xref: /linux/fs/coredump.c (revision 6c363eafc4d637ac4bd83d4a7dd06dd3cfbe7c5f)
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, bool name_only)
157 {
158 	struct file *exe_file;
159 	char *pathbuf, *path, *ptr;
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 	if (name_only) {
179 		ptr = strrchr(path, '/');
180 		if (ptr)
181 			path = ptr + 1;
182 	}
183 	ret = cn_esc_printf(cn, "%s", path);
184 
185 free_buf:
186 	kfree(pathbuf);
187 put_exe_file:
188 	fput(exe_file);
189 	return ret;
190 }
191 
192 /* format_corename will inspect the pattern parameter, and output a
193  * name into corename, which must have space for at least
194  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
195  */
196 static int format_corename(struct core_name *cn, struct coredump_params *cprm,
197 			   size_t **argv, int *argc)
198 {
199 	const struct cred *cred = current_cred();
200 	const char *pat_ptr = core_pattern;
201 	int ispipe = (*pat_ptr == '|');
202 	bool was_space = false;
203 	int pid_in_pattern = 0;
204 	int err = 0;
205 
206 	cn->used = 0;
207 	cn->corename = NULL;
208 	if (expand_corename(cn, core_name_size))
209 		return -ENOMEM;
210 	cn->corename[0] = '\0';
211 
212 	if (ispipe) {
213 		int argvs = sizeof(core_pattern) / 2;
214 		(*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL);
215 		if (!(*argv))
216 			return -ENOMEM;
217 		(*argv)[(*argc)++] = 0;
218 		++pat_ptr;
219 		if (!(*pat_ptr))
220 			return -ENOMEM;
221 	}
222 
223 	/* Repeat as long as we have more pattern to process and more output
224 	   space */
225 	while (*pat_ptr) {
226 		/*
227 		 * Split on spaces before doing template expansion so that
228 		 * %e and %E don't get split if they have spaces in them
229 		 */
230 		if (ispipe) {
231 			if (isspace(*pat_ptr)) {
232 				if (cn->used != 0)
233 					was_space = true;
234 				pat_ptr++;
235 				continue;
236 			} else if (was_space) {
237 				was_space = false;
238 				err = cn_printf(cn, "%c", '\0');
239 				if (err)
240 					return err;
241 				(*argv)[(*argc)++] = cn->used;
242 			}
243 		}
244 		if (*pat_ptr != '%') {
245 			err = cn_printf(cn, "%c", *pat_ptr++);
246 		} else {
247 			switch (*++pat_ptr) {
248 			/* single % at the end, drop that */
249 			case 0:
250 				goto out;
251 			/* Double percent, output one percent */
252 			case '%':
253 				err = cn_printf(cn, "%c", '%');
254 				break;
255 			/* pid */
256 			case 'p':
257 				pid_in_pattern = 1;
258 				err = cn_printf(cn, "%d",
259 					      task_tgid_vnr(current));
260 				break;
261 			/* global pid */
262 			case 'P':
263 				err = cn_printf(cn, "%d",
264 					      task_tgid_nr(current));
265 				break;
266 			case 'i':
267 				err = cn_printf(cn, "%d",
268 					      task_pid_vnr(current));
269 				break;
270 			case 'I':
271 				err = cn_printf(cn, "%d",
272 					      task_pid_nr(current));
273 				break;
274 			/* uid */
275 			case 'u':
276 				err = cn_printf(cn, "%u",
277 						from_kuid(&init_user_ns,
278 							  cred->uid));
279 				break;
280 			/* gid */
281 			case 'g':
282 				err = cn_printf(cn, "%u",
283 						from_kgid(&init_user_ns,
284 							  cred->gid));
285 				break;
286 			case 'd':
287 				err = cn_printf(cn, "%d",
288 					__get_dumpable(cprm->mm_flags));
289 				break;
290 			/* signal that caused the coredump */
291 			case 's':
292 				err = cn_printf(cn, "%d",
293 						cprm->siginfo->si_signo);
294 				break;
295 			/* UNIX time of coredump */
296 			case 't': {
297 				time64_t time;
298 
299 				time = ktime_get_real_seconds();
300 				err = cn_printf(cn, "%lld", time);
