xref: /openbmc/linux/fs/coredump.c (revision 1dff4156)
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 	}
215 
216 	/* Repeat as long as we have more pattern to process and more output
217 	   space */
218 	while (*pat_ptr) {
219 		/*
220 		 * Split on spaces before doing template expansion so that
221 		 * %e and %E don't get split if they have spaces in them
222 		 */
223 		if (ispipe) {
224 			if (isspace(*pat_ptr)) {
225 				was_space = true;
226 				pat_ptr++;
227 				continue;
228 			} else if (was_space) {
229 				was_space = false;
230 				err = cn_printf(cn, "%c", '\0');
231 				if (err)
232 					return err;
233 				(*argv)[(*argc)++] = cn->used;
234 			}
235 		}
236 		if (*pat_ptr != '%') {
237 			err = cn_printf(cn, "%c", *pat_ptr++);
238 		} else {
239 			switch (*++pat_ptr) {
240 			/* single % at the end, drop that */
241 			case 0:
242 				goto out;
243 			/* Double percent, output one percent */
244 			case '%':
245 				err = cn_printf(cn, "%c", '%');
246 				break;
247 			/* pid */
248 			case 'p':
249 				pid_in_pattern = 1;
250 				err = cn_printf(cn, "%d",
251 					      task_tgid_vnr(current));
252 				break;
253 			/* global pid */
254 			case 'P':
255 				err = cn_printf(cn, "%d",
256 					      task_tgid_nr(current));
257 				break;
258 			case 'i':
259 				err = cn_printf(cn, "%d",
260 					      task_pid_vnr(current));
261 				break;
262 			case 'I':
263 				err = cn_printf(cn, "%d",
264 					      task_pid_nr(current));
265 				break;
266 			/* uid */
267 			case 'u':
268 				err = cn_printf(cn, "%u",
269 						from_kuid(&init_user_ns,
270 							  cred->uid));
271 				break;
272 			/* gid */
273 			case 'g':
274 				err = cn_printf(cn, "%u",
275 						from_kgid(&init_user_ns,
276 							  cred->gid));
277 				break;
278 			case 'd':
279 				err = cn_printf(cn, "%d",
280 					__get_dumpable(cprm->mm_flags));
281 				break;
282 			/* signal that caused the coredump */
283 			case 's':
284 				err = cn_printf(cn, "%d",
285 						cprm->siginfo->si_signo);
286 				break;
287 			/* UNIX time of coredump */
288 			case 't': {
289 				time64_t time;
290 
291 				time = ktime_get_real_seconds();
292 				err = cn_printf(cn, "%lld", time);
293 				break;
294 			}
295 			/* hostname */
296 			case 'h':
297 				down_read(&uts_sem);
298 				err = cn_esc_printf(cn, "%s",
299 					      utsname()->nodename);
300 				up_read(&uts_sem);
301 				break;
302 			/* executable */
303 			case 'e':
304 				err = cn_esc_printf(cn, "%s", current->comm);
305 				break;
306 			case 'E':
307 				err = cn_print_exe_file(cn);
308 				break;
309 			/* core limit size */
310 			case 'c':
311 				err = cn_printf(cn, "%lu",
312 					      rlimit(RLIMIT_CORE));
313 				break;
314 			default:
315 				break;
316 			}
317 			++pat_ptr;
318 		}
319 
320 		if (err)
321 			return err;
322 	}
323 
324 out:
325 	/* Backward compatibility with core_uses_pid:
326 	 *
327 	 * If core_pattern does not include a %p (as is the default)
328 	 * and core_uses_pid is set, then .%pid will be appended to
329 	 * the filename. Do not do this for piped commands. */
330 	if (!ispipe && !pid_in_pattern && core_uses_pid) {
331 		err = cn_printf(cn, ".%d", task_tgid_vnr(current));
332 		if (err)
333 			return err;
334 	}
335 	return ispipe;
336 }
337 
338 static int zap_process(struct task_struct *start, int exit_code, int flags)
339 {
340 	struct task_struct *t;
341 	int nr = 0;
342 
343 	/* ignore all signals except SIGKILL, see prepare_signal() */
344 	start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
345 	start->signal->group_exit_code = exit_code;
346 	start->signal->group_stop_count = 0;
347 
348 	for_each_thread(start, t) {
