xref: /linux/fs/coredump.c (revision d8e473182ab9e85708067be81d20424045d939fa)
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/kmod.h>
35 #include <linux/fsnotify.h>
36 #include <linux/fs_struct.h>
37 #include <linux/pipe_fs_i.h>
38 #include <linux/oom.h>
39 #include <linux/compat.h>
40 #include <linux/fs.h>
41 #include <linux/path.h>
42 #include <linux/timekeeping.h>
43 #include <linux/sysctl.h>
44 #include <linux/elf.h>
45 
46 #include <linux/uaccess.h>
47 #include <asm/mmu_context.h>
48 #include <asm/tlb.h>
49 #include <asm/exec.h>
50 
51 #include <trace/events/task.h>
52 #include "internal.h"
53 
54 #include <trace/events/sched.h>
55 
56 static bool dump_vma_snapshot(struct coredump_params *cprm);
57 static void free_vma_snapshot(struct coredump_params *cprm);
58 
59 static int core_uses_pid;
60 static unsigned int core_pipe_limit;
61 static char core_pattern[CORENAME_MAX_SIZE] = "core";
62 static int core_name_size = CORENAME_MAX_SIZE;
63 
64 struct core_name {
65 	char *corename;
66 	int used, size;
67 };
68 
69 static int expand_corename(struct core_name *cn, int size)
70 {
71 	char *corename;
72 
73 	size = kmalloc_size_roundup(size);
74 	corename = krealloc(cn->corename, size, GFP_KERNEL);
75 
76 	if (!corename)
77 		return -ENOMEM;
78 
79 	if (size > core_name_size) /* racy but harmless */
80 		core_name_size = size;
81 
82 	cn->size = size;
83 	cn->corename = corename;
84 	return 0;
85 }
86 
87 static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
88 				     va_list arg)
89 {
90 	int free, need;
91 	va_list arg_copy;
92 
93 again:
94 	free = cn->size - cn->used;
95 
96 	va_copy(arg_copy, arg);
97 	need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
98 	va_end(arg_copy);
99 
100 	if (need < free) {
101 		cn->used += need;
102 		return 0;
103 	}
104 
105 	if (!expand_corename(cn, cn->size + need - free + 1))
106 		goto again;
107 
108 	return -ENOMEM;
109 }
110 
111 static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
112 {
113 	va_list arg;
114 	int ret;
115 
116 	va_start(arg, fmt);
117 	ret = cn_vprintf(cn, fmt, arg);
118 	va_end(arg);
119 
120 	return ret;
121 }
122 
123 static __printf(2, 3)
124 int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
125 {
126 	int cur = cn->used;
127 	va_list arg;
128 	int ret;
129 
130 	va_start(arg, fmt);
131 	ret = cn_vprintf(cn, fmt, arg);
132 	va_end(arg);
133 
134 	if (ret == 0) {
135 		/*
136 		 * Ensure that this coredump name component can't cause the
137 		 * resulting corefile path to consist of a ".." or ".".
138 		 */
139 		if ((cn->used - cur == 1 && cn->corename[cur] == '.') ||
140 				(cn->used - cur == 2 && cn->corename[cur] == '.'
141 				&& cn->corename[cur+1] == '.'))
142 			cn->corename[cur] = '!';
143 
144 		/*
145 		 * Empty names are fishy and could be used to create a "//" in a
146 		 * corefile name, causing the coredump to happen one directory
147 		 * level too high. Enforce that all components of the core
148 		 * pattern are at least one character long.
149 		 */
150 		if (cn->used == cur)
151 			ret = cn_printf(cn, "!");
152 	}
153 
154 	for (; cur < cn->used; ++cur) {
155 		if (cn->corename[cur] == '/')
156 			cn->corename[cur] = '!';
157 	}
158 	return ret;
159 }
160 
161 static int cn_print_exe_file(struct core_name *cn, bool name_only)
162 {
163 	struct file *exe_file;
164 	char *pathbuf, *path, *ptr;
165 	int ret;
166 
167 	exe_file = get_mm_exe_file(current->mm);
168 	if (!exe_file)
169 		return cn_esc_printf(cn, "%s (path unknown)", current->comm);
170 
171 	pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
172 	if (!pathbuf) {
173 		ret = -ENOMEM;
174 		goto put_exe_file;
175 	}
176 
177 	path = file_path(exe_file, pathbuf, PATH_MAX);
178 	if (IS_ERR(path)) {
179 		ret = PTR_ERR(path);
180 		goto free_buf;
181 	}
182 
183 	if (name_only) {
184 		ptr = strrchr(path, '/');
185 		if (ptr)
186 			path = ptr + 1;
187 	}
188 	ret = cn_esc_printf(cn, "%s", path);
189 
190 free_buf:
191 	kfree(pathbuf);
192 put_exe_file:
193 	fput(exe_file);
194 	return ret;
195 }
196 
197 /* format_corename will inspect the pattern parameter, and output a
198  * name into corename, which must have space for at least
