X-Git-Url: http://pilppa.org/gitweb/gitweb.cgi?a=blobdiff_plain;f=net%2Fsctp%2Fulpqueue.c;h=5061a26c5028d927bf40bb51d772fe91691432f4;hb=3dd59e226e01ddb5b041eb0b2e7c7f28b1f730c9;hp=bfb197e37da3784a45c168bf9f87204e717e810b;hpb=7f52a3afc4c02db75b3953fcbbb980abc86873a2;p=linux-2.6-omap-h63xx.git diff --git a/net/sctp/ulpqueue.c b/net/sctp/ulpqueue.c index bfb197e37da..5061a26c502 100644 --- a/net/sctp/ulpqueue.c +++ b/net/sctp/ulpqueue.c @@ -1,4 +1,4 @@ -/* SCTP kernel reference Implementation +/* SCTP kernel implementation * (C) Copyright IBM Corp. 2001, 2004 * Copyright (c) 1999-2000 Cisco, Inc. * Copyright (c) 1999-2001 Motorola, Inc. @@ -8,13 +8,13 @@ * * This abstraction carries sctp events to the ULP (sockets). * - * The SCTP reference implementation is free software; + * This SCTP implementation is free software; * you can redistribute it and/or modify it under the terms of * the GNU General Public License as published by * the Free Software Foundation; either version 2, or (at your option) * any later version. * - * The SCTP reference implementation is distributed in the hope that it + * This SCTP implementation is distributed in the hope that it * will be useful, but WITHOUT ANY WARRANTY; without even the implied * ************************ * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. @@ -53,6 +53,7 @@ static struct sctp_ulpevent * sctp_ulpq_reasm(struct sctp_ulpq *ulpq, struct sctp_ulpevent *); static struct sctp_ulpevent * sctp_ulpq_order(struct sctp_ulpq *, struct sctp_ulpevent *); +static void sctp_ulpq_reasm_drain(struct sctp_ulpq *ulpq); /* 1st Level Abstractions */ @@ -138,26 +139,60 @@ int sctp_ulpq_tail_data(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk, /* Clear the partial delivery mode for this socket. Note: This * assumes that no association is currently in partial delivery mode. */ -int sctp_clear_pd(struct sock *sk) +int sctp_clear_pd(struct sock *sk, struct sctp_association *asoc) { struct sctp_sock *sp = sctp_sk(sk); - sp->pd_mode = 0; - if (!skb_queue_empty(&sp->pd_lobby)) { - struct list_head *list; - sctp_skb_list_tail(&sp->pd_lobby, &sk->sk_receive_queue); - list = (struct list_head *)&sctp_sk(sk)->pd_lobby; - INIT_LIST_HEAD(list); - return 1; + if (atomic_dec_and_test(&sp->pd_mode)) { + /* This means there are no other associations in PD, so + * we can go ahead and clear out the lobby in one shot + */ + if (!skb_queue_empty(&sp->pd_lobby)) { + struct list_head *list; + sctp_skb_list_tail(&sp->pd_lobby, &sk->sk_receive_queue); + list = (struct list_head *)&sctp_sk(sk)->pd_lobby; + INIT_LIST_HEAD(list); + return 1; + } + } else { + /* There are other associations in PD, so we only need to + * pull stuff out of the lobby that belongs to the + * associations that is exiting PD (all of its notifications + * are posted here). + */ + if (!skb_queue_empty(&sp->pd_lobby) && asoc) { + struct sk_buff *skb, *tmp; + struct sctp_ulpevent *event; + + sctp_skb_for_each(skb, &sp->pd_lobby, tmp) { + event = sctp_skb2event(skb); + if (event->asoc == asoc) { + __skb_unlink(skb, &sp->pd_lobby); + __skb_queue_tail(&sk->sk_receive_queue, + skb); + } + } + } } + return 0; } +/* Set the pd_mode on the socket and ulpq */ +static void sctp_ulpq_set_pd(struct sctp_ulpq *ulpq) +{ + struct sctp_sock *sp = sctp_sk(ulpq->asoc->base.sk); + + atomic_inc(&sp->pd_mode); + ulpq->pd_mode = 1; +} + /* Clear the pd_mode and restart any pending messages waiting for delivery. */ static int sctp_ulpq_clear_pd(struct sctp_ulpq *ulpq) { ulpq->pd_mode = 0; - return sctp_clear_pd(ulpq->asoc->base.sk); + sctp_ulpq_reasm_drain(ulpq); + return