[mirror_ubuntu-jammy-kernel.git] / net / strparser / strparser.c

// SPDX-License-Identifier: GPL-2.0-only
/*
 * Stream Parser
 *
 * Copyright (c) 2016 Tom Herbert <tom@herbertland.com>
 */

#include <linux/bpf.h>
#include <linux/errno.h>
#include <linux/errqueue.h>
#include <linux/file.h>
#include <linux/in.h>
#include <linux/kernel.h>
#include <linux/export.h>
#include <linux/init.h>
#include <linux/net.h>
#include <linux/netdevice.h>
#include <linux/poll.h>
#include <linux/rculist.h>
#include <linux/skbuff.h>
#include <linux/socket.h>
#include <linux/uaccess.h>
#include <linux/workqueue.h>
#include <net/strparser.h>
#include <net/netns/generic.h>
#include <net/sock.h>

static struct workqueue_struct *strp_wq;

static inline struct _strp_msg *_strp_msg(struct sk_buff *skb)
{
	return (struct _strp_msg *)((void *)skb->cb +
		offsetof(struct sk_skb_cb, strp));
}

/* Lower lock held */
static void strp_abort_strp(struct strparser *strp, int err)
{
	/* Unrecoverable error in receive */

	cancel_delayed_work(&strp->msg_timer_work);

	if (strp->stopped)
		return;

	strp->stopped = 1;

	if (strp->sk) {
		struct sock *sk = strp->sk;

		/* Report an error on the lower socket */
		sk->sk_err = -err;
		sk_error_report(sk);
	}
}

static void strp_start_timer(struct strparser *strp, long timeo)
{
	if (timeo && timeo != LONG_MAX)
		mod_delayed_work(strp_wq, &strp->msg_timer_work, timeo);
}

/* Lower lock held */
static void strp_parser_err(struct strparser *strp, int err,
			    read_descriptor_t *desc)
{
	desc->error = err;
	kfree_skb(strp->skb_head);
	strp->skb_head = NULL;
	strp->cb.abort_parser(strp, err);
}

static inline int strp_peek_len(struct strparser *strp)
{
	if (strp->sk) {
		struct socket *sock = strp->sk->sk_socket;

		return sock->ops->peek_len(sock);
	}

	/* If we don't have an associated socket there's nothing to peek.
	 * Return int max to avoid stopping the strparser.
	 */

	return INT_MAX;
}

/* Lower socket lock held */
static int __strp_recv(read_descriptor_t *desc, struct sk_buff *orig_skb,
		       unsigned int orig_offset, size_t orig_len,
		       size_t max_msg_size, long timeo)
{
	struct strparser *strp = (struct strparser *)desc->arg.data;
	struct _strp_msg *stm;
	struct sk_buff *head, *skb;
	size_t eaten = 0, cand_len;
	ssize_t extra;
	int err;
	bool cloned_orig = false;

	if (strp->paused)
		return 0;

	head = strp->skb_head;
	if (head) {
		/* Message already in progress */
		if (unlikely(orig_offset)) {
			/* Getting data with a non-zero offset when a message is
			 * in progress is not expected. If it does happen, we
			 * need to clone and pull since we can't deal with
			 * offsets in the skbs for a message expect in the head.
			 */
			orig_skb = skb_clone(orig_skb, GFP_ATOMIC);
			if (!orig_skb) {
				STRP_STATS_INCR(strp->stats.mem_fail);
				desc->error = -ENOMEM;
				return 0;
			}
			if (!pskb_pull(orig_skb, orig_offset)) {
				STRP_STATS_INCR(strp->stats.mem_fail);
				kfree_skb(orig_skb);
				desc->error = -ENOMEM;
				return 0;
			}
			cloned_orig = true;
			orig_offset = 0;
		}

		if (!strp->skb_nextp) {
			/* We are going to append to the frags_list of head.
			 * Need to unshare the frag_list.
			 */
			err = skb_unclone(head, GFP_ATOMIC);
			if (err) {
				STRP_STATS_INCR(strp->stats.mem_fail);
				desc->error = err;
				return 0;
			}

			if (unlikely(skb_shinfo(head)->frag_list)) {
				/* We can't append to an sk_buff that already
				 * has a frag_list. We create a new head, point
				 * the frag_list of that to the old head, and
				 * then are able to use the old head->next for
				 * appending to the message.
				 */
				if (WARN_ON(head->next)) {
					desc->error = -EINVAL;
					return 0;
				}

				skb = alloc_skb_for_msg(head);
				if (!skb) {
					STRP_STATS_INCR(strp->stats.mem_fail);
					desc->error = -ENOMEM;
					return 0;
				}

				strp->skb_nextp = &head->next;
				strp->skb_head = skb;
				head = skb;
			} else {
				strp->skb_nextp =
				    &skb_shinfo(head)->frag_list;
			}
		}
	}

