[mirror_ubuntu-bionic-kernel.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;
};

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(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);
				}

				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 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->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))
		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)
			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	37	int accum_len;
43a0c675 TH	38	};
43a0c675 TH	39
bbb03029	40	static inline struct _strp_msg _strp_msg(struct sk_buff skb)
43a0c675	41	{
bbb03029	42	return (struct _strp_msg )((void )skb->cb +
43a0c675 TH	43	offsetof(struct qdisc_skb_cb, data));
	44	}
	45
	46	/* Lower lock held */
bbb03029	47	static void strp_abort_strp(struct strparser *strp, int err)
43a0c675	48	{
43a0c675 TH	49	/* Unrecoverable error in receive */
43a0c675 TH	50
829385f0	51	cancel_delayed_work(&strp->msg_timer_work);
43a0c675	52
bbb03029	53	if (strp->stopped)
43a0c675 TH	54	return;
43a0c675 TH	55
bbb03029 TH	56	strp->stopped = 1;
	57
	58	if (strp->sk) {
	59	struct sock *sk = strp->sk;
43a0c675	60
bbb03029	61	/* Report an error on the lower socket */
b6d5f4bf	62	sk->sk_err = -err;
bbb03029 TH	63	sk->sk_error_report(sk);
bbb03029 TH	64	}
43a0c675 TH	65	}
43a0c675 TH	66
bbb03029	67	static void strp_start_timer(struct strparser *strp, long timeo)
43a0c675	68	{
c1e45f5b	69	if (timeo && timeo != LONG_MAX)
829385f0	70	mod_delayed_work(strp_wq, &strp->msg_timer_work, timeo);
43a0c675 TH	71	}
	72
	73	/* Lower lock held */
	74	static void strp_parser_err(struct strparser *strp, int err,
	75	read_descriptor_t *desc)
	76	{
	77	desc->error = err;
bbb03029 TH	78	kfree_skb(strp->skb_head);
bbb03029 TH	79	strp->skb_head = NULL;
43a0c675 TH	80	strp->cb.abort_parser(strp, err);
	81	}
	82
96a59083 TH	83	static inline int strp_peek_len(struct strparser *strp)
96a59083 TH	84	{
bbb03029 TH	85	if (strp->sk) {
	86	struct socket *sock = strp->sk->sk_socket;
	87
	88	return sock->ops->peek_len(sock);
	89	}
	90
	91	/* If we don't have an associated socket there's nothing to peek.
	92	* Return int max to avoid stopping the strparser.
	93	*/
96a59083	94
bbb03029	95	return INT_MAX;
96a59083 TH	96	}
96a59083 TH	97
43a0c675	98	/* Lower socket lock held */
bbb03029 TH	99	static int __strp_recv(read_descriptor_t desc, struct sk_buff orig_skb,
	100	unsigned int orig_offset, size_t orig_len,
	101	size_t max_msg_size, long timeo)
43a0c675 TH	102	{
43a0c675 TH	103	struct strparser strp = (struct strparser )desc->arg.data;
bbb03029	104	struct _strp_msg *stm;
43a0c675 TH	105	struct sk_buff head, skb;
	106	size_t eaten = 0, cand_len;
	107	ssize_t extra;
	108	int err;
	109	bool cloned_orig = false;
	110
bbb03029	111	if (strp->paused)
43a0c675 TH	112	return 0;
43a0c675 TH	113
bbb03029	114	head = strp->skb_head;
43a0c675 TH	115	if (head) {
43a0c675 TH	116	/* Message already in progress */
43a0c675 TH	117	if (unlikely(orig_offset)) {
	118	/* Getting data with a non-zero offset when a message is
	119	* in progress is not expected. If it does happen, we
	120	* need to clone and pull since we can't deal with
	121	* offsets in the skbs for a message expect in the head.
	122	*/
	123	orig_skb = skb_clone(orig_skb, GFP_ATOMIC);
	124	if (!orig_skb) {
bbb03029	125	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	126	desc->error = -ENOMEM;
	127	return 0;
	128	}
	129	if (!pskb_pull(orig_skb, orig_offset)) {
bbb03029	130	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	131	kfree_skb(orig_skb);
	132	desc->error = -ENOMEM;
	133	return 0;
	134	}
	135	cloned_orig = true;
	136	orig_offset = 0;
	137	}
	138
bbb03029	139	if (!strp->skb_nextp) {
43a0c675 TH	140	/* We are going to append to the frags_list of head.