301 				break;
302 			}
303 			/* hostname */
304 			case 'h':
305 				down_read(&uts_sem);
306 				err = cn_esc_printf(cn, "%s",
307 					      utsname()->nodename);
308 				up_read(&uts_sem);
309 				break;
310 			/* executable, could be changed by prctl PR_SET_NAME etc */
311 			case 'e':
312 				err = cn_esc_printf(cn, "%s", current->comm);
313 				break;
314 			/* file name of executable */
315 			case 'f':
316 				err = cn_print_exe_file(cn, true);
317 				break;
318 			case 'E':
319 				err = cn_print_exe_file(cn, false);
320 				break;
321 			/* core limit size */
322 			case 'c':
323 				err = cn_printf(cn, "%lu",
324 					      rlimit(RLIMIT_CORE));
325 				break;
326 			default:
327 				break;
328 			}
329 			++pat_ptr;
330 		}
331 
332 		if (err)
333 			return err;
334 	}
335 
336 out:
337 	/* Backward compatibility with core_uses_pid:
338 	 *
339 	 * If core_pattern does not include a %p (as is the default)
340 	 * and core_uses_pid is set, then .%pid will be appended to
341 	 * the filename. Do not do this for piped commands. */
342 	if (!ispipe && !pid_in_pattern && core_uses_pid) {
343 		err = cn_printf(cn, ".%d", task_tgid_vnr(current));
344 		if (err)
345 			return err;
346 	}
347 	return ispipe;
348 }
349 
350 static int zap_process(struct task_struct *start, int exit_code, int flags)
351 {
352 	struct task_struct *t;
353 	int nr = 0;
354 
355 	/* ignore all signals except SIGKILL, see prepare_signal() */
356 	start->signal->flags = SIGNAL_GROUP_COREDUMP | flags;
357 	start->signal->group_exit_code = exit_code;
358 	start->signal->group_stop_count = 0;
359 
360 	for_each_thread(start, t) {
361 		task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
362 		if (t != current && t->mm) {
363 			sigaddset(&t->pending.signal, SIGKILL);
364 			signal_wake_up(t, 1);
365 			nr++;
366 		}
367 	}
368 
369 	return nr;
370 }
371 
372 static int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
373 			struct core_state *core_state, int exit_code)
374 {
375 	struct task_struct *g, *p;
376 	unsigned long flags;
377 	int nr = -EAGAIN;
378 
379 	spin_lock_irq(&tsk->sighand->siglock);
380 	if (!signal_group_exit(tsk->signal)) {
381 		mm->core_state = core_state;
382 		tsk->signal->group_exit_task = tsk;
383 		nr = zap_process(tsk, exit_code, 0);
384 		clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
385 	}
386 	spin_unlock_irq(&tsk->sighand->siglock);
387 	if (unlikely(nr < 0))
388 		return nr;
389 
390 	tsk->flags |= PF_DUMPCORE;
391 	if (atomic_read(&mm->mm_users) == nr + 1)
392 		goto done;
393 	/*
394 	 * We should find and kill all tasks which use this mm, and we should
395 	 * count them correctly into ->nr_threads. We don't take tasklist
396 	 * lock, but this is safe wrt:
397 	 *
398 	 * fork:
399 	 *	None of sub-threads can fork after zap_process(leader). All
400 	 *	processes which were created before this point should be
401 	 *	visible to zap_threads() because copy_process() adds the new
402 	 *	process to the tail of init_task.tasks list, and lock/unlock
403 	 *	of ->siglock provides a memory barrier.
404 	 *
405 	 * do_exit:
406 	 *	The caller holds mm->mmap_lock. This means that the task which
407 	 *	uses this mm can't pass exit_mm(), so it can't exit or clear
408 	 *	its ->mm.
409 	 *
410 	 * de_thread:
411 	 *	It does list_replace_rcu(&leader->tasks, &current->tasks),
412 	 *	we must see either old or new leader, this does not matter.
413 	 *	However, it can change p->sighand, so lock_task_sighand(p)
414 	 *	must be used. Since p->mm != NULL and we hold ->mmap_lock
415 	 *	it can't fail.