	while (eaten < orig_len) {
		/* Always clone since we will consume something */
		skb = skb_clone(orig_skb, GFP_ATOMIC);
		if (!skb) {
			STRP_STATS_INCR(strp->stats.mem_fail);
			desc->error = -ENOMEM;
			break;
		}

		cand_len = orig_len - eaten;

		head = strp->skb_head;
		if (!head) {
			head = skb;
			strp->skb_head = head;
			/* Will set skb_nextp on next packet if needed */
			strp->skb_nextp = NULL;
			stm = _strp_msg(head);
			memset(stm, 0, sizeof(*stm));
			stm->strp.offset = orig_offset + eaten;
		} else {
			/* Unclone if we are appending to an skb that we
			 * already share a frag_list with.
			 */
			if (skb_has_frag_list(skb)) {
				err = skb_unclone(skb, GFP_ATOMIC);
				if (err) {
					STRP_STATS_INCR(strp->stats.mem_fail);
					desc->error = err;
					break;
				}
			}

			stm = _strp_msg(head);
			*strp->skb_nextp = skb;
			strp->skb_nextp = &skb->next;
			head->data_len += skb->len;
			head->len += skb->len;
			head->truesize += skb->truesize;
		}

		if (!stm->strp.full_len) {
			ssize_t len;

			len = (*strp->cb.parse_msg)(strp, head);

			if (!len) {
				/* Need more header to determine length */
				if (!stm->accum_len) {
					/* Start RX timer for new message */
					strp_start_timer(strp, timeo);
				}
				stm->accum_len += cand_len;
				eaten += cand_len;
				STRP_STATS_INCR(strp->stats.need_more_hdr);
				WARN_ON(eaten != orig_len);
				break;
			} else if (len < 0) {
				if (len == -ESTRPIPE && stm->accum_len) {
					len = -ENODATA;
					strp->unrecov_intr = 1;
				} else {
					strp->interrupted = 1;
				}
				strp_parser_err(strp, len, desc);
				break;
			} else if (len > max_msg_size) {
				/* Message length exceeds maximum allowed */
				STRP_STATS_INCR(strp->stats.msg_too_big);
				strp_parser_err(strp, -EMSGSIZE, desc);
				break;
			} else if (len <= (ssize_t)head->len -
					  skb->len - stm->strp.offset) {
				/* Length must be into new skb (and also
				 * greater than zero)
				 */
				STRP_STATS_INCR(strp->stats.bad_hdr_len);
				strp_parser_err(strp, -EPROTO, desc);
				break;
			}

			stm->strp.full_len = len;
		}

		extra = (ssize_t)(stm->accum_len + cand_len) -
			stm->strp.full_len;

		if (extra < 0) {
			/* Message not complete yet. */
			if (stm->strp.full_len - stm->accum_len >
			    strp_peek_len(strp)) {
				/* Don't have the whole message in the socket
				 * buffer. Set strp->need_bytes to wait for
				 * the rest of the message. Also, set "early
				 * eaten" since we've already buffered the skb
				 * but don't consume yet per strp_read_sock.
				 */

				if (!stm->accum_len) {
					/* Start RX timer for new message */
					strp_start_timer(strp, timeo);
				}

				stm->accum_len += cand_len;
				eaten += cand_len;
				strp->need_bytes = stm->strp.full_len -
						       stm->accum_len;
				STRP_STATS_ADD(strp->stats.bytes, cand_len);
				desc->count = 0; /* Stop reading socket */
				break;
			}
			stm->accum_len += cand_len;
			eaten += cand_len;
			WARN_ON(eaten != orig_len);
			break;
		}

		/* Positive extra indicates more bytes than needed for the
		 * message
		 */

		WARN_ON(extra > cand_len);

		eaten += (cand_len - extra);

		/* Hurray, we have a new message! */
		cancel_delayed_work(&strp->msg_timer_work);
		strp->skb_head = NULL;
		strp->need_bytes = 0;
		STRP_STATS_INCR(strp->stats.msgs);

		/* Give skb to upper layer */
		strp->cb.rcv_msg(strp, head);

		if (unlikely(strp->paused)) {
			/* Upper layer paused strp */
			break;
		}
	}

	if (cloned_orig)
		kfree_skb(orig_skb);

	STRP_STATS_ADD(strp->stats.bytes, eaten);

	return eaten;
}

int strp_process(struct strparser *strp, struct sk_buff *orig_skb,
		 unsigned int orig_offset, size_t orig_len,
		 size_t max_msg_size, long timeo)
{
	read_descriptor_t desc; /* Dummy arg to strp_recv */

	desc.arg.data = strp;

	return __strp_recv(&desc, orig_skb, orig_offset, orig_len,
			   max_msg_size, timeo);
}
EXPORT_SYMBOL_GPL(strp_process);

static int strp_recv(read_descriptor_t *desc, struct sk_buff *orig_skb,
		     unsigned int orig_offset, size_t orig_len)
{
	struct strparser *strp = (struct strparser *)desc->arg.data;

	return __strp_recv(desc, orig_skb, orig_offset, orig_len,
			   strp->sk->sk_rcvbuf, strp->sk->sk_rcvtimeo);
}

static int default_read_sock_done(struct strparser *strp, int err)
{
	return err;
}