199  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
200  */
201 static int format_corename(struct core_name *cn, struct coredump_params *cprm,
202 			   size_t **argv, int *argc)
203 {
204 	const struct cred *cred = current_cred();
205 	const char *pat_ptr = core_pattern;
206 	int ispipe = (*pat_ptr == '|');
207 	bool was_space = false;
208 	int pid_in_pattern = 0;
209 	int err = 0;
210 
211 	cn->used = 0;
212 	cn->corename = NULL;
213 	if (expand_corename(cn, core_name_size))
214 		return -ENOMEM;
215 	cn->corename[0] = '\0';
216 
217 	if (ispipe) {
218 		int argvs = sizeof(core_pattern) / 2;
219 		(*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL);
220 		if (!(*argv))
221 			return -ENOMEM;
222 		(*argv)[(*argc)++] = 0;
223 		++pat_ptr;
224 		if (!(*pat_ptr))
225 			return -ENOMEM;
226 	}
227 
228 	/* Repeat as long as we have more pattern to process and more output
229 	   space */
230 	while (*pat_ptr) {
231 		/*
232 		 * Split on spaces before doing template expansion so that
233 		 * %e and %E don't get split if they have spaces in them
234 		 */
235 		if (ispipe) {
236 			if (isspace(*pat_ptr)) {
237 				if (cn->used != 0)
238 					was_space = true;
239 				pat_ptr++;
240 				continue;
241 			} else if (was_space) {
242 				was_space = false;
243 				err = cn_printf(cn, "%c", '\0');
244 				if (err)
245 					return err;
246 				(*argv)[(*argc)++] = cn->used;
247 			}
248 		}
249 		if (*pat_ptr != '%') {
250 			err = cn_printf(cn, "%c", *pat_ptr++);
251 		} else {
252 			switch (*++pat_ptr) {
253 			/* single % at the end, drop that */
254 			case 0:
255 				goto out;
256 			/* Double percent, output one percent */
257 			case '%':
258 				err = cn_printf(cn, "%c", '%');
259 				break;
260 			/* pid */
261 			case 'p':
262 				pid_in_pattern = 1;
263 				err = cn_printf(cn, "%d",
264 					      task_tgid_vnr(current));
265 				break;
266 			/* global pid */
267 			case 'P':
268 				err = cn_printf(cn, "%d",
269 					      task_tgid_nr(current));
270 				break;
271 			case 'i':
272 				err = cn_printf(cn, "%d",
273 					      task_pid_vnr(current));
274 				break;
275 			case 'I':
276 				err = cn_printf(cn, "%d",
277 					      task_pid_nr(current));
278 				break;
279 			/* uid */
280 			case 'u':
281 				err = cn_printf(cn, "%u",
282 						from_kuid(&init_user_ns,
283 							  cred->uid));
284 				break;
285 			/* gid */
286 			case 'g':
287 				err = cn_printf(cn, "%u",
288 						from_kgid(&init_user_ns,
289 							  cred->gid));
290 				break;
291 			case 'd':
292 				err = cn_printf(cn, "%d",
293 					__get_dumpable(cprm->mm_flags));
294 				break;
295 			/* signal that caused the coredump */
296 			case 's':
297 				err = cn_printf(cn, "%d",
298 						cprm->siginfo->si_signo);
299 				break;
300 			/* UNIX time of coredump */
301 			case 't': {
302 				time64_t time;
303 
304 				time = ktime_get_real_seconds();
305 				err = cn_printf(cn, "%lld", time);
306 				break;
307 			}
308 			/* hostname */
309 			case 'h':
310 				down_read(&uts_sem);
311 				err = cn_esc_printf(cn, "%s",
312 					      utsname()->nodename);
313 				up_read(&uts_sem);
314 				break;
315 			/* executable, could be changed by prctl PR_SET_NAME etc */
316 			case 'e':
317 				err = cn_esc_printf(cn, "%s", current->comm);
318 				break;
319 			/* file name of executable */
320 			case 'f':
321 				err = cn_print_exe_file(cn, true);
322 				break;
323 			case 'E':
324 				err = cn_print_exe_file(cn, false);
325 				break;
326 			/* core limit size */
327 			case 'c':
328 				err = cn_printf(cn, "%lu",
329 					      rlimit(RLIMIT_CORE));
330 				break;
331 			/* CPU the task ran on */
332 			case 'C':
333 				err = cn_printf(cn, "%d", cprm->cpu);
334 				break;
335 			default:
336 				break;
337 			}
338 			++pat_ptr;
339 		}
340 
341 		if (err)
342 			return err;
343 	}
344 
345 out:
346 	/* Backward compatibility with core_uses_pid:
347 	 *
348 	 * If core_pattern does not include a %p (as is the default)
349 	 * and core_uses_pid is set, then .%pid will be appended to
350 	 * the filename. Do not do this for piped commands. */
351 	if (!ispipe && !pid_in_pattern && core_uses_pid) {
352 		err = cn_printf(cn, ".%d", task_tgid_vnr(current));