416 	 *
417 	 *	Note also that "g" can be the old leader with ->mm == NULL
418 	 *	and already unhashed and thus removed from ->thread_group.
419 	 *	This is OK, __unhash_process()->list_del_rcu() does not
420 	 *	clear the ->next pointer, we will find the new leader via
421 	 *	next_thread().
422 	 */
423 	rcu_read_lock();
424 	for_each_process(g) {
425 		if (g == tsk->group_leader)
426 			continue;
427 		if (g->flags & PF_KTHREAD)
428 			continue;
429 
430 		for_each_thread(g, p) {
431 			if (unlikely(!p->mm))
432 				continue;
433 			if (unlikely(p->mm == mm)) {
434 				lock_task_sighand(p, &flags);
435 				nr += zap_process(p, exit_code,
436 							SIGNAL_GROUP_EXIT);
437 				unlock_task_sighand(p, &flags);
438 			}
439 			break;
440 		}
441 	}
442 	rcu_read_unlock();
443 done:
444 	atomic_set(&core_state->nr_threads, nr);
445 	return nr;
446 }
447 
448 static int coredump_wait(int exit_code, struct core_state *core_state)
449 {
450 	struct task_struct *tsk = current;
451 	struct mm_struct *mm = tsk->mm;
452 	int core_waiters = -EBUSY;
453 
454 	init_completion(&core_state->startup);
455 	core_state->dumper.task = tsk;
456 	core_state->dumper.next = NULL;
457 
458 	if (mmap_write_lock_killable(mm))
459 		return -EINTR;
460 
461 	if (!mm->core_state)
462 		core_waiters = zap_threads(tsk, mm, core_state, exit_code);
463 	mmap_write_unlock(mm);
464 
465 	if (core_waiters > 0) {
466 		struct core_thread *ptr;
467 
468 		freezer_do_not_count();
469 		wait_for_completion(&core_state->startup);
470 		freezer_count();
471 		/*
472 		 * Wait for all the threads to become inactive, so that
473 		 * all the thread context (extended register state, like
474 		 * fpu etc) gets copied to the memory.
475 		 */
476 		ptr = core_state->dumper.next;
477 		while (ptr != NULL) {
478 			wait_task_inactive(ptr->task, 0);
479 			ptr = ptr->next;
480 		}
481 	}
482 
483 	return core_waiters;
484 }
485 
486 static void coredump_finish(struct mm_struct *mm, bool core_dumped)
487 {
488 	struct core_thread *curr, *next;
489 	struct task_struct *task;
490 
491 	spin_lock_irq(&current->sighand->siglock);
492 	if (core_dumped && !__fatal_signal_pending(current))
493 		current->signal->group_exit_code |= 0x80;
494 	current->signal->group_exit_task = NULL;
495 	current->signal->flags = SIGNAL_GROUP_EXIT;
496 	spin_unlock_irq(&current->sighand->siglock);
497 
498 	next = mm->core_state->dumper.next;
499 	while ((curr = next) != NULL) {
500 		next = curr->next;
501 		task = curr->task;
502 		/*
503 		 * see exit_mm(), curr->task must not see
504 		 * ->task == NULL before we read ->next.
505 		 */
506 		smp_mb();
507 		curr->task = NULL;
508 		wake_up_process(task);
509 	}
510 
511 	mm->core_state = NULL;
512 }
513 
514 static bool dump_interrupted(void)
515 {
516 	/*
517 	 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
518 	 * can do try_to_freeze() and check __fatal_signal_pending(),
519 	 * but then we need to teach dump_write() to restart and clear
520 	 * TIF_SIGPENDING.
521 	 */
522 	return signal_pending(current);
523 }
524 
525 static void wait_for_dump_helpers(struct file *file)
526 {
527 	struct pipe_inode_info *pipe = file->private_data;
528 
529 	pipe_lock(pipe);
530 	pipe->readers++;
531 	pipe->writers--;
532 	wake_up_interruptible_sync(&pipe->rd_wait);
533 	kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
534 	pipe_unlock(pipe);
535 
536 	/*
537 	 * We actually want wait_event_freezable() but then we need
538 	 * to clear TIF_SIGPENDING and improve dump_interrupted().
