[linux-2.6-block.git] / net / strparser / strparser.c

/*
 * Stream Parser
 *
 * Copyright (c) 2016 Tom Herbert <tom@herbertland.com>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2
 * as published by the Free Software Foundation.
 */

#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/module.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;
	int early_eaten;
};

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)
		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 */

		stm = _strp_msg(head);
		if (unlikely(stm->early_eaten)) {
			/* Already some number of bytes on the receive sock
			 * data saved in skb_head, just indicate they
			 * are consumed.
			 */
			eaten = orig_len <= stm->early_eaten ?
				orig_len : stm->early_eaten;
			stm->early_eaten -= eaten;

			return eaten;
		}

		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(0, GFP_ATOMIC);
				if (!skb) {
					STRP_STATS_INCR(strp->stats.mem_fail);
					desc->error = -ENOMEM;
					return 0;
				}
				skb->len = head->len;
				skb->data_len = head->len;
				skb->truesize = head->truesize;
				*_strp_msg(skb) = *_strp_msg(head);
				strp->skb_nextp = &head->next;
				skb_shinfo(skb)->frag_list = head;
				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 since we may be appending to an skb that we
			 * already share a frag_list with.
			 */
			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);
				}

				strp->need_bytes = stm->strp.full_len -
						       stm->accum_len;
				stm->accum_len += cand_len;
				stm->early_eaten = cand_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 ore 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_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))
		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->paused)
		return;

	if (strp->need_bytes) {
		if (strp_peek_len(strp) >= strp->need_bytes)
			strp->need_bytes = 0;
		else
			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)
{
	read_descriptor_t rd_desc;

	/* 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;

	rd_desc.arg.data = strp;

	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);

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_mod_init(void)
{
	strp_wq = create_singlethread_workqueue("kstrp");