353 		if (err)
354 			return err;
355 	}
356 	return ispipe;
357 }
358 
359 static int zap_process(struct task_struct *start, int exit_code)
360 {
361 	struct task_struct *t;
362 	int nr = 0;
363 
364 	/* Allow SIGKILL, see prepare_signal() */
365 	start->signal->flags = SIGNAL_GROUP_EXIT;
366 	start->signal->group_exit_code = exit_code;
367 	start->signal->group_stop_count = 0;
368 
369 	for_each_thread(start, t) {
370 		task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
371 		if (t != current && !(t->flags & PF_POSTCOREDUMP)) {
372 			sigaddset(&t->pending.signal, SIGKILL);
373 			signal_wake_up(t, 1);
374 			nr++;
375 		}
376 	}
377 
378 	return nr;
379 }
380 
381 static int zap_threads(struct task_struct *tsk,
382 			struct core_state *core_state, int exit_code)
383 {
384 	struct signal_struct *signal = tsk->signal;
385 	int nr = -EAGAIN;
386 
387 	spin_lock_irq(&tsk->sighand->siglock);
388 	if (!(signal->flags & SIGNAL_GROUP_EXIT) && !signal->group_exec_task) {
389 		signal->core_state = core_state;
390 		nr = zap_process(tsk, exit_code);
391 		clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
392 		tsk->flags |= PF_DUMPCORE;
393 		atomic_set(&core_state->nr_threads, nr);
394 	}
395 	spin_unlock_irq(&tsk->sighand->siglock);
396 	return nr;
397 }
398 
399 static int coredump_wait(int exit_code, struct core_state *core_state)
400 {
401 	struct task_struct *tsk = current;
402 	int core_waiters = -EBUSY;
403 
404 	init_completion(&core_state->startup);
405 	core_state->dumper.task = tsk;
406 	core_state->dumper.next = NULL;
407 
408 	core_waiters = zap_threads(tsk, core_state, exit_code);
409 	if (core_waiters > 0) {
410 		struct core_thread *ptr;
411 
412 		wait_for_completion_state(&core_state->startup,
413 					  TASK_UNINTERRUPTIBLE|TASK_FREEZABLE);
414 		/*
415 		 * Wait for all the threads to become inactive, so that
416 		 * all the thread context (extended register state, like
417 		 * fpu etc) gets copied to the memory.
418 		 */
419 		ptr = core_state->dumper.next;
420 		while (ptr != NULL) {
421 			wait_task_inactive(ptr->task, TASK_ANY);
422 			ptr = ptr->next;
423 		}
424 	}
425 
426 	return core_waiters;
427 }
428 
429 static void coredump_finish(bool core_dumped)
430 {
431 	struct core_thread *curr, *next;
432 	struct task_struct *task;
433 
434 	spin_lock_irq(&current->sighand->siglock);
435 	if (core_dumped && !__fatal_signal_pending(current))
436 		current->signal->group_exit_code |= 0x80;
437 	next = current->signal->core_state->dumper.next;
438 	current->signal->core_state = NULL;
439 	spin_unlock_irq(&current->sighand->siglock);
440 
441 	while ((curr = next) != NULL) {
442 		next = curr->next;
443 		task = curr->task;
444 		/*
445 		 * see coredump_task_exit(), curr->task must not see
446 		 * ->task == NULL before we read ->next.
447 		 */
448 		smp_mb();
449 		curr->task = NULL;
450 		wake_up_process(task);
451 	}
452 }
453 
454 static bool dump_interrupted(void)
455 {
456 	/*
457 	 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
458 	 * can do try_to_freeze() and check __fatal_signal_pending(),
459 	 * but then we need to teach dump_write() to restart and clear
460 	 * TIF_SIGPENDING.
461 	 */
462 	return fatal_signal_pending(current) || freezing(current);
463 }
464 
465 static void wait_for_dump_helpers(struct file *file)
466 {
467 	struct pipe_inode_info *pipe = file->private_data;
468 
469 	pipe_lock(pipe);
470 	pipe->readers++;
471 	pipe->writers--;
472 	wake_up_interruptible_sync(&pipe->rd_wait);
473 	kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
474 	pipe_unlock(pipe);
475 
476 	/*
477 	 * We actually want wait_event_freezable() but then we need
478 	 * to clear TIF_SIGPENDING and improve dump_interrupted().
479 	 */
480 	wait_event_interruptible(pipe->rd_wait, pipe->readers == 1);
481 
482 	pipe_lock(pipe);
483 	pipe->readers--;
484 	pipe->writers++;
485 	pipe_unlock(pipe);
486 }
487 
488 /*
489  * umh_pipe_setup
490  * helper function to customize the process used
491  * to collect the core in userspace.  Specifically
492  * it sets up a pipe and installs it as fd 0 (stdin)
493  * for the process.  Returns 0 on success, or
494  * PTR_ERR on failure.
