[linux-2.6-block.git] / net / unix / garbage.c

/*
 * NET3:	Garbage Collector For AF_UNIX sockets
 *
 * Garbage Collector:
 *	Copyright (C) Barak A. Pearlmutter.
 *	Released under the GPL version 2 or later.
 *
 * Chopped about by Alan Cox 22/3/96 to make it fit the AF_UNIX socket problem.
 * If it doesn't work blame me, it worked when Barak sent it.
 *
 * Assumptions:
 *
 *  - object w/ a bit
 *  - free list
 *
 * Current optimizations:
 *
 *  - explicit stack instead of recursion
 *  - tail recurse on first born instead of immediate push/pop
 *  - we gather the stuff that should not be killed into tree
 *    and stack is just a path from root to the current pointer.
 *
 *  Future optimizations:
 *
 *  - don't just push entire root set; process in place
 *
 *	This program is free software; you can redistribute it and/or
 *	modify it under the terms of the GNU General Public License
 *	as published by the Free Software Foundation; either version
 *	2 of the License, or (at your option) any later version.
 *
 *  Fixes:
 *	Alan Cox	07 Sept	1997	Vmalloc internal stack as needed.
 *					Cope with changing max_files.
 *	Al Viro		11 Oct 1998
 *		Graph may have cycles. That is, we can send the descriptor
 *		of foo to bar and vice versa. Current code chokes on that.
 *		Fix: move SCM_RIGHTS ones into the separate list and then
 *		skb_free() them all instead of doing explicit fput's.
 *		Another problem: since fput() may block somebody may
 *		create a new unix_socket when we are in the middle of sweep
 *		phase. Fix: revert the logic wrt MARKED. Mark everything
 *		upon the beginning and unmark non-junk ones.
 *
 *		[12 Oct 1998] AAARGH! New code purges all SCM_RIGHTS
 *		sent to connect()'ed but still not accept()'ed sockets.
 *		Fixed. Old code had slightly different problem here:
 *		extra fput() in situation when we passed the descriptor via
 *		such socket and closed it (descriptor). That would happen on
 *		each unix_gc() until the accept(). Since the struct file in
 *		question would go to the free list and might be reused...
 *		That might be the reason of random oopses on filp_close()
 *		in unrelated processes.
 *
 *	AV		28 Feb 1999
 *		Kill the explicit allocation of stack. Now we keep the tree
 *		with root in dummy + pointer (gc_current) to one of the nodes.
 *		Stack is represented as path from gc_current to dummy. Unmark
 *		now means "add to tree". Push == "make it a son of gc_current".
 *		Pop == "move gc_current to parent". We keep only pointers to
 *		parents (->gc_tree).
 *	AV		1 Mar 1999
 *		Damn. Added missing check for ->dead in listen queues scanning.
 *
 *	Miklos Szeredi 25 Jun 2007
 *		Reimplement with a cycle collecting algorithm. This should
 *		solve several problems with the previous code, like being racy
 *		wrt receive and holding up unrelated socket operations.
 */

#include <linux/kernel.h>
#include <linux/string.h>
#include <linux/socket.h>
#include <linux/un.h>
#include <linux/net.h>
#include <linux/fs.h>
#include <linux/skbuff.h>
#include <linux/netdevice.h>
#include <linux/file.h>
#include <linux/proc_fs.h>
#include <linux/mutex.h>
#include <linux/wait.h>

#include <net/sock.h>
#include <net/af_unix.h>
#include <net/scm.h>
#include <net/tcp_states.h>

/* Internal data structures and random procedures: */

static LIST_HEAD(gc_inflight_list);
static LIST_HEAD(gc_candidates);
static DEFINE_SPINLOCK(unix_gc_lock);
static DECLARE_WAIT_QUEUE_HEAD(unix_gc_wait);

unsigned int unix_tot_inflight;

struct sock *unix_get_socket(struct file *filp)
{
	struct sock *u_sock = NULL;
	struct inode *inode = file_inode(filp);

	/* Socket ? */
	if (S_ISSOCK(inode->i_mode) && !(filp->f_mode & FMODE_PATH)) {
		struct socket *sock = SOCKET_I(inode);
		struct sock *s = sock->sk;

		/* PF_UNIX ? */
		if (s && sock->ops && sock->ops->family == PF_UNIX)
			u_sock = s;
	}
	return u_sock;
}

