1 #include <linux/slab.h>
2 #include <linux/file.h>
3 #include <linux/fdtable.h>
5 #include <linux/stat.h>
6 #include <linux/fcntl.h>
7 #include <linux/swap.h>
8 #include <linux/string.h>
9 #include <linux/init.h>
10 #include <linux/pagemap.h>
11 #include <linux/perf_event.h>
12 #include <linux/highmem.h>
13 #include <linux/spinlock.h>
14 #include <linux/key.h>
15 #include <linux/personality.h>
16 #include <linux/binfmts.h>
17 #include <linux/coredump.h>
18 #include <linux/utsname.h>
19 #include <linux/pid_namespace.h>
20 #include <linux/module.h>
21 #include <linux/namei.h>
22 #include <linux/mount.h>
23 #include <linux/security.h>
24 #include <linux/syscalls.h>
25 #include <linux/tsacct_kern.h>
26 #include <linux/cn_proc.h>
27 #include <linux/audit.h>
28 #include <linux/tracehook.h>
29 #include <linux/kmod.h>
30 #include <linux/fsnotify.h>
31 #include <linux/fs_struct.h>
32 #include <linux/pipe_fs_i.h>
33 #include <linux/oom.h>
34 #include <linux/compat.h>
36 #include <asm/uaccess.h>
37 #include <asm/mmu_context.h>
41 #include <trace/events/task.h>
44 #include <trace/events/sched.h>
47 unsigned int core_pipe_limit;
48 char core_pattern[CORENAME_MAX_SIZE] = "core";
49 static int core_name_size = CORENAME_MAX_SIZE;
56 /* The maximal length of core_pattern is also specified in sysctl.c */
58 static int expand_corename(struct core_name *cn, int size)
60 char *corename = krealloc(cn->corename, size, GFP_KERNEL);
65 if (size > core_name_size) /* racy but harmless */
66 core_name_size = size;
68 cn->size = ksize(corename);
69 cn->corename = corename;
73 static int cn_vprintf(struct core_name *cn, const char *fmt, va_list arg)
79 free = cn->size - cn->used;
81 va_copy(arg_copy, arg);
82 need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
90 if (!expand_corename(cn, cn->size + need - free + 1))
96 static int cn_printf(struct core_name *cn, const char *fmt, ...)
102 ret = cn_vprintf(cn, fmt, arg);
108 static int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
115 ret = cn_vprintf(cn, fmt, arg);
118 for (; cur < cn->used; ++cur) {
119 if (cn->corename[cur] == '/')
120 cn->corename[cur] = '!';
125 static int cn_print_exe_file(struct core_name *cn)
127 struct file *exe_file;
128 char *pathbuf, *path;
131 exe_file = get_mm_exe_file(current->mm);
133 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
135 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
141 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
147 ret = cn_esc_printf(cn, "%s", path);
156 /* format_corename will inspect the pattern parameter, and output a
157 * name into corename, which must have space for at least
158 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
160 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
162 const struct cred *cred = current_cred();
163 const char *pat_ptr = core_pattern;
164 int ispipe = (*pat_ptr == '|');
165 int pid_in_pattern = 0;
170 if (expand_corename(cn, core_name_size))
172 cn->corename[0] = '\0';
177 /* Repeat as long as we have more pattern to process and more output
180 if (*pat_ptr != '%') {
181 err = cn_printf(cn, "%c", *pat_ptr++);
183 switch (*++pat_ptr) {
184 /* single % at the end, drop that */
187 /* Double percent, output one percent */
189 err = cn_printf(cn, "%c", '%');
194 err = cn_printf(cn, "%d",
195 task_tgid_vnr(current));
199 err = cn_printf(cn, "%d",
200 task_tgid_nr(current));
203 err = cn_printf(cn, "%d",
204 task_pid_vnr(current));
207 err = cn_printf(cn, "%d",
208 task_pid_nr(current));
212 err = cn_printf(cn, "%d", cred->uid);
216 err = cn_printf(cn, "%d", cred->gid);
219 err = cn_printf(cn, "%d",
220 __get_dumpable(cprm->mm_flags));
222 /* signal that caused the coredump */
224 err = cn_printf(cn, "%ld", cprm->siginfo->si_signo);
226 /* UNIX time of coredump */
229 do_gettimeofday(&tv);