495  * Note that it also sets the core limit to 1.  This
496  * is a special value that we use to trap recursive
497  * core dumps
498  */
499 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
500 {
501 	struct file *files[2];
502 	struct coredump_params *cp = (struct coredump_params *)info->data;
503 	int err = create_pipe_files(files, 0);
504 	if (err)
505 		return err;
506 
507 	cp->file = files[1];
508 
509 	err = replace_fd(0, files[0], 0);
510 	fput(files[0]);
511 	/* and disallow core files too */
512 	current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
513 
514 	return err;
515 }
516 
517 void do_coredump(const kernel_siginfo_t *siginfo)
518 {
519 	struct core_state core_state;
520 	struct core_name cn;
521 	struct mm_struct *mm = current->mm;
522 	struct linux_binfmt * binfmt;
523 	const struct cred *old_cred;
524 	struct cred *cred;
525 	int retval = 0;
526 	int ispipe;
527 	size_t *argv = NULL;
528 	int argc = 0;
529 	/* require nonrelative corefile path and be extra careful */
530 	bool need_suid_safe = false;
531 	bool core_dumped = false;
532 	static atomic_t core_dump_count = ATOMIC_INIT(0);
533 	struct coredump_params cprm = {
534 		.siginfo = siginfo,
535 		.limit = rlimit(RLIMIT_CORE),
536 		/*
537 		 * We must use the same mm->flags while dumping core to avoid
538 		 * inconsistency of bit flags, since this flag is not protected
539 		 * by any locks.
540 		 */
541 		.mm_flags = mm->flags,
542 		.vma_meta = NULL,
543 		.cpu = raw_smp_processor_id(),
544 	};
545 
546 	audit_core_dumps(siginfo->si_signo);
547 
548 	binfmt = mm->binfmt;
549 	if (!binfmt || !binfmt->core_dump)
550 		goto fail;
551 	if (!__get_dumpable(cprm.mm_flags))
552 		goto fail;
553 
554 	cred = prepare_creds();
555 	if (!cred)
556 		goto fail;
557 	/*
558 	 * We cannot trust fsuid as being the "true" uid of the process
559 	 * nor do we know its entire history. We only know it was tainted
560 	 * so we dump it as root in mode 2, and only into a controlled
561 	 * environment (pipe handler or fully qualified path).
562 	 */
563 	if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
564 		/* Setuid core dump mode */
565 		cred->fsuid = GLOBAL_ROOT_UID;	/* Dump root private */
566 		need_suid_safe = true;
567 	}
568 
569 	retval = coredump_wait(siginfo->si_signo, &core_state);
570 	if (retval < 0)
571 		goto fail_creds;
572 
573 	old_cred = override_creds(cred);
574 
575 	ispipe = format_corename(&cn, &cprm, &argv, &argc);
576 
577 	if (ispipe) {
578 		int argi;
579 		int dump_count;
580 		char **helper_argv;
581 		struct subprocess_info *sub_info;
582 
583 		if (ispipe < 0) {
584 			printk(KERN_WARNING "format_corename failed\n");
585 			printk(KERN_WARNING "Aborting core\n");
586 			goto fail_unlock;
587 		}
588 
589 		if (cprm.limit == 1) {
590 			/* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
591 			 *
592 			 * Normally core limits are irrelevant to pipes, since
593 			 * we're not writing to the file system, but we use
594 			 * cprm.limit of 1 here as a special value, this is a
595 			 * consistent way to catch recursive crashes.
596 			 * We can still crash if the core_pattern binary sets
597 			 * RLIM_CORE = !1, but it runs as root, and can do
598 			 * lots of stupid things.
599 			 *
600 			 * Note that we use task_tgid_vnr here to grab the pid
601 			 * of the process group leader.  That way we get the
602 			 * right pid if a thread in a multi-threaded
603 			 * core_pattern process dies.
604 			 */
605 			printk(KERN_WARNING
606 				"Process %d(%s) has RLIMIT_CORE set to 1\n",
607 				task_tgid_vnr(current), current->comm);
608 			printk(KERN_WARNING "Aborting core\n");
609 			goto fail_unlock;
610 		}
611 		cprm.limit = RLIM_INFINITY;
612 
613 		dump_count = atomic_inc_return(&core_dump_count);
614 		if (core_pipe_limit && (core_pipe_limit < dump_count)) {
615 			printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
616 			       task_tgid_vnr(current), current->comm);
617 			printk(KERN_WARNING "Skipping core dump\n");
618 			goto fail_dropcount;
619 		}
620 
621 		helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
622 					    GFP_KERNEL);
623 		if (!helper_argv) {
624 			printk(KERN_WARNING "%s failed to allocate memory\n",
625 			       __func__);
626 			goto fail_dropcount;
627 		}
628 		for (argi = 0; argi < argc; argi++)
629 			helper_argv[argi] = cn.corename + argv[argi];
630 		helper_argv[argi] = NULL;
631 
632 		retval = -ENOMEM;
633 		sub_info = call_usermodehelper_setup(helper_argv[0],
634 						helper_argv, NULL, GFP_KERNEL,
635 						umh_pipe_setup, NULL, &cprm);
636 		if (sub_info)
637 			retval = call_usermodehelper_exec(sub_info,
638 							  UMH_WAIT_EXEC);
639 
640 		kfree(helper_argv);
641 		if (retval) {
642 			printk(KERN_INFO "Core dump to |%s pipe failed\n",
643 			       cn.corename);
644 			goto close_fail;
645 		}
646 	} else {
647 		struct mnt_idmap *idmap;
648 		struct inode *inode;
649 		int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
650 				 O_LARGEFILE | O_EXCL;
651 
652 		if (cprm.limit < binfmt->min_coredump)
653 			goto fail_unlock;
654 
655 		if (need_suid_safe && cn.corename[0] != '/') {
656 			printk(KERN_WARNING "Pid %d(%s) can only dump core "\
657 				"to fully qualified path!\n",
658 				task_tgid_vnr(current), current->comm);
659 			printk(KERN_WARNING "Skipping core dump\n");
660 			goto fail_unlock;
661 		}
662 
663 		/*
664 		 * Unlink the file if it exists unless this is a SUID
665 		 * binary - in that case, we're running around with root
666 		 * privs and don't want to unlink another user's coredump.
