xref: /linux/fs/coredump.c (revision 8f5b5f78113e881cb8570c961b0dc42b218a1b9e)
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 #define CORE_FILE_NOTE_SIZE_DEFAULT (4*1024*1024)
60 /* Define a reasonable max cap */
61 #define CORE_FILE_NOTE_SIZE_MAX (16*1024*1024)
62 
63 static int core_uses_pid;
64 static unsigned int core_pipe_limit;
65 static char core_pattern[CORENAME_MAX_SIZE] = "core";
66 static int core_name_size = CORENAME_MAX_SIZE;
67 unsigned int core_file_note_size_limit = CORE_FILE_NOTE_SIZE_DEFAULT;
68 
69 struct core_name {
70 	char *corename;
71 	int used, size;
72 };
73 
74 static int expand_corename(struct core_name *cn, int size)
75 {
76 	char *corename;
77 
78 	size = kmalloc_size_roundup(size);
79 	corename = krealloc(cn->corename, size, GFP_KERNEL);
80 
81 	if (!corename)
82 		return -ENOMEM;
83 
84 	if (size > core_name_size) /* racy but harmless */
85 		core_name_size = size;
86 
87 	cn->size = size;
88 	cn->corename = corename;
89 	return 0;
90 }
91 
92 static __printf(2, 0) int cn_vprintf(struct core_name *cn, const char *fmt,
93 				     va_list arg)
94 {
95 	int free, need;
96 	va_list arg_copy;
97 
98 again:
99 	free = cn->size - cn->used;
100 
101 	va_copy(arg_copy, arg);
102 	need = vsnprintf(cn->corename + cn->used, free, fmt, arg_copy);
103 	va_end(arg_copy);
104 
105 	if (need < free) {
106 		cn->used += need;
107 		return 0;
108 	}
109 
110 	if (!expand_corename(cn, cn->size + need - free + 1))
111 		goto again;
112 
113 	return -ENOMEM;
114 }
115 
116 static __printf(2, 3) int cn_printf(struct core_name *cn, const char *fmt, ...)
117 {
118 	va_list arg;
119 	int ret;
120 
121 	va_start(arg, fmt);
122 	ret = cn_vprintf(cn, fmt, arg);
123 	va_end(arg);
124 
125 	return ret;
126 }
127 
128 static __printf(2, 3)
129 int cn_esc_printf(struct core_name *cn, const char *fmt, ...)
130 {
131 	int cur = cn->used;
132 	va_list arg;
133 	int ret;
134 
135 	va_start(arg, fmt);
136 	ret = cn_vprintf(cn, fmt, arg);
137 	va_end(arg);
138 
139 	if (ret == 0) {
140 		/*
141 		 * Ensure that this coredump name component can't cause the
142 		 * resulting corefile path to consist of a ".." or ".".
143 		 */
144 		if ((cn->used - cur == 1 && cn->corename[cur] == '.') ||
145 				(cn->used - cur == 2 && cn->corename[cur] == '.'
146 				&& cn->corename[cur+1] == '.'))
147 			cn->corename[cur] = '!';
148 
149 		/*
150 		 * Empty names are fishy and could be used to create a "//" in a
151 		 * corefile name, causing the coredump to happen one directory
152 		 * level too high. Enforce that all components of the core
153 		 * pattern are at least one character long.
154 		 */
155 		if (cn->used == cur)
156 			ret = cn_printf(cn, "!");
157 	}
158 
159 	for (; cur < cn->used; ++cur) {
160 		if (cn->corename[cur] == '/')
161 			cn->corename[cur] = '!';
162 	}
163 	return ret;
164 }
165 
166 static int cn_print_exe_file(struct core_name *cn, bool name_only)
167 {
168 	struct file *exe_file;
169 	char *pathbuf, *path, *ptr;
170 	int ret;
171 
172 	exe_file = get_mm_exe_file(current->mm);
173 	if (!exe_file)
174 		return cn_esc_printf(cn, "%s (path unknown)", current->comm);
175 
176 	pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
177 	if (!pathbuf) {
178 		ret = -ENOMEM;
179 		goto put_exe_file;
180 	}
181 
182 	path = file_path(exe_file, pathbuf, PATH_MAX);
183 	if (IS_ERR(path)) {
184 		ret = PTR_ERR(path);
185 		goto free_buf;
186 	}
187 
188 	if (name_only) {
189 		ptr = strrchr(path, '/');
190 		if (ptr)
191 			path = ptr + 1;
192 	}
193 	ret = cn_esc_printf(cn, "%s", path);
194 
195 free_buf:
196 	kfree(pathbuf);
197 put_exe_file:
198 	fput(exe_file);
199 	return ret;
200 }
201 
202 /* format_corename will inspect the pattern parameter, and output a
203  * name into corename, which must have space for at least
204  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
205  */
206 static int format_corename(struct core_name *cn, struct coredump_params *cprm,
207 			   size_t **argv, int *argc)
208 {
209 	const struct cred *cred = current_cred();
210 	const char *pat_ptr = core_pattern;
211 	int ispipe = (*pat_ptr == '|');
212 	bool was_space = false;
213 	int pid_in_pattern = 0;
214 	int err = 0;
215 
216 	cn->used = 0;
217 	cn->corename = NULL;
218 	if (expand_corename(cn, core_name_size))
219 		return -ENOMEM;
220 	cn->corename[0] = '\0';
221 
222 	if (ispipe) {
223 		int argvs = sizeof(core_pattern) / 2;
224 		(*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL);
225 		if (!(*argv))
226 			return -ENOMEM;
227 		(*argv)[(*argc)++] = 0;
228 		++pat_ptr;
229 		if (!(*pat_ptr))
