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 __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
80 free = cn->size - cn->used;
82 va_copy(arg_copy, arg);
83 need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
91 if (!expand_corename(cn, cn->size + need - free + 1))
97 static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
103 ret = cn_vprintf(cn, fmt, arg);
109 static __printf(2, 3)
110 int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
117 ret = cn_vprintf(cn, fmt, arg);
120 for (; cur < cn->used; ++cur) {
121 if (cn->corename[cur] == '/')
122 cn->corename[cur] = '!';
127 static int cn_print_exe_file(struct core_name *cn)
129 struct file *exe_file;
130 char *pathbuf, *path;
133 exe_file = get_mm_exe_file(current->mm);
135 return cn_esc_printf(cn, "%s (path unknown)", current->comm);
137 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
143 path = file_path(exe_file, pathbuf, PATH_MAX);
149 ret = cn_esc_printf(cn, "%s", path);
158 /* format_corename will inspect the pattern parameter, and output a
159 * name into corename, which must have space for at least
160 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
162 static int format_corename(struct core_name *cn, struct coredump_params *cprm)
164 const struct cred *cred = current_cred();
165 const char *pat_ptr = core_pattern;
166 int ispipe = (*pat_ptr == '|');
167 int pid_in_pattern = 0;
172 if (expand_corename(cn, core_name_size))
174 cn->corename[0] = '\0';
179 /* Repeat as long as we have more pattern to process and more output
182 if (*pat_ptr != '%') {
183 err = cn_printf(cn, "%c", *pat_ptr++);
185 switch (*++pat_ptr) {
186 /* single % at the end, drop that */
189 /* Double percent, output one percent */
191 err = cn_printf(cn, "%c", '%');
196 err = cn_printf(cn, "%d",
197 task_tgid_vnr(current));
201 err = cn_printf(cn, "%d",
202 task_tgid_nr(current));
205 err = cn_printf(cn, "%d",
206 task_pid_vnr(current));
209 err = cn_printf(cn, "%d",
210 task_pid_nr(current));
214 err = cn_printf(cn, "%u",
215 from_kuid(&init_user_ns,
220 err = cn_printf(cn, "%u",
221 from_kgid(&init_user_ns,
225 err = cn_printf(cn, "%d",
226 __get_dumpable(cprm->mm_flags));
228 /* signal that caused the coredump */
230 err = cn_printf(cn, "%d",
231 cprm->siginfo->si_signo);
233 /* UNIX time of coredump */
236 do_gettimeofday(&tv);
237 err = cn_printf(cn, "%lu", tv.tv_sec);
243 err = cn_esc_printf(cn, "%s",
244 utsname()->nodename);
249 err = cn_esc_printf(cn, "%s", current->comm);
252 err = cn_print_exe_file(cn);
254 /* core limit size */
256 err = cn_printf(cn, "%lu",
257 rlimit(RLIMIT_CORE));
270 /* Backward compatibility with core_uses_pid:
272 * If core_pattern does not include a %p (as is the default)
273 * and core_uses_pid is set, then .%pid will be appended to
274 * the filename. Do not do this for piped commands. */
275 if (!ispipe && !pid_in_pattern && core_uses_pid) {
276 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
283 static int zap_process(struct task_struct *start, int exit_code)
285 struct task_struct *t;
288 start->signal->group_exit_code = exit_code;
289 start->signal->group_stop_count = 0;
293 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
294 if (t != current && t->mm) {
295 sigaddset(&t->pending.signal, SIGKILL);
296 signal_wake_up(t, 1);
299 } while_each_thread(start, t);
304 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
305 struct core_state *core_state, int exit_code)
307 struct task_struct *g, *p;
311 spin_lock_irq(&tsk->sighand->siglock);
312 if (!signal_group_exit(tsk->signal)) {
313 mm->core_state = core_state;
314 nr = zap_process(tsk, exit_code);
315 tsk->signal->group_exit_task = tsk;
316 /* ignore all signals except SIGKILL, see prepare_signal() */
317 tsk->signal->flags = SIGNAL_GROUP_COREDUMP;
318 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
320 spin_unlock_irq(&tsk->sighand->siglock);
321 if (unlikely(nr < 0))
324 tsk->flags |= PF_DUMPCORE;
325 if (atomic_read(&mm->mm_users) == nr + 1)
328 * We should find and kill all tasks which use this mm, and we should
329 * count them correctly into ->nr_threads. We don't take tasklist
330 * lock, but this is safe wrt:
333 * None of sub-threads can fork after zap_process(leader). All
334 * processes which were created before this point should be
335 * visible to zap_threads() because copy_process() adds the new
336 * process to the tail of init_task.tasks list, and lock/unlock
337 * of ->siglock provides a memory barrier.
340 * The caller holds mm->mmap_sem. This means that the task which
341 * uses this mm can't pass exit_mm(), so it can't exit or clear
345 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
346 * we must see either old or new leader, this does not matter.