sctp_clear_pd(ulpq->asoc->base.sk, ulpq->asoc); } /* If the SKB of 'event' is on a list, it is the first such member @@ -187,18 +222,35 @@ int sctp_ulpq_tail_event(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event) * the association the cause of the partial delivery. */ - if (!sctp_sk(sk)->pd_mode) { + if (atomic_read(&sctp_sk(sk)->pd_mode) == 0) { queue = &sk->sk_receive_queue; - } else if (ulpq->pd_mode) { - if (event->msg_flags & MSG_NOTIFICATION) - queue = &sctp_sk(sk)->pd_lobby; - else { - clear_pd = event->msg_flags & MSG_EOR; - queue = &sk->sk_receive_queue; + } else { + if (ulpq->pd_mode) { + /* If the association is in partial delivery, we + * need to finish delivering the partially processed + * packet before passing any other data. This is + * because we don't truly support stream interleaving. + */ + if ((event->msg_flags & MSG_NOTIFICATION) || + (SCTP_DATA_NOT_FRAG == + (event->msg_flags & SCTP_DATA_FRAG_MASK))) + queue = &sctp_sk(sk)->pd_lobby; + else { + clear_pd = event->msg_flags & MSG_EOR; + queue = &sk->sk_receive_queue; + } + } else { + /* + * If fragment interleave is enabled, we + * can queue this to the recieve queue instead + * of the lobby. + */ + if (sctp_sk(sk)->frag_interleave) + queue = &sk->sk_receive_queue; + else + queue = &sctp_sk(sk)->pd_lobby; } - } else - queue = &sctp_sk(sk)->pd_lobby; - + } /* If we are harvesting multiple skbs they will be * collected on a list. @@ -231,7 +283,7 @@ out_free: /* 2nd Level Abstractions */ /* Helper function to store chunks that need to be reassembled. */ -static inline void sctp_ulpq_store_reasm(struct sctp_ulpq *ulpq, +static void sctp_ulpq_store_reasm(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event) { struct sk_buff *pos; @@ -341,7 +393,7 @@ static struct sctp_ulpevent *sctp_make_reassembled_event(struct sk_buff_head *qu break; pos->next = pnext; pos = pnext; - }; + } event = sctp_skb2event(f_frag); SCTP_INC_STATS(SCTP_MIB_REASMUSRMSGS); @@ -353,13 +405,18 @@ static struct sctp_ulpevent *sctp_make_reassembled_event(struct sk_buff_head *qu /* Helper function to check if an incoming chunk has filled up the last * missing fragment in a SCTP datagram and return the corresponding event. */ -static inline struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_ulpq *ulpq) +static struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_ulpq *ulpq) { struct sk_buff *pos; struct sctp_ulpevent *cevent; struct sk_buff *first_frag = NULL; __u32 ctsn, next_tsn; struct sctp_ulpevent *retval = NULL; + struct sk_buff *pd_first = NULL; + struct sk_buff *pd_last = NULL; + size_t pd_len = 0; + struct sctp_association *asoc; + u32 pd_point; /* Initialized to 0 just to avoid compiler warning message. Will * never be used with this value. It is referenced only after it @@ -375,6 +432,10 @@ static inline struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_u * we expect to find the remaining middle fragments and the last * fragment in order. If not, first_frag is reset to NULL and we * start the next pass when we find another first fragment. + * + * There is a potential to do partial delivery if user sets + * SCTP_PARTIAL_DELIVERY_POINT option. Lets count some things here + * to see if can do PD. */ skb_queue_walk(&ulpq->reasm, pos) { cevent = sctp_skb2event(pos); @@ -382,14 +443,32 @@ static inline struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_u switch (cevent->msg_flags & SCTP_DATA_FRAG_MASK) { case SCTP_DATA_FIRST_FRAG: + /* If this "FIRST_FRAG" is the first + * element in the queue, then count it towards + * possible PD. + */ + if (pos == ulpq->reasm.next) { + pd_first = pos; + pd_last = pos; + pd_len = pos->len; + } else { + pd_first = NULL; + pd_last = NULL; + pd_len = 0; + } + first_frag = pos; next_tsn = ctsn + 1; break; case SCTP_DATA_MIDDLE_FRAG: - if ((first_frag) && (ctsn == next_tsn)) + if ((first_frag) && (ctsn == next_tsn)) { next_tsn++; - else + if (pd_first) { + pd_last = pos; + pd_len += pos->len; + } + } else first_frag = NULL; break; @@ -399,8 +478,29 @@ static inline struct sctp_ulpevent *sctp_ulpq_retrieve_reassembled(struct sctp_u else first_frag = NULL; break; - }; + } + } + asoc = ulpq->asoc; + if (pd_first) { + /* Make sure we can enter partial deliver. + * We can trigger partial delivery only if framgent + * interleave is set, or the socket is not already + * in partial delivery. + */ + if (!sctp_sk(asoc->base.sk)->frag_interleave && + atomic_read(&sctp_sk(asoc->base.sk)->pd_mode)) + goto done; + + cevent = sctp_skb2event(pd_first); + pd_point = sctp_sk(asoc->base.sk)->pd_point; + if (pd_point && pd_point <= pd_len) { + retval = sctp_make_reassembled_event(&ulpq->reasm, + pd_first, + pd_last); + if (retval) + sctp_ulpq_set_pd(ulpq); + } } done: return retval; @@ -412,7 +512,7 @@ found: } /* Retrieve the next set of fragments of a partial message. */ -static inline struct sctp_ulpevent *sctp_ulpq_retrieve_partial(struct sctp_ulpq *ulpq) +static struct sctp_ulpevent *sctp_ulpq_retrieve_partial(struct sctp_ulpq *ulpq) { struct sk_buff *pos, *last_frag, *first_frag; struct sctp_ulpevent *cevent; @@ -458,7 +558,7 @@ static inline struct sctp_ulpevent *sctp_ulpq_retrieve_partial(struct sctp_ulpq goto done; default: return NULL; - }; + } } /* We have the reassembled event. There is no need to look @@ -506,7 +606,7 @@ static struct sctp_ulpevent *sctp_ulpq_reasm(struct sctp_ulpq *ulpq, } /* Retrieve the first part (sequential fragments) for partial delivery. */ -static inline struct sctp_ulpevent *sctp_ulpq_retrieve_first(struct sctp_ulpq *ulpq) +static struct sctp_ulpevent *sctp_ulpq_retrieve_first(struct sctp_ulpq *ulpq) { struct sk_buff *pos, *last_frag, *first_frag; struct sctp_ulpevent *cevent; @@ -550,7 +650,7 @@ static inline struct sctp_ulpevent *sctp_ulpq_retrieve_first(struct sctp_ulpq *u break; default: return NULL; - }; + } } /* We have the reassembled event. There is no need to look @@ -561,10 +661,81 @@ done: return retval; } +/* + * Flush out stale fragments from the reassembly queue when processing + * a Forward TSN. + * + * RFC 3758, Section 3.6 + * + * After receiving and processing a FORWARD TSN, the data receiver MUST + * take cautions in updating its re-assembly queue. The receiver MUST + * remove any partially reassembled message, which is still missing one + * or more TSNs earlier than or equal to the new cumulative TSN point. + * In the event that the receiver has invoked the partial delivery API, + * a notification SHOULD also be generated to inform the upper layer API + * that the message being partially delivered will NOT be completed. + */ +void sctp_ulpq_reasm_flushtsn(struct sctp_ulpq *ulpq, __u32 fwd_tsn) +{ + struct sk_buff *pos, *tmp; + struct sctp_ulpevent *event; + __u32 tsn; + + if (skb_queue_empty(&ulpq->reasm)) + return; + + skb_queue_walk_safe(&ulpq->reasm, pos, tmp) { + event = sctp_skb2event(pos); + tsn = event->tsn; + + /* Since the entire message must be abandoned by the + * sender (item A3 in Section 3.5, RFC 3758), we can + * free all fragments on the list that are less then + * or equal to ctsn_point + */ + if (TSN_lte(tsn, fwd_tsn)) { + __skb_unlink(pos, &ulpq->reasm); + sctp_ulpevent_free(event); + } else + break; + } +} + +/* + * Drain the reassembly queue. If we just cleared parted delivery, it + * is possible that the reassembly queue will contain already reassembled + * messages. Retrieve any such messages and give them to the user. + */ +static void sctp_ulpq_reasm_drain(struct sctp_ulpq *ulpq) +{ + struct sctp_ulpevent *event = NULL; + struct sk_buff_head temp; + + if (skb_queue_empty(&ulpq->reasm)) + return; + + while ((event = sctp_ulpq_retrieve_reassembled(ulpq)) != NULL) { + /* Do ordering if needed. */ + if ((event) && (event->msg_flags & MSG_EOR)){ + skb_queue_head_init(&temp); + __skb_queue_tail(&temp, sctp_event2skb(event)); + + event = sctp_ulpq_order(ulpq, event); + } + + /* Send event to the ULP. 'event' is the + * sctp_ulpevent for very first SKB on the temp' list. + */ + if (event) + sctp_ulpq_tail_event(ulpq, event); + } +} + + /* Helper function to gather skbs that have possibly become * ordered by an an incoming chunk. */ -static inline void sctp_ulpq_retrieve_ordered(struct sctp_ulpq *ulpq, +static void sctp_ulpq_retrieve_ordered(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event) { struct sk_buff_head *event_list; @@ -608,7 +779,7 @@ static inline void sctp_ulpq_retrieve_ordered(struct sctp_ulpq *ulpq, } /* Helper function to store chunks needing ordering. */ -static inline void sctp_ulpq_store_ordered(struct sctp_ulpq *ulpq, +static void sctp_ulpq_store_ordered(struct sctp_ulpq *ulpq, struct sctp_ulpevent *event) { struct sk_buff *pos; @@ -696,13 +867,14 @@ static struct sctp_ulpevent *sctp_ulpq_order(struct sctp_ulpq *ulpq, /* Helper function to gather skbs that have possibly become * ordered by forward tsn skipping their dependencies. */ -static inline void sctp_ulpq_reap_ordered(struct sctp_ulpq *ulpq) +static void sctp_ulpq_reap_ordered(struct sctp_ulpq *ulpq, __u16 sid) { struct sk_buff *pos, *tmp; struct sctp_ulpevent *cevent; struct sctp_ulpevent *event; struct sctp_stream *in; struct sk_buff_head temp; + struct sk_buff_head *lobby = &ulpq->lobby; __u16 csid, cssn; in = &ulpq->asoc->ssnmap->in; @@ -710,36 +882,61 @@ static inline void sctp_ulpq_reap_ordered(struct sctp_ulpq *ulpq) /* We are holding the chunks by stream, by SSN. */ skb_queue_head_init(&temp); event = NULL; - sctp_skb_for_each(pos, &ulpq->lobby, tmp) { + sctp_skb_for_each(pos, lobby, tmp) { cevent = (struct sctp_ulpevent *) pos->cb; csid = cevent->stream; cssn = cevent->ssn; - if (cssn != sctp_ssn_peek(in, csid)) + /* Have we gone too far? */ + if (csid > sid) break; - /* Found it, so mark in the ssnmap. */ - sctp_ssn_next(in, csid); + /* Have we not gone far enough? */ + if (csid < sid) + continue; - __skb_unlink(pos, &ulpq->lobby); - if (!event) { + /* see if this ssn has been marked by skipping */ + if (!SSN_lt(cssn, sctp_ssn_peek(in, csid))) + break; + + __skb_unlink(pos, lobby); + if (!event) /* Create a temporary list to collect chunks on. */ event = sctp_skb2event(pos); - __skb_queue_tail(&temp, sctp_event2skb(event)); - } else { - /* Attach all gathered skbs to the event. */ + + /* Attach all gathered skbs to the event. */ + __skb_queue_tail(&temp, pos); + } + + /* If we didn't reap any data, see if the next expected SSN + * is next on the queue and if so, use that. + */ + if (event == NULL && pos != (struct sk_buff *)lobby) { + cevent = (struct sctp_ulpevent *) pos->cb; + csid = cevent->stream; + cssn = cevent->ssn; + + if (csid == sid && cssn == sctp_ssn_peek(in, csid)) { + sctp_ssn_next(in, csid); + __skb_unlink(pos, lobby); __skb_queue_tail(&temp, pos); + event = sctp_skb2event(pos); } } /* Send event to the ULP. 