230 			return -ENOMEM;
231 	}
232 
233 	/* Repeat as long as we have more pattern to process and more output
234 	   space */
235 	while (*pat_ptr) {
236 		/*
237 		 * Split on spaces before doing template expansion so that
238 		 * %e and %E don't get split if they have spaces in them
239 		 */
240 		if (ispipe) {
241 			if (isspace(*pat_ptr)) {
242 				if (cn->used != 0)
243 					was_space = true;
244 				pat_ptr++;
245 				continue;
246 			} else if (was_space) {
247 				was_space = false;
248 				err = cn_printf(cn, "%c", '\0');
249 				if (err)
250 					return err;
251 				(*argv)[(*argc)++] = cn->used;
252 			}
253 		}
254 		if (*pat_ptr != '%') {
255 			err = cn_printf(cn, "%c", *pat_ptr++);
256 		} else {
257 			switch (*++pat_ptr) {
258 			/* single % at the end, drop that */
259 			case 0:
260 				goto out;
261 			/* Double percent, output one percent */
262 			case '%':
263 				err = cn_printf(cn, "%c", '%');
264 				break;
265 			/* pid */
266 			case 'p':
267 				pid_in_pattern = 1;
268 				err = cn_printf(cn, "%d",
269 					      task_tgid_vnr(current));
270 				break;
271 			/* global pid */
272 			case 'P':
273 				err = cn_printf(cn, "%d",
274 					      task_tgid_nr(current));
275 				break;
276 			case 'i':
277 				err = cn_printf(cn, "%d",
278 					      task_pid_vnr(current));
279 				break;
280 			case 'I':
281 				err = cn_printf(cn, "%d",
282 					      task_pid_nr(current));
283 				break;
284 			/* uid */
285 			case 'u':
286 				err = cn_printf(cn, "%u",
287 						from_kuid(&init_user_ns,
288 							  cred->uid));
289 				break;
290 			/* gid */
291 			case 'g':
292 				err = cn_printf(cn, "%u",
293 						from_kgid(&init_user_ns,
294 							  cred->gid));
295 				break;
296 			case 'd':
297 				err = cn_printf(cn, "%d",
298 					__get_dumpable(cprm->mm_flags));
299 				break;
300 			/* signal that caused the coredump */
301 			case 's':
302 				err = cn_printf(cn, "%d",
303 						cprm->siginfo->si_signo);
304 				break;
305 			/* UNIX time of coredump */
306 			case 't': {
307 				time64_t time;
308 
309 				time = ktime_get_real_seconds();
310 				err = cn_printf(cn, "%lld", time);
311 				break;
312 			}
313 			/* hostname */
314 			case 'h':
315 				down_read(&uts_sem);
316 				err = cn_esc_printf(cn, "%s",
317 					      utsname()->nodename);
318 				up_read(&uts_sem);
319 				break;
320 			/* executable, could be changed by prctl PR_SET_NAME etc */
321 			case 'e':
322 				err = cn_esc_printf(cn, "%s", current->comm);
323 				break;
324 			/* file name of executable */
325 			case 'f':
326 				err = cn_print_exe_file(cn, true);
327 				break;
328 			case 'E':
329 				err = cn_print_exe_file(cn, false);
330 				break;
331 			/* core limit size */
332 			case 'c':
333 				err = cn_printf(cn, "%lu",
334 					      rlimit(RLIMIT_CORE));
335 				break;
336 			/* CPU the task ran on */
337 			case 'C':
338 				err = cn_printf(cn, "%d", cprm->cpu);
339 				break;
340 			default:
341 				break;
342 			}
343 			++pat_ptr;
344 		}
345 
346 		if (err)
347 			return err;
348 	}
349 
350 out:
351 	/* Backward compatibility with core_uses_pid:
352 	 *
353 	 * If core_pattern does not include a %p (as is the default)
354 	 * and core_uses_pid is set, then .%pid will be appended to
355 	 * the filename. Do not do this for piped commands. */
356 	if (!ispipe && !pid_in_pattern && core_uses_pid) {
357 		err = cn_printf(cn, ".%d", task_tgid_vnr(current));
358 		if (err)
359 			return err;
360 	}
361 	return ispipe;
362 }
363 
364 static int zap_process(struct task_struct *start, int exit_code)
365 {
366 	struct task_struct *t;
367 	int nr = 0;
368 
369 	/* Allow SIGKILL, see prepare_signal() */
370 	start->signal->flags = SIGNAL_GROUP_EXIT;
371 	start->signal->group_exit_code = exit_code;
372 	start->signal->group_stop_count = 0;
373 
374 	for_each_thread(start, t) {
375 		task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
376 		if (t != current && !(t->flags & PF_POSTCOREDUMP)) {
377 			sigaddset(&t->pending.signal, SIGKILL);
378 			signal_wake_up(t, 1);
379 			/* The vhost_worker does not particpate in coredumps */
380 			if ((t->flags & (PF_USER_WORKER | PF_IO_WORKER)) != PF_USER_WORKER)
381 				nr++;
382 		}
383 	}
384 
385 	return nr;
386 }
387 
388 static int zap_threads(struct task_struct *tsk,
389 			struct core_state *core_state, int exit_code)
390 {
391 	struct signal_struct *signal = tsk->signal;
392 	int nr = -EAGAIN;
393 
394 	spin_lock_irq(&tsk->sighand->siglock);
395 	if (!(signal->flags & SIGNAL_GROUP_EXIT) && !signal->group_exec_task) {
396 		signal->core_state = core_state;
