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>
45 #include <trace/events/sched.h>
48 unsigned int core_pipe_limit;
49 char core_pattern[CORENAME_MAX_SIZE] = "core";
50 static int core_name_size = CORENAME_MAX_SIZE;
57 /* The maximal length of core_pattern is also specified in sysctl.c */
59 static int expand_corename(struct core_name *cn, int size)
61 char *corename = krealloc(cn->corename, size, GFP_KERNEL);
66 if (size > core_name_size) /* racy but harmless */
67 core_name_size = size;
69 cn->size = ksize(corename);
70 cn->corename = corename;
74 static int cn_vprintf(struct core_name *cn, const char *fmt, va_list arg)
79 free = cn->size - cn->used;
80 need = vsnprintf(cn->corename + cn->used, free, fmt, arg);
86 if (!expand_corename(cn, cn->size + need - free + 1))
92 static int cn_printf(struct core_name *cn, const char *fmt, ...)
98 ret = cn_vprintf(cn, fmt, arg);
104 static int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
111 ret = cn_vprintf(cn, fmt, arg);
114 for (; cur < cn->used; ++cur) {
115 if (cn->corename[cur] == '/')
116 cn->corename[cur] = '!';
121 static int cn_print_exe_file(struct core_name *cn)
123 struct file *exe_file;
124 char *pathbuf, *path;
127 exe_file = get_mm_exe_file(current->mm);
129 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
131 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
137 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
143 ret = cn_esc_printf(cn, "%s", path);
152 /* format_corename will inspect the pattern parameter, and output a
153 * name into corename, which must have space for at least
154 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
156 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
158 const struct cred *cred = current_cred();
159 const char *pat_ptr = core_pattern;
160 int ispipe = (*pat_ptr == '|');
161 int pid_in_pattern = 0;
166 if (expand_corename(cn, core_name_size))
169 /* Repeat as long as we have more pattern to process and more output
172 if (*pat_ptr != '%') {
175 err = cn_printf(cn, "%c", *pat_ptr++);
177 switch (*++pat_ptr) {
178 /* single % at the end, drop that */
181 /* Double percent, output one percent */
183 err = cn_printf(cn, "%c", '%');
188 err = cn_printf(cn, "%d",
189 task_tgid_vnr(current));
193 err = cn_printf(cn, "%d", cred->uid);
197 err = cn_printf(cn, "%d", cred->gid);
200 err = cn_printf(cn, "%d",
201 __get_dumpable(cprm->mm_flags));
203 /* signal that caused the coredump */
205 err = cn_printf(cn, "%ld", cprm->siginfo->si_signo);
207 /* UNIX time of coredump */
210 do_gettimeofday(&tv);
211 err = cn_printf(cn, "%lu", tv.tv_sec);
217 err = cn_esc_printf(cn, "%s",
218 utsname()->nodename);
223 err = cn_esc_printf(cn, "%s", current->comm);
226 err = cn_print_exe_file(cn);
228 /* core limit size */
230 err = cn_printf(cn, "%lu",
231 rlimit(RLIMIT_CORE));
243 /* Backward compatibility with core_uses_pid:
245 * If core_pattern does not include a %p (as is the default)
246 * and core_uses_pid is set, then .%pid will be appended to
247 * the filename. Do not do this for piped commands. */
248 if (!ispipe && !pid_in_pattern && core_uses_pid) {
249 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
257 static int zap_process(struct task_struct *start, int exit_code)
259 struct task_struct *t;
262 start->signal->group_exit_code = exit_code;
263 start->signal->group_stop_count = 0;
267 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
268 if (t != current && t->mm) {
269 sigaddset(&t->pending.signal, SIGKILL);
270 signal_wake_up(t, 1);
273 } while_each_thread(start, t);
278 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
279 struct core_state *core_state, int exit_code)
281 struct task_struct *g, *p;
285 spin_lock_irq(&tsk->sighand->siglock);
286 if (!signal_group_exit(tsk->signal)) {
287 mm->core_state = core_state;
288 nr = zap_process(tsk, exit_code);
289 tsk->signal->group_exit_task = tsk;
290 /* ignore all signals except SIGKILL, see prepare_signal() */
291 tsk->signal->flags = SIGNAL_GROUP_COREDUMP;
292 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
294 spin_unlock_irq(&tsk->sighand->siglock);
295 if (unlikely(nr < 0))
298 tsk->flags = PF_DUMPCORE;
299 if (atomic_read(&mm->mm_users) == nr + 1)
302 * We should find and kill all tasks which use this mm, and we should
303 * count them correctly into ->nr_threads. We don't take tasklist
304 * lock, but this is safe wrt:
307 * None of sub-threads can fork after zap_process(leader). All
308 * processes which were created before this point should be
309 * visible to zap_threads() because copy_process() adds the new
310 * process to the tail of init_task.tasks list, and lock/unlock
311 * of ->siglock provides a memory barrier.