349 		task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
350 		if (t != current && t->mm) {
351 			sigaddset(&t->pending.signal, SIGKILL);
352 			signal_wake_up(t, 1);
353 			nr++;
354 		}
355 	}
356 
357 	return nr;
358 }
359 
360 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
361 			struct core_state *core_state, int exit_code)
362 {
363 	struct task_struct *g, *p;
364 	unsigned long flags;
365 	int nr = -EAGAIN;
366 
367 	spin_lock_irq(&tsk->sighand->siglock);
368 	if (!signal_group_exit(tsk->signal)) {
369 		mm->core_state = core_state;
370 		tsk->signal->group_exit_task = tsk;
371 		nr = zap_process(tsk, exit_code, 0);
372 		clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
373 	}
374 	spin_unlock_irq(&tsk->sighand->siglock);
375 	if (unlikely(nr < 0))
376 		return nr;
377 
378 	tsk->flags |= PF_DUMPCORE;
379 	if (atomic_read(&mm->mm_users) == nr + 1)
380 		goto done;
381 	/*
382 	 * We should find and kill all tasks which use this mm, and we should
383 	 * count them correctly into ->nr_threads. We don't take tasklist
384 	 * lock, but this is safe wrt:
385 	 *
386 	 * fork:
387 	 *	None of sub-threads can fork after zap_process(leader). All
388 	 *	processes which were created before this point should be
389 	 *	visible to zap_threads() because copy_process() adds the new
390 	 *	process to the tail of init_task.tasks list, and lock/unlock
391 	 *	of ->siglock provides a memory barrier.
392 	 *
393 	 * do_exit:
394 	 *	The caller holds mm->mmap_sem. This means that the task which
395 	 *	uses this mm can't pass exit_mm(), so it can't exit or clear
396 	 *	its ->mm.
397 	 *
398 	 * de_thread:
399 	 *	It does list_replace_rcu(&leader->tasks, &current->tasks),
400 	 *	we must see either old or new leader, this does not matter.
401 	 *	However, it can change p->sighand, so lock_task_sighand(p)
402 	 *	must be used. Since p->mm != NULL and we hold ->mmap_sem
403 	 *	it can't fail.
404 	 *
405 	 *	Note also that "g" can be the old leader with ->mm == NULL
406 	 *	and already unhashed and thus removed from ->thread_group.
407 	 *	This is OK, __unhash_process()->list_del_rcu() does not
408 	 *	clear the ->next pointer, we will find the new leader via
409 	 *	next_thread().
410 	 */
411 	rcu_read_lock();
412 	for_each_process(g) {
413 		if (g == tsk->group_leader)
414 			continue;
415 		if (g->flags & PF_KTHREAD)
416 			continue;
417 
418 		for_each_thread(g, p) {
419 			if (unlikely(!p->mm))
420 				continue;
421 			if (unlikely(p->mm == mm)) {
422 				lock_task_sighand(p, &flags);
423 				nr += zap_process(p, exit_code,
424 							SIGNAL_GROUP_EXIT);
425 				unlock_task_sighand(p, &flags);
426 			}
427 			break;
428 		}
429 	}
430 	rcu_read_unlock();
431 done:
432 	atomic_set(&core_state->nr_threads, nr);
433 	return nr;
434 }
435 
436 static int coredump_wait(int exit_code, struct core_state *core_state)
437 {
438 	struct task_struct *tsk = current;
439 	struct mm_struct *mm = tsk->mm;
440 	int core_waiters = -EBUSY;
441 
442 	init_completion(&core_state->startup);
443 	core_state->dumper.task = tsk;
444 	core_state->dumper.next = NULL;
445 
446 	if (down_write_killable(&mm->mmap_sem))
447 		return -EINTR;
448 
449 	if (!mm->core_state)
450 		core_waiters = zap_threads(tsk, mm, core_state, exit_code);
451 	up_write(&mm->mmap_sem);
452 
453 	if (core_waiters > 0) {
454 		struct core_thread *ptr;
455 
456 		freezer_do_not_count();
457 		wait_for_completion(&core_state->startup);
458 		freezer_count();
459 		/*
460 		 * Wait for all the threads to become inactive, so that
461 		 * all the thread context (extended register state, like
462 		 * fpu etc) gets copied to the memory.
