[linux-2.6-block.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;

struct _strp_msg {
	/* Internal cb structure. struct strp_msg must be first for passing
	 * to upper layer.
	 */
	struct strp_msg strp;
	int accum_len;
};

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

/* 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->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 TH	30	struct _strp_msg {
bbb03029 TH	31	/* Internal cb structure. struct strp_msg must be first for passing
43a0c675 TH	32	* to upper layer.
43a0c675 TH	33	*/
bbb03029	34	struct strp_msg strp;
43a0c675	35	int accum_len;
43a0c675 TH	36	};
43a0c675 TH	37
bbb03029	38	static inline struct _strp_msg _strp_msg(struct sk_buff skb)
43a0c675	39	{
bbb03029	40	return (struct _strp_msg )((void )skb->cb +
43a0c675 TH	41	offsetof(struct qdisc_skb_cb, data));
	42	}
	43
	44	/* Lower lock held */
bbb03029	45	static void strp_abort_strp(struct strparser *strp, int err)
43a0c675	46	{
43a0c675 TH	47	/* Unrecoverable error in receive */
43a0c675 TH	48
829385f0	49	cancel_delayed_work(&strp->msg_timer_work);
43a0c675	50
bbb03029	51	if (strp->stopped)
43a0c675 TH	52	return;
43a0c675 TH	53
bbb03029 TH	54	strp->stopped = 1;
	55
	56	if (strp->sk) {
	57	struct sock *sk = strp->sk;
43a0c675	58
bbb03029	59	/* Report an error on the lower socket */
cd00edc1	60	sk->sk_err = -err;
bbb03029 TH	61	sk->sk_error_report(sk);
bbb03029 TH	62	}
43a0c675 TH	63	}
43a0c675 TH	64
bbb03029	65	static void strp_start_timer(struct strparser *strp, long timeo)
43a0c675	66	{
7c5aba21	67	if (timeo && timeo != LONG_MAX)
829385f0	68	mod_delayed_work(strp_wq, &strp->msg_timer_work, timeo);
43a0c675 TH	69	}
	70
	71	/* Lower lock held */
	72	static void strp_parser_err(struct strparser *strp, int err,
	73	read_descriptor_t *desc)
	74	{
	75	desc->error = err;
bbb03029 TH	76	kfree_skb(strp->skb_head);
bbb03029 TH	77	strp->skb_head = NULL;
43a0c675 TH	78	strp->cb.abort_parser(strp, err);
	79	}
	80
96a59083 TH	81	static inline int strp_peek_len(struct strparser *strp)
96a59083 TH	82	{
bbb03029 TH	83	if (strp->sk) {
	84	struct socket *sock = strp->sk->sk_socket;
	85
	86	return sock->ops->peek_len(sock);
	87	}
	88
	89	/* If we don't have an associated socket there's nothing to peek.
	90	* Return int max to avoid stopping the strparser.
	91	*/
96a59083	92
bbb03029	93	return INT_MAX;
96a59083 TH	94	}
96a59083 TH	95
43a0c675	96	/* Lower socket lock held */
bbb03029 TH	97	static int __strp_recv(read_descriptor_t desc, struct sk_buff orig_skb,
	98	unsigned int orig_offset, size_t orig_len,
	99	size_t max_msg_size, long timeo)
43a0c675 TH	100	{
43a0c675 TH	101	struct strparser strp = (struct strparser )desc->arg.data;
bbb03029	102	struct _strp_msg *stm;
43a0c675 TH	103	struct sk_buff head, skb;
	104	size_t eaten = 0, cand_len;
	105	ssize_t extra;
	106	int err;
	107	bool cloned_orig = false;
	108
bbb03029	109	if (strp->paused)
43a0c675 TH	110	return 0;
43a0c675 TH	111
bbb03029	112	head = strp->skb_head;
43a0c675 TH	113	if (head) {
43a0c675 TH	114	/* Message already in progress */
43a0c675 TH	115	if (unlikely(orig_offset)) {
	116	/* Getting data with a non-zero offset when a message is
	117	* in progress is not expected. If it does happen, we
	118	* need to clone and pull since we can't deal with
	119	* offsets in the skbs for a message expect in the head.
	120	*/
	121	orig_skb = skb_clone(orig_skb, GFP_ATOMIC);
	122	if (!orig_skb) {
bbb03029	123	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	124	desc->error = -ENOMEM;
	125	return 0;
	126	}
	127	if (!pskb_pull(orig_skb, orig_offset)) {
bbb03029	128	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	129	kfree_skb(orig_skb);
	130	desc->error = -ENOMEM;
	131	return 0;
	132	}
	133	cloned_orig = true;
	134	orig_offset = 0;
	135	}
	136
bbb03029	137	if (!strp->skb_nextp) {
43a0c675 TH	138	/* We are going to append to the frags_list of head.