	return 0;
}

static void __exit strp_mod_exit(void)
{
	destroy_workqueue(strp_wq);
}
module_init(strp_mod_init);
module_exit(strp_mod_exit);
MODULE_LICENSE("GPL");
Commit	Line	Data
43a0c675 TH	1	/*
	2	* Stream Parser
	3	*
	4	* Copyright (c) 2016 Tom Herbert <tom@herbertland.com>
	5	*
	6	* This program is free software; you can redistribute it and/or modify
	7	* it under the terms of the GNU General Public License version 2
	8	* as published by the Free Software Foundation.
	9	*/
	10
	11	#include <linux/bpf.h>
	12	#include <linux/errno.h>
	13	#include <linux/errqueue.h>
	14	#include <linux/file.h>
	15	#include <linux/in.h>
	16	#include <linux/kernel.h>
	17	#include <linux/module.h>
	18	#include <linux/net.h>
	19	#include <linux/netdevice.h>
	20	#include <linux/poll.h>
	21	#include <linux/rculist.h>
	22	#include <linux/skbuff.h>
	23	#include <linux/socket.h>
	24	#include <linux/uaccess.h>
	25	#include <linux/workqueue.h>
	26	#include <net/strparser.h>
	27	#include <net/netns/generic.h>
	28	#include <net/sock.h>
43a0c675 TH	29
	30	static struct workqueue_struct *strp_wq;
	31
bbb03029 TH	32	struct _strp_msg {
bbb03029 TH	33	/* Internal cb structure. struct strp_msg must be first for passing
43a0c675 TH	34	* to upper layer.
43a0c675 TH	35	*/
bbb03029	36	struct strp_msg strp;
43a0c675 TH	37	int accum_len;
	38	int early_eaten;
	39	};
	40
bbb03029	41	static inline struct _strp_msg _strp_msg(struct sk_buff skb)
43a0c675	42	{
bbb03029	43	return (struct _strp_msg )((void )skb->cb +
43a0c675 TH	44	offsetof(struct qdisc_skb_cb, data));
	45	}
	46
	47	/* Lower lock held */
bbb03029	48	static void strp_abort_strp(struct strparser *strp, int err)
43a0c675	49	{
43a0c675 TH	50	/* Unrecoverable error in receive */
43a0c675 TH	51
829385f0	52	cancel_delayed_work(&strp->msg_timer_work);
43a0c675	53
bbb03029	54	if (strp->stopped)
43a0c675 TH	55	return;
43a0c675 TH	56
bbb03029 TH	57	strp->stopped = 1;
	58
	59	if (strp->sk) {
	60	struct sock *sk = strp->sk;
43a0c675	61
bbb03029 TH	62	/* Report an error on the lower socket */
	63	sk->sk_err = err;
	64	sk->sk_error_report(sk);
	65	}
43a0c675 TH	66	}
43a0c675 TH	67
bbb03029	68	static void strp_start_timer(struct strparser *strp, long timeo)
43a0c675	69	{
bbb03029	70	if (timeo)
829385f0	71	mod_delayed_work(strp_wq, &strp->msg_timer_work, timeo);
43a0c675 TH	72	}
	73
	74	/* Lower lock held */
	75	static void strp_parser_err(struct strparser *strp, int err,
	76	read_descriptor_t *desc)
	77	{
	78	desc->error = err;
bbb03029 TH	79	kfree_skb(strp->skb_head);
bbb03029 TH	80	strp->skb_head = NULL;
43a0c675 TH	81	strp->cb.abort_parser(strp, err);
	82	}
	83
96a59083 TH	84	static inline int strp_peek_len(struct strparser *strp)
96a59083 TH	85	{
bbb03029 TH	86	if (strp->sk) {
	87	struct socket *sock = strp->sk->sk_socket;
	88
	89	return sock->ops->peek_len(sock);
	90	}
	91
	92	/* If we don't have an associated socket there's nothing to peek.
	93	* Return int max to avoid stopping the strparser.
	94	*/
96a59083	95
bbb03029	96	return INT_MAX;
96a59083 TH	97	}
96a59083 TH	98
43a0c675	99	/* Lower socket lock held */
bbb03029 TH	100	static int __strp_recv(read_descriptor_t desc, struct sk_buff orig_skb,
	101	unsigned int orig_offset, size_t orig_len,
	102	size_t max_msg_size, long timeo)
43a0c675 TH	103	{
43a0c675 TH	104	struct strparser strp = (struct strparser )desc->arg.data;
bbb03029	105	struct _strp_msg *stm;
43a0c675 TH	106	struct sk_buff head, skb;
	107	size_t eaten = 0, cand_len;
	108	ssize_t extra;
	109	int err;
	110	bool cloned_orig = false;
	111
bbb03029	112	if (strp->paused)
43a0c675 TH	113	return 0;
43a0c675 TH	114