/* Called with lock held on lower socket */
static int strp_read_sock(struct strparser *strp)
{
	struct socket *sock = strp->sk->sk_socket;
	read_descriptor_t desc;

	if (unlikely(!sock || !sock->ops || !sock->ops->read_sock))
		return -EBUSY;

	desc.arg.data = strp;
	desc.error = 0;
	desc.count = 1; /* give more than one skb per call */

	/* sk should be locked here, so okay to do read_sock */
	sock->ops->read_sock(strp->sk, &desc, strp_recv);

	desc.error = strp->cb.read_sock_done(strp, desc.error);

	return desc.error;
}

/* Lower sock lock held */
void strp_data_ready(struct strparser *strp)
{
	if (unlikely(strp->stopped) || strp->paused)
		return;

	/* This check is needed to synchronize with do_strp_work.
	 * do_strp_work acquires a process lock (lock_sock) whereas
	 * the lock held here is bh_lock_sock. The two locks can be
	 * held by different threads at the same time, but bh_lock_sock
	 * allows a thread in BH context to safely check if the process
	 * lock is held. In this case, if the lock is held, queue work.
	 */
	if (sock_owned_by_user_nocheck(strp->sk)) {
		queue_work(strp_wq, &strp->work);
		return;
	}

	if (strp->need_bytes) {
		if (strp_peek_len(strp) < strp->need_bytes)
			return;
	}

	if (strp_read_sock(strp) == -ENOMEM)
		queue_work(strp_wq, &strp->work);
}
EXPORT_SYMBOL_GPL(strp_data_ready);

static void do_strp_work(struct strparser *strp)
{
	/* We need the read lock to synchronize with strp_data_ready. We
	 * need the socket lock for calling strp_read_sock.
	 */
	strp->cb.lock(strp);

	if (unlikely(strp->stopped))
		goto out;

	if (strp->paused)
		goto out;

	if (strp_read_sock(strp) == -ENOMEM)
		queue_work(strp_wq, &strp->work);

out:
	strp->cb.unlock(strp);
}

static void strp_work(struct work_struct *w)
{
	do_strp_work(container_of(w, struct strparser, work));
}

static void strp_msg_timeout(struct work_struct *w)
{
	struct strparser *strp = container_of(w, struct strparser,
					      msg_timer_work.work);

	/* Message assembly timed out */
	STRP_STATS_INCR(strp->stats.msg_timeouts);
	strp->cb.lock(strp);
	strp->cb.abort_parser(strp, -ETIMEDOUT);
	strp->cb.unlock(strp);
}

static void strp_sock_lock(struct strparser *strp)
{
	lock_sock(strp->sk);
}

static void strp_sock_unlock(struct strparser *strp)
{
	release_sock(strp->sk);
}

int strp_init(struct strparser *strp, struct sock *sk,
	      const struct strp_callbacks *cb)
{

	if (!cb || !cb->rcv_msg || !cb->parse_msg)
		return -EINVAL;

	/* The sk (sock) arg determines the mode of the stream parser.
	 *
	 * If the sock is set then the strparser is in receive callback mode.
	 * The upper layer calls strp_data_ready to kick receive processing
	 * and strparser calls the read_sock function on the socket to
	 * get packets.
	 *
	 * If the sock is not set then the strparser is in general mode.
	 * The upper layer calls strp_process for each skb to be parsed.
	 */

	if (!sk) {
		if (!cb->lock || !cb->unlock)
			return -EINVAL;
	}

	memset(strp, 0, sizeof(*strp));

	strp->sk = sk;

	strp->cb.lock = cb->lock ? : strp_sock_lock;
	strp->cb.unlock = cb->unlock ? : strp_sock_unlock;
	strp->cb.rcv_msg = cb->rcv_msg;
	strp->cb.parse_msg = cb->parse_msg;
	strp->cb.read_sock_done = cb->read_sock_done ? : default_read_sock_done;
	strp->cb.abort_parser = cb->abort_parser ? : strp_abort_strp;

	INIT_DELAYED_WORK(&strp->msg_timer_work, strp_msg_timeout);
	INIT_WORK(&strp->work, strp_work);

	return 0;
}
EXPORT_SYMBOL_GPL(strp_init);

/* Sock process lock held (lock_sock) */
void __strp_unpause(struct strparser *strp)
{
	strp->paused = 0;

	if (strp->need_bytes) {
		if (strp_peek_len(strp) < strp->need_bytes)
			return;
	}
	strp_read_sock(strp);
}
EXPORT_SYMBOL_GPL(__strp_unpause);

void strp_unpause(struct strparser *strp)
{
	strp->paused = 0;