	139	* Need to unshare the frag_list.
	140	*/
4a9c2e37 JK	141	err = skb_unclone(head, GFP_ATOMIC);
	142	if (err) {
	143	STRP_STATS_INCR(strp->stats.mem_fail);
	144	desc->error = err;
	145	return 0;
43a0c675 TH	146	}
	147
	148	if (unlikely(skb_shinfo(head)->frag_list)) {
	149	/* We can't append to an sk_buff that already
	150	* has a frag_list. We create a new head, point
	151	* the frag_list of that to the old head, and
	152	* then are able to use the old head->next for
	153	* appending to the message.
	154	*/
	155	if (WARN_ON(head->next)) {
	156	desc->error = -EINVAL;
	157	return 0;
	158	}
	159
da29e4b4	160	skb = alloc_skb_for_msg(head);
43a0c675	161	if (!skb) {
bbb03029	162	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	163	desc->error = -ENOMEM;
	164	return 0;
	165	}
da29e4b4	166
bbb03029	167	strp->skb_nextp = &head->next;
bbb03029	168	strp->skb_head = skb;
43a0c675 TH	169	head = skb;
43a0c675 TH	170	} else {
bbb03029	171	strp->skb_nextp =
43a0c675 TH	172	&skb_shinfo(head)->frag_list;
	173	}
	174	}
	175	}
	176
	177	while (eaten < orig_len) {
	178	/* Always clone since we will consume something */
	179	skb = skb_clone(orig_skb, GFP_ATOMIC);
	180	if (!skb) {
bbb03029	181	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	182	desc->error = -ENOMEM;
	183	break;
	184	}
	185
	186	cand_len = orig_len - eaten;
	187
bbb03029	188	head = strp->skb_head;
43a0c675 TH	189	if (!head) {
43a0c675 TH	190	head = skb;
bbb03029 TH	191	strp->skb_head = head;
	192	/* Will set skb_nextp on next packet if needed */
	193	strp->skb_nextp = NULL;
	194	stm = _strp_msg(head);
	195	memset(stm, 0, sizeof(*stm));
	196	stm->strp.offset = orig_offset + eaten;
43a0c675	197	} else {
4e485d06	198	/* Unclone if we are appending to an skb that we
43a0c675 TH	199	* already share a frag_list with.
43a0c675 TH	200	*/
4e485d06 VG	201	if (skb_has_frag_list(skb)) {
	202	err = skb_unclone(skb, GFP_ATOMIC);
	203	if (err) {
	204	STRP_STATS_INCR(strp->stats.mem_fail);
	205	desc->error = err;
	206	break;
	207	}
43a0c675 TH	208	}
43a0c675 TH	209
bbb03029 TH	210	stm = _strp_msg(head);
	211	*strp->skb_nextp = skb;
	212	strp->skb_nextp = &skb->next;
43a0c675 TH	213	head->data_len += skb->len;
	214	head->len += skb->len;
	215	head->truesize += skb->truesize;
	216	}
	217
bbb03029	218	if (!stm->strp.full_len) {
43a0c675 TH	219	ssize_t len;
	220
	221	len = (*strp->cb.parse_msg)(strp, head);
	222
	223	if (!len) {
	224	/* Need more header to determine length */
bbb03029	225	if (!stm->accum_len) {
43a0c675	226	/* Start RX timer for new message */
bbb03029	227	strp_start_timer(strp, timeo);
43a0c675	228	}
bbb03029	229	stm->accum_len += cand_len;
43a0c675	230	eaten += cand_len;
bbb03029	231	STRP_STATS_INCR(strp->stats.need_more_hdr);
43a0c675 TH	232	WARN_ON(eaten != orig_len);
	233	break;
	234	} else if (len < 0) {
bbb03029	235	if (len == -ESTRPIPE && stm->accum_len) {
43a0c675	236	len = -ENODATA;
bbb03029	237	strp->unrecov_intr = 1;
43a0c675	238	} else {
bbb03029	239	strp->interrupted = 1;
43a0c675	240	}
6d3a4c40	241	strp_parser_err(strp, len, desc);
43a0c675	242	break;
bbb03029	243	} else if (len > max_msg_size) {
43a0c675	244	/* Message length exceeds maximum allowed */
bbb03029	245	STRP_STATS_INCR(strp->stats.msg_too_big);
43a0c675 TH	246	strp_parser_err(strp, -EMSGSIZE, desc);
	247	break;
	248	} else if (len <= (ssize_t)head->len -
bbb03029	249	skb->len - stm->strp.offset) {
43a0c675 TH	250	/* Length must be into new skb (and also
	251	* greater than zero)
	252	*/
bbb03029	253	STRP_STATS_INCR(strp->stats.bad_hdr_len);
43a0c675 TH	254	strp_parser_err(strp, -EPROTO, desc);
	255	break;
	256	}
	257
bbb03029	258	stm->strp.full_len = len;
43a0c675 TH	259	}
43a0c675 TH	260
bbb03029 TH	261	extra = (ssize_t)(stm->accum_len + cand_len) -
