xref: /linux/net/unix/garbage.c (revision 870b7fdc660b38c4e1bd8bf48e62aa352ddf8f42)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * NET3:	Garbage Collector For AF_UNIX sockets
4  *
5  * Garbage Collector:
6  *	Copyright (C) Barak A. Pearlmutter.
7  *
8  * Chopped about by Alan Cox 22/3/96 to make it fit the AF_UNIX socket problem.
9  * If it doesn't work blame me, it worked when Barak sent it.
10  *
11  * Assumptions:
12  *
13  *  - object w/ a bit
14  *  - free list
15  *
16  * Current optimizations:
17  *
18  *  - explicit stack instead of recursion
19  *  - tail recurse on first born instead of immediate push/pop
20  *  - we gather the stuff that should not be killed into tree
21  *    and stack is just a path from root to the current pointer.
22  *
23  *  Future optimizations:
24  *
25  *  - don't just push entire root set; process in place
26  *
27  *  Fixes:
28  *	Alan Cox	07 Sept	1997	Vmalloc internal stack as needed.
29  *					Cope with changing max_files.
30  *	Al Viro		11 Oct 1998
31  *		Graph may have cycles. That is, we can send the descriptor
32  *		of foo to bar and vice versa. Current code chokes on that.
33  *		Fix: move SCM_RIGHTS ones into the separate list and then
34  *		skb_free() them all instead of doing explicit fput's.
35  *		Another problem: since fput() may block somebody may
36  *		create a new unix_socket when we are in the middle of sweep
37  *		phase. Fix: revert the logic wrt MARKED. Mark everything
38  *		upon the beginning and unmark non-junk ones.
39  *
40  *		[12 Oct 1998] AAARGH! New code purges all SCM_RIGHTS
41  *		sent to connect()'ed but still not accept()'ed sockets.
42  *		Fixed. Old code had slightly different problem here:
43  *		extra fput() in situation when we passed the descriptor via
44  *		such socket and closed it (descriptor). That would happen on
45  *		each unix_gc() until the accept(). Since the struct file in
46  *		question would go to the free list and might be reused...
47  *		That might be the reason of random oopses on filp_close()
48  *		in unrelated processes.
49  *
50  *	AV		28 Feb 1999
51  *		Kill the explicit allocation of stack. Now we keep the tree
52  *		with root in dummy + pointer (gc_current) to one of the nodes.
53  *		Stack is represented as path from gc_current to dummy. Unmark
54  *		now means "add to tree". Push == "make it a son of gc_current".
55  *		Pop == "move gc_current to parent". We keep only pointers to
56  *		parents (->gc_tree).
57  *	AV		1 Mar 1999
58  *		Damn. Added missing check for ->dead in listen queues scanning.
59  *
60  *	Miklos Szeredi 25 Jun 2007
61  *		Reimplement with a cycle collecting algorithm. This should
62  *		solve several problems with the previous code, like being racy
63  *		wrt receive and holding up unrelated socket operations.
64  */
65 
66 #include <linux/kernel.h>
67 #include <linux/string.h>
68 #include <linux/socket.h>
69 #include <linux/un.h>
70 #include <linux/net.h>
71 #include <linux/fs.h>
72 #include <linux/skbuff.h>
73 #include <linux/netdevice.h>
74 #include <linux/file.h>
75 #include <linux/proc_fs.h>
76 #include <linux/mutex.h>
77 #include <linux/wait.h>
78 
79 #include <net/sock.h>
80 #include <net/af_unix.h>
81 #include <net/scm.h>
82 #include <net/tcp_states.h>
83 
84 struct unix_sock *unix_get_socket(struct file *filp)
85 {
86 	struct inode *inode = file_inode(filp);
87 
88 	/* Socket ? */
89 	if (S_ISSOCK(inode->i_mode) && !(filp->f_mode & FMODE_PATH)) {
90 		struct socket *sock = SOCKET_I(inode);
91 		const struct proto_ops *ops;
92 		struct sock *sk = sock->sk;
93 
94 		ops = READ_ONCE(sock->ops);
95 
96 		/* PF_UNIX ? */
97 		if (sk && ops && ops->family == PF_UNIX)
98 			return unix_sk(sk);
99 	}
100 
101 	return NULL;
102 }
103 
104 static struct unix_vertex *unix_edge_successor(struct unix_edge *edge)
105 {
106 	/* If an embryo socket has a fd,
107 	 * the listener indirectly holds the fd's refcnt.
