe handler or fully qualified path).
2170 if (__get_dumpable(cprm.mm_flags) == 2) {
2171 /* Setuid core dump mode */
2172 cred->fsuid = 0; /* Dump root private */
2173 need_suid_safe = true;
2176 retval = coredump_wait(exit_code, &core_state);
2180 old_cred = override_creds(cred);
2183 * Clear any false indication of pending signals that might
2184 * be seen by the filesystem code called to write the core file.
2186 clear_thread_flag(TIF_SIGPENDING);
2188 ispipe = format_corename(&cn, signr);
2195 printk(KERN_WARNING "format_corename failed\n");
2196 printk(KERN_WARNING "Aborting core\n");
2200 if (cprm.limit == 1) {
2202 * Normally core limits are irrelevant to pipes, since
2203 * we're not writing to the file system, but we use
2204 * cprm.limit of 1 here as a speacial value. Any
2205 * non-1 limit gets set to RLIM_INFINITY below, but
2206 * a limit of 0 skips the dump. This is a consistent
2207 * way to catch recursive crashes. We can still crash
2208 * if the core_pattern binary sets RLIM_CORE = !1
2209 * but it runs as root, and can do lots of stupid things
2210 * Note that we use task_tgid_vnr here to grab the pid
2211 * of the process group leader. That way we get the
2212 * right pid if a thread in a multi-threaded
2213 * core_pattern process dies.
2216 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2217 task_tgid_vnr(current), current->comm);
2218 printk(KERN_WARNING "Aborting core\n");
2221 cprm.limit = RLIM_INFINITY;
2223 dump_count = atomic_inc_return(&core_dump_count);
2224 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2225 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2226 task_tgid_vnr(current), current->comm);
2227 printk(KERN_WARNING "Skipping core dump\n");
2228 goto fail_dropcount;
2231 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2233 printk(KERN_WARNING "%s failed to allocate memory\n",
2235 goto fail_dropcount;
2238 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2239 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2241 argv_free(helper_argv);
2243 printk(KERN_INFO "Core dump to %s pipe failed\n",
2248 struct inode *inode;
2250 if (cprm.limit < binfmt->min_coredump)
2253 if (need_suid_safe && cn.corename[0] != '/') {
2254 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
2255 "to fully qualified path!\n",
2256 task_tgid_vnr(current), current->comm);
2257 printk(KERN_WARNING "Skipping core dump\n");
2262 * Unlink the file if it exists unless this is a SUID
2263 * binary - in that case, we're running around with root
2264 * privs and don't want to unlink another user's coredump.
2266 if (!need_suid_safe) {
2267 mm_segment_t old_fs;
2272 * If it doesn't exist, that's fine. If there's some
2273 * other problem, we'll catch it at the filp_open().
2275 (void) sys_unlink((const char __user *)cn.corename);
2280 * There is a race between unlinking and creating the
2281 * file, but if that causes an EEXIST here, that's
2282 * fine - another process raced with us while creating
2283 * the corefile, and the other process won. To userspace,
2284 * what matters is that at least one of the two processes
2285 * writes its coredump successfully, not which one.
2287 cprm.file = filp_open(cn.corename,
2288 O_CREAT | 2 | O_NOFOLLOW |
2289 O_LARGEFILE | O_EXCL,
2291 if (IS_ERR(cprm.file))
2294 inode = cprm.file->f_path.dentry->d_inode;
2295 if (inode->i_nlink > 1)
2297 if (d_unhashed(cprm.file->f_path.dentry))
2300 * AK: actually i see no reason to not allow this for named
2301 * pipes etc, but keep the previous behaviour for now.
2303 if (!S_ISREG(inode->i_mode))
2306 * Dont allow local users get cute and trick others to coredump
2307 * into their pre-created files.
2309 if (inode->i_uid != current_fsuid())
2311 if (!cprm.file->f_op || !cprm.file->f_op->write)
2313 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2317 retval = binfmt->core_dump(&cprm);
2319 current->signal->group_exit_code |= 0x80;
2321 if (ispipe && core_pipe_limit)
2322 wait_for_dump_helpers(cprm.file);
2325 filp_close(cprm.file, NULL);
2328 atomic_dec(&core_dump_count);
2332 coredump_finish(mm);
2333 revert_creds(old_cred);
2341 * Core dumping helper functions. These are the only things you should
2342 * do on a core-file: use only these functions to write out all the
2345 int dump_write(struct file *file, const void *addr, int nr)
2347 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2349 EXPORT_SYMBOL(dump_write);
2351 int dump_seek(struct file *file, loff_t off)
2355 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2356 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2359 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2364 unsigned long n = off;
2368 if (!dump_write(file, buf, n)) {
2374 free_page((unsigned long)buf);
2378 EXPORT_SYMBOL(dump_seek);