	141	* Need to unshare the frag_list.
	142	*/
	143	err = skb_unclone(head, GFP_ATOMIC);
	144	if (err) {
bbb03029	145	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	146	desc->error = err;
	147	return 0;
	148	}
	149
	150	if (unlikely(skb_shinfo(head)->frag_list)) {
	151	/* We can't append to an sk_buff that already
	152	* has a frag_list. We create a new head, point
	153	* the frag_list of that to the old head, and
	154	* then are able to use the old head->next for
	155	* appending to the message.
	156	*/
	157	if (WARN_ON(head->next)) {
	158	desc->error = -EINVAL;
	159	return 0;
	160	}
	161
	162	skb = alloc_skb(0, GFP_ATOMIC);
	163	if (!skb) {
bbb03029	164	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	165	desc->error = -ENOMEM;
	166	return 0;
	167	}
	168	skb->len = head->len;
	169	skb->data_len = head->len;
	170	skb->truesize = head->truesize;
bbb03029 TH	171	_strp_msg(skb) = _strp_msg(head);
bbb03029 TH	172	strp->skb_nextp = &head->next;
43a0c675	173	skb_shinfo(skb)->frag_list = head;
bbb03029	174	strp->skb_head = skb;
43a0c675 TH	175	head = skb;
43a0c675 TH	176	} else {
bbb03029	177	strp->skb_nextp =
43a0c675 TH	178	&skb_shinfo(head)->frag_list;
	179	}
	180	}
	181	}
	182
	183	while (eaten < orig_len) {
	184	/* Always clone since we will consume something */
	185	skb = skb_clone(orig_skb, GFP_ATOMIC);
	186	if (!skb) {
bbb03029	187	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	188	desc->error = -ENOMEM;
	189	break;
	190	}
	191
	192	cand_len = orig_len - eaten;
	193
bbb03029	194	head = strp->skb_head;
43a0c675 TH	195	if (!head) {
43a0c675 TH	196	head = skb;
bbb03029 TH	197	strp->skb_head = head;
	198	/* Will set skb_nextp on next packet if needed */
	199	strp->skb_nextp = NULL;
	200	stm = _strp_msg(head);
	201	memset(stm, 0, sizeof(*stm));
	202	stm->strp.offset = orig_offset + eaten;
43a0c675 TH	203	} else {
	204	/* Unclone since we may be appending to an skb that we
	205	* already share a frag_list with.
	206	*/
	207	err = skb_unclone(skb, GFP_ATOMIC);
	208	if (err) {
bbb03029	209	STRP_STATS_INCR(strp->stats.mem_fail);
43a0c675 TH	210	desc->error = err;
	211	break;
	212	}
	213
bbb03029 TH	214	stm = _strp_msg(head);
	215	*strp->skb_nextp = skb;
	216	strp->skb_nextp = &skb->next;
43a0c675 TH	217	head->data_len += skb->len;
	218	head->len += skb->len;
	219	head->truesize += skb->truesize;
	220	}
	221
bbb03029	222	if (!stm->strp.full_len) {
43a0c675 TH	223	ssize_t len;
	224
	225	len = (*strp->cb.parse_msg)(strp, head);
	226
	227	if (!len) {
	228	/* Need more header to determine length */
bbb03029	229	if (!stm->accum_len) {
43a0c675	230	/* Start RX timer for new message */
bbb03029	231	strp_start_timer(strp, timeo);
43a0c675	232	}
bbb03029	233	stm->accum_len += cand_len;
43a0c675	234	eaten += cand_len;
bbb03029	235	STRP_STATS_INCR(strp->stats.need_more_hdr);
43a0c675 TH	236	WARN_ON(eaten != orig_len);
	237	break;
	238	} else if (len < 0) {
bbb03029	239	if (len == -ESTRPIPE && stm->accum_len) {
43a0c675	240	len = -ENODATA;
bbb03029	241	strp->unrecov_intr = 1;
43a0c675	242	} else {
bbb03029	243	strp->interrupted = 1;
43a0c675	244	}
6d3a4c40	245	strp_parser_err(strp, len, desc);
43a0c675	246	break;
bbb03029	247	} else if (len > max_msg_size) {
43a0c675	248	/* Message length exceeds maximum allowed */
bbb03029	249	STRP_STATS_INCR(strp->stats.msg_too_big);
43a0c675 TH	250	strp_parser_err(strp, -EMSGSIZE, desc);
	251	break;
	252	} else if (len <= (ssize_t)head->len -
bbb03029	253	skb->len - stm->strp.offset) {
43a0c675 TH	254	/* Length must be into new skb (and also
	255	* greater than zero)
	256	*/
bbb03029	257	STRP_STATS_INCR(strp->stats.bad_hdr_len);
43a0c675 TH	258	strp_parser_err(strp, -EPROTO, desc);
	259	break;
	260	}
	261
bbb03029	262	stm->strp.full_len = len;
43a0c675 TH	263	}
43a0c675 TH	264