397 		nr = zap_process(tsk, exit_code);
398 		clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
399 		tsk->flags |= PF_DUMPCORE;
400 		atomic_set(&core_state->nr_threads, nr);
401 	}
402 	spin_unlock_irq(&tsk->sighand->siglock);
403 	return nr;
404 }
405 
406 static int coredump_wait(int exit_code, struct core_state *core_state)
407 {
408 	struct task_struct *tsk = current;
409 	int core_waiters = -EBUSY;
410 
411 	init_completion(&core_state->startup);
412 	core_state->dumper.task = tsk;
413 	core_state->dumper.next = NULL;
414 
415 	core_waiters = zap_threads(tsk, core_state, exit_code);
416 	if (core_waiters > 0) {
417 		struct core_thread *ptr;
418 
419 		wait_for_completion_state(&core_state->startup,
420 					  TASK_UNINTERRUPTIBLE|TASK_FREEZABLE);
421 		/*
422 		 * Wait for all the threads to become inactive, so that
423 		 * all the thread context (extended register state, like
424 		 * fpu etc) gets copied to the memory.
425 		 */
426 		ptr = core_state->dumper.next;
427 		while (ptr != NULL) {
428 			wait_task_inactive(ptr->task, TASK_ANY);
429 			ptr = ptr->next;
430 		}
431 	}
432 
433 	return core_waiters;
434 }
435 
436 static void coredump_finish(bool core_dumped)
437 {
438 	struct core_thread *curr, *next;
439 	struct task_struct *task;
440 
441 	spin_lock_irq(&current->sighand->siglock);
442 	if (core_dumped && !__fatal_signal_pending(current))
443 		current->signal->group_exit_code |= 0x80;
444 	next = current->signal->core_state->dumper.next;
445 	current->signal->core_state = NULL;
446 	spin_unlock_irq(&current->sighand->siglock);
447 
448 	while ((curr = next) != NULL) {
449 		next = curr->next;
450 		task = curr->task;
451 		/*
452 		 * see coredump_task_exit(), curr->task must not see
453 		 * ->task == NULL before we read ->next.
454 		 */
455 		smp_mb();
456 		curr->task = NULL;
457 		wake_up_process(task);
458 	}
459 }
460 
461 static bool dump_interrupted(void)
462 {
463 	/*
464 	 * SIGKILL or freezing() interrupt the coredumping. Perhaps we
465 	 * can do try_to_freeze() and check __fatal_signal_pending(),
466 	 * but then we need to teach dump_write() to restart and clear
467 	 * TIF_SIGPENDING.
468 	 */
469 	return fatal_signal_pending(current) || freezing(current);
470 }
471 
472 static void wait_for_dump_helpers(struct file *file)
473 {
474 	struct pipe_inode_info *pipe = file->private_data;
475 
476 	pipe_lock(pipe);
477 	pipe->readers++;
478 	pipe->writers--;
479 	wake_up_interruptible_sync(&pipe->rd_wait);
480 	kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
481 	pipe_unlock(pipe);
482 
483 	/*
484 	 * We actually want wait_event_freezable() but then we need
485 	 * to clear TIF_SIGPENDING and improve dump_interrupted().
486 	 */
487 	wait_event_interruptible(pipe->rd_wait, pipe->readers == 1);
488 
489 	pipe_lock(pipe);
490 	pipe->readers--;
491 	pipe->writers++;
492 	pipe_unlock(pipe);
493 }
494 
495 /*
496  * umh_pipe_setup
497  * helper function to customize the process used
498  * to collect the core in userspace.  Specifically
499  * it sets up a pipe and installs it as fd 0 (stdin)
500  * for the process.  Returns 0 on success, or
501  * PTR_ERR on failure.
502  * Note that it also sets the core limit to 1.  This
503  * is a special value that we use to trap recursive
504  * core dumps
505  */
506 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
507 {
508 	struct file *files[2];
509 	struct coredump_params *cp = (struct coredump_params *)info->data;
510 	int err = create_pipe_files(files, 0);
511 	if (err)
512 		return err;
513 
514 	cp->file = files[1];
515 
516 	err = replace_fd(0, files[0], 0);
517 	fput(files[0]);
518 	/* and disallow core files too */
519 	current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
520 
521 	return err;
522 }
523 
524 void do_coredump(const kernel_siginfo_t *siginfo)
525 {
526 	struct core_state core_state;
527 	struct core_name cn;
528 	struct mm_struct *mm = current->mm;
529 	struct linux_binfmt * binfmt;
530 	const struct cred *old_cred;
531 	struct cred *cred;
532 	int retval = 0;
533 	int ispipe;
534 	size_t *argv = NULL;
535 	int argc = 0;
536 	/* require nonrelative corefile path and be extra careful */
537 	bool need_suid_safe = false;
538 	bool core_dumped = false;
539 	static atomic_t core_dump_count = ATOMIC_INIT(0);
540 	struct coredump_params cprm = {
541 		.siginfo = siginfo,
542 		.limit = rlimit(RLIMIT_CORE),
543 		/*
544 		 * We must use the same mm->flags while dumping core to avoid
545 		 * inconsistency of bit flags, since this flag is not protected
546 		 * by any locks.
