xref: /openbmc/linux/fs/coredump.c (revision bd4af432)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/slab.h>
3 #include <linux/file.h>
4 #include <linux/fdtable.h>
5 #include <linux/freezer.h>
6 #include <linux/mm.h>
7 #include <linux/stat.h>
8 #include <linux/fcntl.h>
9 #include <linux/swap.h>
10 #include <linux/ctype.h>
11 #include <linux/string.h>
12 #include <linux/init.h>
13 #include <linux/pagemap.h>
14 #include <linux/perf_event.h>
15 #include <linux/highmem.h>
16 #include <linux/spinlock.h>
17 #include <linux/key.h>
18 #include <linux/personality.h>
19 #include <linux/binfmts.h>
20 #include <linux/coredump.h>
21 #include <linux/sched/coredump.h>
22 #include <linux/sched/signal.h>
23 #include <linux/sched/task_stack.h>
24 #include <linux/utsname.h>
25 #include <linux/pid_namespace.h>
26 #include <linux/module.h>
27 #include <linux/namei.h>
28 #include <linux/mount.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
31 #include <linux/tsacct_kern.h>
32 #include <linux/cn_proc.h>
33 #include <linux/audit.h>
34 #include <linux/tracehook.h>
35 #include <linux/kmod.h>
36 #include <linux/fsnotify.h>
37 #include <linux/fs_struct.h>
38 #include <linux/pipe_fs_i.h>
39 #include <linux/oom.h>
40 #include <linux/compat.h>
41 #include <linux/fs.h>
42 #include <linux/path.h>
43 #include <linux/timekeeping.h>
44 
45 #include <linux/uaccess.h>
46 #include <asm/mmu_context.h>
47 #include <asm/tlb.h>
48 #include <asm/exec.h>
49 
50 #include <trace/events/task.h>
51 #include "internal.h"
52 
53 #include <trace/events/sched.h>
54 
55 int core_uses_pid;
56 unsigned int core_pipe_limit;
57 char core_pattern[CORENAME_MAX_SIZE] = "core";
58 static int core_name_size = CORENAME_MAX_SIZE;
59 
60 struct core_name {
61 	char *corename;
62 	int used, size;
63 };
64 
65 /* The maximal length of core_pattern is also specified in sysctl.c */
66 
67 static int expand_corename(struct core_name *cn, int size)
68 {
69 	char *corename = krealloc(cn->corename, size, GFP_KERNEL);
70 
71 	if (!corename)
72 		return -ENOMEM;
73 
74 	if (size > core_name_size) /* racy but harmless */
75 		core_name_size = size;
76 
77 	cn->size = ksize(corename);
78 	cn->corename = corename;
79 	return 0;
80 }
81 
82 static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
83 				     va_list arg)
84 {
85 	int free, need;
86 	va_list arg_copy;
87 
88 again:
89 	free = cn->size - cn->used;
90 
91 	va_copy(arg_copy, arg);
92 	need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
93 	va_end(arg_copy);
94 
95 	if (need < free) {
96 		cn->used += need;
97 		return 0;
98 	}
99 
100 	if (!expand_corename(cn, cn->size + need - free + 1))
101 		goto again;
102 
103 	return -ENOMEM;
104 }
105 
106 static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
107 {
108 	va_list arg;
109 	int ret;
110 
111 	va_start(arg, fmt);
112 	ret = cn_vprintf(cn, fmt, arg);
113 	va_end(arg);
114 
115 	return ret;
116 }
117 
118 static __printf(2, 3)
119 int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
120 {
121 	int cur = cn->used;
122 	va_list arg;
123 	int ret;
124 
125 	va_start(arg, fmt);
126 	ret = cn_vprintf(cn, fmt, arg);
127 	va_end(arg);
128 
129 	if (ret == 0) {
130 		/*
131 		 * Ensure that this coredump name component can't cause the
132 		 * resulting corefile path to consist of a ".." or ".".
133 		 */
134 		if ((cn->used - cur == 1 && cn->corename[cur] == '.') ||
135 				(cn->used - cur == 2 && cn->corename[cur] == '.'
136 				&& cn->corename[cur+1] == '.'))
137 			cn->corename[cur] = '!';
138 
139 		/*
140 		 * Empty names are fishy and could be used to create a "//" in a
141 		 * corefile name, causing the coredump to happen one directory
142 		 * level too high. Enforce that all components of the core
143 		 * pattern are at least one character long.