230 err = cn_printf(cn, "%lu", tv.tv_sec);
236 err = cn_esc_printf(cn, "%s",
237 utsname()->nodename);
242 err = cn_esc_printf(cn, "%s", current->comm);
245 err = cn_print_exe_file(cn);
247 /* core limit size */
249 err = cn_printf(cn, "%lu",
250 rlimit(RLIMIT_CORE));
263 /* Backward compatibility with core_uses_pid:
265 * If core_pattern does not include a %p (as is the default)
266 * and core_uses_pid is set, then .%pid will be appended to
267 * the filename. Do not do this for piped commands. */
268 if (!ispipe && !pid_in_pattern && core_uses_pid) {
269 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
276 static int zap_process(struct task_struct *start, int exit_code)
278 struct task_struct *t;
281 start->signal->group_exit_code = exit_code;
282 start->signal->group_stop_count = 0;
286 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
287 if (t != current && t->mm) {
288 sigaddset(&t->pending.signal, SIGKILL);
289 signal_wake_up(t, 1);
292 } while_each_thread(start, t);
297 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
298 struct core_state *core_state, int exit_code)
300 struct task_struct *g, *p;
304 spin_lock_irq(&tsk->sighand->siglock);
305 if (!signal_group_exit(tsk->signal)) {
306 mm->core_state = core_state;
307 nr = zap_process(tsk, exit_code);
308 tsk->signal->group_exit_task = tsk;
309 /* ignore all signals except SIGKILL, see prepare_signal() */
310 tsk->signal->flags = SIGNAL_GROUP_COREDUMP;
311 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
313 spin_unlock_irq(&tsk->sighand->siglock);
314 if (unlikely(nr < 0))
317 tsk->flags |= PF_DUMPCORE;
318 if (atomic_read(&mm->mm_users) == nr + 1)
321 * We should find and kill all tasks which use this mm, and we should
322 * count them correctly into ->nr_threads. We don't take tasklist
323 * lock, but this is safe wrt:
326 * None of sub-threads can fork after zap_process(leader). All
327 * processes which were created before this point should be
328 * visible to zap_threads() because copy_process() adds the new
329 * process to the tail of init_task.tasks list, and lock/unlock
330 * of ->siglock provides a memory barrier.
333 * The caller holds mm->mmap_sem. This means that the task which
334 * uses this mm can't pass exit_mm(), so it can't exit or clear
338 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
339 * we must see either old or new leader, this does not matter.
340 * However, it can change p->sighand, so lock_task_sighand(p)
341 * must be used. Since p->mm != NULL and we hold ->mmap_sem
344 * Note also that "g" can be the old leader with ->mm == NULL
345 * and already unhashed and thus removed from ->thread_group.
346 * This is OK, __unhash_process()->list_del_rcu() does not
347 * clear the ->next pointer, we will find the new leader via
351 for_each_process(g) {
352 if (g == tsk->group_leader)
354 if (g->flags & PF_KTHREAD)
359 if (unlikely(p->mm == mm)) {
360 lock_task_sighand(p, &flags);
361 nr += zap_process(p, exit_code);
362 p->signal->flags = SIGNAL_GROUP_EXIT;
363 unlock_task_sighand(p, &flags);
367 } while_each_thread(g, p);
371 atomic_set(&core_state->nr_threads, nr);
375 static int coredump_wait(int exit_code, struct core_state *core_state)
377 struct task_struct *tsk = current;
378 struct mm_struct *mm = tsk->mm;
379 int core_waiters = -EBUSY;
381 init_completion(&core_state->startup);
382 core_state->dumper.task = tsk;
383 core_state->dumper.next = NULL;
385 down_write(&mm->mmap_sem);
387 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
388 up_write(&mm->mmap_sem);
390 if (core_waiters > 0) {
391 struct core_thread *ptr;
393 wait_for_completion(&core_state->startup);
395 * Wait for all the threads to become inactive, so that
396 * all the thread context (extended register state, like
397 * fpu etc) gets copied to the memory.