bbb03029 TH	265	extra = (ssize_t)(stm->accum_len + cand_len) -
bbb03029 TH	266	stm->strp.full_len;
43a0c675 TH	267
	268	if (extra < 0) {
	269	/* Message not complete yet. */
bbb03029	270	if (stm->strp.full_len - stm->accum_len >
96a59083	271	strp_peek_len(strp)) {
bbb03029 TH	272	/* Don't have the whole message in the socket
bbb03029 TH	273	* buffer. Set strp->need_bytes to wait for
43a0c675 TH	274	* the rest of the message. Also, set "early
43a0c675 TH	275	* eaten" since we've already buffered the skb
96a59083	276	* but don't consume yet per strp_read_sock.
43a0c675 TH	277	*/
43a0c675 TH	278
bbb03029	279	if (!stm->accum_len) {
43a0c675	280	/* Start RX timer for new message */
bbb03029	281	strp_start_timer(strp, timeo);
43a0c675 TH	282	}
43a0c675 TH	283
ddd9d854	284	stm->accum_len += cand_len;
753020a3	285	eaten += cand_len;
bbb03029 TH	286	strp->need_bytes = stm->strp.full_len -
bbb03029 TH	287	stm->accum_len;
bbb03029	288	STRP_STATS_ADD(strp->stats.bytes, cand_len);
43a0c675 TH	289	desc->count = 0; /* Stop reading socket */
	290	break;
	291	}
bbb03029	292	stm->accum_len += cand_len;
43a0c675 TH	293	eaten += cand_len;
	294	WARN_ON(eaten != orig_len);
	295	break;
	296	}
	297
	298	/* Positive extra indicates ore bytes than needed for the
	299	* message
	300	*/
	301
	302	WARN_ON(extra > cand_len);
	303
	304	eaten += (cand_len - extra);
	305
	306	/* Hurray, we have a new message! */
829385f0	307	cancel_delayed_work(&strp->msg_timer_work);
bbb03029	308	strp->skb_head = NULL;
ddd9d854	309	strp->need_bytes = 0;
bbb03029	310	STRP_STATS_INCR(strp->stats.msgs);
43a0c675 TH	311
	312	/* Give skb to upper layer */
	313	strp->cb.rcv_msg(strp, head);
	314
bbb03029	315	if (unlikely(strp->paused)) {
43a0c675 TH	316	/* Upper layer paused strp */
	317	break;
	318	}
	319	}
	320
	321	if (cloned_orig)
	322	kfree_skb(orig_skb);
	323
bbb03029	324	STRP_STATS_ADD(strp->stats.bytes, eaten);
43a0c675 TH	325
	326	return eaten;
	327	}
	328
bbb03029 TH	329	int strp_process(struct strparser strp, struct sk_buff orig_skb,
	330	unsigned int orig_offset, size_t orig_len,
	331	size_t max_msg_size, long timeo)
	332	{
	333	read_descriptor_t desc; /* Dummy arg to strp_recv */
	334
	335	desc.arg.data = strp;
	336
	337	return __strp_recv(&desc, orig_skb, orig_offset, orig_len,
	338	max_msg_size, timeo);
	339	}
	340	EXPORT_SYMBOL_GPL(strp_process);
	341
	342	static int strp_recv(read_descriptor_t desc, struct sk_buff orig_skb,
	343	unsigned int orig_offset, size_t orig_len)
	344	{
	345	struct strparser strp = (struct strparser )desc->arg.data;
	346
	347	return __strp_recv(desc, orig_skb, orig_offset, orig_len,
	348	strp->sk->sk_rcvbuf, strp->sk->sk_rcvtimeo);
	349	}
	350
43a0c675 TH	351	static int default_read_sock_done(struct strparser *strp, int err)
	352	{
	353	return err;
	354	}
	355
	356	/* Called with lock held on lower socket */
96a59083	357	static int strp_read_sock(struct strparser *strp)
43a0c675	358	{
96a59083	359	struct socket *sock = strp->sk->sk_socket;
43a0c675 TH	360	read_descriptor_t desc;
43a0c675 TH	361
f26de110 JF	362	if (unlikely(!sock \|\| !sock->ops \|\| !sock->ops->read_sock))
	363	return -EBUSY;
	364
43a0c675 TH	365	desc.arg.data = strp;
	366	desc.error = 0;
	367	desc.count = 1; /* give more than one skb per call */
	368
96a59083 TH	369	/* sk should be locked here, so okay to do read_sock */
96a59083 TH	370	sock->ops->read_sock(strp->sk, &desc, strp_recv);
43a0c675 TH	371
	372	desc.error = strp->cb.read_sock_done(strp, desc.error);
	373
	374	return desc.error;
	375	}
	376
	377	/* Lower sock lock held */
96a59083	378	void strp_data_ready(struct strparser *strp)
43a0c675	379	{
bbb03029	380	if (unlikely(strp->stopped))
43a0c675 TH	381	return;
43a0c675 TH	382
bbb03029 TH	383	/* This check is needed to synchronize with do_strp_work.