314 * The caller holds mm->mmap_sem. This means that the task which
315 * uses this mm can't pass exit_mm(), so it can't exit or clear
319 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
320 * we must see either old or new leader, this does not matter.
321 * However, it can change p->sighand, so lock_task_sighand(p)
322 * must be used. Since p->mm != NULL and we hold ->mmap_sem
325 * Note also that "g" can be the old leader with ->mm == NULL
326 * and already unhashed and thus removed from ->thread_group.
327 * This is OK, __unhash_process()->list_del_rcu() does not
328 * clear the ->next pointer, we will find the new leader via
332 for_each_process(g) {
333 if (g == tsk->group_leader)
335 if (g->flags & PF_KTHREAD)
340 if (unlikely(p->mm == mm)) {
341 lock_task_sighand(p, &flags);
342 nr += zap_process(p, exit_code);
343 p->signal->flags = SIGNAL_GROUP_EXIT;
344 unlock_task_sighand(p, &flags);
348 } while_each_thread(g, p);
352 atomic_set(&core_state->nr_threads, nr);
356 static int coredump_wait(int exit_code, struct core_state *core_state)
358 struct task_struct *tsk = current;
359 struct mm_struct *mm = tsk->mm;
360 int core_waiters = -EBUSY;
362 init_completion(&core_state->startup);
363 core_state->dumper.task = tsk;
364 core_state->dumper.next = NULL;
366 down_write(&mm->mmap_sem);
368 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
369 up_write(&mm->mmap_sem);
371 if (core_waiters > 0) {
372 struct core_thread *ptr;
374 wait_for_completion(&core_state->startup);
376 * Wait for all the threads to become inactive, so that
377 * all the thread context (extended register state, like
378 * fpu etc) gets copied to the memory.
380 ptr = core_state->dumper.next;
381 while (ptr != NULL) {
382 wait_task_inactive(ptr->task, 0);
390 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
392 struct core_thread *curr, *next;
393 struct task_struct *task;
395 spin_lock_irq(¤t->sighand->siglock);
396 if (core_dumped && !__fatal_signal_pending(current))
397 current->signal->group_exit_code |= 0x80;
398 current->signal->group_exit_task = NULL;
399 current->signal->flags = SIGNAL_GROUP_EXIT;
400 spin_unlock_irq(¤t->sighand->siglock);
402 next = mm->core_state->dumper.next;
403 while ((curr = next) != NULL) {
407 * see exit_mm(), curr->task must not see
408 * ->task == NULL before we read ->next.
412 wake_up_process(task);
415 mm->core_state = NULL;
418 static bool dump_interrupted(void)
421 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
422 * can do try_to_freeze() and check __fatal_signal_pending(),
423 * but then we need to teach dump_write() to restart and clear
426 return signal_pending(current);
429 static void wait_for_dump_helpers(struct file *file)
431 struct pipe_inode_info *pipe = file->private_data;
436 wake_up_interruptible_sync(&pipe->wait);
437 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
441 * We actually want wait_event_freezable() but then we need
442 * to clear TIF_SIGPENDING and improve dump_interrupted().
444 wait_event_interruptible(pipe->wait, pipe->readers == 1);
454 * helper function to customize the process used
455 * to collect the core in userspace. Specifically
456 * it sets up a pipe and installs it as fd 0 (stdin)
457 * for the process. Returns 0 on success, or
458 * PTR_ERR on failure.
459 * Note that it also sets the core limit to 1. This
460 * is a special value that we use to trap recursive
463 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
465 struct file *files[2];
466 struct coredump_params *cp = (struct coredump_params *)info->data;
467 int err = create_pipe_files(files, 0);
473 err = replace_fd(0, files[0], 0);
475 /* and disallow core files too */
476 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
481 void do_coredump(siginfo_t *siginfo)
483 struct core_state core_state;
485 struct mm_struct *mm = current->mm;
486 struct linux_binfmt * binfmt;
487 const struct cred *old_cred;
492 struct files_struct *displaced;
493 bool need_nonrelative = false;
494 bool core_dumped = false;
495 static atomic_t core_dump_count = ATOMIC_INIT(0);
496 struct coredump_params cprm = {
498 .regs = signal_pt_regs(),
499 .limit = rlimit(RLIMIT_CORE),
501 * We must use the same mm->flags while dumping core to avoid
502 * inconsistency of bit flags, since this flag is not protected
505 .mm_flags = mm->flags,
508 audit_core_dumps(siginfo->si_signo);
511 if (!binfmt || !binfmt->core_dump)
513 if (!__get_dumpable(cprm.mm_flags))
516 cred = prepare_creds();
520 * We cannot trust fsuid as being the "true" uid of the process
521 * nor do we know its entire history. We only know it was tainted
522 * so we dump it as root in mode 2, and only into a controlled
523 * environment (pipe handler or fully qualified path).