547 		 */
548 		.mm_flags = mm->flags,
549 		.vma_meta = NULL,
550 		.cpu = raw_smp_processor_id(),
551 	};
552 
553 	audit_core_dumps(siginfo->si_signo);
554 
555 	binfmt = mm->binfmt;
556 	if (!binfmt || !binfmt->core_dump)
557 		goto fail;
558 	if (!__get_dumpable(cprm.mm_flags))
559 		goto fail;
560 
561 	cred = prepare_creds();
562 	if (!cred)
563 		goto fail;
564 	/*
565 	 * We cannot trust fsuid as being the "true" uid of the process
566 	 * nor do we know its entire history. We only know it was tainted
567 	 * so we dump it as root in mode 2, and only into a controlled
568 	 * environment (pipe handler or fully qualified path).
569 	 */
570 	if (__get_dumpable(cprm.mm_flags) == SUID_DUMP_ROOT) {
571 		/* Setuid core dump mode */
572 		cred->fsuid = GLOBAL_ROOT_UID;	/* Dump root private */
573 		need_suid_safe = true;
574 	}
575 
576 	retval = coredump_wait(siginfo->si_signo, &core_state);
577 	if (retval < 0)
578 		goto fail_creds;
579 
580 	old_cred = override_creds(cred);
581 
582 	ispipe = format_corename(&cn, &cprm, &argv, &argc);
583 
584 	if (ispipe) {
585 		int argi;
586 		int dump_count;
587 		char **helper_argv;
588 		struct subprocess_info *sub_info;
589 
590 		if (ispipe < 0) {
591 			printk(KERN_WARNING "format_corename failed\n");
592 			printk(KERN_WARNING "Aborting core\n");
593 			goto fail_unlock;
594 		}
595 
596 		if (cprm.limit == 1) {
597 			/* See umh_pipe_setup() which sets RLIMIT_CORE = 1.
598 			 *
599 			 * Normally core limits are irrelevant to pipes, since
600 			 * we're not writing to the file system, but we use
601 			 * cprm.limit of 1 here as a special value, this is a
602 			 * consistent way to catch recursive crashes.
603 			 * We can still crash if the core_pattern binary sets
604 			 * RLIM_CORE = !1, but it runs as root, and can do
605 			 * lots of stupid things.
606 			 *
607 			 * Note that we use task_tgid_vnr here to grab the pid
608 			 * of the process group leader.  That way we get the
609 			 * right pid if a thread in a multi-threaded
610 			 * core_pattern process dies.
611 			 */
612 			printk(KERN_WARNING
613 				"Process %d(%s) has RLIMIT_CORE set to 1\n",
614 				task_tgid_vnr(current), current->comm);
615 			printk(KERN_WARNING "Aborting core\n");
616 			goto fail_unlock;
617 		}
618 		cprm.limit = RLIM_INFINITY;
619 
620 		dump_count = atomic_inc_return(&core_dump_count);
621 		if (core_pipe_limit && (core_pipe_limit < dump_count)) {
622 			printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
623 			       task_tgid_vnr(current), current->comm);
624 			printk(KERN_WARNING "Skipping core dump\n");
625 			goto fail_dropcount;
626 		}
627 
628 		helper_argv = kmalloc_array(argc + 1, sizeof(*helper_argv),
629 					    GFP_KERNEL);
630 		if (!helper_argv) {
631 			printk(KERN_WARNING "%s failed to allocate memory\n",
632 			       __func__);
633 			goto fail_dropcount;
634 		}
635 		for (argi = 0; argi < argc; argi++)
636 			helper_argv[argi] = cn.corename + argv[argi];
637 		helper_argv[argi] = NULL;
638 
639 		retval = -ENOMEM;
640 		sub_info = call_usermodehelper_setup(helper_argv[0],
641 						helper_argv, NULL, GFP_KERNEL,
642 						umh_pipe_setup, NULL, &cprm);
643 		if (sub_info)
644 			retval = call_usermodehelper_exec(sub_info,
645 							  UMH_WAIT_EXEC);
646 
647 		kfree(helper_argv);
648 		if (retval) {
649 			printk(KERN_INFO "Core dump to |%s pipe failed\n",
650 			       cn.corename);
651 			goto close_fail;
652 		}
653 	} else {
654 		struct mnt_idmap *idmap;
655 		struct inode *inode;
656 		int open_flags = O_CREAT | O_WRONLY | O_NOFOLLOW |
657 				 O_LARGEFILE | O_EXCL;
658 
659 		if (cprm.limit < binfmt->min_coredump)
660 			goto fail_unlock;
661 
662 		if (need_suid_safe && cn.corename[0] != '/') {
663 			printk(KERN_WARNING "Pid %d(%s) can only dump core "\
664 				"to fully qualified path!\n",
665 				task_tgid_vnr(current), current->comm);
666 			printk(KERN_WARNING "Skipping core dump\n");
667 			goto fail_unlock;
668 		}
669 
670 		/*
671 		 * Unlink the file if it exists unless this is a SUID
672 		 * binary - in that case, we're running around with root
673 		 * privs and don't want to unlink another user's coredump.