/* Keep the number of times in flight count for the file
 * descriptor if it is for an AF_UNIX socket.
 */

void unix_inflight(struct file *fp)
{
	struct sock *s = unix_get_socket(fp);

	if (s) {
		struct unix_sock *u = unix_sk(s);

		spin_lock(&unix_gc_lock);

		if (atomic_long_inc_return(&u->inflight) == 1) {
			BUG_ON(!list_empty(&u->link));
			list_add_tail(&u->link, &gc_inflight_list);
		} else {
			BUG_ON(list_empty(&u->link));
		}
		unix_tot_inflight++;
		spin_unlock(&unix_gc_lock);
	}
}

void unix_notinflight(struct file *fp)
{
	struct sock *s = unix_get_socket(fp);

	if (s) {
		struct unix_sock *u = unix_sk(s);

		spin_lock(&unix_gc_lock);
		BUG_ON(list_empty(&u->link));

		if (atomic_long_dec_and_test(&u->inflight))
			list_del_init(&u->link);
		unix_tot_inflight--;
		spin_unlock(&unix_gc_lock);
	}
}

static void scan_inflight(struct sock *x, void (*func)(struct unix_sock *),
			  struct sk_buff_head *hitlist)
{
	struct sk_buff *skb;
	struct sk_buff *next;

	spin_lock(&x->sk_receive_queue.lock);
	skb_queue_walk_safe(&x->sk_receive_queue, skb, next) {
		/* Do we have file descriptors ? */
		if (UNIXCB(skb).fp) {
			bool hit = false;
			/* Process the descriptors of this socket */
			int nfd = UNIXCB(skb).fp->count;
			struct file **fp = UNIXCB(skb).fp->fp;

			while (nfd--) {
				/* Get the socket the fd matches if it indeed does so */
				struct sock *sk = unix_get_socket(*fp++);

				if (sk) {
					struct unix_sock *u = unix_sk(sk);

					/* Ignore non-candidates, they could
					 * have been added to the queues after
					 * starting the garbage collection
					 */
					if (test_bit(UNIX_GC_CANDIDATE, &u->gc_flags)) {
						hit = true;

						func(u);
					}
				}
			}
			if (hit && hitlist != NULL) {
				__skb_unlink(skb, &x->sk_receive_queue);
				__skb_queue_tail(hitlist, skb);
			}
		}
	}
	spin_unlock(&x->sk_receive_queue.lock);
}

static void scan_children(struct sock *x, void (*func)(struct unix_sock *),
			  struct sk_buff_head *hitlist)
{
	if (x->sk_state != TCP_LISTEN) {
		scan_inflight(x, func, hitlist);
	} else {
		struct sk_buff *skb;
		struct sk_buff *next;
		struct unix_sock *u;
		LIST_HEAD(embryos);

		/* For a listening socket collect the queued embryos
		 * and perform a scan on them as well.
		 */
		spin_lock(&x->sk_receive_queue.lock);
		skb_queue_walk_safe(&x->sk_receive_queue, skb, next) {
			u = unix_sk(skb->sk);

			/* An embryo cannot be in-flight, so it's safe
			 * to use the list link.
			 */
			BUG_ON(!list_empty(&u->link));
			list_add_tail(&u->link, &embryos);
		}
		spin_unlock(&x->sk_receive_queue.lock);

		while (!list_empty(&embryos)) {
			u = list_entry(embryos.next, struct unix_sock, link);
			scan_inflight(&u->sk, func, hitlist);
			list_del_init(&u->link);
		}
	}
}

static void dec_inflight(struct unix_sock *usk)
{
	atomic_long_dec(&usk->inflight);
}

static void inc_inflight(struct unix_sock *usk)
{
	atomic_long_inc(&usk->inflight);
}

static void inc_inflight_move_tail(struct unix_sock *u)
{
	atomic_long_inc(&u->inflight);
	/* If this still might be part of a cycle, move it to the end
	 * of the list, so that it's checked even if it was already
	 * passed over
	 */
	if (test_bit(UNIX_GC_MAYBE_CYCLE, &u->gc_flags))
		list_move_tail(&u->link, &gc_candidates);
}

static bool gc_in_progress;
#define UNIX_INFLIGHT_TRIGGER_GC 16000

void wait_for_unix_gc(void)
{
	/* If number of inflight sockets is insane,
	 * force a garbage collect right now.
	 */
	if (unix_tot_inflight > UNIX_INFLIGHT_TRIGGER_GC && !gc_in_progress)
		unix_gc();
	wait_event(unix_gc_wait, gc_in_progress == false);
}