463 		 */
464 		ptr = core_state->dumper.next;
465 		while (ptr != NULL) {
466 			wait_task_inactive(ptr->task, 0);
467 			ptr = ptr->next;
468 		}
469 	}
470 
471 	return core_waiters;
472 }
473 
474 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
475 {
476 	struct core_thread *curr, *next;
477 	struct task_struct *task;
478 
479 	spin_lock_irq(&current->sighand->siglock);
480 	if (core_dumped && !__fatal_signal_pending(current))
481 		current->signal->group_exit_code |= 0x80;
482 	current->signal->group_exit_task = NULL;
483 	current->signal->flags = SIGNAL_GROUP_EXIT;
484 	spin_unlock_irq(&current->sighand->siglock);
485 
486 	next = mm->core_state->dumper.next;
487 	while ((curr = next) != NULL) {
488 		next = curr->next;
489 		task = curr->task;
490 		/*
491 		 * see exit_mm(), curr->task must not see
492 		 * ->task == NULL before we read ->next.
493 		 */
494 		smp_mb();
495 		curr->task = NULL;
496 		wake_up_process(task);
497 	}
498 
499 	mm->core_state = NULL;
500 }
501 
502 static bool dump_interrupted(void)
503 {
504 	/*
505 	 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
506 	 * can do try_to_freeze() and check __fatal_signal_pending(),
507 	 * but then we need to teach dump_write() to restart and clear
508 	 * TIF_SIGPENDING.
509 	 */
510 	return signal_pending(current);
511 }
512 
513 static void wait_for_dump_helpers(struct file *file)
514 {
515 	struct pipe_inode_info *pipe = file->private_data;
516 
517 	pipe_lock(pipe);
518 	pipe->readers++;
519 	pipe->writers--;
520 	wake_up_interruptible_sync(&pipe->rd_wait);
521 	kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
522 	pipe_unlock(pipe);
523 
524 	/*
525 	 * We actually want wait_event_freezable() but then we need
526 	 * to clear TIF_SIGPENDING and improve dump_interrupted().
527 	 */
528 	wait_event_interruptible(pipe->rd_wait, pipe->readers == 1);
529 
530 	pipe_lock(pipe);
531 	pipe->readers--;
532 	pipe->writers++;
533 	pipe_unlock(pipe);
534 }
535 
536 /*
537  * umh_pipe_setup
538  * helper function to customize the process used
539  * to collect the core in userspace.  Specifically
540  * it sets up a pipe and installs it as fd 0 (stdin)
541  * for the process.  Returns 0 on success, or
542  * PTR_ERR on failure.