108 	 */
109 	if (edge->successor->listener)
110 		return unix_sk(edge->successor->listener)->vertex;
111 
112 	return edge->successor->vertex;
113 }
114 
115 static bool unix_graph_maybe_cyclic;
116 static bool unix_graph_grouped;
117 
118 static void unix_update_graph(struct unix_vertex *vertex)
119 {
120 	/* If the receiver socket is not inflight, no cyclic
121 	 * reference could be formed.
122 	 */
123 	if (!vertex)
124 		return;
125 
126 	unix_graph_maybe_cyclic = true;
127 	unix_graph_grouped = false;
128 }
129 
130 static LIST_HEAD(unix_unvisited_vertices);
131 
132 enum unix_vertex_index {
133 	UNIX_VERTEX_INDEX_MARK1,
134 	UNIX_VERTEX_INDEX_MARK2,
135 	UNIX_VERTEX_INDEX_START,
136 };
137 
138 static unsigned long unix_vertex_unvisited_index = UNIX_VERTEX_INDEX_MARK1;
139 
140 static void unix_add_edge(struct scm_fp_list *fpl, struct unix_edge *edge)
141 {
142 	struct unix_vertex *vertex = edge->predecessor->vertex;
143 
144 	if (!vertex) {
145 		vertex = list_first_entry(&fpl->vertices, typeof(*vertex), entry);
146 		vertex->index = unix_vertex_unvisited_index;
147 		vertex->out_degree = 0;
148 		INIT_LIST_HEAD(&vertex->edges);
149 		INIT_LIST_HEAD(&vertex->scc_entry);
150 
151 		list_move_tail(&vertex->entry, &unix_unvisited_vertices);
152 		edge->predecessor->vertex = vertex;
153 	}
154 
155 	vertex->out_degree++;
156 	list_add_tail(&edge->vertex_entry, &vertex->edges);
157 
158 	unix_update_graph(unix_edge_successor(edge));
159 }
160 
161 static void unix_del_edge(struct scm_fp_list *fpl, struct unix_edge *edge)
162 {
163 	struct unix_vertex *vertex = edge->predecessor->vertex;
164 
165 	if (!fpl->dead)
166 		unix_update_graph(unix_edge_successor(edge));
167 
168 	list_del(&edge->vertex_entry);
169 	vertex->out_degree--;
170 
171 	if (!vertex->out_degree) {
172 		edge->predecessor->vertex = NULL;
173 		list_move_tail(&vertex->entry, &fpl->vertices);
174 	}
175 }
176 
177 static void unix_free_vertices(struct scm_fp_list *fpl)
178 {
179 	struct unix_vertex *vertex, *next_vertex;
180 
181 	list_for_each_entry_safe(vertex, next_vertex, &fpl->vertices, entry) {
182 		list_del(&vertex->entry);
183 		kfree(vertex);
184 	}
185 }
186 
187 static DEFINE_SPINLOCK(unix_gc_lock);
188 unsigned int unix_tot_inflight;
189 
190 void unix_add_edges(struct scm_fp_list *fpl, struct unix_sock *receiver)
191 {
192 	int i = 0, j = 0;
193 
194 	spin_lock(&unix_gc_lock);
195 
196 	if (!fpl->count_unix)
197 		goto out;
198 
199 	do {
200 		struct unix_sock *inflight = unix_get_socket(fpl->fp[j++]);
201 		struct unix_edge *edge;
202 
203 		if (!inflight)
204 			continue;
205 
206 		edge = fpl->edges + i++;
207 		edge->predecessor = inflight;
208 		edge->successor = receiver;
209 
210 		unix_add_edge(fpl, edge);
211 	} while (i < fpl->count_unix);
212 
213 	receiver->scm_stat.nr_unix_fds += fpl->count_unix;
214 	WRITE_ONCE(unix_tot_inflight, unix_tot_inflight + fpl->count_unix);
215 out:
216 	WRITE_ONCE(fpl->user->unix_inflight, fpl->user->unix_inflight + fpl->count);
217 
218 	spin_unlock(&unix_gc_lock);
219 
220 	fpl->inflight = true;
221 
222 	unix_free_vertices(fpl);
223 }
224 
225 void unix_del_edges(struct scm_fp_list *fpl)
226 {
227 	struct unix_sock *receiver;
228 	int i = 0;
229 
230 	spin_lock(&unix_gc_lock);
231 
232 	if (!fpl->count_unix)
233 		goto out;
234 
235 	do {
236 		struct unix_edge *edge = fpl->edges + i++;
237 
238 		unix_del_edge(fpl, edge);
239 	} while (i < fpl->count_unix);
240 
241 	if (!fpl->dead) {
242 		receiver = fpl->edges[0].successor;
243 		receiver->scm_stat.nr_unix_fds -= fpl->count_unix;
244 	}
245 	WRITE_ONCE(unix_tot_inflight, unix_tot_inflight - fpl->count_unix);
246 out:
247 	WRITE_ONCE(fpl->user->unix_inflight, fpl->user->unix_inflight - fpl->count);
248 
249 	spin_unlock(&unix_gc_lock);
250 
251 	fpl->inflight = false;
252 }
253 
254 void unix_update_edges(struct unix_sock *receiver)
255 {
256 	/* nr_unix_fds is only updated under unix_state_lock().