399 ptr = core_state->dumper.next;
400 while (ptr != NULL) {
401 wait_task_inactive(ptr->task, 0);
409 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
411 struct core_thread *curr, *next;
412 struct task_struct *task;
414 spin_lock_irq(¤t->sighand->siglock);
415 if (core_dumped && !__fatal_signal_pending(current))
416 current->signal->group_exit_code |= 0x80;
417 current->signal->group_exit_task = NULL;
418 current->signal->flags = SIGNAL_GROUP_EXIT;
419 spin_unlock_irq(¤t->sighand->siglock);
421 next = mm->core_state->dumper.next;
422 while ((curr = next) != NULL) {
426 * see exit_mm(), curr->task must not see
427 * ->task == NULL before we read ->next.
431 wake_up_process(task);
434 mm->core_state = NULL;
437 static bool dump_interrupted(void)
440 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
441 * can do try_to_freeze() and check __fatal_signal_pending(),
442 * but then we need to teach dump_write() to restart and clear
445 return signal_pending(current);
448 static void wait_for_dump_helpers(struct file *file)
450 struct pipe_inode_info *pipe = file->private_data;
455 wake_up_interruptible_sync(&pipe->wait);
456 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
460 * We actually want wait_event_freezable() but then we need
461 * to clear TIF_SIGPENDING and improve dump_interrupted().
463 wait_event_interruptible(pipe->wait, pipe->readers == 1);
473 * helper function to customize the process used
474 * to collect the core in userspace. Specifically
475 * it sets up a pipe and installs it as fd 0 (stdin)
476 * for the process. Returns 0 on success, or
477 * PTR_ERR on failure.
478 * Note that it also sets the core limit to 1. This
479 * is a special value that we use to trap recursive
482 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
484 struct file *files[2];
485 struct coredump_params *cp = (struct coredump_params *)info->data;
486 int err = create_pipe_files(files, 0);
492 err = replace_fd(0, files[0], 0);
494 /* and disallow core files too */
495 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
500 void do_coredump(const siginfo_t *siginfo)
502 struct core_state core_state;
504 struct mm_struct *mm = current->mm;
505 struct linux_binfmt * binfmt;
506 const struct cred *old_cred;
511 struct files_struct *displaced;
512 bool need_nonrelative = false;
513 bool core_dumped = false;
514 static atomic_t core_dump_count = ATOMIC_INIT(0);
515 struct coredump_params cprm = {
517 .regs = signal_pt_regs(),
518 .limit = rlimit(RLIMIT_CORE),
520 * We must use the same mm->flags while dumping core to avoid
521 * inconsistency of bit flags, since this flag is not protected
524 .mm_flags = mm->flags,
527 audit_core_dumps(siginfo->si_signo);
530 if (!binfmt || !binfmt->core_dump)
532 if (!__get_dumpable(cprm.mm_flags))
535 cred = prepare_creds();
539 * We cannot trust fsuid as being the "true" uid of the process
540 * nor do we know its entire history. We only know it was tainted
541 * so we dump it as root in mode 2, and only into a controlled
542 * environment (pipe handler or fully qualified path).
544 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
545 /* Setuid core dump mode */
546 flag = O_EXCL; /* Stop rewrite attacks */
547 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
548 need_nonrelative = true;
551 retval = coredump_wait(siginfo->si_signo, &core_state);
555 old_cred = override_creds(cred);
557 ispipe = format_corename(&cn, &cprm);
562 struct subprocess_info *sub_info;
565 printk(KERN_WARNING "format_corename failed\n");
566 printk(KERN_WARNING "Aborting core\n");
570 if (cprm.limit == 1) {
571 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
573 * Normally core limits are irrelevant to pipes, since
574 * we're not writing to the file system, but we use
575 * cprm.limit of 1 here as a special value, this is a
576 * consistent way to catch recursive crashes.