543  * Note that it also sets the core limit to 1.  This
544  * is a special value that we use to trap recursive
545  * core dumps
546  */
547 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
548 {
549 	struct file *files[2];
550 	struct coredump_params *cp = (struct coredump_params *)info->data;
551 	int err = create_pipe_files(files, 0);
552 	if (err)
553 		return err;
554 
555 	cp->file = files[1];
556 
557 	err = replace_fd(0, files[0], 0);
558 	fput(files[0]);
559 	/* and disallow core files too */
560 	current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
561 
562 	return err;
563 }
564 
565 void do_coredump(const kernel_siginfo_t *siginfo)
566 {
567 	struct core_state core_state;
568 	struct core_name cn;
569 	struct mm_struct *mm = current->mm;
570 	struct linux_binfmt * binfmt;
571 	const struct cred *old_cred;
572 	struct cred *cred;
573 	int retval = 0;
574 	int ispipe;
575 	size_t *argv = NULL;
576 	int argc = 0;
577 	struct files_struct *displaced;
578 	/* require nonrelative corefile path and be extra careful */
579 	bool need_suid_safe = false;
580 	bool core_dumped = false;
581 	static atomic_t core_dump_count = ATOMIC_INIT(0);
582 	struct coredump_params cprm = {
583 		.siginfo = siginfo,
584 		.regs = signal_pt_regs(),
585 		.limit = rlimit(RLIMIT_CORE),
586 		/*
587 		 * We must use the same mm->flags while dumping core to avoid
588 		 * inconsistency of bit flags, since this flag is not protected
589 		 * by any locks.
590 		 */
591 		.mm_flags = mm->flags,
592 	};
593 
594 	audit_core_dumps(siginfo->si_signo);
595 
596 	binfmt = mm->binfmt;
597 	if (!binfmt || !binfmt->core_dump)
598 		goto fail;
599 	if (!__get_dumpable(cprm.mm_flags))
600 		goto fail;
601 
602 	cred = prepare_creds();
603 	if (!cred)
604 		goto fail;
605 	/*
606 	 * We cannot trust fsuid as being the "true" uid of the process
607 	 * nor do we know its entire history. We only know it was tainted
608 	 * so we dump it as root in mode 2, and only into a controlled
609 	 * environment (pipe handler or fully qualified path).
610 	 */
611 	if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
612 		/* Setuid core dump mode */
613 		cred->fsuid = GLOBAL_ROOT_UID;	/* Dump root private */
614 		need_suid_safe = true;
615 	}
616 
617 	retval = coredump_wait(siginfo->si_signo, &core_state);
618 	if (retval < 0)
619 		goto fail_creds;
620 
621 	old_cred = override_creds(cred);
622 
623 	ispipe = format_corename(&cn, &cprm, &argv, &argc);
624 
625 	if (ispipe) {
626 		int argi;
627 		int dump_count;
628 		char **helper_argv;
629 		struct subprocess_info *sub_info;
630 
631 		if (ispipe < 0) {
632 			printk(KERN_WARNING "format_corename failed\n");
633 			printk(KERN_WARNING "Aborting core\n");
634 			goto fail_unlock;
635 		}
636 
637 		if (cprm.limit == 1) {
638 			/* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
639 			 *
640 			 * Normally core limits are irrelevant to pipes, since
641 			 * we're not writing to the file system, but we use
642 			 * cprm.limit of 1 here as a special value, this is a
643 			 * consistent way to catch recursive crashes.
644 			 * We can still crash if the core_pattern binary sets
645 			 * RLIM_CORE = !1, but it runs as root, and can do
646 			 * lots of stupid things.
647 			 *
648 			 * Note that we use task_tgid_vnr here to grab the pid
649 			 * of the process group leader.  That way we get the
650 			 * right pid if a thread in a multi-threaded
651 			 * core_pattern process dies.