347 * However, it can change p->sighand, so lock_task_sighand(p)
348 * must be used. Since p->mm != NULL and we hold ->mmap_sem
351 * Note also that "g" can be the old leader with ->mm == NULL
352 * and already unhashed and thus removed from ->thread_group.
353 * This is OK, __unhash_process()->list_del_rcu() does not
354 * clear the ->next pointer, we will find the new leader via
358 for_each_process(g) {
359 if (g == tsk->group_leader)
361 if (g->flags & PF_KTHREAD)
366 if (unlikely(p->mm == mm)) {
367 lock_task_sighand(p, &flags);
368 nr += zap_process(p, exit_code);
369 p->signal->flags = SIGNAL_GROUP_EXIT;
370 unlock_task_sighand(p, &flags);
374 } while_each_thread(g, p);
378 atomic_set(&core_state->nr_threads, nr);
382 static int coredump_wait(int exit_code, struct core_state *core_state)
384 struct task_struct *tsk = current;
385 struct mm_struct *mm = tsk->mm;
386 int core_waiters = -EBUSY;
388 init_completion(&core_state->startup);
389 core_state->dumper.task = tsk;
390 core_state->dumper.next = NULL;
392 down_write(&mm->mmap_sem);
394 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
395 up_write(&mm->mmap_sem);
397 if (core_waiters > 0) {
398 struct core_thread *ptr;
400 wait_for_completion(&core_state->startup);
402 * Wait for all the threads to become inactive, so that
403 * all the thread context (extended register state, like
404 * fpu etc) gets copied to the memory.
406 ptr = core_state->dumper.next;
407 while (ptr != NULL) {
408 wait_task_inactive(ptr->task, 0);
416 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
418 struct core_thread *curr, *next;
419 struct task_struct *task;
421 spin_lock_irq(¤t->sighand->siglock);
422 if (core_dumped && !__fatal_signal_pending(current))
423 current->signal->group_exit_code |= 0x80;
424 current->signal->group_exit_task = NULL;
425 current->signal->flags = SIGNAL_GROUP_EXIT;
426 spin_unlock_irq(¤t->sighand->siglock);
428 next = mm->core_state->dumper.next;
429 while ((curr = next) != NULL) {
433 * see exit_mm(), curr->task must not see
434 * ->task == NULL before we read ->next.
438 wake_up_process(task);
441 mm->core_state = NULL;
444 static bool dump_interrupted(void)
447 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
448 * can do try_to_freeze() and check __fatal_signal_pending(),
449 * but then we need to teach dump_write() to restart and clear
452 return signal_pending(current);
455 static void wait_for_dump_helpers(struct file *file)
457 struct pipe_inode_info *pipe = file->private_data;
462 wake_up_interruptible_sync(&pipe->wait);
463 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
467 * We actually want wait_event_freezable() but then we need
468 * to clear TIF_SIGPENDING and improve dump_interrupted().
470 wait_event_interruptible(pipe->wait, pipe->readers == 1);
480 * helper function to customize the process used
481 * to collect the core in userspace. Specifically
482 * it sets up a pipe and installs it as fd 0 (stdin)
483 * for the process. Returns 0 on success, or
484 * PTR_ERR on failure.
485 * Note that it also sets the core limit to 1. This
486 * is a special value that we use to trap recursive
489 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
491 struct file *files[2];
492 struct coredump_params *cp = (struct coredump_params *)info->data;
493 int err = create_pipe_files(files, 0);
499 err = replace_fd(0, files[0], 0);
501 /* and disallow core files too */
502 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
507 void do_coredump(const siginfo_t *siginfo)
509 struct core_state core_state;
511 struct mm_struct *mm = current->mm;
512 struct linux_binfmt * binfmt;
513 const struct cred *old_cred;
517 struct files_struct *displaced;
518 /* require nonrelative corefile path and be extra careful */
519 bool need_suid_safe = false;
520 bool core_dumped = false;
521 static atomic_t core_dump_count = ATOMIC_INIT(0);
522 struct coredump_params cprm = {
524 .regs = signal_pt_regs(),
525 .limit = rlimit(RLIMIT_CORE),
527 * We must use the same mm->flags while dumping core to avoid
528 * inconsistency of bit flags, since this flag is not protected
531 .mm_flags = mm->flags,
534 audit_core_dumps(siginfo->si_signo);
537 if (!binfmt || !binfmt->core_dump)
539 if (!__get_dumpable(cprm.mm_flags))
542 cred = prepare_creds();
546 * We cannot trust fsuid as being the "true" uid of the process
547 * nor do we know its entire history. We only know it was tainted
548 * so we dump it as root in mode 2, and only into a controlled
549 * environment (pipe handler or fully qualified path).
551 if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
552 /* Setuid core dump mode */
553 cred->fsuid = GLOBAL_ROOT_UID; /* Dump root private */
554 need_suid_safe = true;
557 retval = coredump_wait(siginfo->si_signo, &core_state);
561 old_cred = override_creds(cred);
563 ispipe = format_corename(&cn, &cprm);
568 struct subprocess_info *sub_info;
571 printk(KERN_WARNING "format_corename failed\n");
572 printk(KERN_WARNING "Aborting core\n");
576 if (cprm.limit == 1) {
577 /* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
579 * Normally core limits are irrelevant to pipes, since
580 * we're not writing to the file system, but we use
581 * cprm.limit of 1 here as a special value, this is a
582 * consistent way to catch recursive crashes.