bbb03029 TH	384	* do_strp_work acquires a process lock (lock_sock) whereas
43a0c675 TH	385	* the lock held here is bh_lock_sock. The two locks can be
	386	* held by different threads at the same time, but bh_lock_sock
	387	* allows a thread in BH context to safely check if the process
	388	* lock is held. In this case, if the lock is held, queue work.
	389	*/
d66fa9ec	390	if (sock_owned_by_user_nocheck(strp->sk)) {
bbb03029	391	queue_work(strp_wq, &strp->work);
43a0c675 TH	392	return;
	393	}
	394
bbb03029	395	if (strp->paused)
43a0c675 TH	396	return;
43a0c675 TH	397
bbb03029	398	if (strp->need_bytes) {
ddd9d854	399	if (strp_peek_len(strp) < strp->need_bytes)
43a0c675 TH	400	return;
	401	}
	402
96a59083	403	if (strp_read_sock(strp) == -ENOMEM)
bbb03029	404	queue_work(strp_wq, &strp->work);
43a0c675	405	}
96a59083	406	EXPORT_SYMBOL_GPL(strp_data_ready);
43a0c675	407
bbb03029	408	static void do_strp_work(struct strparser *strp)
43a0c675 TH	409	{
43a0c675 TH	410	read_descriptor_t rd_desc;
43a0c675	411
96a59083 TH	412	/* We need the read lock to synchronize with strp_data_ready. We
96a59083 TH	413	* need the socket lock for calling strp_read_sock.
43a0c675	414	*/
bbb03029	415	strp->cb.lock(strp);
43a0c675	416
bbb03029	417	if (unlikely(strp->stopped))
43a0c675 TH	418	goto out;
43a0c675 TH	419
bbb03029	420	if (strp->paused)
43a0c675 TH	421	goto out;
	422
	423	rd_desc.arg.data = strp;
	424
96a59083	425	if (strp_read_sock(strp) == -ENOMEM)
bbb03029	426	queue_work(strp_wq, &strp->work);
43a0c675 TH	427
43a0c675 TH	428	out:
bbb03029	429	strp->cb.unlock(strp);
43a0c675 TH	430	}
43a0c675 TH	431
bbb03029	432	static void strp_work(struct work_struct *w)
43a0c675	433	{
bbb03029	434	do_strp_work(container_of(w, struct strparser, work));
43a0c675 TH	435	}
43a0c675 TH	436
829385f0	437	static void strp_msg_timeout(struct work_struct *w)
43a0c675	438	{
829385f0 TH	439	struct strparser *strp = container_of(w, struct strparser,
829385f0 TH	440	msg_timer_work.work);
43a0c675 TH	441
43a0c675 TH	442	/* Message assembly timed out */
bbb03029 TH	443	STRP_STATS_INCR(strp->stats.msg_timeouts);
bbb03029 TH	444	strp->cb.lock(strp);
b6d5f4bf	445	strp->cb.abort_parser(strp, -ETIMEDOUT);
bbb03029 TH	446	strp->cb.unlock(strp);
	447	}
	448
	449	static void strp_sock_lock(struct strparser *strp)
	450	{
	451	lock_sock(strp->sk);
	452	}
	453
	454	static void strp_sock_unlock(struct strparser *strp)
	455	{
43a0c675 TH	456	release_sock(strp->sk);
	457	}
	458
bbb03029	459	int strp_init(struct strparser strp, struct sock sk,
3fd87127	460	const struct strp_callbacks *cb)
43a0c675	461	{
96a59083	462
43a0c675 TH	463	if (!cb \|\| !cb->rcv_msg \|\| !cb->parse_msg)
	464	return -EINVAL;
	465
bbb03029 TH	466	/* The sk (sock) arg determines the mode of the stream parser.