bbb03029	115	head = strp->skb_head;
43a0c675 TH	116	if (head) {
	117	/* Message already in progress */
	118
bbb03029 TH	119	stm = _strp_msg(head);
bbb03029 TH	120	if (unlikely(stm->early_eaten)) {
43a0c675	121	/* Already some number of bytes on the receive sock
bbb03029	122	* data saved in skb_head, just indicate they
43a0c675 TH	123	* are consumed.
43a0c675 TH	124	*/
bbb03029 TH	125	eaten = orig_len <= stm->early_eaten ?
	126	orig_len : stm->early_eaten;
	127	stm->early_eaten -= eaten;
43a0c675 TH	128
	129	return eaten;
	130	}
	131
	132	if (unlikely(orig_offset)) {
	133	/* Getting data with a non-zero offset when a message is
	134	* in progress is not expected. If it does happen, we
	135	* need to clone and pull since we can't deal with
	136	* offsets in the skbs for a message expect in the head.
	137	*/
	138	orig_skb = skb_clone(orig_skb, GFP_ATOMIC);
	139	if (!orig_skb) {
bbb03029	140	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	141	desc->error = -ENOMEM;
	142	return 0;
	143	}
	144	if (!pskb_pull(orig_skb, orig_offset)) {
bbb03029	145	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	146	kfree_skb(orig_skb);
	147	desc->error = -ENOMEM;
	148	return 0;
	149	}
	150	cloned_orig = true;
	151	orig_offset = 0;
	152	}
	153
bbb03029	154	if (!strp->skb_nextp) {
43a0c675 TH	155	/* We are going to append to the frags_list of head.
	156	* Need to unshare the frag_list.
	157	*/
	158	err = skb_unclone(head, GFP_ATOMIC);
	159	if (err) {
bbb03029	160	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	161	desc->error = err;
	162	return 0;
	163	}
	164
	165	if (unlikely(skb_shinfo(head)->frag_list)) {
	166	/* We can't append to an sk_buff that already
	167	* has a frag_list. We create a new head, point
	168	* the frag_list of that to the old head, and
	169	* then are able to use the old head->next for
	170	* appending to the message.
	171	*/
	172	if (WARN_ON(head->next)) {
	173	desc->error = -EINVAL;
	174	return 0;
	175	}
	176
	177	skb = alloc_skb(0, GFP_ATOMIC);
	178	if (!skb) {
bbb03029	179	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	180	desc->error = -ENOMEM;
	181	return 0;
	182	}
	183	skb->len = head->len;
	184	skb->data_len = head->len;
	185	skb->truesize = head->truesize;
bbb03029 TH	186	_strp_msg(skb) = _strp_msg(head);
bbb03029 TH	187	strp->skb_nextp = &head->next;
43a0c675	188	skb_shinfo(skb)->frag_list = head;
bbb03029	189	strp->skb_head = skb;
43a0c675 TH	190	head = skb;
43a0c675 TH	191	} else {
bbb03029	192	strp->skb_nextp =
43a0c675 TH	193	&skb_shinfo(head)->frag_list;
	194	}
	195	}
	196	}
	197
	198	while (eaten < orig_len) {
	199	/* Always clone since we will consume something */
	200	skb = skb_clone(orig_skb, GFP_ATOMIC);
	201	if (!skb) {
bbb03029	202	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	203	desc->error = -ENOMEM;
	204	break;
	205	}
	206
	207	cand_len = orig_len - eaten;
	208
bbb03029	209	head = strp->skb_head;
43a0c675 TH	210	if (!head) {
43a0c675 TH	211	head = skb;
bbb03029 TH	212	strp->skb_head = head;
	213	/* Will set skb_nextp on next packet if needed */
	214	strp->skb_nextp = NULL;
	215	stm = _strp_msg(head);
	216	memset(stm, 0, sizeof(*stm));
	217	stm->strp.offset = orig_offset + eaten;
43a0c675 TH	218	} else {
	219	/* Unclone since we may be appending to an skb that we
	220	* already share a frag_list with.
	221	*/
	222	err = skb_unclone(skb, GFP_ATOMIC);
	223	if (err) {
bbb03029	224	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	225	desc->error = err;
	226	break;
	227	}
	228
bbb03029 TH	229	stm = _strp_msg(head);
	230	*strp->skb_nextp = skb;
	231	strp->skb_nextp = &skb->next;