bbb03029 TH	262	stm->strp.full_len;
43a0c675 TH	263
	264	if (extra < 0) {
	265	/* Message not complete yet. */
bbb03029	266	if (stm->strp.full_len - stm->accum_len >
96a59083	267	strp_peek_len(strp)) {
bbb03029 TH	268	/* Don't have the whole message in the socket
bbb03029 TH	269	* buffer. Set strp->need_bytes to wait for
43a0c675 TH	270	* the rest of the message. Also, set "early
43a0c675 TH	271	* eaten" since we've already buffered the skb
96a59083	272	* but don't consume yet per strp_read_sock.
43a0c675 TH	273	*/
43a0c675 TH	274
bbb03029	275	if (!stm->accum_len) {
43a0c675	276	/* Start RX timer for new message */
bbb03029	277	strp_start_timer(strp, timeo);
43a0c675 TH	278	}
43a0c675 TH	279
9d0c75bf	280	stm->accum_len += cand_len;
977c7114	281	eaten += cand_len;
bbb03029 TH	282	strp->need_bytes = stm->strp.full_len -
bbb03029 TH	283	stm->accum_len;
bbb03029	284	STRP_STATS_ADD(strp->stats.bytes, cand_len);
43a0c675 TH	285	desc->count = 0; /* Stop reading socket */
	286	break;
	287	}
bbb03029	288	stm->accum_len += cand_len;
43a0c675 TH	289	eaten += cand_len;
	290	WARN_ON(eaten != orig_len);
	291	break;
	292	}
	293
93e21254	294	/* Positive extra indicates more bytes than needed for the
43a0c675 TH	295	* message
	296	*/
	297
	298	WARN_ON(extra > cand_len);
	299
	300	eaten += (cand_len - extra);
	301
	302	/* Hurray, we have a new message! */
829385f0	303	cancel_delayed_work(&strp->msg_timer_work);
bbb03029	304	strp->skb_head = NULL;
9d0c75bf	305	strp->need_bytes = 0;
bbb03029	306	STRP_STATS_INCR(strp->stats.msgs);
43a0c675 TH	307
	308	/* Give skb to upper layer */
	309	strp->cb.rcv_msg(strp, head);
	310
bbb03029	311	if (unlikely(strp->paused)) {
43a0c675 TH	312	/* Upper layer paused strp */
	313	break;
	314	}
	315	}
	316
	317	if (cloned_orig)
	318	kfree_skb(orig_skb);
	319
bbb03029	320	STRP_STATS_ADD(strp->stats.bytes, eaten);
43a0c675 TH	321
	322	return eaten;
	323	}
	324
bbb03029 TH	325	int strp_process(struct strparser strp, struct sk_buff orig_skb,
	326	unsigned int orig_offset, size_t orig_len,
	327	size_t max_msg_size, long timeo)
	328	{
	329	read_descriptor_t desc; /* Dummy arg to strp_recv */
	330
	331	desc.arg.data = strp;
	332
	333	return __strp_recv(&desc, orig_skb, orig_offset, orig_len,
	334	max_msg_size, timeo);
	335	}
	336	EXPORT_SYMBOL_GPL(strp_process);
	337
	338	static int strp_recv(read_descriptor_t desc, struct sk_buff orig_skb,
	339	unsigned int orig_offset, size_t orig_len)
	340	{
	341	struct strparser strp = (struct strparser )desc->arg.data;
	342
	343	return __strp_recv(desc, orig_skb, orig_offset, orig_len,
	344	strp->sk->sk_rcvbuf, strp->sk->sk_rcvtimeo);
	345	}
	346
43a0c675 TH	347	static int default_read_sock_done(struct strparser *strp, int err)
	348	{
	349	return err;
	350	}
	351
	352	/* Called with lock held on lower socket */
96a59083	353	static int strp_read_sock(struct strparser *strp)
43a0c675	354	{
96a59083	355	struct socket *sock = strp->sk->sk_socket;
43a0c675 TH	356	read_descriptor_t desc;
43a0c675 TH	357
f26de110 JF	358	if (unlikely(!sock \|\| !sock->ops \|\| !sock->ops->read_sock))
	359	return -EBUSY;
	360
43a0c675 TH	361	desc.arg.data = strp;
	362	desc.error = 0;
	363	desc.count = 1; /* give more than one skb per call */
	364
96a59083 TH	365	/* sk should be locked here, so okay to do read_sock */
96a59083 TH	366	sock->ops->read_sock(strp->sk, &desc, strp_recv);
43a0c675 TH	367
	368	desc.error = strp->cb.read_sock_done(strp, desc.error);
	369
	370	return desc.error;
	371	}
	372
	373	/* Lower sock lock held */
96a59083	374	void strp_data_ready(struct strparser *strp)
43a0c675	375	{
456488cd	376	if (unlikely(strp->stopped) \|\| strp->paused)
43a0c675 TH	377	return;
43a0c675 TH	378
bbb03029 TH	379	/* This check is needed to synchronize with do_strp_work.