257 	 * If it's 0 here, the embryo socket is not part of the
258 	 * inflight graph, and GC will not see it, so no lock needed.
259 	 */
260 	if (!receiver->scm_stat.nr_unix_fds) {
261 		receiver->listener = NULL;
262 	} else {
263 		spin_lock(&unix_gc_lock);
264 		unix_update_graph(unix_sk(receiver->listener)->vertex);
265 		receiver->listener = NULL;
266 		spin_unlock(&unix_gc_lock);
267 	}
268 }
269 
270 int unix_prepare_fpl(struct scm_fp_list *fpl)
271 {
272 	struct unix_vertex *vertex;
273 	int i;
274 
275 	if (!fpl->count_unix)
276 		return 0;
277 
278 	for (i = 0; i < fpl->count_unix; i++) {
279 		vertex = kmalloc(sizeof(*vertex), GFP_KERNEL);
280 		if (!vertex)
281 			goto err;
282 
283 		list_add(&vertex->entry, &fpl->vertices);
284 	}
285 
286 	fpl->edges = kvmalloc_array(fpl->count_unix, sizeof(*fpl->edges),
287 				    GFP_KERNEL_ACCOUNT);
288 	if (!fpl->edges)
289 		goto err;
290 
291 	return 0;
292 
293 err:
294 	unix_free_vertices(fpl);
295 	return -ENOMEM;
296 }
297 
298 void unix_destroy_fpl(struct scm_fp_list *fpl)
299 {
300 	if (fpl->inflight)
301 		unix_del_edges(fpl);
302 
303 	kvfree(fpl->edges);
304 	unix_free_vertices(fpl);
305 }
306 
307 static bool unix_vertex_dead(struct unix_vertex *vertex)
308 {
309 	struct unix_edge *edge;
310 	struct unix_sock *u;
311 	long total_ref;
312 
313 	list_for_each_entry(edge, &vertex->edges, vertex_entry) {
314 		struct unix_vertex *next_vertex = unix_edge_successor(edge);
315 
316 		/* The vertex's fd can be received by a non-inflight socket. */
317 		if (!next_vertex)
318 			return false;
319 
320 		/* The vertex's fd can be received by an inflight socket in
321 		 * another SCC.
322 		 */
323 		if (next_vertex->scc_index != vertex->scc_index)
324 			return false;
325 	}
326 
327 	/* No receiver exists out of the same SCC. */
328 
329 	edge = list_first_entry(&vertex->edges, typeof(*edge), vertex_entry);
330 	u = edge->predecessor;
331 	total_ref = file_count(u->sk.sk_socket->file);
332 
333 	/* If not close()d, total_ref > out_degree. */
334 	if (total_ref != vertex->out_degree)
335 		return false;
336 
337 	return true;
338 }
339 
340 static void unix_collect_skb(struct list_head *scc, struct sk_buff_head *hitlist)
341 {
342 	struct unix_vertex *vertex;
343 
344 	list_for_each_entry_reverse(vertex, scc, scc_entry) {
345 		struct sk_buff_head *queue;
346 		struct unix_edge *edge;
347 		struct unix_sock *u;
348 
349 		edge = list_first_entry(&vertex->edges, typeof(*edge), vertex_entry);
350 		u = edge->predecessor;
351 		queue = &u->sk.sk_receive_queue;
352 
353 		spin_lock(&queue->lock);
354 
355 		if (u->sk.sk_state == TCP_LISTEN) {
356 			struct sk_buff *skb;
357 
358 			skb_queue_walk(queue, skb) {
359 				struct sk_buff_head *embryo_queue = &skb->sk->sk_receive_queue;
360 
361 				spin_lock(&embryo_queue->lock);
362 				skb_queue_splice_init(embryo_queue, hitlist);
363 				spin_unlock(&embryo_queue->lock);
364 			}
365 		} else {
366 			skb_queue_splice_init(queue, hitlist);
367 		}
368 
369 		spin_unlock(&queue->lock);
370 	}
371 }
372 
373 static bool unix_scc_cyclic(struct list_head *scc)
374 {
375 	struct unix_vertex *vertex;
376 	struct unix_edge *edge;
377 
378 	/* SCC containing multiple vertices ? */
379 	if (!list_is_singular(scc))
380 		return true;
381 
382 	vertex = list_first_entry(scc, typeof(*vertex), scc_entry);
383 
384 	/* Self-reference or a embryo-listener circle ? */
385 	list_for_each_entry(edge, &vertex->edges, vertex_entry) {
386 		if (unix_edge_successor(edge) == vertex)
387 			return true;
388 	}
389 
390 	return false;
391 }
392 
393 static LIST_HEAD(unix_visited_vertices);
394 static unsigned long unix_vertex_grouped_index = UNIX_VERTEX_INDEX_MARK2;
395 
396 static void __unix_walk_scc(struct unix_vertex *vertex, unsigned long *last_index,
397 			    struct sk_buff_head *hitlist)
398 {
399 	LIST_HEAD(vertex_stack);
400 	struct unix_edge *edge;
401 	LIST_HEAD(edge_stack);
402 
403 next_vertex:
404 	/* Push vertex to vertex_stack and mark it as on-stack
405 	 * (index >= UNIX_VERTEX_INDEX_START).