539 	 */
540 	wait_event_interruptible(pipe->rd_wait, pipe->readers == 1);
541 
542 	pipe_lock(pipe);
543 	pipe->readers--;
544 	pipe->writers++;
545 	pipe_unlock(pipe);
546 }
547 
548 /*
549  * umh_pipe_setup
550  * helper function to customize the process used
551  * to collect the core in userspace.  Specifically
552  * it sets up a pipe and installs it as fd 0 (stdin)
553  * for the process.  Returns 0 on success, or
554  * PTR_ERR on failure.
555  * Note that it also sets the core limit to 1.  This
556  * is a special value that we use to trap recursive
557  * core dumps
558  */
559 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
560 {
561 	struct file *files[2];
562 	struct coredump_params *cp = (struct coredump_params *)info->data;
563 	int err = create_pipe_files(files, 0);
564 	if (err)
565 		return err;
566 
567 	cp->file = files[1];
568 
569 	err = replace_fd(0, files[0], 0);
570 	fput(files[0]);
571 	/* and disallow core files too */
572 	current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
573 
574 	return err;
575 }
576 
577 void do_coredump(const kernel_siginfo_t *siginfo)
578 {
579 	struct core_state core_state;
580 	struct core_name cn;
581 	struct mm_struct *mm = current->mm;
582 	struct linux_binfmt * binfmt;
583 	const struct cred *old_cred;
584 	struct cred *cred;
585 	int retval = 0;
586 	int ispipe;
587 	size_t *argv = NULL;
588 	int argc = 0;
589 	/* require nonrelative corefile path and be extra careful */
590 	bool need_suid_safe = false;
591 	bool core_dumped = false;
592 	static atomic_t core_dump_count = ATOMIC_INIT(0);
593 	struct coredump_params cprm = {
594 		.siginfo = siginfo,
595 		.regs = signal_pt_regs(),
596 		.limit = rlimit(RLIMIT_CORE),
597 		/*
598 		 * We must use the same mm->flags while dumping core to avoid
599 		 * inconsistency of bit flags, since this flag is not protected
600 		 * by any locks.
601 		 */
602 		.mm_flags = mm->flags,
603 	};
604 
605 	audit_core_dumps(siginfo->si_signo);
606 
607 	binfmt = mm->binfmt;
608 	if (!binfmt || !binfmt->core_dump)
609 		goto fail;
610 	if (!__get_dumpable(cprm.mm_flags))
611 		goto fail;
612 
613 	cred = prepare_creds();
614 	if (!cred)
615 		goto fail;
616 	/*
617 	 * We cannot trust fsuid as being the "true" uid of the process
618 	 * nor do we know its entire history. We only know it was tainted
619 	 * so we dump it as root in mode 2, and only into a controlled
620 	 * environment (pipe handler or fully qualified path).
621 	 */
622 	if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
623 		/* Setuid core dump mode */
624 		cred->fsuid = GLOBAL_ROOT_UID;	/* Dump root private */
625 		need_suid_safe = true;
626 	}
627 
628 	retval = coredump_wait(siginfo->si_signo, &core_state);
629 	if (retval < 0)
630 		goto fail_creds;
631 
632 	old_cred = override_creds(cred);
633 
634 	ispipe = format_corename(&cn, &cprm, &argv, &argc);
635 
636 	if (ispipe) {
637 		int argi;
638 		int dump_count;
639 		char **helper_argv;
640 		struct subprocess_info *sub_info;
641 
642 		if (ispipe < 0) {
643 			printk(KERN_WARNING "format_corename failed\n");
644 			printk(KERN_WARNING "Aborting core\n");
645 			goto fail_unlock;
646 		}
647 
648 		if (cprm.limit == 1) {
649 			/* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
650 			 *
651 			 * Normally core limits are irrelevant to pipes, since
652 			 * we're not writing to the file system, but we use
653 			 * cprm.limit of 1 here as a special value, this is a
654 			 * consistent way to catch recursive crashes.
