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 char core_pattern[CORENAME_MAX_SIZE] = "core";
49 unsigned int core_pipe_limit;
55 static atomic_t call_count = ATOMIC_INIT(1);
57 /* The maximal length of core_pattern is also specified in sysctl.c */
59 static int expand_corename(struct core_name *cn)
61 int size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
62 char *corename = krealloc(cn->corename, size, GFP_KERNEL);
68 cn->corename = corename;
72 static int cn_printf(struct core_name *cn, const char *fmt, ...)
80 need = vsnprintf(NULL, 0, fmt, arg);
83 if (likely(need < cn->size - cn->used - 1))
86 ret = expand_corename(cn);
91 cur = cn->corename + cn->used;
93 vsnprintf(cur, need + 1, fmt, arg);
102 static void cn_escape(char *str)
109 static int cn_print_exe_file(struct core_name *cn)
111 struct file *exe_file;
112 char *pathbuf, *path;
115 exe_file = get_mm_exe_file(current->mm);
117 char *commstart = cn->corename + cn->used;
118 ret = cn_printf(cn, "%s (path unknown)", current->comm);
119 cn_escape(commstart);
123 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
129 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
137 ret = cn_printf(cn, "%s", path);
146 /* format_corename will inspect the pattern parameter, and output a
147 * name into corename, which must have space for at least
148 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
150 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
152 const struct cred *cred = current_cred();
153 const char *pat_ptr = core_pattern;
154 int ispipe = (*pat_ptr == '|');
155 int pid_in_pattern = 0;
159 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
160 cn->corename = kmalloc(cn->size, GFP_KERNEL);
164 /* Repeat as long as we have more pattern to process and more output
167 if (*pat_ptr != '%') {
170 err = cn_printf(cn, "%c", *pat_ptr++);
172 switch (*++pat_ptr) {
173 /* single % at the end, drop that */
176 /* Double percent, output one percent */
178 err = cn_printf(cn, "%c", '%');
183 err = cn_printf(cn, "%d",
184 task_tgid_vnr(current));
188 err = cn_printf(cn, "%d", cred->uid);
192 err = cn_printf(cn, "%d", cred->gid);
195 err = cn_printf(cn, "%d",
196 __get_dumpable(cprm->mm_flags));
198 /* signal that caused the coredump */
200 err = cn_printf(cn, "%ld", cprm->siginfo->si_signo);
202 /* UNIX time of coredump */
205 do_gettimeofday(&tv);
206 err = cn_printf(cn, "%lu", tv.tv_sec);
211 char *namestart = cn->corename + cn->used;
213 err = cn_printf(cn, "%s",
214 utsname()->nodename);
216 cn_escape(namestart);
221 char *commstart = cn->corename + cn->used;
222 err = cn_printf(cn, "%s", current->comm);
223 cn_escape(commstart);
227 err = cn_print_exe_file(cn);
229 /* core limit size */
231 err = cn_printf(cn, "%lu",
232 rlimit(RLIMIT_CORE));
244 /* Backward compatibility with core_uses_pid:
246 * If core_pattern does not include a %p (as is the default)
247 * and core_uses_pid is set, then .%pid will be appended to
248 * the filename. Do not do this for piped commands. */
249 if (!ispipe && !pid_in_pattern && core_uses_pid) {
250 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
258 static int zap_process(struct task_struct *start, int exit_code)
260 struct task_struct *t;
263 start->signal->group_exit_code = exit_code;
264 start->signal->group_stop_count = 0;
268 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
269 if (t != current && t->mm) {
270 sigaddset(&t->pending.signal, SIGKILL);
271 signal_wake_up(t, 1);
274 } while_each_thread(start, t);
279 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
280 struct core_state *core_state, int exit_code)
282 struct task_struct *g, *p;
286 spin_lock_irq(&tsk->sighand->siglock);
287 if (!signal_group_exit(tsk->signal)) {
288 mm->core_state = core_state;
289 nr = zap_process(tsk, exit_code);
290 tsk->signal->group_exit_task = tsk;
291 /* ignore all signals except SIGKILL, see prepare_signal() */
292 tsk->signal->flags = SIGNAL_GROUP_COREDUMP;
293 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
295 spin_unlock_irq(&tsk->sighand->siglock);
296 if (unlikely(nr < 0))
299 tsk->flags = PF_DUMPCORE;
300 if (atomic_read(&mm->mm_users) == nr + 1)
303 * We should find and kill all tasks which use this mm, and we should
304 * count them correctly into ->nr_threads. We don't take tasklist
305 * lock, but this is safe wrt:
308 * None of sub-threads can fork after zap_process(leader). All
309 * processes which were created before this point should be
310 * visible to zap_threads() because copy_process() adds the new
311 * process to the tail of init_task.tasks list, and lock/unlock
312 * of ->siglock provides a memory barrier.
