freebsd-skq/sys/netinet/sctp_indata.c

5571 lines
170 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2001-2007, by Cisco Systems, Inc. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* a) Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
*
* b) Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the distribution.
*
* c) Neither the name of Cisco Systems, Inc. nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
* THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
* THE POSSIBILITY OF SUCH DAMAGE.
*/
/* $KAME: sctp_indata.c,v 1.36 2005/03/06 16:04:17 itojun Exp $ */
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <netinet/sctp_os.h>
#include <netinet/sctp_var.h>
#include <netinet/sctp_sysctl.h>
#include <netinet/sctp_pcb.h>
#include <netinet/sctp_header.h>
#include <netinet/sctputil.h>
#include <netinet/sctp_output.h>
#include <netinet/sctp_input.h>
#include <netinet/sctp_indata.h>
#include <netinet/sctp_uio.h>
#include <netinet/sctp_timer.h>
/*
* NOTES: On the outbound side of things I need to check the sack timer to
* see if I should generate a sack into the chunk queue (if I have data to
* send that is and will be sending it .. for bundling.
*
* The callback in sctp_usrreq.c will get called when the socket is read from.
* This will cause sctp_service_queues() to get called on the top entry in
* the list.
*/
void
sctp_set_rwnd(struct sctp_tcb *stcb, struct sctp_association *asoc)
{
asoc->my_rwnd = sctp_calc_rwnd(stcb, asoc);
}
/* Calculate what the rwnd would be */
uint32_t
sctp_calc_rwnd(struct sctp_tcb *stcb, struct sctp_association *asoc)
{
uint32_t calc = 0;
/*
* This is really set wrong with respect to a 1-2-m socket. Since
* the sb_cc is the count that everyone as put up. When we re-write
* sctp_soreceive then we will fix this so that ONLY this
* associations data is taken into account.
*/
if (stcb->sctp_socket == NULL)
return (calc);
if (stcb->asoc.sb_cc == 0 &&
asoc->size_on_reasm_queue == 0 &&
asoc->size_on_all_streams == 0) {
/* Full rwnd granted */
calc = max(SCTP_SB_LIMIT_RCV(stcb->sctp_socket), SCTP_MINIMAL_RWND);
return (calc);
}
/* get actual space */
calc = (uint32_t) sctp_sbspace(&stcb->asoc, &stcb->sctp_socket->so_rcv);
/*
* take out what has NOT been put on socket queue and we yet hold
* for putting up.
*/
calc = sctp_sbspace_sub(calc, (uint32_t) asoc->size_on_reasm_queue);
calc = sctp_sbspace_sub(calc, (uint32_t) asoc->size_on_all_streams);
if (calc == 0) {
/* out of space */
return (calc);
}
/* what is the overhead of all these rwnd's */
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
calc = sctp_sbspace_sub(calc, stcb->asoc.my_rwnd_control_len);
/*
* If the window gets too small due to ctrl-stuff, reduce it to 1,
* even it is 0. SWS engaged
*/
if (calc < stcb->asoc.my_rwnd_control_len) {
calc = 1;
}
return (calc);
}
/*
* Build out our readq entry based on the incoming packet.
*/
struct sctp_queued_to_read *
sctp_build_readq_entry(struct sctp_tcb *stcb,
struct sctp_nets *net,
uint32_t tsn, uint32_t ppid,
uint32_t context, uint16_t stream_no,
uint16_t stream_seq, uint8_t flags,
struct mbuf *dm)
{
struct sctp_queued_to_read *read_queue_e = NULL;
sctp_alloc_a_readq(stcb, read_queue_e);
if (read_queue_e == NULL) {
goto failed_build;
}
read_queue_e->sinfo_stream = stream_no;
read_queue_e->sinfo_ssn = stream_seq;
read_queue_e->sinfo_flags = (flags << 8);
read_queue_e->sinfo_ppid = ppid;
read_queue_e->sinfo_context = stcb->asoc.context;
read_queue_e->sinfo_timetolive = 0;
read_queue_e->sinfo_tsn = tsn;
read_queue_e->sinfo_cumtsn = tsn;
read_queue_e->sinfo_assoc_id = sctp_get_associd(stcb);
read_queue_e->whoFrom = net;
read_queue_e->length = 0;
atomic_add_int(&net->ref_count, 1);
read_queue_e->data = dm;
read_queue_e->spec_flags = 0;
read_queue_e->tail_mbuf = NULL;
read_queue_e->aux_data = NULL;
read_queue_e->stcb = stcb;
read_queue_e->port_from = stcb->rport;
read_queue_e->do_not_ref_stcb = 0;
read_queue_e->end_added = 0;
read_queue_e->some_taken = 0;
read_queue_e->pdapi_aborted = 0;
failed_build:
return (read_queue_e);
}
/*
* Build out our readq entry based on the incoming packet.