144 		 */
145 		if (cn->used == cur)
146 			ret = cn_printf(cn, "!");
147 	}
148 
149 	for (; cur < cn->used; ++cur) {
150 		if (cn->corename[cur] == '/')
151 			cn->corename[cur] = '!';
152 	}
153 	return ret;
154 }
155 
156 static int cn_print_exe_file(struct core_name *cn)
157 {
158 	struct file *exe_file;
159 	char *pathbuf, *path;
160 	int ret;
161 
162 	exe_file = get_mm_exe_file(current->mm);
163 	if (!exe_file)
164 		return cn_esc_printf(cn, "%s (path unknown)", current->comm);
165 
166 	pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
167 	if (!pathbuf) {
168 		ret = -ENOMEM;
169 		goto put_exe_file;
170 	}
171 
172 	path = file_path(exe_file, pathbuf, PATH_MAX);
173 	if (IS_ERR(path)) {
174 		ret = PTR_ERR(path);
175 		goto free_buf;
176 	}
177 
178 	ret = cn_esc_printf(cn, "%s", path);
179 
180 free_buf:
181 	kfree(pathbuf);
182 put_exe_file:
183 	fput(exe_file);
184 	return ret;
185 }
186 
187 /* format_corename will inspect the pattern parameter, and output a
188  * name into corename, which must have space for at least
189  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
190  */
191 static int format_corename(struct core_name *cn, struct coredump_params *cprm,
192 			   size_t **argv, int *argc)
193 {
194 	const struct cred *cred = current_cred();
195 	const char *pat_ptr = core_pattern;
196 	int ispipe = (*pat_ptr == '|');
197 	bool was_space = false;
198 	int pid_in_pattern = 0;
199 	int err = 0;
200 
201 	cn->used = 0;
202 	cn->corename = NULL;
203 	if (expand_corename(cn, core_name_size))
204 		return -ENOMEM;
205 	cn->corename[0] = '\0';
206 
207 	if (ispipe) {
208 		int argvs = sizeof(core_pattern) / 2;
209 		(*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL);
210 		if (!(*argv))
211 			return -ENOMEM;
212 		(*argv)[(*argc)++] = 0;
213 		++pat_ptr;
214 		if (!(*pat_ptr))
215 			return -ENOMEM;
216 	}
217 
218 	/* Repeat as long as we have more pattern to process and more output
219 	   space */
220 	while (*pat_ptr) {
221 		/*
222 		 * Split on spaces before doing template expansion so that
223 		 * %e and %E don't get split if they have spaces in them
224 		 */
225 		if (ispipe) {
226 			if (isspace(*pat_ptr)) {
227 				was_space = true;
228 				pat_ptr++;
229 				continue;
230 			} else if (was_space) {
231 				was_space = false;
232 				err = cn_printf(cn, "%c", '\0');
233 				if (err)
234 					return err;
235 				(*argv)[(*argc)++] = cn->used;
236 			}
237 		}
238 		if (*pat_ptr != '%') {
239 			err = cn_printf(cn, "%c", *pat_ptr++);
240 		} else {
241 			switch (*++pat_ptr) {
242 			/* single % at the end, drop that */
243 			case 0:
244 				goto out;
245 			/* Double percent, output one percent */
246 			case '%':
247 				err = cn_printf(cn, "%c", '%');
248 				break;
249 			/* pid */
250 			case 'p':
251 				pid_in_pattern = 1;
252 				err = cn_printf(cn, "%d",
253 					      task_tgid_vnr(current));
254 				break;
255 			/* global pid */
256 			case 'P':
257 				err = cn_printf(cn, "%d",
258 					      task_tgid_nr(current));
259 				break;
260 			case 'i':
261 				err = cn_printf(cn, "%d",
262 					      task_pid_vnr(current));
263 				break;
264 			case 'I':
265 				err = cn_printf(cn, "%d",
266 					      task_pid_nr(current));
267 				break;
268 			/* uid */
269 			case 'u':
270 				err = cn_printf(cn, "%u",
271 						from_kuid(&init_user_ns,
272 							  cred->uid));
273 				break;
274 			/* gid */
275 			case 'g':
276 				err = cn_printf(cn, "%u",
277 						from_kgid(&init_user_ns,
278 							  cred->gid));
279 				break;
280 			case 'd':
281 				err = cn_printf(cn, "%d",