667 		 */
668 		if (!need_suid_safe) {
669 			/*
670 			 * If it doesn't exist, that's fine. If there's some
671 			 * other problem, we'll catch it at the filp_open().
672 			 */
673 			do_unlinkat(AT_FDCWD, getname_kernel(cn.corename));
674 		}
675 
676 		/*
677 		 * There is a race between unlinking and creating the
678 		 * file, but if that causes an EEXIST here, that's
679 		 * fine - another process raced with us while creating
680 		 * the corefile, and the other process won. To userspace,
681 		 * what matters is that at least one of the two processes
682 		 * writes its coredump successfully, not which one.
683 		 */
684 		if (need_suid_safe) {
685 			/*
686 			 * Using user namespaces, normal user tasks can change
687 			 * their current->fs->root to point to arbitrary
688 			 * directories. Since the intention of the "only dump
689 			 * with a fully qualified path" rule is to control where
690 			 * coredumps may be placed using root privileges,
691 			 * current->fs->root must not be used. Instead, use the
692 			 * root directory of init_task.
693 			 */
694 			struct path root;
695 
696 			task_lock(&init_task);
697 			get_fs_root(init_task.fs, &root);
698 			task_unlock(&init_task);
699 			cprm.file = file_open_root(&root, cn.corename,
700 						   open_flags, 0600);
701 			path_put(&root);
702 		} else {
703 			cprm.file = filp_open(cn.corename, open_flags, 0600);
704 		}
705 		if (IS_ERR(cprm.file))
706 			goto fail_unlock;
707 
708 		inode = file_inode(cprm.file);
709 		if (inode->i_nlink > 1)
710 			goto close_fail;
711 		if (d_unhashed(cprm.file->f_path.dentry))
712 			goto close_fail;
713 		/*
714 		 * AK: actually i see no reason to not allow this for named
715 		 * pipes etc, but keep the previous behaviour for now.
716 		 */
717 		if (!S_ISREG(inode->i_mode))
718 			goto close_fail;
719 		/*
720 		 * Don't dump core if the filesystem changed owner or mode
721 		 * of the file during file creation. This is an issue when
722 		 * a process dumps core while its cwd is e.g. on a vfat
723 		 * filesystem.
724 		 */
725 		idmap = file_mnt_idmap(cprm.file);
726 		if (!vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode),
727 				    current_fsuid())) {
728 			pr_info_ratelimited("Core dump to %s aborted: cannot preserve file owner\n",
729 					    cn.corename);
730 			goto close_fail;
731 		}
732 		if ((inode->i_mode & 0677) != 0600) {
733 			pr_info_ratelimited("Core dump to %s aborted: cannot preserve file permissions\n",
734 					    cn.corename);
735 			goto close_fail;
736 		}
737 		if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
738 			goto close_fail;
739 		if (do_truncate(idmap, cprm.file->f_path.dentry,
740 				0, 0, cprm.file))
741 			goto close_fail;
742 	}
743 
744 	/* get us an unshared descriptor table; almost always a no-op */
745 	/* The cell spufs coredump code reads the file descriptor tables */
746 	retval = unshare_files();
747 	if (retval)
748 		goto close_fail;
749 	if (!dump_interrupted()) {
750 		/*
751 		 * umh disabled with CONFIG_STATIC_USERMODEHELPER_PATH="" would
752 		 * have this set to NULL.
753 		 */
754 		if (!cprm.file) {
755 			pr_info("Core dump to |%s disabled\n", cn.corename);
756 			goto close_fail;
757 		}
758 		if (!dump_vma_snapshot(&cprm))
759 			goto close_fail;
760 
761 		file_start_write(cprm.file);
762 		core_dumped = binfmt->core_dump(&cprm);
763 		/*
764 		 * Ensures that file size is big enough to contain the current
765 		 * file postion. This prevents gdb from complaining about
766 		 * a truncated file if the last "write" to the file was
767 		 * dump_skip.