*/
static struct sctp_queued_to_read *
sctp_build_readq_entry_chk(struct sctp_tcb *stcb,
struct sctp_tmit_chunk *chk)
{
struct sctp_queued_to_read *read_queue_e = NULL;
sctp_alloc_a_readq(stcb, read_queue_e);
if (read_queue_e == NULL) {
goto failed_build;
}
read_queue_e->sinfo_stream = chk->rec.data.stream_number;
read_queue_e->sinfo_ssn = chk->rec.data.stream_seq;
read_queue_e->sinfo_flags = (chk->rec.data.rcv_flags << 8);
read_queue_e->sinfo_ppid = chk->rec.data.payloadtype;
read_queue_e->sinfo_context = stcb->asoc.context;
read_queue_e->sinfo_timetolive = 0;
read_queue_e->sinfo_tsn = chk->rec.data.TSN_seq;
read_queue_e->sinfo_cumtsn = chk->rec.data.TSN_seq;
read_queue_e->sinfo_assoc_id = sctp_get_associd(stcb);
read_queue_e->whoFrom = chk->whoTo;
read_queue_e->aux_data = NULL;
read_queue_e->length = 0;
atomic_add_int(&chk->whoTo->ref_count, 1);
read_queue_e->data = chk->data;
read_queue_e->tail_mbuf = NULL;
read_queue_e->stcb = stcb;
read_queue_e->port_from = stcb->rport;
read_queue_e->spec_flags = 0;
read_queue_e->do_not_ref_stcb = 0;
read_queue_e->end_added = 0;
read_queue_e->some_taken = 0;
read_queue_e->pdapi_aborted = 0;
failed_build:
return (read_queue_e);
}
struct mbuf *
sctp_build_ctl_nchunk(struct sctp_inpcb *inp,
struct sctp_sndrcvinfo *sinfo)
{
struct sctp_sndrcvinfo *outinfo;
struct cmsghdr *cmh;
struct mbuf *ret;
int len;
int use_extended = 0;
if (sctp_is_feature_off(inp, SCTP_PCB_FLAGS_RECVDATAIOEVNT)) {
/* user does not want the sndrcv ctl */
return (NULL);
}
if (sctp_is_feature_on(inp, SCTP_PCB_FLAGS_EXT_RCVINFO)) {
use_extended = 1;
len = CMSG_LEN(sizeof(struct sctp_extrcvinfo));
} else {
len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo));
}
ret = sctp_get_mbuf_for_msg(len,
0, M_DONTWAIT, 1, MT_DATA);
if (ret == NULL) {
/* No space */
return (ret);
}
/* We need a CMSG header followed by the struct */
cmh = mtod(ret, struct cmsghdr *);
outinfo = (struct sctp_sndrcvinfo *)CMSG_DATA(cmh);
cmh->cmsg_level = IPPROTO_SCTP;
if (use_extended) {
cmh->cmsg_type = SCTP_EXTRCV;
cmh->cmsg_len = len;
memcpy(outinfo, sinfo, len);
} else {
cmh->cmsg_type = SCTP_SNDRCV;
cmh->cmsg_len = len;
*outinfo = *sinfo;
}
SCTP_BUF_LEN(ret) = cmh->cmsg_len;
return (ret);
}
char *
sctp_build_ctl_cchunk(struct sctp_inpcb *inp,
int *control_len,
struct sctp_sndrcvinfo *sinfo)
{
struct sctp_sndrcvinfo *outinfo;
struct cmsghdr *cmh;
char *buf;
int len;
int use_extended = 0;
if (sctp_is_feature_off(inp, SCTP_PCB_FLAGS_RECVDATAIOEVNT)) {
/* user does not want the sndrcv ctl */
return (NULL);
}
if (sctp_is_feature_on(inp, SCTP_PCB_FLAGS_EXT_RCVINFO)) {
use_extended = 1;
len = CMSG_LEN(sizeof(struct sctp_extrcvinfo));
} else {
len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo));
}
SCTP_MALLOC(buf, char *, len, SCTP_M_CMSG);
if (buf == NULL) {
/* No space */
return (buf);
}
/* We need a CMSG header followed by the struct */
cmh = (struct cmsghdr *)buf;
outinfo = (struct sctp_sndrcvinfo *)CMSG_DATA(cmh);
cmh->cmsg_level = IPPROTO_SCTP;
if (use_extended) {
cmh->cmsg_type = SCTP_EXTRCV;
cmh->cmsg_len = len;
memcpy(outinfo, sinfo, len);
} else {
cmh->cmsg_type = SCTP_SNDRCV;
cmh->cmsg_len = len;
*outinfo = *sinfo;
}
*control_len = len;
return (buf);
}
/*
* We are delivering currently from the reassembly queue. We must continue to
* deliver until we either: 1) run out of space. 2) run out of sequential
* TSN's 3) hit the SCTP_DATA_LAST_FRAG flag.