282 					__get_dumpable(cprm->mm_flags));
283 				break;
284 			/* signal that caused the coredump */
285 			case 's':
286 				err = cn_printf(cn, "%d",
287 						cprm->siginfo->si_signo);
288 				break;
289 			/* UNIX time of coredump */
290 			case 't': {
291 				time64_t time;
292 
293 				time = ktime_get_real_seconds();
294 				err = cn_printf(cn, "%lld", time);
295 				break;
296 			}
297 			/* hostname */
298 			case 'h':
299 				down_read(&uts_sem);
300 				err = cn_esc_printf(cn, "%s",
301 					      utsname()->nodename);
302 				up_read(&uts_sem);
303 				break;
304 			/* executable */
305 			case 'e':
306 				err = cn_esc_printf(cn, "%s", current->comm);
307 				break;
308 			case 'E':
309 				err = cn_print_exe_file(cn);
310 				break;
311 			/* core limit size */
312 			case 'c':
313 				err = cn_printf(cn, "%lu",
314 					      rlimit(RLIMIT_CORE));
315 				break;
316 			default:
317 				break;
318 			}
319 			++pat_ptr;
320 		}
321 
322 		if (err)
323 			return err;
324 	}
325 
326 out:
327 	/* Backward compatibility with core_uses_pid:
328 	 *
329 	 * If core_pattern does not include a %p (as is the default)
330 	 * and core_uses_pid is set, then .%pid will be appended to
331 	 * the filename. Do not do this for piped commands. */
332 	if (!ispipe && !pid_in_pattern && core_uses_pid) {
333 		err = cn_printf(cn, ".%d", task_tgid_vnr(current));
334 		if (err)
335 			return err;
336 	}
337 	return ispipe;
338 }
339 
340 static int zap_process(struct task_struct *start, int exit_code, int flags)
341 {
342 	struct task_struct *t;
343 	int nr = 0;
344 
345 	/* ignore all signals except SIGKILL, see prepare_signal() */
346 	start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
347 	start->signal->group_exit_code = exit_code;
348 	start->signal->group_stop_count = 0;
349 
350 	for_each_thread(start, t) {
351 		task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
352 		if (t != current && t->mm) {
353 			sigaddset(&t->pending.signal, SIGKILL);
354 			signal_wake_up(t, 1);
355 			nr++;
356 		}
357 	}
358 
359 	return nr;
360 }
361 
362 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
363 			struct core_state *core_state, int exit_code)
364 {
365 	struct task_struct *g, *p;
366 	unsigned long flags;
367 	int nr = -EAGAIN;
368 
369 	spin_lock_irq(&tsk->sighand->siglock);
370 	if (!signal_group_exit(tsk->signal)) {
371 		mm->core_state = core_state;
372 		tsk->signal->group_exit_task = tsk;
373 		nr = zap_process(tsk, exit_code, 0);
374 		clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
375 	}
376 	spin_unlock_irq(&tsk->sighand->siglock);
377 	if (unlikely(nr < 0))
378 		return nr;
379 
380 	tsk->flags |= PF_DUMPCORE;
381 	if (atomic_read(&mm->mm_users) == nr + 1)
382 		goto done;
383 	/*
384 	 * We should find and kill all tasks which use this mm, and we should
385 	 * count them correctly into ->nr_threads. We don't take tasklist
386 	 * lock, but this is safe wrt:
387 	 *
388 	 * fork:
389 	 *	None of sub-threads can fork after zap_process(leader). All
390 	 *	processes which were created before this point should be
391 	 *	visible to zap_threads() because copy_process() adds the new
392 	 *	process to the tail of init_task.tasks list, and lock/unlock
393 	 *	of ->siglock provides a memory barrier.
394 	 *
395 	 * do_exit:
396 	 *	The caller holds mm->mmap_sem. This means that the task which
397 	 *	uses this mm can't pass exit_mm(), so it can't exit or clear
398 	 *	its ->mm.
399 	 *
400 	 * de_thread:
401 	 *	It does list_replace_rcu(&leader->tasks, &current->tasks),
402 	 *	we must see either old or new leader, this does not matter.