768 		 */
769 		if (cprm.to_skip) {
770 			cprm.to_skip--;
771 			dump_emit(&cprm, "", 1);
772 		}
773 		file_end_write(cprm.file);
774 		free_vma_snapshot(&cprm);
775 	}
776 	if (ispipe && core_pipe_limit)
777 		wait_for_dump_helpers(cprm.file);
778 close_fail:
779 	if (cprm.file)
780 		filp_close(cprm.file, NULL);
781 fail_dropcount:
782 	if (ispipe)
783 		atomic_dec(&core_dump_count);
784 fail_unlock:
785 	kfree(argv);
786 	kfree(cn.corename);
787 	coredump_finish(core_dumped);
788 	revert_creds(old_cred);
789 fail_creds:
790 	put_cred(cred);
791 fail:
792 	return;
793 }
794 
795 /*
796  * Core dumping helper functions.  These are the only things you should
797  * do on a core-file: use only these functions to write out all the
798  * necessary info.
799  */
800 static int __dump_emit(struct coredump_params *cprm, const void *addr, int nr)
801 {
802 	struct file *file = cprm->file;
803 	loff_t pos = file->f_pos;
804 	ssize_t n;
805 	if (cprm->written + nr > cprm->limit)
806 		return 0;
807 
808 
809 	if (dump_interrupted())
810 		return 0;
811 	n = __kernel_write(file, addr, nr, &pos);
812 	if (n != nr)
813 		return 0;
814 	file->f_pos = pos;
815 	cprm->written += n;
816 	cprm->pos += n;
817 
818 	return 1;
819 }
820 
821 static int __dump_skip(struct coredump_params *cprm, size_t nr)
822 {
823 	static char zeroes[PAGE_SIZE];
824 	struct file *file = cprm->file;
825 	if (file->f_mode & FMODE_LSEEK) {
826 		if (dump_interrupted() ||
827 		    vfs_llseek(file, nr, SEEK_CUR) < 0)
828 			return 0;
829 		cprm->pos += nr;
830 		return 1;
831 	} else {
832 		while (nr > PAGE_SIZE) {
833 			if (!__dump_emit(cprm, zeroes, PAGE_SIZE))
834 				return 0;
835 			nr -= PAGE_SIZE;
836 		}
837 		return __dump_emit(cprm, zeroes, nr);
838 	}
839 }
840 
841 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
842 {
843 	if (cprm->to_skip) {
844 		if (!__dump_skip(cprm, cprm->to_skip))
845 			return 0;
846 		cprm->to_skip = 0;
847 	}
848 	return __dump_emit(cprm, addr, nr);
849 }
850 EXPORT_SYMBOL(dump_emit);
851 
852 void dump_skip_to(struct coredump_params *cprm, unsigned long pos)
853 {
854 	cprm->to_skip = pos - cprm->pos;
855 }
856 EXPORT_SYMBOL(dump_skip_to);
857 
858 void dump_skip(struct coredump_params *cprm, size_t nr)
859 {
860 	cprm->to_skip += nr;
861 }
862 EXPORT_SYMBOL(dump_skip);
863 
864 #ifdef CONFIG_ELF_CORE
865 static int dump_emit_page(struct coredump_params *cprm, struct page *page)
866 {
867 	struct bio_vec bvec;
868 	struct iov_iter iter;
869 	struct file *file = cprm->file;
870 	loff_t pos;
871 	ssize_t n;
872 
873 	if (cprm->to_skip) {
874 		if (!__dump_skip(cprm, cprm->to_skip))
875 			return 0;
876 		cprm->to_skip = 0;
877 	}
878 	if (cprm->written + PAGE_SIZE > cprm->limit)
879 		return 0;
880 	if (dump_interrupted())
881 		return 0;
882 	pos = file->f_pos;
883 	bvec_set_page(&bvec, page, PAGE_SIZE, 0);
884 	iov_iter_bvec(&iter, ITER_SOURCE, &bvec, 1, PAGE_SIZE);
885 	n = __kernel_write_iter(cprm->file, &iter, &pos);
886 	if (n != PAGE_SIZE)
887 		return 0;
888 	file->f_pos = pos;
889 	cprm->written += PAGE_SIZE;
890 	cprm->pos += PAGE_SIZE;
891 
892 	return 1;
893 }
894 
895 int dump_user_range(struct coredump_params *cprm, unsigned long start,
896 		    unsigned long len)
897 {
898 	unsigned long addr;
899 
900 	for (addr = start; addr < start + len; addr += PAGE_SIZE) {
901 		struct page *page;
902 
903 		/*
904 		 * To avoid having to allocate page tables for virtual address
905 		 * ranges that have never been used yet, and also to make it
906 		 * easy to generate sparse core files, use a helper that returns
907 		 * NULL when encountering an empty page table entry that would
908 		 * otherwise have been filled with the zero page.