406 	 * The vertex will be popped when finalising SCC later.
407 	 */
408 	list_add(&vertex->scc_entry, &vertex_stack);
409 
410 	vertex->index = *last_index;
411 	vertex->scc_index = *last_index;
412 	(*last_index)++;
413 
414 	/* Explore neighbour vertices (receivers of the current vertex's fd). */
415 	list_for_each_entry(edge, &vertex->edges, vertex_entry) {
416 		struct unix_vertex *next_vertex = unix_edge_successor(edge);
417 
418 		if (!next_vertex)
419 			continue;
420 
421 		if (next_vertex->index == unix_vertex_unvisited_index) {
422 			/* Iterative deepening depth first search
423 			 *
424 			 *   1. Push a forward edge to edge_stack and set
425 			 *      the successor to vertex for the next iteration.
426 			 */
427 			list_add(&edge->stack_entry, &edge_stack);
428 
429 			vertex = next_vertex;
430 			goto next_vertex;
431 
432 			/*   2. Pop the edge directed to the current vertex
433 			 *      and restore the ancestor for backtracking.
434 			 */
435 prev_vertex:
436 			edge = list_first_entry(&edge_stack, typeof(*edge), stack_entry);
437 			list_del_init(&edge->stack_entry);
438 
439 			next_vertex = vertex;
440 			vertex = edge->predecessor->vertex;
441 
442 			/* If the successor has a smaller scc_index, two vertices
443 			 * are in the same SCC, so propagate the smaller scc_index
444 			 * to skip SCC finalisation.
445 			 */
446 			vertex->scc_index = min(vertex->scc_index, next_vertex->scc_index);
447 		} else if (next_vertex->index != unix_vertex_grouped_index) {
448 			/* Loop detected by a back/cross edge.
449 			 *
450 			 * The successor is on vertex_stack, so two vertices are in
451 			 * the same SCC.  If the successor has a smaller *scc_index*,
452 			 * propagate it to skip SCC finalisation.
453 			 */
454 			vertex->scc_index = min(vertex->scc_index, next_vertex->scc_index);
455 		} else {
456 			/* The successor was already grouped as another SCC */
457 		}
458 	}
459 
460 	if (vertex->index == vertex->scc_index) {
461 		struct unix_vertex *v;
462 		struct list_head scc;
463 		bool scc_dead = true;
464 
465 		/* SCC finalised.
466 		 *
467 		 * If the scc_index was not updated, all the vertices above on
468 		 * vertex_stack are in the same SCC.  Group them using scc_entry.
469 		 */
470 		__list_cut_position(&scc, &vertex_stack, &vertex->scc_entry);
471 
472 		list_for_each_entry_reverse(v, &scc, scc_entry) {
473 			/* Don't restart DFS from this vertex in unix_walk_scc(). */
474 			list_move_tail(&v->entry, &unix_visited_vertices);
475 
476 			/* Mark vertex as off-stack. */
477 			v->index = unix_vertex_grouped_index;
478 
479 			if (scc_dead)
480 				scc_dead = unix_vertex_dead(v);
481 		}
482 
483 		if (scc_dead)
484 			unix_collect_skb(&scc, hitlist);
485 		else if (!unix_graph_maybe_cyclic)
486 			unix_graph_maybe_cyclic = unix_scc_cyclic(&scc);
487 
488 		list_del(&scc);
489 	}
490 
491 	/* Need backtracking ? */
492 	if (!list_empty(&edge_stack))
493 		goto prev_vertex;
494 }
495 
496 static void unix_walk_scc(struct sk_buff_head *hitlist)
497 {
498 	unsigned long last_index = UNIX_VERTEX_INDEX_START;
499 
500 	unix_graph_maybe_cyclic = false;
501 
502 	/* Visit every vertex exactly once.