403 	 *	However, it can change p->sighand, so lock_task_sighand(p)
404 	 *	must be used. Since p->mm != NULL and we hold ->mmap_sem
405 	 *	it can't fail.
406 	 *
407 	 *	Note also that "g" can be the old leader with ->mm == NULL
408 	 *	and already unhashed and thus removed from ->thread_group.
409 	 *	This is OK, __unhash_process()->list_del_rcu() does not
410 	 *	clear the ->next pointer, we will find the new leader via
411 	 *	next_thread().
412 	 */
413 	rcu_read_lock();
414 	for_each_process(g) {
415 		if (g == tsk->group_leader)
416 			continue;
417 		if (g->flags & PF_KTHREAD)
418 			continue;
419 
420 		for_each_thread(g, p) {
421 			if (unlikely(!p->mm))
422 				continue;
423 			if (unlikely(p->mm == mm)) {
424 				lock_task_sighand(p, &flags);
425 				nr += zap_process(p, exit_code,
426 							SIGNAL_GROUP_EXIT);
427 				unlock_task_sighand(p, &flags);
428 			}
429 			break;
430 		}
431 	}
432 	rcu_read_unlock();
433 done:
434 	atomic_set(&core_state->nr_threads, nr);
435 	return nr;
436 }
437 
438 static int coredump_wait(int exit_code, struct core_state *core_state)
439 {
440 	struct task_struct *tsk = current;
441 	struct mm_struct *mm = tsk->mm;
442 	int core_waiters = -EBUSY;
443 
444 	init_completion(&core_state->startup);
445 	core_state->dumper.task = tsk;
446 	core_state->dumper.next = NULL;
447 
448 	if (down_write_killable(&mm->mmap_sem))
449 		return -EINTR;
450 
451 	if (!mm->core_state)
452 		core_waiters = zap_threads(tsk, mm, core_state, exit_code);
453 	up_write(&mm->mmap_sem);
454 
455 	if (core_waiters > 0) {
456 		struct core_thread *ptr;
457 
458 		freezer_do_not_count();
459 		wait_for_completion(&core_state->startup);
460 		freezer_count();
461 		/*
462 		 * Wait for all the threads to become inactive, so that
463 		 * all the thread context (extended register state, like
464 		 * fpu etc) gets copied to the memory.
465 		 */
466 		ptr = core_state->dumper.next;
467 		while (ptr != NULL) {
468 			wait_task_inactive(ptr->task, 0);
469 			ptr = ptr->next;
470 		}
471 	}
472 
473 	return core_waiters;
474 }
475 
476 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
477 {
478 	struct core_thread *curr, *next;
479 	struct task_struct *task;
480 
481 	spin_lock_irq(&current->sighand->siglock);
482 	if (core_dumped && !__fatal_signal_pending(current))
483 		current->signal->group_exit_code |= 0x80;
484 	current->signal->group_exit_task = NULL;
485 	current->signal->flags = SIGNAL_GROUP_EXIT;
486 	spin_unlock_irq(&current->sighand->siglock);
487 
488 	next = mm->core_state->dumper.next;
489 	while ((curr = next) != NULL) {
490 		next = curr->next;
491 		task = curr->task;
492 		/*
493 		 * see exit_mm(), curr->task must not see
494 		 * ->task == NULL before we read ->next.
495 		 */
496 		smp_mb();
497 		curr->task = NULL;
498 		wake_up_process(task);
499 	}
500 
501 	mm->core_state = NULL;
502 }
503 
504 static bool dump_interrupted(void)
505 {
506 	/*
507 	 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
508 	 * can do try_to_freeze() and check __fatal_signal_pending(),
509 	 * but then we need to teach dump_write() to restart and clear
510 	 * TIF_SIGPENDING.
511 	 */
512 	return signal_pending(current);
513 }
514 
515 static void wait_for_dump_helpers(struct file *file)
516 {
517 	struct pipe_inode_info *pipe = file->private_data;
518 
519 	pipe_lock(pipe);
520 	pipe->readers++;
521 	pipe->writers--;
522 	wake_up_interruptible_sync(&pipe->rd_wait);
523 	kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
524 	pipe_unlock(pipe);
525 
526 	/*
527 	 * We actually want wait_event_freezable() but then we need
528 	 * to clear TIF_SIGPENDING and improve dump_interrupted().