bbb03029 TH	380	* do_strp_work acquires a process lock (lock_sock) whereas
43a0c675 TH	381	* the lock held here is bh_lock_sock. The two locks can be
	382	* held by different threads at the same time, but bh_lock_sock
	383	* allows a thread in BH context to safely check if the process
	384	* lock is held. In this case, if the lock is held, queue work.
	385	*/
d66fa9ec	386	if (sock_owned_by_user_nocheck(strp->sk)) {
bbb03029	387	queue_work(strp_wq, &strp->work);
43a0c675 TH	388	return;
	389	}
	390
bbb03029	391	if (strp->need_bytes) {
9d0c75bf	392	if (strp_peek_len(strp) < strp->need_bytes)
43a0c675 TH	393	return;
	394	}
	395
96a59083	396	if (strp_read_sock(strp) == -ENOMEM)
bbb03029	397	queue_work(strp_wq, &strp->work);
43a0c675	398	}
96a59083	399	EXPORT_SYMBOL_GPL(strp_data_ready);
43a0c675	400
bbb03029	401	static void do_strp_work(struct strparser *strp)
43a0c675	402	{
96a59083 TH	403	/* We need the read lock to synchronize with strp_data_ready. We
96a59083 TH	404	* need the socket lock for calling strp_read_sock.
43a0c675	405	*/
bbb03029	406	strp->cb.lock(strp);
43a0c675	407
bbb03029	408	if (unlikely(strp->stopped))
43a0c675 TH	409	goto out;
43a0c675 TH	410
bbb03029	411	if (strp->paused)
43a0c675 TH	412	goto out;
43a0c675 TH	413
96a59083	414	if (strp_read_sock(strp) == -ENOMEM)
bbb03029	415	queue_work(strp_wq, &strp->work);
43a0c675 TH	416
43a0c675 TH	417	out:
bbb03029	418	strp->cb.unlock(strp);
43a0c675 TH	419	}
43a0c675 TH	420
bbb03029	421	static void strp_work(struct work_struct *w)
43a0c675	422	{
bbb03029	423	do_strp_work(container_of(w, struct strparser, work));
43a0c675 TH	424	}
43a0c675 TH	425
829385f0	426	static void strp_msg_timeout(struct work_struct *w)
43a0c675	427	{
829385f0 TH	428	struct strparser *strp = container_of(w, struct strparser,
829385f0 TH	429	msg_timer_work.work);
43a0c675 TH	430
43a0c675 TH	431	/* Message assembly timed out */
bbb03029 TH	432	STRP_STATS_INCR(strp->stats.msg_timeouts);
bbb03029 TH	433	strp->cb.lock(strp);
cd00edc1	434	strp->cb.abort_parser(strp, -ETIMEDOUT);
bbb03029 TH	435	strp->cb.unlock(strp);
	436	}
	437
	438	static void strp_sock_lock(struct strparser *strp)
	439	{
	440	lock_sock(strp->sk);
	441	}
	442
	443	static void strp_sock_unlock(struct strparser *strp)
	444	{
43a0c675 TH	445	release_sock(strp->sk);
	446	}
	447
bbb03029	448	int strp_init(struct strparser strp, struct sock sk,
3fd87127	449	const struct strp_callbacks *cb)
43a0c675	450	{
96a59083	451
43a0c675 TH	452	if (!cb \|\| !cb->rcv_msg \|\| !cb->parse_msg)
	453	return -EINVAL;
	454
bbb03029 TH	455	/* The sk (sock) arg determines the mode of the stream parser.