/* The external entry point: unix_gc() */
void unix_gc(void)
{
	struct unix_sock *u;
	struct unix_sock *next;
	struct sk_buff_head hitlist;
	struct list_head cursor;
	LIST_HEAD(not_cycle_list);

	spin_lock(&unix_gc_lock);

	/* Avoid a recursive GC. */
	if (gc_in_progress)
		goto out;

	gc_in_progress = true;
	/* First, select candidates for garbage collection.  Only
	 * in-flight sockets are considered, and from those only ones
	 * which don't have any external reference.
	 *
	 * Holding unix_gc_lock will protect these candidates from
	 * being detached, and hence from gaining an external
	 * reference.  Since there are no possible receivers, all
	 * buffers currently on the candidates' queues stay there
	 * during the garbage collection.
	 *
	 * We also know that no new candidate can be added onto the
	 * receive queues.  Other, non candidate sockets _can_ be
	 * added to queue, so we must make sure only to touch
	 * candidates.
	 */
	list_for_each_entry_safe(u, next, &gc_inflight_list, link) {
		long total_refs;
		long inflight_refs;

		total_refs = file_count(u->sk.sk_socket->file);
		inflight_refs = atomic_long_read(&u->inflight);

		BUG_ON(inflight_refs < 1);
		BUG_ON(total_refs < inflight_refs);
		if (total_refs == inflight_refs) {
			list_move_tail(&u->link, &gc_candidates);
			__set_bit(UNIX_GC_CANDIDATE, &u->gc_flags);
			__set_bit(UNIX_GC_MAYBE_CYCLE, &u->gc_flags);
		}
	}

	/* Now remove all internal in-flight reference to children of
	 * the candidates.
	 */
	list_for_each_entry(u, &gc_candidates, link)
		scan_children(&u->sk, dec_inflight, NULL);

	/* Restore the references for children of all candidates,
	 * which have remaining references.  Do this recursively, so
	 * only those remain, which form cyclic references.
	 *
	 * Use a "cursor" link, to make the list traversal safe, even
	 * though elements might be moved about.
	 */
	list_add(&cursor, &gc_candidates);
	while (cursor.next != &gc_candidates) {
		u = list_entry(cursor.next, struct unix_sock, link);

		/* Move cursor to after the current position. */
		list_move(&cursor, &u->link);

		if (atomic_long_read(&u->inflight) > 0) {
			list_move_tail(&u->link, &not_cycle_list);
			__clear_bit(UNIX_GC_MAYBE_CYCLE, &u->gc_flags);
			scan_children(&u->sk, inc_inflight_move_tail, NULL);
		}
	}
	list_del(&cursor);

	/* not_cycle_list contains those sockets which do not make up a
	 * cycle.  Restore these to the inflight list.
	 */
	while (!list_empty(&not_cycle_list)) {
		u = list_entry(not_cycle_list.next, struct unix_sock, link);
		__clear_bit(UNIX_GC_CANDIDATE, &u->gc_flags);
		list_move_tail(&u->link, &gc_inflight_list);
	}

	/* Now gc_candidates contains only garbage.  Restore original
	 * inflight counters for these as well, and remove the skbuffs
	 * which are creating the cycle(s).
	 */
	skb_queue_head_init(&hitlist);
	list_for_each_entry(u, &gc_candidates, link)
	scan_children(&u->sk, inc_inflight, &hitlist);

	spin_unlock(&unix_gc_lock);

	/* Here we are. Hitlist is filled. Die. */
	__skb_queue_purge(&hitlist);

	spin_lock(&unix_gc_lock);