577 * We can still crash if the core_pattern binary sets
578 * RLIM_CORE = !1, but it runs as root, and can do
579 * lots of stupid things.
581 * Note that we use task_tgid_vnr here to grab the pid
582 * of the process group leader. That way we get the
583 * right pid if a thread in a multi-threaded
584 * core_pattern process dies.
587 "Process %d(%s) has RLIMIT_CORE set to 1\n",
588 task_tgid_vnr(current), current->comm);
589 printk(KERN_WARNING "Aborting core\n");
592 cprm.limit = RLIM_INFINITY;
594 dump_count = atomic_inc_return(&core_dump_count);
595 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
596 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
597 task_tgid_vnr(current), current->comm);
598 printk(KERN_WARNING "Skipping core dump\n");
602 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
604 printk(KERN_WARNING "%s failed to allocate memory\n",
610 sub_info = call_usermodehelper_setup(helper_argv[0],
611 helper_argv, NULL, GFP_KERNEL,
612 umh_pipe_setup, NULL, &cprm);
614 retval = call_usermodehelper_exec(sub_info,
617 argv_free(helper_argv);
619 printk(KERN_INFO "Core dump to |%s pipe failed\n",
626 if (cprm.limit < binfmt->min_coredump)
629 if (need_nonrelative && cn.corename[0] != '/') {
630 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
631 "to fully qualified path!\n",
632 task_tgid_vnr(current), current->comm);
633 printk(KERN_WARNING "Skipping core dump\n");
637 cprm.file = filp_open(cn.corename,
638 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
640 if (IS_ERR(cprm.file))
643 inode = file_inode(cprm.file);
644 if (inode->i_nlink > 1)
646 if (d_unhashed(cprm.file->f_path.dentry))
649 * AK: actually i see no reason to not allow this for named
650 * pipes etc, but keep the previous behaviour for now.
652 if (!S_ISREG(inode->i_mode))
655 * Dont allow local users get cute and trick others to coredump
656 * into their pre-created files.
658 if (!uid_eq(inode->i_uid, current_fsuid()))
660 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
662 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
666 /* get us an unshared descriptor table; almost always a no-op */
667 retval = unshare_files(&displaced);
671 put_files_struct(displaced);
672 if (!dump_interrupted()) {
673 file_start_write(cprm.file);
674 core_dumped = binfmt->core_dump(&cprm);
675 file_end_write(cprm.file);
677 if (ispipe && core_pipe_limit)
678 wait_for_dump_helpers(cprm.file);
681 filp_close(cprm.file, NULL);
684 atomic_dec(&core_dump_count);
687 coredump_finish(mm, core_dumped);
688 revert_creds(old_cred);
696 * Core dumping helper functions. These are the only things you should
697 * do on a core-file: use only these functions to write out all the
700 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
702 struct file *file = cprm->file;
703 loff_t pos = file->f_pos;
705 if (cprm->written + nr > cprm->limit)
708 if (dump_interrupted())
710 n = __kernel_write(file, addr, nr, &pos);
719 EXPORT_SYMBOL(dump_emit);
721 int dump_skip(struct coredump_params *cprm, size_t nr)
723 static char zeroes[PAGE_SIZE];
724 struct file *file = cprm->file;
725 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
726 if (cprm->written + nr > cprm->limit)
728 if (dump_interrupted() ||
729 file->f_op->llseek(file, nr, SEEK_CUR) < 0)
734 while (nr > PAGE_SIZE) {
735 if (!dump_emit(cprm, zeroes, PAGE_SIZE))
739 return dump_emit(cprm, zeroes, nr);
742 EXPORT_SYMBOL(dump_skip);
744 int dump_align(struct coredump_params *cprm, int align)
746 unsigned mod = cprm->written & (align - 1);
747 if (align & (align - 1))
749 return mod ? dump_skip(cprm, align - mod) : 1;
751 EXPORT_SYMBOL(dump_align);