*/
static void
sctp_service_reassembly(struct sctp_tcb *stcb, struct sctp_association *asoc)
{
struct sctp_tmit_chunk *chk;
uint16_t nxt_todel;
uint16_t stream_no;
int end = 0;
int cntDel;
struct sctp_queued_to_read *control, *ctl, *ctlat;
if (stcb == NULL)
return;
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
cntDel = stream_no = 0;
if ((stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_SOCKET_GONE) ||
- send call has a reference to uio->uio_resid in the recent send code, but uio may be NULL on sendfile calls. Change to use sndlen variable. - EMSGSIZE is not being returned in non-blocking mode and needs a small tweak to look if the msg would ever fit when returning EWOULDBLOCK. - FWD-TSN has a bug in stream processing which could cause a panic. This is a follow on to the codenomicon fix. - PDAPI level 1 and 2 do not work unless the reader gets his returned buffer full. Fix so we can break out when at level 1 or 2. - Fix fast-handoff features to copy across properly on accepted sockets - Fix sctp_peeloff() system call when no true system call exists to screen arguments for errors. In cases where a real system call exists the system call itself does this. - Fix raddr leak in recent add-ip code change for bundled asconfs (even when non-bundled asconfs are received) - Make sure ipi_addr lock is held when walking global addr list. Need to change this lock type to a rwlock(). - Add don't wake flag on both input and output when the socket is closing. - When deleting an address verify the interface is correct before allowing the delete to process. This protects panda and unnumbered. - Clean up old sysctl stuff and get rid of the old Open/Net BSD structures. - Add a function to watch the ranges in the sysctl sets. - When appending in the reassembly queue, validate that the assoc has not gone to about to be freed. If so (in the middle) abort out. Note this especially effects MAC I think due to the lock/unlock they do (or with LOCK testing in place). - Netstat patch to get rid of warnings. - Make sure that no data gets queued to inactive/unconfirmed destinations. This especially effect CMT but also makes a impact on regular SCTP as well. - During init collision when we detect seq number out of sync we need to treat it like Case C and discard the cookie (no invarient needed here). - Atomic access to the random store. - When we declare a vtag good, we need to shove it into the time wait hash to prevent further use. When the tag is put into the assoc hash, we need to remove it from the twait hash (where it will surely be). This prevents duplicate tag assignments. - Move decr-ref count to better protect sysctl out of data. - ltrace error corrections in sctp6_usrreq.c - Add hook for interface up/down to be sent to us. - Make sysctl() exported structures independent of processor architecture. - Fix route and src addr cache clearing for delete address case. - Make sure address marked SCTP_DEL_IP_ADDRESS is never selected as src addr. - in icmp handling fixed so we actually look at the icmp codes to figure out what to do. - Modified mobility code. Reception of DELETE IP ADDRESS for a primary destination and SET PRIMARY for a new primary destination is used for retransmission trigger to the new primary destination. Also, in this case, destination of chunks in send_queue are changed to the new primary destination. - Fix so that we disallow sending by mbuf to ever have EEOR mode set upon it. Approved by: re@freebsd.org (B Mah)
2007-09-08 17:48:46 +00:00
(stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) ||
(stcb->asoc.state & SCTP_STATE_CLOSED_SOCKET)) {
- send call has a reference to uio->uio_resid in the recent send code, but uio may be NULL on sendfile calls. Change to use sndlen variable. - EMSGSIZE is not being returned in non-blocking mode and needs a small tweak to look if the msg would ever fit when returning EWOULDBLOCK. - FWD-TSN has a bug in stream processing which could cause a panic. This is a follow on to the codenomicon fix. - PDAPI level 1 and 2 do not work unless the reader gets his returned buffer full. Fix so we can break out when at level 1 or 2. - Fix fast-handoff features to copy across properly on accepted sockets - Fix sctp_peeloff() system call when no true system call exists