	/* All candidates should have been detached by now. */
	BUG_ON(!list_empty(&gc_candidates));
	gc_in_progress = false;
	wake_up(&unix_gc_wait);

 out:
	spin_unlock(&unix_gc_lock);
}
Commit	Line	Data
1da177e4 LT	1	/*
	2	* NET3: Garbage Collector For AF_UNIX sockets
	3	*
	4	* Garbage Collector:
	5	* Copyright (C) Barak A. Pearlmutter.
	6	* Released under the GPL version 2 or later.
	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	* This program is free software; you can redistribute it and/or
	28	* modify it under the terms of the GNU General Public License
	29	* as published by the Free Software Foundation; either version
	30	* 2 of the License, or (at your option) any later version.
	31	*
	32	* Fixes:
	33	* Alan Cox 07 Sept 1997 Vmalloc internal stack as needed.
	34	* Cope with changing max_files.
	35	* Al Viro 11 Oct 1998
	36	* Graph may have cycles. That is, we can send the descriptor
	37	* of foo to bar and vice versa. Current code chokes on that.
	38	* Fix: move SCM_RIGHTS ones into the separate list and then
	39	* skb_free() them all instead of doing explicit fput's.
	40	* Another problem: since fput() may block somebody may
	41	* create a new unix_socket when we are in the middle of sweep
	42	* phase. Fix: revert the logic wrt MARKED. Mark everything
	43	* upon the beginning and unmark non-junk ones.
	44	*
	45	* [12 Oct 1998] AAARGH! New code purges all SCM_RIGHTS
	46	* sent to connect()'ed but still not accept()'ed sockets.
	47	* Fixed. Old code had slightly different problem here:
	48	* extra fput() in situation when we passed the descriptor via
	49	* such socket and closed it (descriptor). That would happen on
	50	* each unix_gc() until the accept(). Since the struct file in
	51	* question would go to the free list and might be reused...
	52	* That might be the reason of random oopses on filp_close()
	53	* in unrelated processes.
	54	*
	55	* AV 28 Feb 1999
	56	* Kill the explicit allocation of stack. Now we keep the tree
	57	* with root in dummy + pointer (gc_current) to one of the nodes.
	58	* Stack is represented as path from gc_current to dummy. Unmark
	59	* now means "add to tree". Push == "make it a son of gc_current".
	60	* Pop == "move gc_current to parent". We keep only pointers to
	61	* parents (->gc_tree).
	62	* AV 1 Mar 1999
	63	* Damn. Added missing check for ->dead in listen queues scanning.
	64	*
1fd05ba5 MS	65	* Miklos Szeredi 25 Jun 2007
	66	* Reimplement with a cycle collecting algorithm. This should
	67	* solve several problems with the previous code, like being racy
	68	* wrt receive and holding up unrelated socket operations.
1da177e4	69	*/
ac7bfa62	70
1da177e4	71	#include <linux/kernel.h>
1da177e4 LT	72	#include <linux/string.h>
	73	#include <linux/socket.h>
	74	#include <linux/un.h>
	75	#include <linux/net.h>
	76	#include <linux/fs.h>
1da177e4 LT	77	#include <linux/skbuff.h>
	78	#include <linux/netdevice.h>
	79	#include <linux/file.h>
	80	#include <linux/proc_fs.h>
4a3e2f71	81	#include <linux/mutex.h>
5f23b734	82	#include <linux/wait.h>
1da177e4 LT	83
	84	#include <net/sock.h>
	85	#include <net/af_unix.h>
	86	#include <net/scm.h>
c752f073	87	#include <net/tcp_states.h>
1da177e4 LT	88
	89	/* Internal data structures and random procedures: */
	90
1fd05ba5 MS	91	static LIST_HEAD(gc_inflight_list);
	92	static LIST_HEAD(gc_candidates);