43a0c675 TH	232	head->data_len += skb->len;
	233	head->len += skb->len;
	234	head->truesize += skb->truesize;
	235	}
	236
bbb03029	237	if (!stm->strp.full_len) {
43a0c675 TH	238	ssize_t len;
	239
	240	len = (*strp->cb.parse_msg)(strp, head);
	241
	242	if (!len) {
	243	/* Need more header to determine length */
bbb03029	244	if (!stm->accum_len) {
43a0c675	245	/* Start RX timer for new message */
bbb03029	246	strp_start_timer(strp, timeo);
43a0c675	247	}
bbb03029	248	stm->accum_len += cand_len;
43a0c675	249	eaten += cand_len;
bbb03029	250	STRP_STATS_INCR(strp->stats.need_more_hdr);
43a0c675 TH	251	WARN_ON(eaten != orig_len);
	252	break;
	253	} else if (len < 0) {
bbb03029	254	if (len == -ESTRPIPE && stm->accum_len) {
43a0c675	255	len = -ENODATA;
bbb03029	256	strp->unrecov_intr = 1;
43a0c675	257	} else {
bbb03029	258	strp->interrupted = 1;
43a0c675	259	}
6d3a4c40	260	strp_parser_err(strp, len, desc);
43a0c675	261	break;
bbb03029	262	} else if (len > max_msg_size) {
43a0c675	263	/* Message length exceeds maximum allowed */
bbb03029	264	STRP_STATS_INCR(strp->stats.msg_too_big);
43a0c675 TH	265	strp_parser_err(strp, -EMSGSIZE, desc);
	266	break;
	267	} else if (len <= (ssize_t)head->len -
bbb03029	268	skb->len - stm->strp.offset) {
43a0c675 TH	269	/* Length must be into new skb (and also
	270	* greater than zero)
	271	*/
bbb03029	272	STRP_STATS_INCR(strp->stats.bad_hdr_len);
43a0c675 TH	273	strp_parser_err(strp, -EPROTO, desc);
	274	break;
	275	}
	276
bbb03029	277	stm->strp.full_len = len;
43a0c675 TH	278	}
43a0c675 TH	279
bbb03029 TH	280	extra = (ssize_t)(stm->accum_len + cand_len) -
bbb03029 TH	281	stm->strp.full_len;
43a0c675 TH	282
	283	if (extra < 0) {
	284	/* Message not complete yet. */
bbb03029	285	if (stm->strp.full_len - stm->accum_len >
96a59083	286	strp_peek_len(strp)) {
bbb03029 TH	287	/* Don't have the whole message in the socket
bbb03029 TH	288	* buffer. Set strp->need_bytes to wait for
43a0c675 TH	289	* the rest of the message. Also, set "early
43a0c675 TH	290	* eaten" since we've already buffered the skb
96a59083	291	* but don't consume yet per strp_read_sock.
43a0c675 TH	292	*/
43a0c675 TH	293
bbb03029	294	if (!stm->accum_len) {
43a0c675	295	/* Start RX timer for new message */
bbb03029	296	strp_start_timer(strp, timeo);
43a0c675 TH	297	}
43a0c675 TH	298
bbb03029 TH	299	strp->need_bytes = stm->strp.full_len -
	300	stm->accum_len;
	301	stm->accum_len += cand_len;
	302	stm->early_eaten = cand_len;
	303	STRP_STATS_ADD(strp->stats.bytes, cand_len);
43a0c675 TH	304	desc->count = 0; /* Stop reading socket */
	305	break;
	306	}
bbb03029	307	stm->accum_len += cand_len;
43a0c675 TH	308	eaten += cand_len;
	309	WARN_ON(eaten != orig_len);
	310	break;
	311	}
	312
	313	/* Positive extra indicates ore bytes than needed for the
	314	* message
	315	*/
	316
	317	WARN_ON(extra > cand_len);
	318
	319	eaten += (cand_len - extra);
	320
	321	/* Hurray, we have a new message! */
829385f0	322	cancel_delayed_work(&strp->msg_timer_work);
bbb03029 TH	323	strp->skb_head = NULL;
bbb03029 TH	324	STRP_STATS_INCR(strp->stats.msgs);
43a0c675 TH	325
	326	/* Give skb to upper layer */
	327	strp->cb.rcv_msg(strp, head);
	328
bbb03029	329	if (unlikely(strp->paused)) {
43a0c675 TH	330	/* Upper layer paused strp */
	331	break;
	332	}
	333	}
	334
	335	if (cloned_orig)
	336	kfree_skb(orig_skb);
	337
bbb03029	338	STRP_STATS_ADD(strp->stats.bytes, eaten);
43a0c675 TH	339
	340	return eaten;
	341	}
	342
bbb03029 TH	343	int strp_process(struct strparser strp, struct sk_buff orig_skb,
	344	unsigned int orig_offset, size_t orig_len,
	345	size_t max_msg_size, long timeo)
	346	{