674 		 */
675 		if (!need_suid_safe) {
676 			/*
677 			 * If it doesn't exist, that's fine. If there's some
678 			 * other problem, we'll catch it at the filp_open().
679 			 */
680 			do_unlinkat(AT_FDCWD, getname_kernel(cn.corename));
681 		}
682 
683 		/*
684 		 * There is a race between unlinking and creating the
685 		 * file, but if that causes an EEXIST here, that's
686 		 * fine - another process raced with us while creating
687 		 * the corefile, and the other process won. To userspace,
688 		 * what matters is that at least one of the two processes
689 		 * writes its coredump successfully, not which one.
690 		 */
691 		if (need_suid_safe) {
692 			/*
693 			 * Using user namespaces, normal user tasks can change
694 			 * their current->fs->root to point to arbitrary
695 			 * directories. Since the intention of the "only dump
696 			 * with a fully qualified path" rule is to control where
697 			 * coredumps may be placed using root privileges,
698 			 * current->fs->root must not be used. Instead, use the
699 			 * root directory of init_task.
700 			 */
701 			struct path root;
702 
703 			task_lock(&init_task);
704 			get_fs_root(init_task.fs, &root);
705 			task_unlock(&init_task);
706 			cprm.file = file_open_root(&root, cn.corename,
707 						   open_flags, 0600);
708 			path_put(&root);
709 		} else {
710 			cprm.file = filp_open(cn.corename, open_flags, 0600);
711 		}
712 		if (IS_ERR(cprm.file))
713 			goto fail_unlock;
714 
715 		inode = file_inode(cprm.file);
716 		if (inode->i_nlink > 1)
717 			goto close_fail;
718 		if (d_unhashed(cprm.file->f_path.dentry))
719 			goto close_fail;
720 		/*
721 		 * AK: actually i see no reason to not allow this for named
722 		 * pipes etc, but keep the previous behaviour for now.
723 		 */
724 		if (!S_ISREG(inode->i_mode))
725 			goto close_fail;
726 		/*
727 		 * Don't dump core if the filesystem changed owner or mode
728 		 * of the file during file creation. This is an issue when
729 		 * a process dumps core while its cwd is e.g. on a vfat
730 		 * filesystem.
731 		 */
732 		idmap = file_mnt_idmap(cprm.file);
733 		if (!vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode),
734 				    current_fsuid())) {
735 			pr_info_ratelimited("Core dump to %s aborted: cannot preserve file owner\n",
736 					    cn.corename);
737 			goto close_fail;
738 		}
739 		if ((inode->i_mode & 0677) != 0600) {
740 			pr_info_ratelimited("Core dump to %s aborted: cannot preserve file permissions\n",
741 					    cn.corename);
742 			goto close_fail;
743 		}
744 		if (!(cprm.file->f_mode & FMODE_CAN_WRITE))
745 			goto close_fail;
746 		if (do_truncate(idmap, cprm.file->f_path.dentry,
747 				0, 0, cprm.file))
748 			goto close_fail;
749 	}
750 
751 	/* get us an unshared descriptor table; almost always a no-op */
752 	/* The cell spufs coredump code reads the file descriptor tables */
753 	retval = unshare_files();
754 	if (retval)
755 		goto close_fail;
756 	if (!dump_interrupted()) {
757 		/*
758 		 * umh disabled with CONFIG_STATIC_USERMODEHELPER_PATH="" would
759 		 * have this set to NULL.
760 		 */
761 		if (!cprm.file) {
762 			pr_info("Core dump to |%s disabled\n", cn.corename);
763 			goto close_fail;
764 		}
765 		if (!dump_vma_snapshot(&cprm))
766 			goto close_fail;
767 
768 		file_start_write(cprm.file);
769 		core_dumped = binfmt->core_dump(&cprm);
770 		/*
771 		 * Ensures that file size is big enough to contain the current
772 		 * file postion. This prevents gdb from complaining about
773 		 * a truncated file if the last "write" to the file was
774 		 * dump_skip.
775 		 */
776 		if (cprm.to_skip) {
777 			cprm.to_skip--;
778 			dump_emit(&cprm, "", 1);
779 		}
780 		file_end_write(cprm.file);
781 		free_vma_snapshot(&cprm);
782 	}
783 	if (ispipe && core_pipe_limit)
784 		wait_for_dump_helpers(cprm.file);
785 close_fail:
786 	if (cprm.file)
787 		filp_close(cprm.file, NULL);
788 fail_dropcount:
789 	if (ispipe)
790 		atomic_dec(&core_dump_count);
791 fail_unlock:
792 	kfree(argv);
793 	kfree(cn.corename);
794 	coredump_finish(core_dumped);
795 	revert_creds(old_cred);
796 fail_creds:
797 	put_cred(cred);
798 fail:
799 	return;
800 }
801 
802 /*
803  * Core dumping helper functions.  These are the only things you should
804  * do on a core-file: use only these functions to write out all the
805  * necessary info.