503 	 * __unix_walk_scc() moves visited vertices to unix_visited_vertices.
504 	 */
505 	while (!list_empty(&unix_unvisited_vertices)) {
506 		struct unix_vertex *vertex;
507 
508 		vertex = list_first_entry(&unix_unvisited_vertices, typeof(*vertex), entry);
509 		__unix_walk_scc(vertex, &last_index, hitlist);
510 	}
511 
512 	list_replace_init(&unix_visited_vertices, &unix_unvisited_vertices);
513 	swap(unix_vertex_unvisited_index, unix_vertex_grouped_index);
514 
515 	unix_graph_grouped = true;
516 }
517 
518 static void unix_walk_scc_fast(struct sk_buff_head *hitlist)
519 {
520 	unix_graph_maybe_cyclic = false;
521 
522 	while (!list_empty(&unix_unvisited_vertices)) {
523 		struct unix_vertex *vertex;
524 		struct list_head scc;
525 		bool scc_dead = true;
526 
527 		vertex = list_first_entry(&unix_unvisited_vertices, typeof(*vertex), entry);
528 		list_add(&scc, &vertex->scc_entry);
529 
530 		list_for_each_entry_reverse(vertex, &scc, scc_entry) {
531 			list_move_tail(&vertex->entry, &unix_visited_vertices);
532 
533 			if (scc_dead)
534 				scc_dead = unix_vertex_dead(vertex);
535 		}
536 
537 		if (scc_dead)
538 			unix_collect_skb(&scc, hitlist);
539 		else if (!unix_graph_maybe_cyclic)
540 			unix_graph_maybe_cyclic = unix_scc_cyclic(&scc);
541 
542 		list_del(&scc);
543 	}
544 
545 	list_replace_init(&unix_visited_vertices, &unix_unvisited_vertices);
546 }
547 
548 static bool gc_in_progress;
549 
550 static void __unix_gc(struct work_struct *work)
551 {
552 	struct sk_buff_head hitlist;
553 	struct sk_buff *skb;
554 
555 	spin_lock(&unix_gc_lock);
556 
557 	if (!unix_graph_maybe_cyclic) {
558 		spin_unlock(&unix_gc_lock);
559 		goto skip_gc;
560 	}
561 
562 	__skb_queue_head_init(&hitlist);
563 
564 	if (unix_graph_grouped)
565 		unix_walk_scc_fast(&hitlist);
566 	else
567 		unix_walk_scc(&hitlist);
568 
569 	spin_unlock(&unix_gc_lock);
570 
571 	skb_queue_walk(&hitlist, skb) {
572 		if (UNIXCB(skb).fp)
573 			UNIXCB(skb).fp->dead = true;
574 	}
575 
576 	__skb_queue_purge(&hitlist);
577 skip_gc:
578 	WRITE_ONCE(gc_in_progress, false);
579 }
580 
581 static DECLARE_WORK(unix_gc_work, __unix_gc);
582 
583 void unix_gc(void)
584 {
585 	WRITE_ONCE(gc_in_progress, true);
586 	queue_work(system_unbound_wq, &unix_gc_work);
587 }
588 
589 #define UNIX_INFLIGHT_TRIGGER_GC 16000
590 #define UNIX_INFLIGHT_SANE_USER (SCM_MAX_FD * 8)
591 
592 void wait_for_unix_gc(struct scm_fp_list *fpl)
593 {
594 	/* If number of inflight sockets is insane,
595 	 * force a garbage collect right now.
596 	 *
597 	 * Paired with the WRITE_ONCE() in unix_inflight(),
598 	 * unix_notinflight(), and __unix_gc().
599 	 */
600 	if (READ_ONCE(unix_tot_inflight) > UNIX_INFLIGHT_TRIGGER_GC &&
601 	    !READ_ONCE(gc_in_progress))
602 		unix_gc();
603 
604 	/* Penalise users who want to send AF_UNIX sockets
605 	 * but whose sockets have not been received yet.
606 	 */
607 	if (!fpl || !fpl->count_unix ||
608 	    READ_ONCE(fpl->user->unix_inflight) < UNIX_INFLIGHT_SANE_USER)
609 		return;
610 
611 	if (READ_ONCE(gc_in_progress))
612 		flush_work(&unix_gc_work);
613 }
614