529 	 */
530 	wait_event_interruptible(pipe->rd_wait, pipe->readers == 1);
531 
532 	pipe_lock(pipe);
533 	pipe->readers--;
534 	pipe->writers++;
535 	pipe_unlock(pipe);
536 }
537 
538 /*
539  * umh_pipe_setup
540  * helper function to customize the process used
541  * to collect the core in userspace.  Specifically
542  * it sets up a pipe and installs it as fd 0 (stdin)
543  * for the process.  Returns 0 on success, or
544  * PTR_ERR on failure.
545  * Note that it also sets the core limit to 1.  This
546  * is a special value that we use to trap recursive
547  * core dumps
548  */
549 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
550 {
551 	struct file *files[2];
552 	struct coredump_params *cp = (struct coredump_params *)info->data;
553 	int err = create_pipe_files(files, 0);
554 	if (err)
555 		return err;
556 
557 	cp->file = files[1];
558 
559 	err = replace_fd(0, files[0], 0);
560 	fput(files[0]);
561 	/* and disallow core files too */
562 	current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
563 
564 	return err;
565 }
566 
567 void do_coredump(const kernel_siginfo_t *siginfo)
568 {
569 	struct core_state core_state;
570 	struct core_name cn;
571 	struct mm_struct *mm = current->mm;
572 	struct linux_binfmt * binfmt;
573 	const struct cred *old_cred;
574 	struct cred *cred;
575 	int retval = 0;
576 	int ispipe;
577 	size_t *argv = NULL;
578 	int argc = 0;
579 	struct files_struct *displaced;
580 	/* require nonrelative corefile path and be extra careful */
581 	bool need_suid_safe = false;
582 	bool core_dumped = false;
583 	static atomic_t core_dump_count = ATOMIC_INIT(0);
584 	struct coredump_params cprm = {
585 		.siginfo = siginfo,
586 		.regs = signal_pt_regs(),
587 		.limit = rlimit(RLIMIT_CORE),
588 		/*
589 		 * We must use the same mm->flags while dumping core to avoid
590 		 * inconsistency of bit flags, since this flag is not protected
591 		 * by any locks.
592 		 */
593 		.mm_flags = mm->flags,
594 	};
595 
596 	audit_core_dumps(siginfo->si_signo);
597 
598 	binfmt = mm->binfmt;
599 	if (!binfmt || !binfmt->core_dump)
600 		goto fail;
601 	if (!__get_dumpable(cprm.mm_flags))
602 		goto fail;
603 
604 	cred = prepare_creds();
605 	if (!cred)
606 		goto fail;
607 	/*
608 	 * We cannot trust fsuid as being the "true" uid of the process
609 	 * nor do we know its entire history. We only know it was tainted
610 	 * so we dump it as root in mode 2, and only into a controlled
611 	 * environment (pipe handler or fully qualified path).
612 	 */
613 	if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
614 		/* Setuid core dump mode */
615 		cred->fsuid = GLOBAL_ROOT_UID;	/* Dump root private */
616 		need_suid_safe = true;
617 	}
618 
619 	retval = coredump_wait(siginfo->si_signo, &core_state);
620 	if (retval < 0)
621 		goto fail_creds;
622 
623 	old_cred = override_creds(cred);
624 
625 	ispipe = format_corename(&cn, &cprm, &argv, &argc);
626 
627 	if (ispipe) {
628 		int argi;
629 		int dump_count;
630 		char **helper_argv;
631 		struct subprocess_info *sub_info;
632 
633 		if (ispipe < 0) {
634 			printk(KERN_WARNING "format_corename failed\n");
635 			printk(KERN_WARNING "Aborting core\n");
636 			goto fail_unlock;
637 		}
638 
639 		if (cprm.limit == 1) {
640 			/* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
641 			 *
642 			 * Normally core limits are irrelevant to pipes, since
643 			 * we're not writing to the file system, but we use
644 			 * cprm.limit of 1 here as a special value, this is a
645 			 * consistent way to catch recursive crashes.
646 			 * We can still crash if the core_pattern binary sets
647 			 * RLIM_CORE = !1, but it runs as root, and can do
648 			 * lots of stupid things.
649 			 *
650 			 * Note that we use task_tgid_vnr here to grab the pid
651 			 * of the process group leader.  That way we get the
652 			 * right pid if a thread in a multi-threaded
653 			 * core_pattern process dies.