	347	read_descriptor_t desc; /* Dummy arg to strp_recv */
	348
	349	desc.arg.data = strp;
	350
	351	return __strp_recv(&desc, orig_skb, orig_offset, orig_len,
	352	max_msg_size, timeo);
	353	}
	354	EXPORT_SYMBOL_GPL(strp_process);
	355
	356	static int strp_recv(read_descriptor_t desc, struct sk_buff orig_skb,
	357	unsigned int orig_offset, size_t orig_len)
	358	{
	359	struct strparser strp = (struct strparser )desc->arg.data;
	360
	361	return __strp_recv(desc, orig_skb, orig_offset, orig_len,
	362	strp->sk->sk_rcvbuf, strp->sk->sk_rcvtimeo);
	363	}
	364
43a0c675 TH	365	static int default_read_sock_done(struct strparser *strp, int err)
	366	{
	367	return err;
	368	}
	369
	370	/* Called with lock held on lower socket */
96a59083	371	static int strp_read_sock(struct strparser *strp)
43a0c675	372	{
96a59083	373	struct socket *sock = strp->sk->sk_socket;
43a0c675 TH	374	read_descriptor_t desc;
43a0c675 TH	375
f26de110 JF	376	if (unlikely(!sock \|\| !sock->ops \|\| !sock->ops->read_sock))
	377	return -EBUSY;
	378
43a0c675 TH	379	desc.arg.data = strp;
	380	desc.error = 0;
	381	desc.count = 1; /* give more than one skb per call */
	382
96a59083 TH	383	/* sk should be locked here, so okay to do read_sock */
96a59083 TH	384	sock->ops->read_sock(strp->sk, &desc, strp_recv);
43a0c675 TH	385
	386	desc.error = strp->cb.read_sock_done(strp, desc.error);
	387
	388	return desc.error;
	389	}
	390
	391	/* Lower sock lock held */
96a59083	392	void strp_data_ready(struct strparser *strp)
43a0c675	393	{
bbb03029	394	if (unlikely(strp->stopped))
43a0c675 TH	395	return;
43a0c675 TH	396
bbb03029 TH	397	/* This check is needed to synchronize with do_strp_work.
bbb03029 TH	398	* do_strp_work acquires a process lock (lock_sock) whereas
43a0c675 TH	399	* the lock held here is bh_lock_sock. The two locks can be
	400	* held by different threads at the same time, but bh_lock_sock
	401	* allows a thread in BH context to safely check if the process
	402	* lock is held. In this case, if the lock is held, queue work.
	403	*/
d66fa9ec	404	if (sock_owned_by_user_nocheck(strp->sk)) {
bbb03029	405	queue_work(strp_wq, &strp->work);
43a0c675 TH	406	return;
	407	}
	408
bbb03029	409	if (strp->paused)
43a0c675 TH	410	return;
43a0c675 TH	411
bbb03029 TH	412	if (strp->need_bytes) {
	413	if (strp_peek_len(strp) >= strp->need_bytes)
	414	strp->need_bytes = 0;
43a0c675 TH	415	else
	416	return;
	417	}
	418
96a59083	419	if (strp_read_sock(strp) == -ENOMEM)
bbb03029	420	queue_work(strp_wq, &strp->work);
43a0c675	421	}
96a59083	422	EXPORT_SYMBOL_GPL(strp_data_ready);
43a0c675	423
bbb03029	424	static void do_strp_work(struct strparser *strp)
43a0c675 TH	425	{
43a0c675 TH	426	read_descriptor_t rd_desc;
43a0c675	427
96a59083 TH	428	/* We need the read lock to synchronize with strp_data_ready. We
96a59083 TH	429	* need the socket lock for calling strp_read_sock.
43a0c675	430	*/
bbb03029	431	strp->cb.lock(strp);
43a0c675	432
bbb03029	433	if (unlikely(strp->stopped))
43a0c675 TH	434	goto out;
43a0c675 TH	435
bbb03029	436	if (strp->paused)
43a0c675 TH	437	goto out;
	438
	439	rd_desc.arg.data = strp;
	440
96a59083	441	if (strp_read_sock(strp) == -ENOMEM)
bbb03029	442	queue_work(strp_wq, &strp->work);
43a0c675 TH	443
43a0c675 TH	444	out:
bbb03029	445	strp->cb.unlock(strp);
43a0c675 TH	446	}
43a0c675 TH	447
bbb03029	448	static void strp_work(struct work_struct *w)
43a0c675	449	{
bbb03029	450	do_strp_work(container_of(w, struct strparser, work));
43a0c675 TH	451	}
43a0c675 TH	452
829385f0	453	static void strp_msg_timeout(struct work_struct *w)
43a0c675	454	{
829385f0 TH	455	struct strparser *strp = container_of(w, struct strparser,