	/* Sync setting paused with RX work */
	smp_mb();

	queue_work(strp_wq, &strp->work);
}
EXPORT_SYMBOL_GPL(strp_unpause);

/* strp must already be stopped so that strp_recv will no longer be called.
 * Note that strp_done is not called with the lower socket held.
 */
void strp_done(struct strparser *strp)
{
	WARN_ON(!strp->stopped);

	cancel_delayed_work_sync(&strp->msg_timer_work);
	cancel_work_sync(&strp->work);

	if (strp->skb_head) {
		kfree_skb(strp->skb_head);
		strp->skb_head = NULL;
	}
}
EXPORT_SYMBOL_GPL(strp_done);

void strp_stop(struct strparser *strp)
{
	strp->stopped = 1;
}
EXPORT_SYMBOL_GPL(strp_stop);

void strp_check_rcv(struct strparser *strp)
{
	queue_work(strp_wq, &strp->work);
}
EXPORT_SYMBOL_GPL(strp_check_rcv);

static int __init strp_dev_init(void)
{
	strp_wq = create_singlethread_workqueue("kstrp");
	if (unlikely(!strp_wq))
		return -ENOMEM;

	return 0;
}
device_initcall(strp_dev_init);
Commit	Line	Data
d2912cb1	1	// SPDX-License-Identifier: GPL-2.0-only
43a0c675 TH	2	/*
	3	* Stream Parser
	4	*
	5	* Copyright (c) 2016 Tom Herbert <tom@herbertland.com>
43a0c675 TH	6	*/
	7
	8	#include <linux/bpf.h>
	9	#include <linux/errno.h>
	10	#include <linux/errqueue.h>
	11	#include <linux/file.h>
	12	#include <linux/in.h>
	13	#include <linux/kernel.h>
15253b4a PG	14	#include <linux/export.h>
15253b4a PG	15	#include <linux/init.h>
43a0c675 TH	16	#include <linux/net.h>
	17	#include <linux/netdevice.h>
	18	#include <linux/poll.h>
	19	#include <linux/rculist.h>
	20	#include <linux/skbuff.h>
	21	#include <linux/socket.h>
	22	#include <linux/uaccess.h>
	23	#include <linux/workqueue.h>
	24	#include <net/strparser.h>
	25	#include <net/netns/generic.h>
	26	#include <net/sock.h>
43a0c675 TH	27
	28	static struct workqueue_struct *strp_wq;
	29
bbb03029	30	static inline struct _strp_msg _strp_msg(struct sk_buff skb)
43a0c675	31	{
bbb03029	32	return (struct _strp_msg )((void )skb->cb +
b0e43904	33	offsetof(struct sk_skb_cb, strp));
43a0c675 TH	34	}
	35
	36	/* Lower lock held */
bbb03029	37	static void strp_abort_strp(struct strparser *strp, int err)
43a0c675	38	{
43a0c675 TH	39	/* Unrecoverable error in receive */
43a0c675 TH	40
829385f0	41	cancel_delayed_work(&strp->msg_timer_work);
43a0c675	42
bbb03029	43	if (strp->stopped)
43a0c675 TH	44	return;
43a0c675 TH	45
bbb03029 TH	46	strp->stopped = 1;
	47
	48	if (strp->sk) {
	49	struct sock *sk = strp->sk;
43a0c675	50
bbb03029	51	/* Report an error on the lower socket */
cd00edc1	52	sk->sk_err = -err;
e3ae2365	53	sk_error_report(sk);
bbb03029	54	}
43a0c675 TH	55	}
43a0c675 TH	56
bbb03029	57	static void strp_start_timer(struct strparser *strp, long timeo)
43a0c675	58	{
7c5aba21	59	if (timeo && timeo != LONG_MAX)
829385f0	60	mod_delayed_work(strp_wq, &strp->msg_timer_work, timeo);
43a0c675 TH	61	}
	62
	63	/* Lower lock held */
	64	static void strp_parser_err(struct strparser *strp, int err,
	65	read_descriptor_t *desc)
	66	{
	67	desc->error = err;
bbb03029 TH	68	kfree_skb(strp->skb_head);
bbb03029 TH	69	strp->skb_head = NULL;
43a0c675 TH	70	strp->cb.abort_parser(strp, err);
	71	}
	72
96a59083 TH	73	static inline int strp_peek_len(struct strparser *strp)
96a59083 TH	74	{
bbb03029 TH	75	if (strp->sk) {
	76	struct socket *sock = strp->sk->sk_socket;
	77
	78	return sock->ops->peek_len(sock);
	79	}
	80
	81	/* If we don't have an associated socket there's nothing to peek.
	82	* Return int max to avoid stopping the strparser.
	83	*/
96a59083	84
bbb03029	85	return INT_MAX;
96a59083 TH	86	}
96a59083 TH	87
43a0c675	88	/* Lower socket lock held */
bbb03029 TH	89	static int __strp_recv(read_descriptor_t desc, struct sk_buff orig_skb,