806  */
807 static int __dump_emit(struct coredump_params *cprm, const void *addr, int nr)
808 {
809 	struct file *file = cprm->file;
810 	loff_t pos = file->f_pos;
811 	ssize_t n;
812 	if (cprm->written + nr > cprm->limit)
813 		return 0;
814 
815 
816 	if (dump_interrupted())
817 		return 0;
818 	n = __kernel_write(file, addr, nr, &pos);
819 	if (n != nr)
820 		return 0;
821 	file->f_pos = pos;
822 	cprm->written += n;
823 	cprm->pos += n;
824 
825 	return 1;
826 }
827 
828 static int __dump_skip(struct coredump_params *cprm, size_t nr)
829 {
830 	static char zeroes[PAGE_SIZE];
831 	struct file *file = cprm->file;
832 	if (file->f_mode & FMODE_LSEEK) {
833 		if (dump_interrupted() ||
834 		    vfs_llseek(file, nr, SEEK_CUR) < 0)
835 			return 0;
836 		cprm->pos += nr;
837 		return 1;
838 	} else {
839 		while (nr > PAGE_SIZE) {
840 			if (!__dump_emit(cprm, zeroes, PAGE_SIZE))
841 				return 0;
842 			nr -= PAGE_SIZE;
843 		}
844 		return __dump_emit(cprm, zeroes, nr);
845 	}
846 }
847 
848 int dump_emit(struct coredump_params *cprm, const void *addr, int nr)
849 {
850 	if (cprm->to_skip) {
851 		if (!__dump_skip(cprm, cprm->to_skip))
852 			return 0;
853 		cprm->to_skip = 0;
854 	}
855 	return __dump_emit(cprm, addr, nr);
856 }
857 EXPORT_SYMBOL(dump_emit);
858 
859 void dump_skip_to(struct coredump_params *cprm, unsigned long pos)
860 {
861 	cprm->to_skip = pos - cprm->pos;
862 }
863 EXPORT_SYMBOL(dump_skip_to);
864 
865 void dump_skip(struct coredump_params *cprm, size_t nr)
866 {
867 	cprm->to_skip += nr;
868 }
869 EXPORT_SYMBOL(dump_skip);
870 
871 #ifdef CONFIG_ELF_CORE
872 static int dump_emit_page(struct coredump_params *cprm, struct page *page)
873 {
874 	struct bio_vec bvec;
875 	struct iov_iter iter;
876 	struct file *file = cprm->file;
877 	loff_t pos;
878 	ssize_t n;
879 
880 	if (!page)
881 		return 0;
882 
883 	if (cprm->to_skip) {
884 		if (!__dump_skip(cprm, cprm->to_skip))
885 			return 0;
886 		cprm->to_skip = 0;
887 	}
888 	if (cprm->written + PAGE_SIZE > cprm->limit)
889 		return 0;
890 	if (dump_interrupted())
891 		return 0;
892 	pos = file->f_pos;
893 	bvec_set_page(&bvec, page, PAGE_SIZE, 0);
894 	iov_iter_bvec(&iter, ITER_SOURCE, &bvec, 1, PAGE_SIZE);
895 	n = __kernel_write_iter(cprm->file, &iter, &pos);
896 	if (n != PAGE_SIZE)
897 		return 0;
898 	file->f_pos = pos;
899 	cprm->written += PAGE_SIZE;
900 	cprm->pos += PAGE_SIZE;
901 
902 	return 1;
903 }
904 
905 /*
906  * If we might get machine checks from kernel accesses during the
907  * core dump, let's get those errors early rather than during the
908  * IO. This is not performance-critical enough to warrant having
909  * all the machine check logic in the iovec paths.
910  */
911 #ifdef copy_mc_to_kernel
912 
913 #define dump_page_alloc() alloc_page(GFP_KERNEL)
914 #define dump_page_free(x) __free_page(x)
915 static struct page *dump_page_copy(struct page *src, struct page *dst)
916 {
917 	void *buf = kmap_local_page(src);
918 	size_t left = copy_mc_to_kernel(page_address(dst), buf, PAGE_SIZE);
919 	kunmap_local(buf);
920 	return left ? NULL : dst;
921 }
922 
923 #else
924 
925 /* We just want to return non-NULL; it's never used. */
926 #define dump_page_alloc() ERR_PTR(-EINVAL)
927 #define dump_page_free(x) ((void)(x))
928 static inline struct page *dump_page_copy(struct page *src, struct page *dst)
929 {
930 	return src;
931 }
932 #endif
933 
934 int dump_user_range(struct coredump_params *cprm, unsigned long start,
935 		    unsigned long len)
936 {
937 	unsigned long addr;
938 	struct page *dump_page;
939 
940 	dump_page = dump_page_alloc();
941 	if (!dump_page)
942 		return 0;
943 
944 	for (addr = start; addr < start + len; addr += PAGE_SIZE) {
945 		struct page *page;
946 
947 		/*
948 		 * To avoid having to allocate page tables for virtual address
949 		 * ranges that have never been used yet, and also to make it
950 		 * easy to generate sparse core files, use a helper that returns
951 		 * NULL when encountering an empty page table entry that would
952 		 * otherwise have been filled with the zero page.