829385f0 TH	456	msg_timer_work.work);
43a0c675 TH	457
43a0c675 TH	458	/* Message assembly timed out */
bbb03029 TH	459	STRP_STATS_INCR(strp->stats.msg_timeouts);
bbb03029 TH	460	strp->cb.lock(strp);
43a0c675	461	strp->cb.abort_parser(strp, ETIMEDOUT);
bbb03029 TH	462	strp->cb.unlock(strp);
	463	}
	464
	465	static void strp_sock_lock(struct strparser *strp)
	466	{
	467	lock_sock(strp->sk);
	468	}
	469
	470	static void strp_sock_unlock(struct strparser *strp)
	471	{
43a0c675 TH	472	release_sock(strp->sk);
	473	}
	474
bbb03029	475	int strp_init(struct strparser strp, struct sock sk,
3fd87127	476	const struct strp_callbacks *cb)
43a0c675	477	{
96a59083	478
43a0c675 TH	479	if (!cb \|\| !cb->rcv_msg \|\| !cb->parse_msg)
	480	return -EINVAL;
	481
bbb03029 TH	482	/* The sk (sock) arg determines the mode of the stream parser.
	483	*
	484	* If the sock is set then the strparser is in receive callback mode.
	485	* The upper layer calls strp_data_ready to kick receive processing
	486	* and strparser calls the read_sock function on the socket to
	487	* get packets.
	488	*
	489	* If the sock is not set then the strparser is in general mode.
	490	* The upper layer calls strp_process for each skb to be parsed.
	491	*/
96a59083	492
f26de110	493	if (!sk) {
bbb03029 TH	494	if (!cb->lock \|\| !cb->unlock)
	495	return -EINVAL;
	496	}
43a0c675	497
bbb03029	498	memset(strp, 0, sizeof(*strp));
43a0c675	499
bbb03029	500	strp->sk = sk;
43a0c675	501
bbb03029 TH	502	strp->cb.lock = cb->lock ? : strp_sock_lock;
bbb03029 TH	503	strp->cb.unlock = cb->unlock ? : strp_sock_unlock;
43a0c675 TH	504	strp->cb.rcv_msg = cb->rcv_msg;
	505	strp->cb.parse_msg = cb->parse_msg;
	506	strp->cb.read_sock_done = cb->read_sock_done ? : default_read_sock_done;
bbb03029 TH	507	strp->cb.abort_parser = cb->abort_parser ? : strp_abort_strp;
bbb03029 TH	508
829385f0	509	INIT_DELAYED_WORK(&strp->msg_timer_work, strp_msg_timeout);
bbb03029	510	INIT_WORK(&strp->work, strp_work);
43a0c675 TH	511
	512	return 0;
	513	}
	514	EXPORT_SYMBOL_GPL(strp_init);
	515
cff6a334 TH	516	void strp_unpause(struct strparser *strp)
cff6a334 TH	517	{
bbb03029	518	strp->paused = 0;
cff6a334	519
bbb03029	520	/* Sync setting paused with RX work */
cff6a334 TH	521	smp_mb();
cff6a334 TH	522
bbb03029	523	queue_work(strp_wq, &strp->work);
cff6a334 TH	524	}
	525	EXPORT_SYMBOL_GPL(strp_unpause);
	526
96a59083	527	/* strp must already be stopped so that strp_recv will no longer be called.
43a0c675 TH	528	* Note that strp_done is not called with the lower socket held.
	529	*/
	530	void strp_done(struct strparser *strp)
	531	{
bbb03029	532	WARN_ON(!strp->stopped);
43a0c675	533
829385f0	534	cancel_delayed_work_sync(&strp->msg_timer_work);
bbb03029	535	cancel_work_sync(&strp->work);
43a0c675	536
bbb03029 TH	537	if (strp->skb_head) {
	538	kfree_skb(strp->skb_head);
	539	strp->skb_head = NULL;
43a0c675 TH	540	}
	541	}
	542	EXPORT_SYMBOL_GPL(strp_done);
	543
	544	void strp_stop(struct strparser *strp)
	545	{
bbb03029	546	strp->stopped = 1;
43a0c675 TH	547	}
	548	EXPORT_SYMBOL_GPL(strp_stop);
	549
	550	void strp_check_rcv(struct strparser *strp)
	551	{
bbb03029	552	queue_work(strp_wq, &strp->work);
43a0c675 TH	553	}
	554	EXPORT_SYMBOL_GPL(strp_check_rcv);
	555
	556	static int __init strp_mod_init(void)
	557	{
	558	strp_wq = create_singlethread_workqueue("kstrp");
	559
	560	return 0;
	561	}
	562
	563	static void __exit strp_mod_exit(void)
	564	{
f78ef7cd	565	destroy_workqueue(strp_wq);
43a0c675 TH	566	}
	567	module_init(strp_mod_init);
	568	module_exit(strp_mod_exit);
	569	MODULE_LICENSE("GPL");