	456	*
	457	* If the sock is set then the strparser is in receive callback mode.
	458	* The upper layer calls strp_data_ready to kick receive processing
	459	* and strparser calls the read_sock function on the socket to
	460	* get packets.
	461	*
	462	* If the sock is not set then the strparser is in general mode.
	463	* The upper layer calls strp_process for each skb to be parsed.
	464	*/
96a59083	465
f26de110	466	if (!sk) {
bbb03029 TH	467	if (!cb->lock \|\| !cb->unlock)
	468	return -EINVAL;
	469	}
43a0c675	470
bbb03029	471	memset(strp, 0, sizeof(*strp));
43a0c675	472
bbb03029	473	strp->sk = sk;
43a0c675	474
bbb03029 TH	475	strp->cb.lock = cb->lock ? : strp_sock_lock;
bbb03029 TH	476	strp->cb.unlock = cb->unlock ? : strp_sock_unlock;
43a0c675 TH	477	strp->cb.rcv_msg = cb->rcv_msg;
	478	strp->cb.parse_msg = cb->parse_msg;
	479	strp->cb.read_sock_done = cb->read_sock_done ? : default_read_sock_done;
bbb03029 TH	480	strp->cb.abort_parser = cb->abort_parser ? : strp_abort_strp;
bbb03029 TH	481
829385f0	482	INIT_DELAYED_WORK(&strp->msg_timer_work, strp_msg_timeout);
bbb03029	483	INIT_WORK(&strp->work, strp_work);
43a0c675 TH	484
	485	return 0;
	486	}
	487	EXPORT_SYMBOL_GPL(strp_init);
	488
7170e604 DRK	489	/* Sock process lock held (lock_sock) */
	490	void __strp_unpause(struct strparser *strp)
	491	{
	492	strp->paused = 0;
	493
	494	if (strp->need_bytes) {
	495	if (strp_peek_len(strp) < strp->need_bytes)
	496	return;
	497	}
	498	strp_read_sock(strp);
	499	}
	500	EXPORT_SYMBOL_GPL(__strp_unpause);
	501
cff6a334 TH	502	void strp_unpause(struct strparser *strp)
cff6a334 TH	503	{
bbb03029	504	strp->paused = 0;
cff6a334	505
bbb03029	506	/* Sync setting paused with RX work */
cff6a334 TH	507	smp_mb();
cff6a334 TH	508
bbb03029	509	queue_work(strp_wq, &strp->work);
cff6a334 TH	510	}
	511	EXPORT_SYMBOL_GPL(strp_unpause);
	512
96a59083	513	/* strp must already be stopped so that strp_recv will no longer be called.
43a0c675 TH	514	* Note that strp_done is not called with the lower socket held.
	515	*/
	516	void strp_done(struct strparser *strp)
	517	{
bbb03029	518	WARN_ON(!strp->stopped);
43a0c675	519
829385f0	520	cancel_delayed_work_sync(&strp->msg_timer_work);
bbb03029	521	cancel_work_sync(&strp->work);
43a0c675	522
bbb03029 TH	523	if (strp->skb_head) {
	524	kfree_skb(strp->skb_head);
	525	strp->skb_head = NULL;
43a0c675 TH	526	}
	527	}
	528	EXPORT_SYMBOL_GPL(strp_done);
	529
	530	void strp_stop(struct strparser *strp)
	531	{
bbb03029	532	strp->stopped = 1;
43a0c675 TH	533	}
	534	EXPORT_SYMBOL_GPL(strp_stop);
	535
	536	void strp_check_rcv(struct strparser *strp)
	537	{
bbb03029	538	queue_work(strp_wq, &strp->work);
43a0c675 TH	539	}
	540	EXPORT_SYMBOL_GPL(strp_check_rcv);
	541
15253b4a	542	static int __init strp_dev_init(void)
43a0c675 TH	543	{
43a0c675 TH	544	strp_wq = create_singlethread_workqueue("kstrp");
228cd2db KL	545	if (unlikely(!strp_wq))
228cd2db KL	546	return -ENOMEM;
43a0c675 TH	547
	548	return 0;
	549	}
15253b4a	550	device_initcall(strp_dev_init);