	93	static DEFINE_SPINLOCK(unix_gc_lock);
5f23b734	94	static DECLARE_WAIT_QUEUE_HEAD(unix_gc_wait);
1da177e4	95
9305cfa4	96	unsigned int unix_tot_inflight;
1da177e4	97
25888e30	98	struct sock unix_get_socket(struct file filp)
1da177e4 LT	99	{
1da177e4 LT	100	struct sock *u_sock = NULL;
496ad9aa	101	struct inode *inode = file_inode(filp);
1da177e4	102
d1ab39f1	103	/* Socket ? */
326be7b4	104	if (S_ISSOCK(inode->i_mode) && !(filp->f_mode & FMODE_PATH)) {
e27dfcea JK	105	struct socket *sock = SOCKET_I(inode);
e27dfcea JK	106	struct sock *s = sock->sk;
1da177e4	107
d1ab39f1	108	/* PF_UNIX ? */
1da177e4 LT	109	if (s && sock->ops && sock->ops->family == PF_UNIX)
	110	u_sock = s;
	111	}
	112	return u_sock;
	113	}
	114
d1ab39f1 JE	115	/* Keep the number of times in flight count for the file
d1ab39f1 JE	116	* descriptor if it is for an AF_UNIX socket.
1da177e4	117	*/
ac7bfa62	118
1da177e4 LT	119	void unix_inflight(struct file *fp)
	120	{
	121	struct sock *s = unix_get_socket(fp);
d1ab39f1	122
e27dfcea	123	if (s) {
1fd05ba5	124	struct unix_sock *u = unix_sk(s);
d1ab39f1	125
1fd05ba5	126	spin_lock(&unix_gc_lock);
d1ab39f1	127
516e0cc5	128	if (atomic_long_inc_return(&u->inflight) == 1) {
1fd05ba5 MS	129	BUG_ON(!list_empty(&u->link));
	130	list_add_tail(&u->link, &gc_inflight_list);
	131	} else {
	132	BUG_ON(list_empty(&u->link));
	133	}
9305cfa4	134	unix_tot_inflight++;
1fd05ba5	135	spin_unlock(&unix_gc_lock);
1da177e4 LT	136	}
	137	}
	138
	139	void unix_notinflight(struct file *fp)
	140	{
	141	struct sock *s = unix_get_socket(fp);
d1ab39f1	142
e27dfcea	143	if (s) {
1fd05ba5	144	struct unix_sock *u = unix_sk(s);
d1ab39f1	145
1fd05ba5 MS	146	spin_lock(&unix_gc_lock);
1fd05ba5 MS	147	BUG_ON(list_empty(&u->link));
d1ab39f1	148
516e0cc5	149	if (atomic_long_dec_and_test(&u->inflight))
1fd05ba5	150	list_del_init(&u->link);
9305cfa4	151	unix_tot_inflight--;
1fd05ba5	152	spin_unlock(&unix_gc_lock);
1da177e4 LT	153	}
	154	}
	155
5c80f1ae	156	static void scan_inflight(struct sock x, void (func)(struct unix_sock *),
1fd05ba5	157	struct sk_buff_head *hitlist)
1da177e4	158	{
1fd05ba5 MS	159	struct sk_buff *skb;
	160	struct sk_buff *next;
	161
	162	spin_lock(&x->sk_receive_queue.lock);
a2f3be17	163	skb_queue_walk_safe(&x->sk_receive_queue, skb, next) {
d1ab39f1	164	/* Do we have file descriptors ? */
1fd05ba5 MS	165	if (UNIXCB(skb).fp) {
1fd05ba5 MS	166	bool hit = false;
d1ab39f1	167	/* Process the descriptors of this socket */
1fd05ba5 MS	168	int nfd = UNIXCB(skb).fp->count;
1fd05ba5 MS	169	struct file **fp = UNIXCB(skb).fp->fp;
d1ab39f1	170
1fd05ba5	171	while (nfd--) {
d1ab39f1	172	/* Get the socket the fd matches if it indeed does so */
1fd05ba5	173	struct sock sk = unix_get_socket(fp++);
d1ab39f1	174
5c80f1ae	175	if (sk) {
6209344f MS	176	struct unix_sock *u = unix_sk(sk);
6209344f MS	177
d1ab39f1	178	/* Ignore non-candidates, they could
6209344f MS	179	* have been added to the queues after
	180	* starting the garbage collection
	181	*/
60bc851a	182	if (test_bit(UNIX_GC_CANDIDATE, &u->gc_flags)) {
6209344f	183	hit = true;
d1ab39f1	184
6209344f MS	185	func(u);
6209344f MS	186	}
1fd05ba5 MS	187	}
	188	}
	189	if (hit && hitlist != NULL) {
	190	__skb_unlink(skb, &x->sk_receive_queue);
	191	__skb_queue_tail(hitlist, skb);
	192	}
	193	}
	194	}
	195	spin_unlock(&x->sk_receive_queue.lock);