'event' is the sctp_ulpevent for * very first SKB on the 'temp' list. */ - if (event) + if (event) { + /* see if we have more ordered that we can deliver */ + sctp_ulpq_retrieve_ordered(ulpq, event); sctp_ulpq_tail_event(ulpq, event); + } } -/* Skip over an SSN. */ +/* Skip over an SSN. This is used during the processing of + * Forwared TSN chunk to skip over the abandoned ordered data + */ void sctp_ulpq_skip(struct sctp_ulpq *ulpq, __u16 sid, __u16 ssn) { struct sctp_stream *in; @@ -757,12 +954,12 @@ void sctp_ulpq_skip(struct sctp_ulpq *ulpq, __u16 sid, __u16 ssn) /* Go find any other chunks that were waiting for * ordering and deliver them if needed. */ - sctp_ulpq_reap_ordered(ulpq); + sctp_ulpq_reap_ordered(ulpq, sid); return; } -/* Renege 'needed' bytes from the ordering queue. */ -static __u16 sctp_ulpq_renege_order(struct sctp_ulpq *ulpq, __u16 needed) +static __u16 sctp_ulpq_renege_list(struct sctp_ulpq *ulpq, + struct sk_buff_head *list, __u16 needed) { __u16 freed = 0; __u32 tsn; @@ -772,7 +969,7 @@ static __u16 sctp_ulpq_renege_order(struct sctp_ulpq *ulpq, __u16 needed) tsnmap = &ulpq->asoc->peer.tsn_map; - while ((skb = __skb_dequeue_tail(&ulpq->lobby)) != NULL) { + while ((skb = __skb_dequeue_tail(list)) != NULL) { freed += skb_headlen(skb); event = sctp_skb2event(skb); tsn = event->tsn; @@ -786,30 +983,16 @@ static __u16 sctp_ulpq_renege_order(struct sctp_ulpq *ulpq, __u16 needed) return freed; } +/* Renege 'needed' bytes from the ordering queue. */ +static __u16 sctp_ulpq_renege_order(struct sctp_ulpq *ulpq, __u16 needed) +{ + return sctp_ulpq_renege_list(ulpq, &ulpq->lobby, needed); +} + /* Renege 'needed' bytes from the reassembly queue. */ static __u16 sctp_ulpq_renege_frags(struct sctp_ulpq *ulpq, __u16 needed) { - __u16 freed = 0; - __u32 tsn; - struct sk_buff *skb; - struct sctp_ulpevent *event; - struct sctp_tsnmap *tsnmap; - - tsnmap = &ulpq->asoc->peer.tsn_map; - - /* Walk backwards through the list, reneges the newest tsns. */ - while ((skb = __skb_dequeue_tail(&ulpq->reasm)) != NULL) { - freed += skb_headlen(skb); - event = sctp_skb2event(skb); - tsn = event->tsn; - - sctp_ulpevent_free(event); - sctp_tsnmap_renege(tsnmap, tsn); - if (freed >= needed) - return freed; - } - - return freed; + return sctp_ulpq_renege_list(ulpq, &ulpq->reasm, needed); } /* Partial deliver the first message as there is pressure on rwnd. */ @@ -819,19 +1002,29 @@ void sctp_ulpq_partial_delivery(struct sctp_ulpq *ulpq, { struct sctp_ulpevent *event; struct sctp_association *asoc; + struct sctp_sock *sp; asoc = ulpq->asoc; + sp = sctp_sk(asoc->base.sk); - /* Are we already in partial delivery mode? */ - if (!sctp_sk(asoc->base.sk)->pd_mode) { + /* If the association is already in Partial Delivery mode + * we have noting to do. + */ + if (ulpq->pd_mode) + return; + /* If the user enabled fragment interleave socket option, + * multiple associations can enter partial delivery. + * Otherwise, we can only enter partial delivery if the + * socket is not in partial deliver mode. + */ + if (sp->frag_interleave || atomic_read(&sp->pd_mode) == 0) { /* Is partial delivery possible? */ event = sctp_ulpq_retrieve_first(ulpq); /* Send event to the ULP. */ if (event) { sctp_ulpq_tail_event(ulpq, event); - sctp_sk(asoc->base.sk)->pd_mode = 1; - ulpq->pd_mode = 1; + sctp_ulpq_set_pd(ulpq); return; } } @@ -870,6 +1063,7 @@ void sctp_ulpq_renege(struct sctp_ulpq *ulpq, struct sctp_chunk *chunk, sctp_ulpq_partial_delivery(ulpq, chunk, gfp); } + sk_mem_reclaim(asoc->base.sk); return; }