	90	unsigned int orig_offset, size_t orig_len,
	91	size_t max_msg_size, long timeo)
43a0c675 TH	92	{
43a0c675 TH	93	struct strparser strp = (struct strparser )desc->arg.data;
bbb03029	94	struct _strp_msg *stm;
43a0c675 TH	95	struct sk_buff head, skb;
	96	size_t eaten = 0, cand_len;
	97	ssize_t extra;
	98	int err;
	99	bool cloned_orig = false;
	100
bbb03029	101	if (strp->paused)
43a0c675 TH	102	return 0;
43a0c675 TH	103
bbb03029	104	head = strp->skb_head;
43a0c675 TH	105	if (head) {
43a0c675 TH	106	/* Message already in progress */
43a0c675 TH	107	if (unlikely(orig_offset)) {
	108	/* Getting data with a non-zero offset when a message is
	109	* in progress is not expected. If it does happen, we
	110	* need to clone and pull since we can't deal with
	111	* offsets in the skbs for a message expect in the head.
	112	*/
	113	orig_skb = skb_clone(orig_skb, GFP_ATOMIC);
	114	if (!orig_skb) {
bbb03029	115	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	116	desc->error = -ENOMEM;
	117	return 0;
	118	}
	119	if (!pskb_pull(orig_skb, orig_offset)) {
bbb03029	120	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	121	kfree_skb(orig_skb);
	122	desc->error = -ENOMEM;
	123	return 0;
	124	}
	125	cloned_orig = true;
	126	orig_offset = 0;
	127	}
	128
bbb03029	129	if (!strp->skb_nextp) {
43a0c675 TH	130	/* We are going to append to the frags_list of head.
	131	* Need to unshare the frag_list.
	132	*/
4a9c2e37 JK	133	err = skb_unclone(head, GFP_ATOMIC);
	134	if (err) {
	135	STRP_STATS_INCR(strp->stats.mem_fail);
	136	desc->error = err;
	137	return 0;
43a0c675 TH	138	}
	139
	140	if (unlikely(skb_shinfo(head)->frag_list)) {
	141	/* We can't append to an sk_buff that already
	142	* has a frag_list. We create a new head, point
	143	* the frag_list of that to the old head, and
	144	* then are able to use the old head->next for
	145	* appending to the message.
	146	*/
	147	if (WARN_ON(head->next)) {
	148	desc->error = -EINVAL;
	149	return 0;
	150	}
	151
da29e4b4	152	skb = alloc_skb_for_msg(head);
43a0c675	153	if (!skb) {
bbb03029	154	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	155	desc->error = -ENOMEM;
	156	return 0;
	157	}
da29e4b4	158
bbb03029	159	strp->skb_nextp = &head->next;
bbb03029	160	strp->skb_head = skb;
43a0c675 TH	161	head = skb;
43a0c675 TH	162	} else {
bbb03029	163	strp->skb_nextp =
43a0c675 TH	164	&skb_shinfo(head)->frag_list;
	165	}
	166	}
	167	}
	168
	169	while (eaten < orig_len) {
	170	/* Always clone since we will consume something */
	171	skb = skb_clone(orig_skb, GFP_ATOMIC);
	172	if (!skb) {
bbb03029	173	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	174	desc->error = -ENOMEM;
	175	break;
	176	}
	177
	178	cand_len = orig_len - eaten;
	179
bbb03029	180	head = strp->skb_head;
43a0c675 TH	181	if (!head) {
43a0c675 TH	182	head = skb;
bbb03029 TH	183	strp->skb_head = head;
	184	/* Will set skb_nextp on next packet if needed */
	185	strp->skb_nextp = NULL;
	186	stm = _strp_msg(head);
	187	memset(stm, 0, sizeof(*stm));
	188	stm->strp.offset = orig_offset + eaten;
43a0c675	189	} else {
4e485d06	190	/* Unclone if we are appending to an skb that we
43a0c675 TH	191	* already share a frag_list with.
43a0c675 TH	192	*/
4e485d06 VG	193	if (skb_has_frag_list(skb)) {
	194	err = skb_unclone(skb, GFP_ATOMIC);
	195	if (err) {
	196	STRP_STATS_INCR(strp->stats.mem_fail);
	197	desc->error = err;
	198	break;
	199	}
43a0c675 TH	200	}
43a0c675 TH	201
bbb03029 TH	202	stm = _strp_msg(head);
	203	*strp->skb_nextp = skb;
	204	strp->skb_nextp = &skb->next;
43a0c675 TH	205	head->data_len += skb->len;
	206	head->len += skb->len;
	207	head->truesize += skb->truesize;