655 			 * We can still crash if the core_pattern binary sets
656 			 * RLIM_CORE = !1, but it runs as root, and can do
657 			 * lots of stupid things.
658 			 *
659 			 * Note that we use task_tgid_vnr here to grab the pid
660 			 * of the process group leader.  That way we get the
661 			 * right pid if a thread in a multi-threaded
662 			 * core_pattern process dies.
663 			 */
664 			printk(KERN_WARNING
665 				"Process %d(%s) has RLIMIT_CORE set to 1\n",
666 				task_tgid_vnr(current), current->comm);
667 			printk(KERN_WARNING "Aborting core\n");
668 			goto fail_unlock;
669 		}
670 		cprm.limit = RLIM_INFINITY;
671 
672 		dump_count = atomic_inc_return(&core_dump_count);
673 		if (core_pipe_limit && (core_pipe_limit < dump_count)) {
674 			printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
675 			       task_tgid_vnr(current), current->comm);
676 			printk(KERN_WARNING "Skipping core dump\n");
677 			goto fail_dropcount;
678 		}
679 
680 		helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
681 					    GFP_KERNEL);
682 		if (!helper_argv) {
683 			printk(KERN_WARNING "%s failed to allocate memory\n",
684 			       __func__);
685 			goto fail_dropcount;
686 		}
687 		for (argi = 0; argi < argc; argi++)
688 			helper_argv[argi] = cn.corename + argv[argi];
689 		helper_argv[argi] = NULL;
690 
691 		retval = -ENOMEM;
692 		sub_info = call_usermodehelper_setup(helper_argv[0],
693 						helper_argv, NULL, GFP_KERNEL,
694 						umh_pipe_setup, NULL, &cprm);
695 		if (sub_info)
696 			retval = call_usermodehelper_exec(sub_info,
697 							  UMH_WAIT_EXEC);
698 
699 		kfree(helper_argv);
700 		if (retval) {
701 			printk(KERN_INFO "Core dump to |%s pipe failed\n",
702 			       cn.corename);
703 			goto close_fail;
704 		}
705 	} else {
706 		struct user_namespace *mnt_userns;
707 		struct inode *inode;
708 		int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
709 				 O_LARGEFILE | O_EXCL;
710 
711 		if (cprm.limit < binfmt->min_coredump)
712 			goto fail_unlock;
713 
714 		if (need_suid_safe && cn.corename[0] != '/') {
715 			printk(KERN_WARNING "Pid %d(%s) can only dump core "\
716 				"to fully qualified path!\n",
717 				task_tgid_vnr(current), current->comm);
718 			printk(KERN_WARNING "Skipping core dump\n");
719 			goto fail_unlock;
720 		}
721 
722 		/*
723 		 * Unlink the file if it exists unless this is a SUID
724 		 * binary - in that case, we're running around with root
725 		 * privs and don't want to unlink another user's coredump.
726 		 */
727 		if (!need_suid_safe) {
728 			/*
729 			 * If it doesn't exist, that's fine. If there's some
730 			 * other problem, we'll catch it at the filp_open().
731 			 */
732 			do_unlinkat(AT_FDCWD, getname_kernel(cn.corename));
733 		}
734 
735 		/*
736 		 * There is a race between unlinking and creating the
737 		 * file, but if that causes an EEXIST here, that's
738 		 * fine - another process raced with us while creating
739 		 * the corefile, and the other process won. To userspace,
740 		 * what matters is that at least one of the two processes
741 		 * writes its coredump successfully, not which one.
742 		 */
743 		if (need_suid_safe) {
744 			/*
745 			 * Using user namespaces, normal user tasks can change
746 			 * their current->fs->root to point to arbitrary
747 			 * directories. Since the intention of the "only dump
748 			 * with a fully qualified path" rule is to control where
749 			 * coredumps may be placed using root privileges,
750 			 * current->fs->root must not be used. Instead, use the
751 			 * root directory of init_task.