	467	*
	468	* If the sock is set then the strparser is in receive callback mode.
	469	* The upper layer calls strp_data_ready to kick receive processing
	470	* and strparser calls the read_sock function on the socket to
	471	* get packets.
	472	*
	473	* If the sock is not set then the strparser is in general mode.
	474	* The upper layer calls strp_process for each skb to be parsed.
	475	*/
96a59083	476
f26de110	477	if (!sk) {
bbb03029 TH	478	if (!cb->lock \|\| !cb->unlock)
	479	return -EINVAL;
	480	}
43a0c675	481
bbb03029	482	memset(strp, 0, sizeof(*strp));
43a0c675	483
bbb03029	484	strp->sk = sk;
43a0c675	485
bbb03029 TH	486	strp->cb.lock = cb->lock ? : strp_sock_lock;
bbb03029 TH	487	strp->cb.unlock = cb->unlock ? : strp_sock_unlock;
43a0c675 TH	488	strp->cb.rcv_msg = cb->rcv_msg;
	489	strp->cb.parse_msg = cb->parse_msg;
	490	strp->cb.read_sock_done = cb->read_sock_done ? : default_read_sock_done;
bbb03029 TH	491	strp->cb.abort_parser = cb->abort_parser ? : strp_abort_strp;
bbb03029 TH	492
829385f0	493	INIT_DELAYED_WORK(&strp->msg_timer_work, strp_msg_timeout);
bbb03029	494	INIT_WORK(&strp->work, strp_work);
43a0c675 TH	495
	496	return 0;
	497	}
	498	EXPORT_SYMBOL_GPL(strp_init);
	499
cff6a334 TH	500	void strp_unpause(struct strparser *strp)
cff6a334 TH	501	{
bbb03029	502	strp->paused = 0;
cff6a334	503
bbb03029	504	/* Sync setting paused with RX work */
cff6a334 TH	505	smp_mb();
cff6a334 TH	506
bbb03029	507	queue_work(strp_wq, &strp->work);
cff6a334 TH	508	}
	509	EXPORT_SYMBOL_GPL(strp_unpause);
	510
96a59083	511	/* strp must already be stopped so that strp_recv will no longer be called.
43a0c675 TH	512	* Note that strp_done is not called with the lower socket held.
	513	*/
	514	void strp_done(struct strparser *strp)
	515	{
bbb03029	516	WARN_ON(!strp->stopped);
43a0c675	517
829385f0	518	cancel_delayed_work_sync(&strp->msg_timer_work);
bbb03029	519	cancel_work_sync(&strp->work);
43a0c675	520
bbb03029 TH	521	if (strp->skb_head) {
	522	kfree_skb(strp->skb_head);
	523	strp->skb_head = NULL;
43a0c675 TH	524	}
	525	}
	526	EXPORT_SYMBOL_GPL(strp_done);
	527
	528	void strp_stop(struct strparser *strp)
	529	{
bbb03029	530	strp->stopped = 1;
43a0c675 TH	531	}
	532	EXPORT_SYMBOL_GPL(strp_stop);
	533
	534	void strp_check_rcv(struct strparser *strp)
	535	{
bbb03029	536	queue_work(strp_wq, &strp->work);
43a0c675 TH	537	}
	538	EXPORT_SYMBOL_GPL(strp_check_rcv);
	539
	540	static int __init strp_mod_init(void)
	541	{
	542	strp_wq = create_singlethread_workqueue("kstrp");
	543
	544	return 0;
	545	}
	546
	547	static void __exit strp_mod_exit(void)
	548	{
f78ef7cd	549	destroy_workqueue(strp_wq);
43a0c675 TH	550	}
	551	module_init(strp_mod_init);
	552	module_exit(strp_mod_exit);
	553	MODULE_LICENSE("GPL");