315 * The caller holds mm->mmap_sem. This means that the task which
316 * uses this mm can't pass exit_mm(), so it can't exit or clear
320 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
321 * we must see either old or new leader, this does not matter.
322 * However, it can change p->sighand, so lock_task_sighand(p)
323 * must be used. Since p->mm != NULL and we hold ->mmap_sem
326 * Note also that "g" can be the old leader with ->mm == NULL
327 * and already unhashed and thus removed from ->thread_group.
328 * This is OK, __unhash_process()->list_del_rcu() does not
329 * clear the ->next pointer, we will find the new leader via
333 for_each_process(g) {
334 if (g == tsk->group_leader)
336 if (g->flags & PF_KTHREAD)
341 if (unlikely(p->mm == mm)) {
342 lock_task_sighand(p, &flags);
343 nr += zap_process(p, exit_code);
344 p->signal->flags = SIGNAL_GROUP_EXIT;
345 unlock_task_sighand(p, &flags);
349 } while_each_thread(g, p);
353 atomic_set(&core_state->nr_threads, nr);
357 static int coredump_wait(int exit_code, struct core_state *core_state)
359 struct task_struct *tsk = current;
360 struct mm_struct *mm = tsk->mm;
361 int core_waiters = -EBUSY;
363 init_completion(&core_state->startup);
364 core_state->dumper.task = tsk;
365 core_state->dumper.next = NULL;
367 down_write(&mm->mmap_sem);
369 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
370 up_write(&mm->mmap_sem);
372 if (core_waiters > 0) {
373 struct core_thread *ptr;
375 wait_for_completion(&core_state->startup);
377 * Wait for all the threads to become inactive, so that
378 * all the thread context (extended register state, like
379 * fpu etc) gets copied to the memory.
381 ptr = core_state->dumper.next;
382 while (ptr != NULL) {
383 wait_task_inactive(ptr->task, 0);
391 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
393 struct core_thread *curr, *next;
394 struct task_struct *task;
396 spin_lock_irq(¤t->sighand->siglock);
397 if (core_dumped && !__fatal_signal_pending(current))
398 current->signal->group_exit_code |= 0x80;
399 current->signal->group_exit_task = NULL;
400 current->signal->flags = SIGNAL_GROUP_EXIT;
401 spin_unlock_irq(¤t->sighand->siglock);
403 next = mm->core_state->dumper.next;
404 while ((curr = next) != NULL) {
408 * see exit_mm(), curr->task must not see
409 * ->task == NULL before we read ->next.
413 wake_up_process(task);
416 mm->core_state = NULL;
419 static bool dump_interrupted(void)
422 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
423 * can do try_to_freeze() and check __fatal_signal_pending(),
424 * but then we need to teach dump_write() to restart and clear
427 return signal_pending(current);
430 static void wait_for_dump_helpers(struct file *file)
432 struct pipe_inode_info *pipe = file->private_data;
437 wake_up_interruptible_sync(&pipe->wait);
438 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
442 * We actually want wait_event_freezable() but then we need
443 * to clear TIF_SIGPENDING and improve dump_interrupted().
445 wait_event_interruptible(pipe->wait, pipe->readers == 1);
455 * helper function to customize the process used
456 * to collect the core in userspace. Specifically
457 * it sets up a pipe and installs it as fd 0 (stdin)
458 * for the process. Returns 0 on success, or
459 * PTR_ERR on failure.
460 * Note that it also sets the core limit to 1. This
461 * is a special value that we use to trap recursive
464 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
466 struct file *files[2];
467 struct coredump_params *cp = (struct coredump_params *)info->data;
468 int err = create_pipe_files(files, 0);
474 err = replace_fd(0, files[0], 0);
476 /* and disallow core files too */
477 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
482 void do_coredump(siginfo_t *siginfo)
484 struct core_state core_state;
486 struct mm_struct *mm = current->mm;
487 struct linux_binfmt * binfmt;
488 const struct cred *old_cred;
493 struct files_struct *displaced;
494 bool need_nonrelative = false;
495 bool core_dumped = false;
496 static atomic_t core_dump_count = ATOMIC_INIT(0);
497 struct coredump_params cprm = {
499 .regs = signal_pt_regs(),
500 .limit = rlimit(RLIMIT_CORE),
502 * We must use the same mm->flags while dumping core to avoid
503 * inconsistency of bit flags, since this flag is not protected
506 .mm_flags = mm->flags,
509 audit_core_dumps(siginfo->si_signo);
512 if (!binfmt || !binfmt->core_dump)
514 if (!__get_dumpable(cprm.mm_flags))
517 cred = prepare_creds();
521 * We cannot trust fsuid as being the "true" uid of the process
522 * nor do we know its entire history. We only know it was tainted
523 * so we dump it as root in mode 2, and only into a controlled
524 * environment (pipe handler or fully qualified path).