909 		 */
910 		page = get_dump_page(addr);
911 		if (page) {
912 			int stop = !dump_emit_page(cprm, page);
913 			put_page(page);
914 			if (stop)
915 				return 0;
916 		} else {
917 			dump_skip(cprm, PAGE_SIZE);
918 		}
919 	}
920 	return 1;
921 }
922 #endif
923 
924 int dump_align(struct coredump_params *cprm, int align)
925 {
926 	unsigned mod = (cprm->pos + cprm->to_skip) & (align - 1);
927 	if (align & (align - 1))
928 		return 0;
929 	if (mod)
930 		cprm->to_skip += align - mod;
931 	return 1;
932 }
933 EXPORT_SYMBOL(dump_align);
934 
935 #ifdef CONFIG_SYSCTL
936 
937 void validate_coredump_safety(void)
938 {
939 	if (suid_dumpable == SUID_DUMP_ROOT &&
940 	    core_pattern[0] != '/' && core_pattern[0] != '|') {
941 		pr_warn(
942 "Unsafe core_pattern used with fs.suid_dumpable=2.\n"
943 "Pipe handler or fully qualified core dump path required.\n"
944 "Set kernel.core_pattern before fs.suid_dumpable.\n"
945 		);
946 	}
947 }
948 
949 static int proc_dostring_coredump(struct ctl_table *table, int write,
950 		  void *buffer, size_t *lenp, loff_t *ppos)
951 {
952 	int error = proc_dostring(table, write, buffer, lenp, ppos);
953 
954 	if (!error)
955 		validate_coredump_safety();
956 	return error;
957 }
958 
959 static struct ctl_table coredump_sysctls[] = {
960 	{
961 		.procname	= "core_uses_pid",
962 		.data		= &core_uses_pid,
963 		.maxlen		= sizeof(int),
964 		.mode		= 0644,
965 		.proc_handler	= proc_dointvec,
966 	},
967 	{
968 		.procname	= "core_pattern",
969 		.data		= core_pattern,
970 		.maxlen		= CORENAME_MAX_SIZE,
971 		.mode		= 0644,
972 		.proc_handler	= proc_dostring_coredump,
973 	},
974 	{
975 		.procname	= "core_pipe_limit",
976 		.data		= &core_pipe_limit,
977 		.maxlen		= sizeof(unsigned int),
978 		.mode		= 0644,
979 		.proc_handler	= proc_dointvec,
980 	},
981 	{ }
982 };
983 
984 static int __init init_fs_coredump_sysctls(void)
985 {
986 	register_sysctl_init("kernel", coredump_sysctls);
987 	return 0;
988 }
989 fs_initcall(init_fs_coredump_sysctls);
990 #endif /* CONFIG_SYSCTL */
991 
992 /*
993  * The purpose of always_dump_vma() is to make sure that special kernel mappings
994  * that are useful for post-mortem analysis are included in every core dump.
995  * In that way we ensure that the core dump is fully interpretable later
996  * without matching up the same kernel and hardware config to see what PC values
997  * meant. These special mappings include - vDSO, vsyscall, and other
998  * architecture specific mappings
999  */
1000 static bool always_dump_vma(struct vm_area_struct *vma)
1001 {
1002 	/* Any vsyscall mappings? */
1003 	if (vma == get_gate_vma(vma->vm_mm))
1004 		return true;
1005 
1006 	/*
1007 	 * Assume that all vmas with a .name op should always be dumped.
1008 	 * If this changes, a new vm_ops field can easily be added.
1009 	 */
1010 	if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1011 		return true;
1012 
1013 	/*
1014 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1015 	 * such as vDSO sections.
1016 	 */
1017 	if (arch_vma_name(vma))
1018 		return true;
1019 
1020 	return false;
1021 }
1022 
1023 #define DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER 1
1024 
1025 /*
1026  * Decide how much of @vma's contents should be included in a core dump.
1027  */
1028 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1029 				   unsigned long mm_flags)
1030 {
1031 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1032 
1033 	/* always dump the vdso and vsyscall sections */
1034 	if (always_dump_vma(vma))
1035 		goto whole;
1036 
1037 	if (vma->vm_flags & VM_DONTDUMP)
1038 		return 0;
1039 
1040 	/* support for DAX */
1041 	if (vma_is_dax(vma)) {
1042 		if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1043 			goto whole;
1044 		if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1045 			goto whole;
1046 		return 0;
1047 	}
1048 
1049 	/* Hugetlb memory check */
1050 	if (is_vm_hugetlb_page(vma)) {
1051 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1052 			goto whole;
1053 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1054 			goto whole;
1055 		return 0;
1056 	}
1057 
1058 	/* Do not dump I/O mapped devices or special mappings */
1059 	if (vma->vm_flags & VM_IO)
1060 		return 0;
1061 
1062 	/* By default, dump shared memory if mapped from an anonymous file. */
1063 	if (vma->vm_flags & VM_SHARED) {
1064 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1065 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1066 			goto whole;
1067 		return 0;
1068 	}
1069 
1070 	/* Dump segments that have been written to.  */
1071 	if ((!IS_ENABLED(CONFIG_MMU) || vma->anon_vma) && FILTER(ANON_PRIVATE))
1072 		goto whole;
1073 	if (vma->vm_file == NULL)
1074 		return 0;
1075 
1076 	if (FILTER(MAPPED_PRIVATE))
1077 		goto whole;
1078 
1079 	/*
1080 	 * If this is the beginning of an executable file mapping,
1081 	 * dump the first page to aid in determining what was mapped here.