583 * We can still crash if the core_pattern binary sets
584 * RLIM_CORE = !1, but it runs as root, and can do
585 * lots of stupid things.
587 * Note that we use task_tgid_vnr here to grab the pid
588 * of the process group leader. That way we get the
589 * right pid if a thread in a multi-threaded
590 * core_pattern process dies.
593 "Process %d(%s) has RLIMIT_CORE set to 1\n",
594 task_tgid_vnr(current), current->comm);
595 printk(KERN_WARNING "Aborting core\n");
598 cprm.limit = RLIM_INFINITY;
600 dump_count = atomic_inc_return(&core_dump_count);
601 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
602 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
603 task_tgid_vnr(current), current->comm);
604 printk(KERN_WARNING "Skipping core dump\n");
608 helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
610 printk(KERN_WARNING "%s failed to allocate memory\n",
616 sub_info = call_usermodehelper_setup(helper_argv[0],
617 helper_argv, NULL, GFP_KERNEL,
618 umh_pipe_setup, NULL, &cprm);
620 retval = call_usermodehelper_exec(sub_info,
623 argv_free(helper_argv);
625 printk(KERN_INFO "Core dump to |%s pipe failed\n",
632 if (cprm.limit < binfmt->min_coredump)
635 if (need_suid_safe && cn.corename[0] != '/') {
636 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
637 "to fully qualified path!\n",
638 task_tgid_vnr(current), current->comm);
639 printk(KERN_WARNING "Skipping core dump\n");
644 * Unlink the file if it exists unless this is a SUID
645 * binary - in that case, we're running around with root
646 * privs and don't want to unlink another user's coredump.
648 if (!need_suid_safe) {
654 * If it doesn't exist, that's fine. If there's some
655 * other problem, we'll catch it at the filp_open().
657 (void) sys_unlink((const char __user *)cn.corename);
662 * There is a race between unlinking and creating the
663 * file, but if that causes an EEXIST here, that's
664 * fine - another process raced with us while creating
665 * the corefile, and the other process won. To userspace,
666 * what matters is that at least one of the two processes
667 * writes its coredump successfully, not which one.
669 cprm.file = filp_open(cn.corename,
670 O_CREAT | 2 | O_NOFOLLOW |
671 O_LARGEFILE | O_EXCL,
673 if (IS_ERR(cprm.file))
676 inode = file_inode(cprm.file);
677 if (inode->i_nlink > 1)
679 if (d_unhashed(cprm.file->f_path.dentry))
682 * AK: actually i see no reason to not allow this for named
683 * pipes etc, but keep the previous behaviour for now.
685 if (!S_ISREG(inode->i_mode))
688 * Don't dump core if the filesystem changed owner or mode
689 * of the file during file creation. This is an issue when
690 * a process dumps core while its cwd is e.g. on a vfat
693 if (!uid_eq(inode->i_uid, current_fsuid()))
695 if ((inode->i_mode & 0677) != 0600)
697 if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
699 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
703 /* get us an unshared descriptor table; almost always a no-op */
704 retval = unshare_files(&displaced);
708 put_files_struct(displaced);
709 if (!dump_interrupted()) {
710 file_start_write(cprm.file);
711 core_dumped = binfmt->core_dump(&cprm);
712 file_end_write(cprm.file);
714 if (ispipe && core_pipe_limit)
715 wait_for_dump_helpers(cprm.file);
718 filp_close(cprm.file, NULL);
721 atomic_dec(&core_dump_count);
724 coredump_finish(mm, core_dumped);
725 revert_creds(old_cred);
733 * Core dumping helper functions. These are the only things you should
734 * do on a core-file: use only these functions to write out all the
737 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
739 struct file *file = cprm->file;
740 loff_t pos = file->f_pos;
742 if (cprm->written + nr > cprm->limit)
745 if (dump_interrupted())
747 n = __kernel_write(file, addr, nr, &pos);
756 EXPORT_SYMBOL(dump_emit);
758 int dump_skip(struct coredump_params *cprm, size_t nr)
760 static char zeroes[PAGE_SIZE];
761 struct file *file = cprm->file;
762 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
763 if (cprm->written + nr > cprm->limit)
765 if (dump_interrupted() ||
766 file->f_op->llseek(file, nr, SEEK_CUR) < 0)
771 while (nr > PAGE_SIZE) {
772 if (!dump_emit(cprm, zeroes, PAGE_SIZE))
776 return dump_emit(cprm, zeroes, nr);
779 EXPORT_SYMBOL(dump_skip);
781 int dump_align(struct coredump_params *cprm, int align)
783 unsigned mod = cprm->written & (align - 1);
784 if (align & (align - 1))
786 return mod ? dump_skip(cprm, align - mod) : 1;
788 EXPORT_SYMBOL(dump_align);