526 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
527 /* Setuid core dump mode */
528 flag = O_EXCL; /* Stop rewrite attacks */
529 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
530 need_nonrelative = true;
533 retval = coredump_wait(siginfo->si_signo, &core_state);
537 old_cred = override_creds(cred);
539 ispipe = format_corename(&cn, &cprm);
544 struct subprocess_info *sub_info;
547 printk(KERN_WARNING "format_corename failed\n");
548 printk(KERN_WARNING "Aborting core\n");
552 if (cprm.limit == 1) {
553 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
555 * Normally core limits are irrelevant to pipes, since
556 * we're not writing to the file system, but we use
557 * cprm.limit of 1 here as a speacial value, this is a
558 * consistent way to catch recursive crashes.
559 * We can still crash if the core_pattern binary sets
560 * RLIM_CORE = !1, but it runs as root, and can do
561 * lots of stupid things.
563 * Note that we use task_tgid_vnr here to grab the pid
564 * of the process group leader. That way we get the
565 * right pid if a thread in a multi-threaded
566 * core_pattern process dies.
569 "Process %d(%s) has RLIMIT_CORE set to 1\n",
570 task_tgid_vnr(current), current->comm);
571 printk(KERN_WARNING "Aborting core\n");
574 cprm.limit = RLIM_INFINITY;
576 dump_count = atomic_inc_return(&core_dump_count);
577 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
578 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
579 task_tgid_vnr(current), current->comm);
580 printk(KERN_WARNING "Skipping core dump\n");
584 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
586 printk(KERN_WARNING "%s failed to allocate memory\n",
592 sub_info = call_usermodehelper_setup(helper_argv[0],
593 helper_argv, NULL, GFP_KERNEL,
594 umh_pipe_setup, NULL, &cprm);
596 retval = call_usermodehelper_exec(sub_info,
599 argv_free(helper_argv);
601 printk(KERN_INFO "Core dump to %s pipe failed\n",
608 if (cprm.limit < binfmt->min_coredump)
611 if (need_nonrelative && cn.corename[0] != '/') {
612 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
613 "to fully qualified path!\n",
614 task_tgid_vnr(current), current->comm);
615 printk(KERN_WARNING "Skipping core dump\n");
619 cprm.file = filp_open(cn.corename,
620 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
622 if (IS_ERR(cprm.file))
625 inode = file_inode(cprm.file);
626 if (inode->i_nlink > 1)
628 if (d_unhashed(cprm.file->f_path.dentry))
631 * AK: actually i see no reason to not allow this for named
632 * pipes etc, but keep the previous behaviour for now.
634 if (!S_ISREG(inode->i_mode))
637 * Dont allow local users get cute and trick others to coredump
638 * into their pre-created files.
640 if (!uid_eq(inode->i_uid, current_fsuid()))
642 if (!cprm.file->f_op || !cprm.file->f_op->write)
644 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
648 /* get us an unshared descriptor table; almost always a no-op */
649 retval = unshare_files(&displaced);
653 put_files_struct(displaced);
654 if (!dump_interrupted()) {
655 file_start_write(cprm.file);
656 core_dumped = binfmt->core_dump(&cprm);
657 file_end_write(cprm.file);
659 if (ispipe && core_pipe_limit)
660 wait_for_dump_helpers(cprm.file);
663 filp_close(cprm.file, NULL);
666 atomic_dec(&core_dump_count);
669 coredump_finish(mm, core_dumped);
670 revert_creds(old_cred);
678 * Core dumping helper functions. These are the only things you should
679 * do on a core-file: use only these functions to write out all the
682 int dump_write(struct file *file, const void *addr, int nr)
684 return !dump_interrupted() &&
685 access_ok(VERIFY_READ, addr, nr) &&
686 file->f_op->write(file, addr, nr, &file->f_pos) == nr;
688 EXPORT_SYMBOL(dump_write);
690 int dump_seek(struct file *file, loff_t off)
694 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
695 if (dump_interrupted() ||
696 file->f_op->llseek(file, off, SEEK_CUR) < 0)
699 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
704 unsigned long n = off;
708 if (!dump_write(file, buf, n)) {
714 free_page((unsigned long)buf);
718 EXPORT_SYMBOL(dump_seek);