to screen arguments for errors. In cases where a real system call exists the system call itself does this. - Fix raddr leak in recent add-ip code change for bundled asconfs (even when non-bundled asconfs are received) - Make sure ipi_addr lock is held when walking global addr list. Need to change this lock type to a rwlock(). - Add don't wake flag on both input and output when the socket is closing. - When deleting an address verify the interface is correct before allowing the delete to process. This protects panda and unnumbered. - Clean up old sysctl stuff and get rid of the old Open/Net BSD structures. - Add a function to watch the ranges in the sysctl sets. - When appending in the reassembly queue, validate that the assoc has not gone to about to be freed. If so (in the middle) abort out. Note this especially effects MAC I think due to the lock/unlock they do (or with LOCK testing in place). - Netstat patch to get rid of warnings. - Make sure that no data gets queued to inactive/unconfirmed destinations. This especially effect CMT but also makes a impact on regular SCTP as well. - During init collision when we detect seq number out of sync we need to treat it like Case C and discard the cookie (no invarient needed here). - Atomic access to the random store. - When we declare a vtag good, we need to shove it into the time wait hash to prevent further use. When the tag is put into the assoc hash, we need to remove it from the twait hash (where it will surely be). This prevents duplicate tag assignments. - Move decr-ref count to better protect sysctl out of data. - ltrace error corrections in sctp6_usrreq.c - Add hook for interface up/down to be sent to us. - Make sysctl() exported structures independent of processor architecture. - Fix route and src addr cache clearing for delete address case. - Make sure address marked SCTP_DEL_IP_ADDRESS is never selected as src addr. - in icmp handling fixed so we actually look at the icmp codes to figure out what to do. - Modified mobility code. Reception of DELETE IP ADDRESS for a primary destination and SET PRIMARY for a new primary destination is used for retransmission trigger to the new primary destination. Also, in this case, destination of chunks in send_queue are changed to the new primary destination. - Fix so that we disallow sending by mbuf to ever have EEOR mode set upon it. Approved by: re@freebsd.org (B Mah)
2007-09-08 17:48:46 +00:00
/* socket above is long gone or going.. */
abandon:
asoc->fragmented_delivery_inprogress = 0;
chk = TAILQ_FIRST(&asoc->reasmqueue);
while (chk) {
TAILQ_REMOVE(&asoc->reasmqueue, chk, sctp_next);
asoc->size_on_reasm_queue -= chk->send_size;
sctp_ucount_decr(asoc->cnt_on_reasm_queue);
/*
* Lose the data pointer, since its in the socket
* buffer
*/
if (chk->data) {
sctp_m_freem(chk->data);
chk->data = NULL;
}
/* Now free the address and data */
sctp_free_a_chunk(stcb, chk);
/* sa_ignore FREED_MEMORY */
chk = TAILQ_FIRST(&asoc->reasmqueue);
}
return;
}
SCTP_TCB_LOCK_ASSERT(stcb);
do {
chk = TAILQ_FIRST(&asoc->reasmqueue);
if (chk == NULL) {
return;
}
if (chk->rec.data.TSN_seq != (asoc->tsn_last_delivered + 1)) {
/* Can't deliver more :< */
return;
}
stream_no = chk->rec.data.stream_number;
nxt_todel = asoc->strmin[stream_no].last_sequence_delivered + 1;
if (nxt_todel != chk->rec.data.stream_seq &&
(chk->rec.data.rcv_flags & SCTP_DATA_UNORDERED) == 0) {
/*
* Not the next sequence to deliver in its stream OR
* unordered
*/
return;
}
if (chk->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG) {
control = sctp_build_readq_entry_chk(stcb, chk);
if (control == NULL) {
/* out of memory? */
return;
}
/* save it off for our future deliveries */
stcb->asoc.control_pdapi = control;
if (chk->rec.data.rcv_flags & SCTP_DATA_LAST_FRAG)
end = 1;
else
end = 0;
sctp_add_to_readq(stcb->sctp_ep,
stcb, control, &stcb->sctp_socket->so_rcv, end, SCTP_SO_NOT_LOCKED);
cntDel++;
} else {
if (chk->rec.data.rcv_flags & SCTP_DATA_LAST_FRAG)
end = 1;
else
end = 0;
if (sctp_append_to_readq(stcb->sctp_ep, stcb,
stcb->asoc.control_pdapi,
chk->data, end, chk->rec.data.TSN_seq,
&stcb->sctp_socket->so_rcv)) {
/*
* something is very wrong, either
* control_pdapi is NULL, or the tail_mbuf
* is corrupt, or there is a EOM already on
* the mbuf chain.