1da177e4 LT	196	}
1da177e4 LT	197
5c80f1ae	198	static void scan_children(struct sock x, void (func)(struct unix_sock *),
1fd05ba5	199	struct sk_buff_head *hitlist)
1da177e4	200	{
d1ab39f1	201	if (x->sk_state != TCP_LISTEN) {
1fd05ba5	202	scan_inflight(x, func, hitlist);
d1ab39f1	203	} else {
1fd05ba5 MS	204	struct sk_buff *skb;
	205	struct sk_buff *next;
	206	struct unix_sock *u;
	207	LIST_HEAD(embryos);
	208
d1ab39f1	209	/* For a listening socket collect the queued embryos
1fd05ba5 MS	210	* and perform a scan on them as well.
	211	*/
	212	spin_lock(&x->sk_receive_queue.lock);
a2f3be17	213	skb_queue_walk_safe(&x->sk_receive_queue, skb, next) {
1fd05ba5 MS	214	u = unix_sk(skb->sk);
1fd05ba5 MS	215
d1ab39f1	216	/* An embryo cannot be in-flight, so it's safe
1fd05ba5 MS	217	* to use the list link.
	218	*/
	219	BUG_ON(!list_empty(&u->link));
	220	list_add_tail(&u->link, &embryos);
	221	}
	222	spin_unlock(&x->sk_receive_queue.lock);
	223
	224	while (!list_empty(&embryos)) {
	225	u = list_entry(embryos.next, struct unix_sock, link);
	226	scan_inflight(&u->sk, func, hitlist);
	227	list_del_init(&u->link);
	228	}
	229	}
1da177e4 LT	230	}
1da177e4 LT	231
5c80f1ae	232	static void dec_inflight(struct unix_sock *usk)
1da177e4	233	{
516e0cc5	234	atomic_long_dec(&usk->inflight);
1fd05ba5	235	}
1da177e4	236
5c80f1ae	237	static void inc_inflight(struct unix_sock *usk)
1fd05ba5	238	{
516e0cc5	239	atomic_long_inc(&usk->inflight);
1da177e4 LT	240	}
1da177e4 LT	241
5c80f1ae	242	static void inc_inflight_move_tail(struct unix_sock *u)
1fd05ba5	243	{
516e0cc5	244	atomic_long_inc(&u->inflight);
d1ab39f1	245	/* If this still might be part of a cycle, move it to the end
6209344f MS	246	* of the list, so that it's checked even if it was already
6209344f MS	247	* passed over
1fd05ba5	248	*/
60bc851a	249	if (test_bit(UNIX_GC_MAYBE_CYCLE, &u->gc_flags))
1fd05ba5 MS	250	list_move_tail(&u->link, &gc_candidates);
1fd05ba5 MS	251	}
1da177e4	252
505e907d	253	static bool gc_in_progress;
9915672d	254	#define UNIX_INFLIGHT_TRIGGER_GC 16000
1da177e4	255
5f23b734	256	void wait_for_unix_gc(void)
1da177e4	257	{
d1ab39f1	258	/* If number of inflight sockets is insane,
9915672d ED	259	* force a garbage collect right now.
	260	*/
	261	if (unix_tot_inflight > UNIX_INFLIGHT_TRIGGER_GC && !gc_in_progress)
	262	unix_gc();
5f23b734	263	wait_event(unix_gc_wait, gc_in_progress == false);
5f23b734	264	}
1da177e4	265
5f23b734	266	/* The external entry point: unix_gc() */
	267	void unix_gc(void)
	268	{
1fd05ba5 MS	269	struct unix_sock *u;
	270	struct unix_sock *next;
	271	struct sk_buff_head hitlist;
	272	struct list_head cursor;
6209344f	273	LIST_HEAD(not_cycle_list);
1da177e4	274
1fd05ba5	275	spin_lock(&unix_gc_lock);
1da177e4	276
1fd05ba5 MS	277	/* Avoid a recursive GC. */
	278	if (gc_in_progress)
	279	goto out;
1da177e4	280
1fd05ba5	281	gc_in_progress = true;
d1ab39f1	282	/* First, select candidates for garbage collection. Only
1fd05ba5 MS	283	* in-flight sockets are considered, and from those only ones
	284	* which don't have any external reference.
	285	*
	286	* Holding unix_gc_lock will protect these candidates from
	287	* being detached, and hence from gaining an external
6209344f MS	288	* reference. Since there are no possible receivers, all
	289	* buffers currently on the candidates' queues stay there