953 		 */
954 		page = get_dump_page(addr);
955 		if (page) {
956 			int stop = !dump_emit_page(cprm, dump_page_copy(page, dump_page));
957 			put_page(page);
958 			if (stop) {
959 				dump_page_free(dump_page);
960 				return 0;
961 			}
962 		} else {
963 			dump_skip(cprm, PAGE_SIZE);
964 		}
965 	}
966 	dump_page_free(dump_page);
967 	return 1;
968 }
969 #endif
970 
971 int dump_align(struct coredump_params *cprm, int align)
972 {
973 	unsigned mod = (cprm->pos + cprm->to_skip) & (align - 1);
974 	if (align & (align - 1))
975 		return 0;
976 	if (mod)
977 		cprm->to_skip += align - mod;
978 	return 1;
979 }
980 EXPORT_SYMBOL(dump_align);
981 
982 #ifdef CONFIG_SYSCTL
983 
984 void validate_coredump_safety(void)
985 {
986 	if (suid_dumpable == SUID_DUMP_ROOT &&
987 	    core_pattern[0] != '/' && core_pattern[0] != '|') {
988 		pr_warn(
989 "Unsafe core_pattern used with fs.suid_dumpable=2.\n"
990 "Pipe handler or fully qualified core dump path required.\n"
991 "Set kernel.core_pattern before fs.suid_dumpable.\n"
992 		);
993 	}
994 }
995 
996 static int proc_dostring_coredump(struct ctl_table *table, int write,
997 		  void *buffer, size_t *lenp, loff_t *ppos)
998 {
999 	int error = proc_dostring(table, write, buffer, lenp, ppos);
1000 
1001 	if (!error)
1002 		validate_coredump_safety();
1003 	return error;
1004 }
1005 
1006 static const unsigned int core_file_note_size_min = CORE_FILE_NOTE_SIZE_DEFAULT;
1007 static const unsigned int core_file_note_size_max = CORE_FILE_NOTE_SIZE_MAX;
1008 
1009 static struct ctl_table coredump_sysctls[] = {
1010 	{
1011 		.procname	= "core_uses_pid",
1012 		.data		= &core_uses_pid,
1013 		.maxlen		= sizeof(int),
1014 		.mode		= 0644,
1015 		.proc_handler	= proc_dointvec,
1016 	},
1017 	{
1018 		.procname	= "core_pattern",
1019 		.data		= core_pattern,
1020 		.maxlen		= CORENAME_MAX_SIZE,
1021 		.mode		= 0644,
1022 		.proc_handler	= proc_dostring_coredump,
1023 	},
1024 	{
1025 		.procname	= "core_pipe_limit",
1026 		.data		= &core_pipe_limit,
1027 		.maxlen		= sizeof(unsigned int),
1028 		.mode		= 0644,
1029 		.proc_handler	= proc_dointvec,
1030 	},
1031 	{
1032 		.procname       = "core_file_note_size_limit",
1033 		.data           = &core_file_note_size_limit,
1034 		.maxlen         = sizeof(unsigned int),
1035 		.mode           = 0644,
1036 		.proc_handler	= proc_douintvec_minmax,
1037 		.extra1		= (unsigned int *)&core_file_note_size_min,
1038 		.extra2		= (unsigned int *)&core_file_note_size_max,
1039 	},
1040 };
1041 
1042 static int __init init_fs_coredump_sysctls(void)
1043 {
1044 	register_sysctl_init("kernel", coredump_sysctls);
1045 	return 0;
1046 }
1047 fs_initcall(init_fs_coredump_sysctls);
1048 #endif /* CONFIG_SYSCTL */
1049 
1050 /*
1051  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1052  * that are useful for post-mortem analysis are included in every core dump.
1053  * In that way we ensure that the core dump is fully interpretable later
1054  * without matching up the same kernel and hardware config to see what PC values
1055  * meant. These special mappings include - vDSO, vsyscall, and other
1056  * architecture specific mappings
1057  */
1058 static bool always_dump_vma(struct vm_area_struct *vma)
1059 {
1060 	/* Any vsyscall mappings? */
1061 	if (vma == get_gate_vma(vma->vm_mm))
1062 		return true;
1063 
1064 	/*
1065 	 * Assume that all vmas with a .name op should always be dumped.
1066 	 * If this changes, a new vm_ops field can easily be added.
1067 	 */
1068 	if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1069 		return true;
1070 
1071 	/*
1072 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1073 	 * such as vDSO sections.
1074 	 */
1075 	if (arch_vma_name(vma))
1076 		return true;
1077 
1078 	return false;
1079 }
1080 
1081 #define DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER 1
1082 
1083 /*
1084  * Decide how much of @vma's contents should be included in a core dump.
1085  */
1086 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1087 				   unsigned long mm_flags)
1088 {
1089 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1090 
1091 	/* always dump the vdso and vsyscall sections */
1092 	if (always_dump_vma(vma))
1093 		goto whole;
1094 
1095 	if (vma->vm_flags & VM_DONTDUMP)
1096 		return 0;
1097 
1098 	/* support for DAX */
1099 	if (vma_is_dax(vma)) {
1100 		if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1101 			goto whole;
1102 		if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1103 			goto whole;
1104 		return 0;
1105 	}
1106 
1107 	/* Hugetlb memory check */
1108 	if (is_vm_hugetlb_page(vma)) {
1109 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1110 			goto whole;
1111 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1112 			goto whole;
1113 		return 0;
1114 	}
1115 
1116 	/* Do not dump I/O mapped devices or special mappings */
1117 	if (vma->vm_flags & VM_IO)
1118 		return 0;
1119 
1120 	/* By default, dump shared memory if mapped from an anonymous file. */
1121 	if (vma->vm_flags & VM_SHARED) {
1122 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1123 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1124 			goto whole;
1125 		return 0;
1126 	}
1127 
1128 	/* Dump segments that have been written to.  */
1129 	if ((!IS_ENABLED(CONFIG_MMU) || vma->anon_vma) && FILTER(ANON_PRIVATE))
1130 		goto whole;
1131 	if (vma->vm_file == NULL)
1132 		return 0;
1133 
1134 	if (FILTER(MAPPED_PRIVATE))
1135 		goto whole;
1136 
1137 	/*
1138 	 * If this is the beginning of an executable file mapping,
1139 	 * dump the first page to aid in determining what was mapped here.