*/
- send call has a reference to uio->uio_resid in the recent send code, but uio may be NULL on sendfile calls. Change to use sndlen variable. - EMSGSIZE is not being returned in non-blocking mode and needs a small tweak to look if the msg would ever fit when returning EWOULDBLOCK. - FWD-TSN has a bug in stream processing which could cause a panic. This is a follow on to the codenomicon fix. - PDAPI level 1 and 2 do not work unless the reader gets his returned buffer full. Fix so we can break out when at level 1 or 2. - Fix fast-handoff features to copy across properly on accepted sockets - Fix sctp_peeloff() system call when no true system call exists to screen arguments for errors. In cases where a real system call exists the system call itself does this. - Fix raddr leak in recent add-ip code change for bundled asconfs (even when non-bundled asconfs are received) - Make sure ipi_addr lock is held when walking global addr list. Need to change this lock type to a rwlock(). - Add don't wake flag on both input and output when the socket is closing. - When deleting an address verify the interface is correct before allowing the delete to process. This protects panda and unnumbered. - Clean up old sysctl stuff and get rid of the old Open/Net BSD structures. - Add a function to watch the ranges in the sysctl sets. - When appending in the reassembly queue, validate that the assoc has not gone to about to be freed. If so (in the middle) abort out. Note this especially effects MAC I think due to the lock/unlock they do (or with LOCK testing in place). - Netstat patch to get rid of warnings. - Make sure that no data gets queued to inactive/unconfirmed destinations. This especially effect CMT but also makes a impact on regular SCTP as well. - During init collision when we detect seq number out of sync we need to treat it like Case C and discard the cookie (no invarient needed here). - Atomic access to the random store. - When we declare a vtag good, we need to shove it into the time wait hash to prevent further use. When the tag is put into the assoc hash, we need to remove it from the twait hash (where it will surely be). This prevents duplicate tag assignments. - Move decr-ref count to better protect sysctl out of data. - ltrace error corrections in sctp6_usrreq.c - Add hook for interface up/down to be sent to us. - Make sysctl() exported structures independent of processor architecture. - Fix route and src addr cache clearing for delete address case. - Make sure address marked SCTP_DEL_IP_ADDRESS is never selected as src addr. - in icmp handling fixed so we actually look at the icmp codes to figure out what to do. - Modified mobility code. Reception of DELETE IP ADDRESS for a primary destination and SET PRIMARY for a new primary destination is used for retransmission trigger to the new primary destination. Also, in this case, destination of chunks in send_queue are changed to the new primary destination. - Fix so that we disallow sending by mbuf to ever have EEOR mode set upon it. Approved by: re@freebsd.org (B Mah)
2007-09-08 17:48:46 +00:00
if (stcb->asoc.state & SCTP_STATE_ABOUT_TO_BE_FREED) {
goto abandon;
} else {
#ifdef INVARIANTS