752 			 */
753 			struct path root;
754 
755 			task_lock(&init_task);
756 			get_fs_root(init_task.fs, &root);
757 			task_unlock(&init_task);
758 			cprm.file = file_open_root(root.dentry, root.mnt,
759 				cn.corename, open_flags, 0600);
760 			path_put(&root);
761 		} else {
762 			cprm.file = filp_open(cn.corename, open_flags, 0600);
763 		}
764 		if (IS_ERR(cprm.file))
765 			goto fail_unlock;
766 
767 		inode = file_inode(cprm.file);
768 		if (inode->i_nlink > 1)
769 			goto close_fail;
770 		if (d_unhashed(cprm.file->f_path.dentry))
771 			goto close_fail;
772 		/*
773 		 * AK: actually i see no reason to not allow this for named
774 		 * pipes etc, but keep the previous behaviour for now.
775 		 */
776 		if (!S_ISREG(inode->i_mode))
777 			goto close_fail;
778 		/*
779 		 * Don't dump core if the filesystem changed owner or mode
780 		 * of the file during file creation. This is an issue when
781 		 * a process dumps core while its cwd is e.g. on a vfat
782 		 * filesystem.
783 		 */
784 		mnt_userns = file_mnt_user_ns(cprm.file);
785 		if (!uid_eq(i_uid_into_mnt(mnt_userns, inode), current_fsuid()))
786 			goto close_fail;
787 		if ((inode->i_mode & 0677) != 0600)
788 			goto close_fail;
789 		if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
790 			goto close_fail;
791 		if (do_truncate(mnt_userns, cprm.file->f_path.dentry,
792 				0, 0, cprm.file))
793 			goto close_fail;
794 	}
795 
796 	/* get us an unshared descriptor table; almost always a no-op */
797 	/* The cell spufs coredump code reads the file descriptor tables */
798 	retval = unshare_files();
799 	if (retval)
800 		goto close_fail;
801 	if (!dump_interrupted()) {
802 		/*
803 		 * umh disabled with CONFIG_STATIC_USERMODEHELPER_PATH="" would
804 		 * have this set to NULL.
805 		 */
806 		if (!cprm.file) {
807 			pr_info("Core dump to |%s disabled\n", cn.corename);
808 			goto close_fail;
809 		}
810 		file_start_write(cprm.file);
811 		core_dumped = binfmt->core_dump(&cprm);
812 		file_end_write(cprm.file);
813 	}
814 	if (ispipe && core_pipe_limit)
815 		wait_for_dump_helpers(cprm.file);
816 close_fail:
817 	if (cprm.file)
818 		filp_close(cprm.file, NULL);
819 fail_dropcount:
820 	if (ispipe)
821 		atomic_dec(&core_dump_count);
822 fail_unlock:
823 	kfree(argv);
824 	kfree(cn.corename);
825 	coredump_finish(mm, core_dumped);
826 	revert_creds(old_cred);
827 fail_creds:
828 	put_cred(cred);
829 fail:
830 	return;
831 }
832 
833 /*
834  * Core dumping helper functions.  These are the only things you should
835  * do on a core-file: use only these functions to write out all the
836  * necessary info.
837  */
838 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
839 {
840 	struct file *file = cprm->file;
841 	loff_t pos = file->f_pos;
842 	ssize_t n;
843 	if (cprm->written + nr > cprm->limit)
844 		return 0;
845 
846 
847 	if (dump_interrupted())
848 		return 0;
849 	n = __kernel_write(file, addr, nr, &pos);
850 	if (n != nr)
851 		return 0;
852 	file->f_pos = pos;
853 	cprm->written += n;
854 	cprm->pos += n;
855 
856 	return 1;
857 }
858 EXPORT_SYMBOL(dump_emit);
859 
860 int dump_skip(struct coredump_params *cprm, size_t nr)
861 {
862 	static char zeroes[PAGE_SIZE];
863 	struct file *file = cprm->file;
864 	if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
865 		if (dump_interrupted() ||
866 		    file->f_op->llseek(file, nr, SEEK_CUR) < 0)
867 			return 0;
868 		cprm->pos += nr;
869 		return 1;
870 	} else {
871 		while (nr > PAGE_SIZE) {
872 			if (!dump_emit(cprm, zeroes, PAGE_SIZE))
873 				return 0;
874 			nr -= PAGE_SIZE;
875 		}
876 		return dump_emit(cprm, zeroes, nr);
877 	}
878 }
879 EXPORT_SYMBOL(dump_skip);
880 
881 #ifdef CONFIG_ELF_CORE
882 int dump_user_range(struct coredump_params *cprm, unsigned long start,
883 		    unsigned long len)
884 {
885 	unsigned long addr;
886 
887 	for (addr = start; addr < start + len; addr += PAGE_SIZE) {
888 		struct page *page;
889 		int stop;
890 
891 		/*
892 		 * To avoid having to allocate page tables for virtual address
893 		 * ranges that have never been used yet, and also to make it
894 		 * easy to generate sparse core files, use a helper that returns
895 		 * NULL when encountering an empty page table entry that would
896 		 * otherwise have been filled with the zero page.