652 			 */
653 			printk(KERN_WARNING
654 				"Process %d(%s) has RLIMIT_CORE set to 1\n",
655 				task_tgid_vnr(current), current->comm);
656 			printk(KERN_WARNING "Aborting core\n");
657 			goto fail_unlock;
658 		}
659 		cprm.limit = RLIM_INFINITY;
660 
661 		dump_count = atomic_inc_return(&core_dump_count);
662 		if (core_pipe_limit && (core_pipe_limit < dump_count)) {
663 			printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
664 			       task_tgid_vnr(current), current->comm);
665 			printk(KERN_WARNING "Skipping core dump\n");
666 			goto fail_dropcount;
667 		}
668 
669 		helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
670 					    GFP_KERNEL);
671 		if (!helper_argv) {
672 			printk(KERN_WARNING "%s failed to allocate memory\n",
673 			       __func__);
674 			goto fail_dropcount;
675 		}
676 		for (argi = 0; argi < argc; argi++)
677 			helper_argv[argi] = cn.corename + argv[argi];
678 		helper_argv[argi] = NULL;
679 
680 		retval = -ENOMEM;
681 		sub_info = call_usermodehelper_setup(helper_argv[0],
682 						helper_argv, NULL, GFP_KERNEL,
683 						umh_pipe_setup, NULL, &cprm);
684 		if (sub_info)
685 			retval = call_usermodehelper_exec(sub_info,
686 							  UMH_WAIT_EXEC);
687 
688 		kfree(helper_argv);
689 		if (retval) {
690 			printk(KERN_INFO "Core dump to |%s pipe failed\n",
691 			       cn.corename);
692 			goto close_fail;
693 		}
694 	} else {
695 		struct inode *inode;
696 		int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
697 				 O_LARGEFILE | O_EXCL;
698 
699 		if (cprm.limit < binfmt->min_coredump)
700 			goto fail_unlock;
701 
702 		if (need_suid_safe && cn.corename[0] != '/') {
703 			printk(KERN_WARNING "Pid %d(%s) can only dump core "\
704 				"to fully qualified path!\n",
705 				task_tgid_vnr(current), current->comm);
706 			printk(KERN_WARNING "Skipping core dump\n");
707 			goto fail_unlock;
708 		}
709 
710 		/*
711 		 * Unlink the file if it exists unless this is a SUID
712 		 * binary - in that case, we're running around with root
713 		 * privs and don't want to unlink another user's coredump.
714 		 */
715 		if (!need_suid_safe) {
716 			/*
717 			 * If it doesn't exist, that's fine. If there's some
718 			 * other problem, we'll catch it at the filp_open().
719 			 */
720 			do_unlinkat(AT_FDCWD, getname_kernel(cn.corename));
721 		}
722 
723 		/*
724 		 * There is a race between unlinking and creating the
725 		 * file, but if that causes an EEXIST here, that's
726 		 * fine - another process raced with us while creating
727 		 * the corefile, and the other process won. To userspace,
728 		 * what matters is that at least one of the two processes
729 		 * writes its coredump successfully, not which one.
730 		 */
731 		if (need_suid_safe) {
732 			/*
733 			 * Using user namespaces, normal user tasks can change
734 			 * their current->fs->root to point to arbitrary
735 			 * directories. Since the intention of the "only dump
736 			 * with a fully qualified path" rule is to control where
737 			 * coredumps may be placed using root privileges,
738 			 * current->fs->root must not be used. Instead, use the
739 			 * root directory of init_task.
740 			 */
741 			struct path root;
742 
743 			task_lock(&init_task);
744 			get_fs_root(init_task.fs, &root);
745 			task_unlock(&init_task);
746 			cprm.file = file_open_root(root.dentry, root.mnt,
747 				cn.corename, open_flags, 0600);
748 			path_put(&root);
749 		} else {
750 			cprm.file = filp_open(cn.corename, open_flags, 0600);
751 		}
752 		if (IS_ERR(cprm.file))
753 			goto fail_unlock;
754 
755 		inode = file_inode(cprm.file);
756 		if (inode->i_nlink > 1)
757 			goto close_fail;
758 		if (d_unhashed(cprm.file->f_path.dentry))
759 			goto close_fail;
760 		/*
761 		 * AK: actually i see no reason to not allow this for named
762 		 * pipes etc, but keep the previous behaviour for now.
763 		 */
764 		if (!S_ISREG(inode->i_mode))
765 			goto close_fail;
766 		/*
767 		 * Don't dump core if the filesystem changed owner or mode
768 		 * of the file during file creation. This is an issue when
769 		 * a process dumps core while its cwd is e.g. on a vfat
770 		 * filesystem.