	208	}
	209
bbb03029	210	if (!stm->strp.full_len) {
43a0c675 TH	211	ssize_t len;
	212
	213	len = (*strp->cb.parse_msg)(strp, head);
	214
	215	if (!len) {
	216	/* Need more header to determine length */
bbb03029	217	if (!stm->accum_len) {
43a0c675	218	/* Start RX timer for new message */
bbb03029	219	strp_start_timer(strp, timeo);
43a0c675	220	}
bbb03029	221	stm->accum_len += cand_len;
43a0c675	222	eaten += cand_len;
bbb03029	223	STRP_STATS_INCR(strp->stats.need_more_hdr);
43a0c675 TH	224	WARN_ON(eaten != orig_len);
	225	break;
	226	} else if (len < 0) {
bbb03029	227	if (len == -ESTRPIPE && stm->accum_len) {
43a0c675	228	len = -ENODATA;
bbb03029	229	strp->unrecov_intr = 1;
43a0c675	230	} else {
bbb03029	231	strp->interrupted = 1;
43a0c675	232	}
6d3a4c40	233	strp_parser_err(strp, len, desc);
43a0c675	234	break;
bbb03029	235	} else if (len > max_msg_size) {
43a0c675	236	/* Message length exceeds maximum allowed */
bbb03029	237	STRP_STATS_INCR(strp->stats.msg_too_big);
43a0c675 TH	238	strp_parser_err(strp, -EMSGSIZE, desc);
	239	break;
	240	} else if (len <= (ssize_t)head->len -
bbb03029	241	skb->len - stm->strp.offset) {
43a0c675 TH	242	/* Length must be into new skb (and also
	243	* greater than zero)
	244	*/
bbb03029	245	STRP_STATS_INCR(strp->stats.bad_hdr_len);
43a0c675 TH	246	strp_parser_err(strp, -EPROTO, desc);
	247	break;
	248	}
	249
bbb03029	250	stm->strp.full_len = len;
43a0c675 TH	251	}
43a0c675 TH	252
bbb03029 TH	253	extra = (ssize_t)(stm->accum_len + cand_len) -
bbb03029 TH	254	stm->strp.full_len;
43a0c675 TH	255
	256	if (extra < 0) {
	257	/* Message not complete yet. */
bbb03029	258	if (stm->strp.full_len - stm->accum_len >
96a59083	259	strp_peek_len(strp)) {
bbb03029 TH	260	/* Don't have the whole message in the socket
bbb03029 TH	261	* buffer. Set strp->need_bytes to wait for
43a0c675 TH	262	* the rest of the message. Also, set "early
43a0c675 TH	263	* eaten" since we've already buffered the skb
96a59083	264	* but don't consume yet per strp_read_sock.
43a0c675 TH	265	*/
43a0c675 TH	266
bbb03029	267	if (!stm->accum_len) {
43a0c675	268	/* Start RX timer for new message */
bbb03029	269	strp_start_timer(strp, timeo);
43a0c675 TH	270	}
43a0c675 TH	271
9d0c75bf	272	stm->accum_len += cand_len;
977c7114	273	eaten += cand_len;
bbb03029 TH	274	strp->need_bytes = stm->strp.full_len -
bbb03029 TH	275	stm->accum_len;
bbb03029	276	STRP_STATS_ADD(strp->stats.bytes, cand_len);
43a0c675 TH	277	desc->count = 0; /* Stop reading socket */
	278	break;
	279	}
bbb03029	280	stm->accum_len += cand_len;
43a0c675 TH	281	eaten += cand_len;
	282	WARN_ON(eaten != orig_len);
	283	break;
	284	}
	285
93e21254	286	/* Positive extra indicates more bytes than needed for the
43a0c675 TH	287	* message
	288	*/
	289
	290	WARN_ON(extra > cand_len);
	291
	292	eaten += (cand_len - extra);
	293
	294	/* Hurray, we have a new message! */
829385f0	295	cancel_delayed_work(&strp->msg_timer_work);
bbb03029	296	strp->skb_head = NULL;
9d0c75bf	297	strp->need_bytes = 0;
bbb03029	298	STRP_STATS_INCR(strp->stats.msgs);
43a0c675 TH	299
	300	/* Give skb to upper layer */
	301	strp->cb.rcv_msg(strp, head);
	302
bbb03029	303	if (unlikely(strp->paused)) {
43a0c675 TH	304	/* Upper layer paused strp */
	305	break;
	306	}
	307	}
	308
	309	if (cloned_orig)
	310	kfree_skb(orig_skb);
	311
bbb03029	312	STRP_STATS_ADD(strp->stats.bytes, eaten);
43a0c675 TH	313
	314	return eaten;
	315	}
	316
bbb03029 TH	317	int strp_process(struct strparser strp, struct sk_buff orig_skb,
	318	unsigned int orig_offset, size_t orig_len,
	319	size_t max_msg_size, long timeo)
	320	{
	321	read_descriptor_t desc; /* Dummy arg to strp_recv */