1140 	 */
1141 	if (FILTER(ELF_HEADERS) &&
1142 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1143 		if ((READ_ONCE(file_inode(vma->vm_file)->i_mode) & 0111) != 0)
1144 			return PAGE_SIZE;
1145 
1146 		/*
1147 		 * ELF libraries aren't always executable.
1148 		 * We'll want to check whether the mapping starts with the ELF
1149 		 * magic, but not now - we're holding the mmap lock,
1150 		 * so copy_from_user() doesn't work here.
1151 		 * Use a placeholder instead, and fix it up later in
1152 		 * dump_vma_snapshot().
1153 		 */
1154 		return DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER;
1155 	}
1156 
1157 #undef	FILTER
1158 
1159 	return 0;
1160 
1161 whole:
1162 	return vma->vm_end - vma->vm_start;
1163 }
1164 
1165 /*
1166  * Helper function for iterating across a vma list.  It ensures that the caller
1167  * will visit `gate_vma' prior to terminating the search.
1168  */
1169 static struct vm_area_struct *coredump_next_vma(struct vma_iterator *vmi,
1170 				       struct vm_area_struct *vma,
1171 				       struct vm_area_struct *gate_vma)
1172 {
1173 	if (gate_vma && (vma == gate_vma))
1174 		return NULL;
1175 
1176 	vma = vma_next(vmi);
1177 	if (vma)
1178 		return vma;
1179 	return gate_vma;
1180 }
1181 
1182 static void free_vma_snapshot(struct coredump_params *cprm)
1183 {
1184 	if (cprm->vma_meta) {
1185 		int i;
1186 		for (i = 0; i < cprm->vma_count; i++) {
1187 			struct file *file = cprm->vma_meta[i].file;
1188 			if (file)
1189 				fput(file);
1190 		}
1191 		kvfree(cprm->vma_meta);
1192 		cprm->vma_meta = NULL;
1193 	}
1194 }
1195 
1196 /*
1197  * Under the mmap_lock, take a snapshot of relevant information about the task's
1198  * VMAs.
1199  */
1200 static bool dump_vma_snapshot(struct coredump_params *cprm)
1201 {
1202 	struct vm_area_struct *gate_vma, *vma = NULL;
1203 	struct mm_struct *mm = current->mm;
1204 	VMA_ITERATOR(vmi, mm, 0);
1205 	int i = 0;
1206 
1207 	/*
1208 	 * Once the stack expansion code is fixed to not change VMA bounds
1209 	 * under mmap_lock in read mode, this can be changed to take the
1210 	 * mmap_lock in read mode.
1211 	 */
1212 	if (mmap_write_lock_killable(mm))
1213 		return false;
1214 
1215 	cprm->vma_data_size = 0;
1216 	gate_vma = get_gate_vma(mm);
1217 	cprm->vma_count = mm->map_count + (gate_vma ? 1 : 0);
1218 
1219 	cprm->vma_meta = kvmalloc_array(cprm->vma_count, sizeof(*cprm->vma_meta), GFP_KERNEL);
1220 	if (!cprm->vma_meta) {
1221 		mmap_write_unlock(mm);
1222 		return false;
1223 	}
1224 
1225 	while ((vma = coredump_next_vma(&vmi, vma, gate_vma)) != NULL) {
1226 		struct core_vma_metadata *m = cprm->vma_meta + i;
1227 
1228 		m->start = vma->vm_start;
1229 		m->end = vma->vm_end;
1230 		m->flags = vma->vm_flags;
1231 		m->dump_size = vma_dump_size(vma, cprm->mm_flags);
1232 		m->pgoff = vma->vm_pgoff;
1233 		m->file = vma->vm_file;
1234 		if (m->file)
1235 			get_file(m->file);
1236 		i++;
1237 	}
1238 
1239 	mmap_write_unlock(mm);
1240 
1241 	for (i = 0; i < cprm->vma_count; i++) {
1242 		struct core_vma_metadata *m = cprm->vma_meta + i;
1243 
1244 		if (m->dump_size == DUMP_SIZE_MAYBE_ELFHDR_PLACEHOLDER) {
1245 			char elfmag[SELFMAG];
1246 
1247 			if (copy_from_user(elfmag, (void __user *)m->start, SELFMAG) ||
1248 					memcmp(elfmag, ELFMAG, SELFMAG) != 0) {
1249 				m->dump_size = 0;
1250 			} else {
1251 				m->dump_size = PAGE_SIZE;
1252 			}
1253 		}
1254 
1255 		cprm->vma_data_size += m->dump_size;
1256 	}
1257 
1258 	return true;
1259 }
1260