- send call has a reference to uio->uio_resid in the recent send code, but uio may be NULL on sendfile calls. Change to use sndlen variable. - EMSGSIZE is not being returned in non-blocking mode and needs a small tweak to look if the msg would ever fit when returning EWOULDBLOCK. - FWD-TSN has a bug in stream processing which could cause a panic. This is a follow on to the codenomicon fix. - PDAPI level 1 and 2 do not work unless the reader gets his returned buffer full. Fix so we can break out when at level 1 or 2. - Fix fast-handoff features to copy across properly on accepted sockets - Fix sctp_peeloff() system call when no true system call exists to screen arguments for errors. In cases where a real system call exists the system call itself does this. - Fix raddr leak in recent add-ip code change for bundled asconfs (even when non-bundled asconfs are received) - Make sure ipi_addr lock is held when walking global addr list. Need to change this lock type to a rwlock(). - Add don't wake flag on both input and output when the socket is closing. - When deleting an address verify the interface is correct before allowing the delete to process. This protects panda and unnumbered. - Clean up old sysctl stuff and get rid of the old Open/Net BSD structures. - Add a function to watch the ranges in the sysctl sets. - When appending in the reassembly queue, validate that the assoc has not gone to about to be freed. If so (in the middle) abort out. Note this especially effects MAC I think due to the lock/unlock they do (or with LOCK testing in place). - Netstat patch to get rid of warnings. - Make sure that no data gets queued to inactive/unconfirmed destinations. This especially effect CMT but also makes a impact on regular SCTP as well. - During init collision when we detect seq number out of sync we need to treat it like Case C and discard the cookie (no invarient needed here). - Atomic access to the random store. - When we declare a vtag good, we need to shove it into the time wait hash to prevent further use. When the tag is put into the assoc hash, we need to remove it from the twait hash (where it will surely be). This prevents duplicate tag assignments. - Move decr-ref count to better protect sysctl out of data. - ltrace error corrections in sctp6_usrreq.c - Add hook for interface up/down to be sent to us. - Make sysctl() exported structures independent of processor architecture. - Fix route and src addr cache clearing for delete address case. - Make sure address marked SCTP_DEL_IP_ADDRESS is never selected as src addr. - in icmp handling fixed so we actually look at the icmp codes to figure out what to do. - Modified mobility code. Reception of DELETE IP ADDRESS for a primary destination and SET PRIMARY for a new primary destination is used for retransmission trigger to the new primary destination. Also, in this case, destination of chunks in send_queue are changed to the new primary destination. - Fix so that we disallow sending by mbuf to ever have EEOR mode set upon it. Approved by: re@freebsd.org (B Mah)
2007-09-08 17:48:46 +00:00
if ((stcb->asoc.control_pdapi == NULL) || (stcb->asoc.control_pdapi->tail_mbuf == NULL)) {
panic("This should not happen control_pdapi NULL?");
}
/* if we did not panic, it was a EOM */
panic("Bad chunking ??");
#else
if ((stcb->asoc.control_pdapi == NULL) || (stcb->asoc.control_pdapi->tail_mbuf == NULL)) {
SCTP_PRINTF("This should not happen control_pdapi NULL?\n");
}
SCTP_PRINTF("Bad chunking ??\n");
SCTP_PRINTF("Dumping re-assembly queue this will probably hose the association\n");
#endif
goto abandon;
}
}
cntDel++;
}
/* pull it we did it */
TAILQ_REMOVE(&asoc->reasmqueue, chk, sctp_next);
if (chk->rec.data.rcv_flags & SCTP_DATA_LAST_FRAG) {
asoc->fragmented_delivery_inprogress = 0;
if ((chk->rec.data.rcv_flags & SCTP_DATA_UNORDERED) == 0) {
asoc->strmin[stream_no].last_sequence_delivered++;
}
if ((chk->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG) == 0) {
SCTP_STAT_INCR_COUNTER64(sctps_reasmusrmsgs);
}
} else if (chk->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG) {
/*
* turn the flag back on since we just delivered
* yet another one.