	322
	323	desc.arg.data = strp;
	324
	325	return __strp_recv(&desc, orig_skb, orig_offset, orig_len,
	326	max_msg_size, timeo);
	327	}
	328	EXPORT_SYMBOL_GPL(strp_process);
	329
	330	static int strp_recv(read_descriptor_t desc, struct sk_buff orig_skb,
	331	unsigned int orig_offset, size_t orig_len)
	332	{
	333	struct strparser strp = (struct strparser )desc->arg.data;
	334
	335	return __strp_recv(desc, orig_skb, orig_offset, orig_len,
	336	strp->sk->sk_rcvbuf, strp->sk->sk_rcvtimeo);
	337	}
	338
43a0c675 TH	339	static int default_read_sock_done(struct strparser *strp, int err)
	340	{
	341	return err;
	342	}
	343
	344	/* Called with lock held on lower socket */
96a59083	345	static int strp_read_sock(struct strparser *strp)
43a0c675	346	{
96a59083	347	struct socket *sock = strp->sk->sk_socket;
43a0c675 TH	348	read_descriptor_t desc;
43a0c675 TH	349
f26de110 JF	350	if (unlikely(!sock \|\| !sock->ops \|\| !sock->ops->read_sock))
	351	return -EBUSY;
	352
43a0c675 TH	353	desc.arg.data = strp;
	354	desc.error = 0;
	355	desc.count = 1; /* give more than one skb per call */
	356
96a59083 TH	357	/* sk should be locked here, so okay to do read_sock */
96a59083 TH	358	sock->ops->read_sock(strp->sk, &desc, strp_recv);
43a0c675 TH	359
	360	desc.error = strp->cb.read_sock_done(strp, desc.error);
	361
	362	return desc.error;
	363	}
	364
	365	/* Lower sock lock held */
96a59083	366	void strp_data_ready(struct strparser *strp)
43a0c675	367	{
456488cd	368	if (unlikely(strp->stopped) \|\| strp->paused)
43a0c675 TH	369	return;
43a0c675 TH	370
bbb03029 TH	371	/* This check is needed to synchronize with do_strp_work.
bbb03029 TH	372	* do_strp_work acquires a process lock (lock_sock) whereas
43a0c675 TH	373	* the lock held here is bh_lock_sock. The two locks can be
	374	* held by different threads at the same time, but bh_lock_sock
	375	* allows a thread in BH context to safely check if the process
	376	* lock is held. In this case, if the lock is held, queue work.
	377	*/
d66fa9ec	378	if (sock_owned_by_user_nocheck(strp->sk)) {
bbb03029	379	queue_work(strp_wq, &strp->work);
43a0c675 TH	380	return;
	381	}
	382
bbb03029	383	if (strp->need_bytes) {
9d0c75bf	384	if (strp_peek_len(strp) < strp->need_bytes)
43a0c675 TH	385	return;
	386	}
	387
96a59083	388	if (strp_read_sock(strp) == -ENOMEM)
bbb03029	389	queue_work(strp_wq, &strp->work);
43a0c675	390	}
96a59083	391	EXPORT_SYMBOL_GPL(strp_data_ready);
43a0c675	392
bbb03029	393	static void do_strp_work(struct strparser *strp)
43a0c675	394	{
96a59083 TH	395	/* We need the read lock to synchronize with strp_data_ready. We
96a59083 TH	396	* need the socket lock for calling strp_read_sock.
43a0c675	397	*/
bbb03029	398	strp->cb.lock(strp);
43a0c675	399
bbb03029	400	if (unlikely(strp->stopped))
43a0c675 TH	401	goto out;
43a0c675 TH	402
bbb03029	403	if (strp->paused)
43a0c675 TH	404	goto out;
43a0c675 TH	405
96a59083	406	if (strp_read_sock(strp) == -ENOMEM)
bbb03029	407	queue_work(strp_wq, &strp->work);
43a0c675 TH	408
43a0c675 TH	409	out:
bbb03029	410	strp->cb.unlock(strp);
43a0c675 TH	411	}
43a0c675 TH	412
bbb03029	413	static void strp_work(struct work_struct *w)
43a0c675	414	{
bbb03029	415	do_strp_work(container_of(w, struct strparser, work));
43a0c675 TH	416	}
43a0c675 TH	417
829385f0	418	static void strp_msg_timeout(struct work_struct *w)
43a0c675	419	{
829385f0 TH	420	struct strparser *strp = container_of(w, struct strparser,
829385f0 TH	421	msg_timer_work.work);
43a0c675 TH	422
43a0c675 TH	423	/* Message assembly timed out */
bbb03029 TH	424	STRP_STATS_INCR(strp->stats.msg_timeouts);
bbb03029 TH	425	strp->cb.lock(strp);