654 			 */
655 			printk(KERN_WARNING
656 				"Process %d(%s) has RLIMIT_CORE set to 1\n",
657 				task_tgid_vnr(current), current->comm);
658 			printk(KERN_WARNING "Aborting core\n");
659 			goto fail_unlock;
660 		}
661 		cprm.limit = RLIM_INFINITY;
662 
663 		dump_count = atomic_inc_return(&core_dump_count);
664 		if (core_pipe_limit && (core_pipe_limit < dump_count)) {
665 			printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
666 			       task_tgid_vnr(current), current->comm);
667 			printk(KERN_WARNING "Skipping core dump\n");
668 			goto fail_dropcount;
669 		}
670 
671 		helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
672 					    GFP_KERNEL);
673 		if (!helper_argv) {
674 			printk(KERN_WARNING "%s failed to allocate memory\n",
675 			       __func__);
676 			goto fail_dropcount;
677 		}
678 		for (argi = 0; argi < argc; argi++)
679 			helper_argv[argi] = cn.corename + argv[argi];
680 		helper_argv[argi] = NULL;
681 
682 		retval = -ENOMEM;
683 		sub_info = call_usermodehelper_setup(helper_argv[0],
684 						helper_argv, NULL, GFP_KERNEL,
685 						umh_pipe_setup, NULL, &cprm);
686 		if (sub_info)
687 			retval = call_usermodehelper_exec(sub_info,
688 							  UMH_WAIT_EXEC);
689 
690 		kfree(helper_argv);
691 		if (retval) {
692 			printk(KERN_INFO "Core dump to |%s pipe failed\n",
693 			       cn.corename);
694 			goto close_fail;
695 		}
696 	} else {
697 		struct inode *inode;
698 		int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
699 				 O_LARGEFILE | O_EXCL;
700 
701 		if (cprm.limit < binfmt->min_coredump)
702 			goto fail_unlock;
703 
704 		if (need_suid_safe && cn.corename[0] != '/') {
705 			printk(KERN_WARNING "Pid %d(%s) can only dump core "\
706 				"to fully qualified path!\n",
707 				task_tgid_vnr(current), current->comm);
708 			printk(KERN_WARNING "Skipping core dump\n");
709 			goto fail_unlock;
710 		}
711 
712 		/*
713 		 * Unlink the file if it exists unless this is a SUID
714 		 * binary - in that case, we're running around with root
715 		 * privs and don't want to unlink another user's coredump.
716 		 */
717 		if (!need_suid_safe) {
718 			/*
719 			 * If it doesn't exist, that's fine. If there's some
720 			 * other problem, we'll catch it at the filp_open().
721 			 */
722 			do_unlinkat(AT_FDCWD, getname_kernel(cn.corename));
723 		}
724 
725 		/*
726 		 * There is a race between unlinking and creating the
727 		 * file, but if that causes an EEXIST here, that's
728 		 * fine - another process raced with us while creating
729 		 * the corefile, and the other process won. To userspace,
730 		 * what matters is that at least one of the two processes
731 		 * writes its coredump successfully, not which one.
732 		 */
733 		if (need_suid_safe) {
734 			/*
735 			 * Using user namespaces, normal user tasks can change
736 			 * their current->fs->root to point to arbitrary
737 			 * directories. Since the intention of the "only dump
738 			 * with a fully qualified path" rule is to control where
739 			 * coredumps may be placed using root privileges,
740 			 * current->fs->root must not be used. Instead, use the
741 			 * root directory of init_task.
742 			 */
743 			struct path root;
744 
745 			task_lock(&init_task);
746 			get_fs_root(init_task.fs, &root);
747 			task_unlock(&init_task);
748 			cprm.file = file_open_root(root.dentry, root.mnt,
749 				cn.corename, open_flags, 0600);
750 			path_put(&root);
751 		} else {
752 			cprm.file = filp_open(cn.corename, open_flags, 0600);
753 		}
754 		if (IS_ERR(cprm.file))
755 			goto fail_unlock;
756 
757 		inode = file_inode(cprm.file);
758 		if (inode->i_nlink > 1)
759 			goto close_fail;
760 		if (d_unhashed(cprm.file->f_path.dentry))
761 			goto close_fail;
762 		/*
763 		 * AK: actually i see no reason to not allow this for named
764 		 * pipes etc, but keep the previous behaviour for now.