1082 	 */
1083 	if (FILTER(ELF_HEADERS) &&
1084 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1085 		if ((READ_ONCE(file_inode(vma->vm_file)->i_mode) & 0111) != 0)
1086 			return PAGE_SIZE;
1087 
1088 		/*
1089 		 * ELF libraries aren't always executable.
1090 		 * We'll want to check whether the mapping starts with the ELF
1091 		 * magic, but not now - we're holding the mmap lock,
1092 		 * so copy_from_user() doesn't work here.
1093 		 * Use a placeholder instead, and fix it up later in
1094 		 * dump_vma_snapshot().
1095 		 */
1096 		return DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER;
1097 	}
1098 
1099 #undef	FILTER
1100 
1101 	return 0;
1102 
1103 whole:
1104 	return vma->vm_end - vma->vm_start;
1105 }
1106 
1107 /*
1108  * Helper function for iterating across a vma list.  It ensures that the caller
1109  * will visit `gate_vma' prior to terminating the search.
1110  */
1111 static struct vm_area_struct *coredump_next_vma(struct vma_iterator *vmi,
1112 				       struct vm_area_struct *vma,
1113 				       struct vm_area_struct *gate_vma)
1114 {
1115 	if (gate_vma && (vma == gate_vma))
1116 		return NULL;
1117 
1118 	vma = vma_next(vmi);
1119 	if (vma)
1120 		return vma;
1121 	return gate_vma;
1122 }
1123 
1124 static void free_vma_snapshot(struct coredump_params *cprm)
1125 {
1126 	if (cprm->vma_meta) {
1127 		int i;
1128 		for (i = 0; i < cprm->vma_count; i++) {
1129 			struct file *file = cprm->vma_meta[i].file;
1130 			if (file)
1131 				fput(file);
1132 		}
1133 		kvfree(cprm->vma_meta);
1134 		cprm->vma_meta = NULL;
1135 	}
1136 }
1137 
1138 /*
1139  * Under the mmap_lock, take a snapshot of relevant information about the task's
1140  * VMAs.
1141  */
1142 static bool dump_vma_snapshot(struct coredump_params *cprm)
1143 {
1144 	struct vm_area_struct *gate_vma, *vma = NULL;
1145 	struct mm_struct *mm = current->mm;
1146 	VMA_ITERATOR(vmi, mm, 0);
1147 	int i = 0;
1148 
1149 	/*
1150 	 * Once the stack expansion code is fixed to not change VMA bounds
1151 	 * under mmap_lock in read mode, this can be changed to take the
1152 	 * mmap_lock in read mode.
1153 	 */
1154 	if (mmap_write_lock_killable(mm))
1155 		return false;
1156 
1157 	cprm->vma_data_size = 0;
1158 	gate_vma = get_gate_vma(mm);
1159 	cprm->vma_count = mm->map_count + (gate_vma ? 1 : 0);
1160 
1161 	cprm->vma_meta = kvmalloc_array(cprm->vma_count, sizeof(*cprm->vma_meta), GFP_KERNEL);
1162 	if (!cprm->vma_meta) {
1163 		mmap_write_unlock(mm);
1164 		return false;
1165 	}
1166 
1167 	while ((vma = coredump_next_vma(&vmi, vma, gate_vma)) != NULL) {
1168 		struct core_vma_metadata *m = cprm->vma_meta + i;
1169 
1170 		m->start = vma->vm_start;
1171 		m->end = vma->vm_end;
1172 		m->flags = vma->vm_flags;
1173 		m->dump_size = vma_dump_size(vma, cprm->mm_flags);
1174 		m->pgoff = vma->vm_pgoff;
1175 		m->file = vma->vm_file;
1176 		if (m->file)
1177 			get_file(m->file);
1178 		i++;
1179 	}
1180 
1181 	mmap_write_unlock(mm);
1182 
1183 	for (i = 0; i < cprm->vma_count; i++) {
1184 		struct core_vma_metadata *m = cprm->vma_meta + i;
1185 
1186 		if (m->dump_size == DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER) {
1187 			char elfmag[SELFMAG];
1188 
1189 			if (copy_from_user(elfmag, (void __user *)m->start, SELFMAG) ||
1190 					memcmp(elfmag, ELFMAG, SELFMAG) != 0) {
1191 				m->dump_size = 0;
1192 			} else {
1193 				m->dump_size = PAGE_SIZE;
1194 			}
1195 		}
1196 
1197 		cprm->vma_data_size += m->dump_size;
1198 	}
1199 
1200 	return true;
1201 }
1202