	290	* during the garbage collection.
	291	*
	292	* We also know that no new candidate can be added onto the
	293	* receive queues. Other, non candidate sockets _can_ be
	294	* added to queue, so we must make sure only to touch
	295	* candidates.
1da177e4	296	*/
1fd05ba5	297	list_for_each_entry_safe(u, next, &gc_inflight_list, link) {
516e0cc5 AV	298	long total_refs;
516e0cc5 AV	299	long inflight_refs;
1fd05ba5 MS	300
1fd05ba5 MS	301	total_refs = file_count(u->sk.sk_socket->file);
516e0cc5	302	inflight_refs = atomic_long_read(&u->inflight);
1fd05ba5 MS	303
	304	BUG_ON(inflight_refs < 1);
	305	BUG_ON(total_refs < inflight_refs);
	306	if (total_refs == inflight_refs) {
	307	list_move_tail(&u->link, &gc_candidates);
60bc851a ED	308	__set_bit(UNIX_GC_CANDIDATE, &u->gc_flags);
60bc851a ED	309	__set_bit(UNIX_GC_MAYBE_CYCLE, &u->gc_flags);
1fd05ba5 MS	310	}
1fd05ba5 MS	311	}
1da177e4	312
d1ab39f1	313	/* Now remove all internal in-flight reference to children of
1fd05ba5	314	* the candidates.
1da177e4	315	*/
1fd05ba5 MS	316	list_for_each_entry(u, &gc_candidates, link)
1fd05ba5 MS	317	scan_children(&u->sk, dec_inflight, NULL);
1da177e4	318
d1ab39f1	319	/* Restore the references for children of all candidates,
1fd05ba5 MS	320	* which have remaining references. Do this recursively, so
	321	* only those remain, which form cyclic references.
	322	*
	323	* Use a "cursor" link, to make the list traversal safe, even
	324	* though elements might be moved about.
1da177e4	325	*/
1fd05ba5 MS	326	list_add(&cursor, &gc_candidates);
	327	while (cursor.next != &gc_candidates) {
	328	u = list_entry(cursor.next, struct unix_sock, link);
1da177e4	329
1fd05ba5 MS	330	/* Move cursor to after the current position. */
1fd05ba5 MS	331	list_move(&cursor, &u->link);
ac7bfa62	332
516e0cc5	333	if (atomic_long_read(&u->inflight) > 0) {
6209344f	334	list_move_tail(&u->link, &not_cycle_list);
60bc851a	335	__clear_bit(UNIX_GC_MAYBE_CYCLE, &u->gc_flags);
1fd05ba5	336	scan_children(&u->sk, inc_inflight_move_tail, NULL);
1da177e4	337	}
1da177e4	338	}
1fd05ba5	339	list_del(&cursor);
1da177e4	340
d1ab39f1	341	/* not_cycle_list contains those sockets which do not make up a
6209344f MS	342	* cycle. Restore these to the inflight list.
	343	*/
	344	while (!list_empty(&not_cycle_list)) {
	345	u = list_entry(not_cycle_list.next, struct unix_sock, link);
60bc851a	346	__clear_bit(UNIX_GC_CANDIDATE, &u->gc_flags);
6209344f MS	347	list_move_tail(&u->link, &gc_inflight_list);
	348	}
	349
d1ab39f1	350	/* Now gc_candidates contains only garbage. Restore original
1fd05ba5 MS	351	* inflight counters for these as well, and remove the skbuffs
	352	* which are creating the cycle(s).
	353	*/
1da177e4	354	skb_queue_head_init(&hitlist);
1fd05ba5	355	list_for_each_entry(u, &gc_candidates, link)
e27dfcea	356	scan_children(&u->sk, inc_inflight, &hitlist);
1da177e4	357
1fd05ba5	358	spin_unlock(&unix_gc_lock);
1da177e4	359
1fd05ba5 MS	360	/* Here we are. Hitlist is filled. Die. */
1fd05ba5 MS	361	__skb_queue_purge(&hitlist);
1da177e4	362
1fd05ba5	363	spin_lock(&unix_gc_lock);
1da177e4	364
1fd05ba5 MS	365	/* All candidates should have been detached by now. */
	366	BUG_ON(!list_empty(&gc_candidates));
	367	gc_in_progress = false;
5f23b734	368	wake_up(&unix_gc_wait);
1da177e4	369
1fd05ba5 MS	370	out:
1fd05ba5 MS	371	spin_unlock(&unix_gc_lock);
1da177e4	372	}