771 		 */
772 		if (!uid_eq(inode->i_uid, current_fsuid()))
773 			goto close_fail;
774 		if ((inode->i_mode & 0677) != 0600)
775 			goto close_fail;
776 		if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
777 			goto close_fail;
778 		if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
779 			goto close_fail;
780 	}
781 
782 	/* get us an unshared descriptor table; almost always a no-op */
783 	retval = unshare_files(&displaced);
784 	if (retval)
785 		goto close_fail;
786 	if (displaced)
787 		put_files_struct(displaced);
788 	if (!dump_interrupted()) {
789 		file_start_write(cprm.file);
790 		core_dumped = binfmt->core_dump(&cprm);
791 		file_end_write(cprm.file);
792 	}
793 	if (ispipe && core_pipe_limit)
794 		wait_for_dump_helpers(cprm.file);
795 close_fail:
796 	if (cprm.file)
797 		filp_close(cprm.file, NULL);
798 fail_dropcount:
799 	if (ispipe)
800 		atomic_dec(&core_dump_count);
801 fail_unlock:
802 	kfree(argv);
803 	kfree(cn.corename);
804 	coredump_finish(mm, core_dumped);
805 	revert_creds(old_cred);
806 fail_creds:
807 	put_cred(cred);
808 fail:
809 	return;
810 }
811 
812 /*
813  * Core dumping helper functions.  These are the only things you should
814  * do on a core-file: use only these functions to write out all the
815  * necessary info.
816  */
817 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
818 {
819 	struct file *file = cprm->file;
820 	loff_t pos = file->f_pos;
821 	ssize_t n;
822 	if (cprm->written + nr > cprm->limit)
823 		return 0;
824 	while (nr) {
825 		if (dump_interrupted())
826 			return 0;
827 		n = __kernel_write(file, addr, nr, &pos);
828 		if (n <= 0)
829 			return 0;
830 		file->f_pos = pos;
831 		cprm->written += n;
832 		cprm->pos += n;
833 		nr -= n;
834 	}
835 	return 1;
836 }
837 EXPORT_SYMBOL(dump_emit);
838 
839 int dump_skip(struct coredump_params *cprm, size_t nr)
840 {
841 	static char zeroes[PAGE_SIZE];
842 	struct file *file = cprm->file;
843 	if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
844 		if (dump_interrupted() ||
845 		    file->f_op->llseek(file, nr, SEEK_CUR) < 0)
846 			return 0;
847 		cprm->pos += nr;
848 		return 1;
849 	} else {
850 		while (nr > PAGE_SIZE) {
851 			if (!dump_emit(cprm, zeroes, PAGE_SIZE))
852 				return 0;
853 			nr -= PAGE_SIZE;
854 		}
855 		return dump_emit(cprm, zeroes, nr);
856 	}
857 }
858 EXPORT_SYMBOL(dump_skip);
859 
860 int dump_align(struct coredump_params *cprm, int align)
861 {
862 	unsigned mod = cprm->pos & (align - 1);
863 	if (align & (align - 1))
864 		return 0;
865 	return mod ? dump_skip(cprm, align - mod) : 1;
866 }
867 EXPORT_SYMBOL(dump_align);
868 
869 /*
870  * Ensures that file size is big enough to contain the current file
871  * postion. This prevents gdb from complaining about a truncated file
872  * if the last "write" to the file was dump_skip.
873  */
874 void dump_truncate(struct coredump_params *cprm)
875 {
876 	struct file *file = cprm->file;
877 	loff_t offset;
878 
879 	if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
880 		offset = file->f_op->llseek(file, 0, SEEK_CUR);
881 		if (i_size_read(file->f_mapping->host) < offset)
882 			do_truncate(file->f_path.dentry, offset, 0, file);
883 	}
884 }
885 EXPORT_SYMBOL(dump_truncate);
886