765 		 */
766 		if (!S_ISREG(inode->i_mode))
767 			goto close_fail;
768 		/*
769 		 * Don't dump core if the filesystem changed owner or mode
770 		 * of the file during file creation. This is an issue when
771 		 * a process dumps core while its cwd is e.g. on a vfat
772 		 * filesystem.
773 		 */
774 		if (!uid_eq(inode->i_uid, current_fsuid()))
775 			goto close_fail;
776 		if ((inode->i_mode & 0677) != 0600)
777 			goto close_fail;
778 		if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
779 			goto close_fail;
780 		if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
781 			goto close_fail;
782 	}
783 
784 	/* get us an unshared descriptor table; almost always a no-op */
785 	retval = unshare_files(&displaced);
786 	if (retval)
787 		goto close_fail;
788 	if (displaced)
789 		put_files_struct(displaced);
790 	if (!dump_interrupted()) {
791 		file_start_write(cprm.file);
792 		core_dumped = binfmt->core_dump(&cprm);
793 		file_end_write(cprm.file);
794 	}
795 	if (ispipe && core_pipe_limit)
796 		wait_for_dump_helpers(cprm.file);
797 close_fail:
798 	if (cprm.file)
799 		filp_close(cprm.file, NULL);
800 fail_dropcount:
801 	if (ispipe)
802 		atomic_dec(&core_dump_count);
803 fail_unlock:
804 	kfree(argv);
805 	kfree(cn.corename);
806 	coredump_finish(mm, core_dumped);
807 	revert_creds(old_cred);
808 fail_creds:
809 	put_cred(cred);
810 fail:
811 	return;
812 }
813 
814 /*
815  * Core dumping helper functions.  These are the only things you should
816  * do on a core-file: use only these functions to write out all the
817  * necessary info.
818  */
819 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
820 {
821 	struct file *file = cprm->file;
822 	loff_t pos = file->f_pos;
823 	ssize_t n;
824 	if (cprm->written + nr > cprm->limit)
825 		return 0;
826 	while (nr) {
827 		if (dump_interrupted())
828 			return 0;
829 		n = __kernel_write(file, addr, nr, &pos);
830 		if (n <= 0)
831 			return 0;
832 		file->f_pos = pos;
833 		cprm->written += n;
834 		cprm->pos += n;
835 		nr -= n;
836 	}
837 	return 1;
838 }
839 EXPORT_SYMBOL(dump_emit);
840 
841 int dump_skip(struct coredump_params *cprm, size_t nr)
842 {
843 	static char zeroes[PAGE_SIZE];
844 	struct file *file = cprm->file;
845 	if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
846 		if (dump_interrupted() ||
847 		    file->f_op->llseek(file, nr, SEEK_CUR) < 0)
848 			return 0;
849 		cprm->pos += nr;
850 		return 1;
851 	} else {
852 		while (nr > PAGE_SIZE) {
853 			if (!dump_emit(cprm, zeroes, PAGE_SIZE))
854 				return 0;
855 			nr -= PAGE_SIZE;
856 		}
857 		return dump_emit(cprm, zeroes, nr);
858 	}
859 }
860 EXPORT_SYMBOL(dump_skip);
861 
862 int dump_align(struct coredump_params *cprm, int align)
863 {
864 	unsigned mod = cprm->pos & (align - 1);
865 	if (align & (align - 1))
866 		return 0;
867 	return mod ? dump_skip(cprm, align - mod) : 1;
868 }
869 EXPORT_SYMBOL(dump_align);
870 
871 /*
872  * Ensures that file size is big enough to contain the current file
873  * postion. This prevents gdb from complaining about a truncated file
874  * if the last "write" to the file was dump_skip.
875  */
876 void dump_truncate(struct coredump_params *cprm)
877 {
878 	struct file *file = cprm->file;
879 	loff_t offset;
880 
881 	if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
882 		offset = file->f_op->llseek(file, 0, SEEK_CUR);
883 		if (i_size_read(file->f_mapping->host) < offset)
884 			do_truncate(file->f_path.dentry, offset, 0, file);
885 	}
886 }
887 EXPORT_SYMBOL(dump_truncate);
888