897 		 */
898 		page = get_dump_page(addr);
899 		if (page) {
900 			void *kaddr = kmap_local_page(page);
901 
902 			stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
903 			kunmap_local(kaddr);
904 			put_page(page);
905 		} else {
906 			stop = !dump_skip(cprm, PAGE_SIZE);
907 		}
908 		if (stop)
909 			return 0;
910 	}
911 	return 1;
912 }
913 #endif
914 
915 int dump_align(struct coredump_params *cprm, int align)
916 {
917 	unsigned mod = cprm->pos & (align - 1);
918 	if (align & (align - 1))
919 		return 0;
920 	return mod ? dump_skip(cprm, align - mod) : 1;
921 }
922 EXPORT_SYMBOL(dump_align);
923 
924 /*
925  * Ensures that file size is big enough to contain the current file
926  * postion. This prevents gdb from complaining about a truncated file
927  * if the last "write" to the file was dump_skip.
928  */
929 void dump_truncate(struct coredump_params *cprm)
930 {
931 	struct file *file = cprm->file;
932 	loff_t offset;
933 
934 	if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
935 		offset = file->f_op->llseek(file, 0, SEEK_CUR);
936 		if (i_size_read(file->f_mapping->host) < offset)
937 			do_truncate(file_mnt_user_ns(file), file->f_path.dentry,
938 				    offset, 0, file);
939 	}
940 }
941 EXPORT_SYMBOL(dump_truncate);
942 
943 /*
944  * The purpose of always_dump_vma() is to make sure that special kernel mappings
945  * that are useful for post-mortem analysis are included in every core dump.
946  * In that way we ensure that the core dump is fully interpretable later
947  * without matching up the same kernel and hardware config to see what PC values
948  * meant. These special mappings include - vDSO, vsyscall, and other
949  * architecture specific mappings
950  */
951 static bool always_dump_vma(struct vm_area_struct *vma)
952 {
953 	/* Any vsyscall mappings? */
954 	if (vma == get_gate_vma(vma->vm_mm))
955 		return true;
956 
957 	/*
958 	 * Assume that all vmas with a .name op should always be dumped.
959 	 * If this changes, a new vm_ops field can easily be added.
960 	 */
961 	if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
962 		return true;
963 
964 	/*
965 	 * arch_vma_name() returns non-NULL for special architecture mappings,
966 	 * such as vDSO sections.
967 	 */
968 	if (arch_vma_name(vma))
969 		return true;
970 
971 	return false;
972 }
973 
974 /*
975  * Decide how much of @vma's contents should be included in a core dump.