cd00edc1	426	strp->cb.abort_parser(strp, -ETIMEDOUT);
bbb03029 TH	427	strp->cb.unlock(strp);
	428	}
	429
	430	static void strp_sock_lock(struct strparser *strp)
	431	{
	432	lock_sock(strp->sk);
	433	}
	434
	435	static void strp_sock_unlock(struct strparser *strp)
	436	{
43a0c675 TH	437	release_sock(strp->sk);
	438	}
	439
bbb03029	440	int strp_init(struct strparser strp, struct sock sk,
3fd87127	441	const struct strp_callbacks *cb)
43a0c675	442	{
96a59083	443
43a0c675 TH	444	if (!cb \|\| !cb->rcv_msg \|\| !cb->parse_msg)
	445	return -EINVAL;
	446
bbb03029 TH	447	/* The sk (sock) arg determines the mode of the stream parser.
	448	*
	449	* If the sock is set then the strparser is in receive callback mode.
	450	* The upper layer calls strp_data_ready to kick receive processing
	451	* and strparser calls the read_sock function on the socket to
	452	* get packets.
	453	*
	454	* If the sock is not set then the strparser is in general mode.
	455	* The upper layer calls strp_process for each skb to be parsed.
	456	*/
96a59083	457
f26de110	458	if (!sk) {
bbb03029 TH	459	if (!cb->lock \|\| !cb->unlock)
	460	return -EINVAL;
	461	}
43a0c675	462
bbb03029	463	memset(strp, 0, sizeof(*strp));
43a0c675	464
bbb03029	465	strp->sk = sk;
43a0c675	466
bbb03029 TH	467	strp->cb.lock = cb->lock ? : strp_sock_lock;
bbb03029 TH	468	strp->cb.unlock = cb->unlock ? : strp_sock_unlock;
43a0c675 TH	469	strp->cb.rcv_msg = cb->rcv_msg;
	470	strp->cb.parse_msg = cb->parse_msg;
	471	strp->cb.read_sock_done = cb->read_sock_done ? : default_read_sock_done;
bbb03029 TH	472	strp->cb.abort_parser = cb->abort_parser ? : strp_abort_strp;
bbb03029 TH	473
829385f0	474	INIT_DELAYED_WORK(&strp->msg_timer_work, strp_msg_timeout);
bbb03029	475	INIT_WORK(&strp->work, strp_work);
43a0c675 TH	476
	477	return 0;
	478	}
	479	EXPORT_SYMBOL_GPL(strp_init);
	480
7170e604 DRK	481	/* Sock process lock held (lock_sock) */
	482	void __strp_unpause(struct strparser *strp)
	483	{
	484	strp->paused = 0;
	485
	486	if (strp->need_bytes) {
	487	if (strp_peek_len(strp) < strp->need_bytes)
	488	return;
	489	}
	490	strp_read_sock(strp);
	491	}
	492	EXPORT_SYMBOL_GPL(__strp_unpause);
	493
cff6a334 TH	494	void strp_unpause(struct strparser *strp)
cff6a334 TH	495	{
bbb03029	496	strp->paused = 0;
cff6a334	497
bbb03029	498	/* Sync setting paused with RX work */
cff6a334 TH	499	smp_mb();
cff6a334 TH	500
bbb03029	501	queue_work(strp_wq, &strp->work);
cff6a334 TH	502	}
	503	EXPORT_SYMBOL_GPL(strp_unpause);
	504
96a59083	505	/* strp must already be stopped so that strp_recv will no longer be called.
43a0c675 TH	506	* Note that strp_done is not called with the lower socket held.
	507	*/
	508	void strp_done(struct strparser *strp)
	509	{
bbb03029	510	WARN_ON(!strp->stopped);
43a0c675	511
829385f0	512	cancel_delayed_work_sync(&strp->msg_timer_work);
bbb03029	513	cancel_work_sync(&strp->work);
43a0c675	514
bbb03029 TH	515	if (strp->skb_head) {
	516	kfree_skb(strp->skb_head);
	517	strp->skb_head = NULL;
43a0c675 TH	518	}
	519	}
	520	EXPORT_SYMBOL_GPL(strp_done);
	521
	522	void strp_stop(struct strparser *strp)
	523	{
bbb03029	524	strp->stopped = 1;
43a0c675 TH	525	}
	526	EXPORT_SYMBOL_GPL(strp_stop);
	527
	528	void strp_check_rcv(struct strparser *strp)
	529	{
bbb03029	530	queue_work(strp_wq, &strp->work);
43a0c675 TH	531	}
	532	EXPORT_SYMBOL_GPL(strp_check_rcv);
	533
15253b4a	534	static int __init strp_dev_init(void)
43a0c675 TH	535	{
43a0c675 TH	536	strp_wq = create_singlethread_workqueue("kstrp");
228cd2db KL	537	if (unlikely(!strp_wq))
228cd2db KL	538	return -ENOMEM;
43a0c675 TH	539
	540	return 0;
	541	}
15253b4a	542	device_initcall(strp_dev_init);