525 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
526 /* Setuid core dump mode */
527 flag = O_EXCL; /* Stop rewrite attacks */
528 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
529 need_nonrelative = true;
532 retval = coredump_wait(siginfo->si_signo, &core_state);
536 old_cred = override_creds(cred);
538 ispipe = format_corename(&cn, &cprm);
543 struct subprocess_info *sub_info;
546 printk(KERN_WARNING "format_corename failed\n");
547 printk(KERN_WARNING "Aborting core\n");
551 if (cprm.limit == 1) {
552 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
554 * Normally core limits are irrelevant to pipes, since
555 * we're not writing to the file system, but we use
556 * cprm.limit of 1 here as a speacial value, this is a
557 * consistent way to catch recursive crashes.
558 * We can still crash if the core_pattern binary sets
559 * RLIM_CORE = !1, but it runs as root, and can do
560 * lots of stupid things.
562 * Note that we use task_tgid_vnr here to grab the pid
563 * of the process group leader. That way we get the
564 * right pid if a thread in a multi-threaded
565 * core_pattern process dies.
568 "Process %d(%s) has RLIMIT_CORE set to 1\n",
569 task_tgid_vnr(current), current->comm);
570 printk(KERN_WARNING "Aborting core\n");
573 cprm.limit = RLIM_INFINITY;
575 dump_count = atomic_inc_return(&core_dump_count);
576 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
577 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
578 task_tgid_vnr(current), current->comm);
579 printk(KERN_WARNING "Skipping core dump\n");
583 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
585 printk(KERN_WARNING "%s failed to allocate memory\n",
591 sub_info = call_usermodehelper_setup(helper_argv[0],
592 helper_argv, NULL, GFP_KERNEL,
593 umh_pipe_setup, NULL, &cprm);
595 retval = call_usermodehelper_exec(sub_info,
598 argv_free(helper_argv);
600 printk(KERN_INFO "Core dump to %s pipe failed\n",
607 if (cprm.limit < binfmt->min_coredump)
610 if (need_nonrelative && cn.corename[0] != '/') {
611 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
612 "to fully qualified path!\n",
613 task_tgid_vnr(current), current->comm);
614 printk(KERN_WARNING "Skipping core dump\n");
618 cprm.file = filp_open(cn.corename,
619 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
621 if (IS_ERR(cprm.file))
624 inode = file_inode(cprm.file);
625 if (inode->i_nlink > 1)
627 if (d_unhashed(cprm.file->f_path.dentry))
630 * AK: actually i see no reason to not allow this for named
631 * pipes etc, but keep the previous behaviour for now.
633 if (!S_ISREG(inode->i_mode))
636 * Dont allow local users get cute and trick others to coredump
637 * into their pre-created files.
639 if (!uid_eq(inode->i_uid, current_fsuid()))
641 if (!cprm.file->f_op || !cprm.file->f_op->write)
643 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
647 /* get us an unshared descriptor table; almost always a no-op */
648 retval = unshare_files(&displaced);
652 put_files_struct(displaced);
653 if (!dump_interrupted()) {
654 file_start_write(cprm.file);
655 core_dumped = binfmt->core_dump(&cprm);
656 file_end_write(cprm.file);
658 if (ispipe && core_pipe_limit)
659 wait_for_dump_helpers(cprm.file);
662 filp_close(cprm.file, NULL);
665 atomic_dec(&core_dump_count);
668 coredump_finish(mm, core_dumped);
669 revert_creds(old_cred);
677 * Core dumping helper functions. These are the only things you should
678 * do on a core-file: use only these functions to write out all the
681 int dump_write(struct file *file, const void *addr, int nr)
683 return !dump_interrupted() &&
684 access_ok(VERIFY_READ, addr, nr) &&
685 file->f_op->write(file, addr, nr, &file->f_pos) == nr;
687 EXPORT_SYMBOL(dump_write);
689 int dump_seek(struct file *file, loff_t off)
693 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
694 if (dump_interrupted() ||
695 file->f_op->llseek(file, off, SEEK_CUR) < 0)
698 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
703 unsigned long n = off;
707 if (!dump_write(file, buf, n)) {
713 free_page((unsigned long)buf);
717 EXPORT_SYMBOL(dump_seek);