976  */
977 static unsigned long vma_dump_size(struct vm_area_struct *vma,
978 				   unsigned long mm_flags)
979 {
980 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
981 
982 	/* always dump the vdso and vsyscall sections */
983 	if (always_dump_vma(vma))
984 		goto whole;
985 
986 	if (vma->vm_flags & VM_DONTDUMP)
987 		return 0;
988 
989 	/* support for DAX */
990 	if (vma_is_dax(vma)) {
991 		if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
992 			goto whole;
993 		if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
994 			goto whole;
995 		return 0;
996 	}
997 
998 	/* Hugetlb memory check */
999 	if (is_vm_hugetlb_page(vma)) {
1000 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1001 			goto whole;
1002 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1003 			goto whole;
1004 		return 0;
1005 	}
1006 
1007 	/* Do not dump I/O mapped devices or special mappings */
1008 	if (vma->vm_flags & VM_IO)
1009 		return 0;
1010 
1011 	/* By default, dump shared memory if mapped from an anonymous file. */
1012 	if (vma->vm_flags & VM_SHARED) {
1013 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1014 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1015 			goto whole;
1016 		return 0;
1017 	}
1018 
1019 	/* Dump segments that have been written to.  */
1020 	if ((!IS_ENABLED(CONFIG_MMU) || vma->anon_vma) && FILTER(ANON_PRIVATE))
1021 		goto whole;
1022 	if (vma->vm_file == NULL)
1023 		return 0;
1024 
1025 	if (FILTER(MAPPED_PRIVATE))
1026 		goto whole;
1027 
1028 	/*
1029 	 * If this is the beginning of an executable file mapping,
1030 	 * dump the first page to aid in determining what was mapped here.
1031 	 */
1032 	if (FILTER(ELF_HEADERS) &&
1033 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ) &&
1034 	    (READ_ONCE(file_inode(vma->vm_file)->i_mode) & 0111) != 0)
1035 		return PAGE_SIZE;
1036 
1037 #undef	FILTER
1038 
1039 	return 0;
1040 
1041 whole:
1042 	return vma->vm_end - vma->vm_start;
1043 }
1044 
1045 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1046 					struct vm_area_struct *gate_vma)
1047 {
1048 	struct vm_area_struct *ret = tsk->mm->mmap;
1049 
1050 	if (ret)
1051 		return ret;
1052 	return gate_vma;
1053 }
1054 
1055 /*
1056  * Helper function for iterating across a vma list.  It ensures that the caller
1057  * will visit `gate_vma' prior to terminating the search.
1058  */
1059 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1060 				       struct vm_area_struct *gate_vma)
1061 {
1062 	struct vm_area_struct *ret;
1063 
1064 	ret = this_vma->vm_next;
1065 	if (ret)
1066 		return ret;
1067 	if (this_vma == gate_vma)
1068 		return NULL;
1069 	return gate_vma;
1070 }
1071 
1072 /*
1073  * Under the mmap_lock, take a snapshot of relevant information about the task's
1074  * VMAs.
1075  */
1076 int dump_vma_snapshot(struct coredump_params *cprm, int *vma_count,
1077 		      struct core_vma_metadata **vma_meta,
1078 		      size_t *vma_data_size_ptr)
1079 {
1080 	struct vm_area_struct *vma, *gate_vma;
1081 	struct mm_struct *mm = current->mm;
1082 	int i;
1083 	size_t vma_data_size = 0;
1084 
1085 	/*
1086 	 * Once the stack expansion code is fixed to not change VMA bounds
1087 	 * under mmap_lock in read mode, this can be changed to take the
1088 	 * mmap_lock in read mode.
1089 	 */
1090 	if (mmap_write_lock_killable(mm))
1091 		return -EINTR;
1092 
1093 	gate_vma = get_gate_vma(mm);
1094 	*vma_count = mm->map_count + (gate_vma ? 1 : 0);
1095 
1096 	*vma_meta = kvmalloc_array(*vma_count, sizeof(**vma_meta), GFP_KERNEL);
1097 	if (!*vma_meta) {
1098 		mmap_write_unlock(mm);
1099 		return -ENOMEM;
1100 	}
1101 
1102 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
1103 			vma = next_vma(vma, gate_vma), i++) {
1104 		struct core_vma_metadata *m = (*vma_meta) + i;
1105 
1106 		m->start = vma->vm_start;
1107 		m->end = vma->vm_end;
1108 		m->flags = vma->vm_flags;
1109 		m->dump_size = vma_dump_size(vma, cprm->mm_flags);
1110 
1111 		vma_data_size += m->dump_size;
1112 	}
1113 
1114 	mmap_write_unlock(mm);
1115 
1116 	if (WARN_ON(i != *vma_count))
1117 		return -EFAULT;
1118 
1119 	*vma_data_size_ptr = vma_data_size;
1120 	return 0;
1121 }
1122