*/
asoc->fragmented_delivery_inprogress = 1;
}
asoc->tsn_of_pdapi_last_delivered = chk->rec.data.TSN_seq;
asoc->last_flags_delivered = chk->rec.data.rcv_flags;
asoc->last_strm_seq_delivered = chk->rec.data.stream_seq;
asoc->last_strm_no_delivered = chk->rec.data.stream_number;
asoc->tsn_last_delivered = chk->rec.data.TSN_seq;
asoc->size_on_reasm_queue -= chk->send_size;
sctp_ucount_decr(asoc->cnt_on_reasm_queue);
/* free up the chk */
chk->data = NULL;
sctp_free_a_chunk(stcb, chk);
if (asoc->fragmented_delivery_inprogress == 0) {
/*
* Now lets see if we can deliver the next one on
* the stream
*/
struct sctp_stream_in *strm;
strm = &asoc->strmin[stream_no];
nxt_todel = strm->last_sequence_delivered + 1;
ctl = TAILQ_FIRST(&strm->inqueue);
if (ctl && (nxt_todel == ctl->sinfo_ssn)) {
while (ctl != NULL) {
/* Deliver more if we can. */
if (nxt_todel == ctl->sinfo_ssn) {
ctlat = TAILQ_NEXT(ctl, next);
TAILQ_REMOVE(&strm->inqueue, ctl, next);
asoc->size_on_all_streams -= ctl->length;
sctp_ucount_decr(asoc->cnt_on_all_streams);
strm->last_sequence_delivered++;
sctp_add_to_readq(stcb->sctp_ep, stcb,
ctl,
&stcb->sctp_socket->so_rcv, 1, SCTP_SO_NOT_LOCKED);
ctl = ctlat;
} else {
break;
}
nxt_todel = strm->last_sequence_delivered + 1;
}
}
break;
}
/* sa_ignore FREED_MEMORY */
chk = TAILQ_FIRST(&asoc->reasmqueue);
} while (chk);
}
/*
* Queue the chunk either right into the socket buffer if it is the next one
* to go OR put it in the correct place in the delivery queue. If we do
* append to the so_buf, keep doing so until we are out of order. One big
* question still remains, what to do when the socket buffer is FULL??
*/
static void
sctp_queue_data_to_stream(struct sctp_tcb *stcb, struct sctp_association *asoc,
struct sctp_queued_to_read *control, int *abort_flag)
{
/*
* FIX-ME maybe? What happens when the ssn wraps? If we are getting
* all the data in one stream this could happen quite rapidly. One
* could use the TSN to keep track of things, but this scheme breaks
* down in the other type of stream useage that could occur. Send a
* single msg to stream 0, send 4Billion messages to stream 1, now
* send a message to stream 0. You have a situation where the TSN
* has wrapped but not in the stream. Is this worth worrying about
* or should we just change our queue sort at the bottom to be by
* TSN.
*
* Could it also be legal for a peer to send ssn 1 with TSN 2 and ssn 2
* with TSN 1? If the peer is doing some sort of funky TSN/SSN
* assignment this could happen... and I don't see how this would be
* a violation. So for now I am undecided an will leave the sort by
* SSN alone. Maybe a hybred approach is the answer
*
*/
struct sctp_stream_in *strm;
struct sctp_queued_to_read *at;
int queue_needed;
uint16_t nxt_todel;
struct mbuf *oper;
queue_needed = 1;
asoc->size_on_all_streams += control->length;
sctp_ucount_incr(asoc->cnt_on_all_streams);
strm = &asoc->strmin[control->sinfo_stream];
nxt_todel = strm->last_sequence_delivered + 1;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_STR_LOGGING_ENABLE) {
sctp_log_strm_del(control, NULL, SCTP_STR_LOG_FROM_INTO_STRD);
}
SCTPDBG(SCTP_DEBUG_INDATA1,
"queue to stream called for ssn:%u lastdel:%u nxt:%u\n",
(uint32_t) control->sinfo_stream,
(uint32_t) strm->last_sequence_delivered,
(uint32_t) nxt_todel);
if (compare_with_wrap(strm->last_sequence_delivered,
control->sinfo_ssn, MAX_SEQ) ||
(strm->last_sequence_delivered == control->sinfo_ssn)) {
/* The incoming sseq is behind where we last delivered? */
SCTPDBG(SCTP_DEBUG_INDATA1, "Duplicate S-SEQ:%d delivered:%d from peer, Abort association\n",
control->sinfo_ssn, strm->last_sequence_delivered);
protocol_error:
/*
* throw it in the stream so it gets cleaned up in
* association destruction
*/
TAILQ_INSERT_HEAD(&strm->inqueue, control, next);
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) = sizeof(struct sctp_paramhdr) +
(sizeof(uint32_t) * 3);
ph = mtod(oper, struct sctp_paramhdr *);
ph->param_type = htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length = htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_1);
ippp++;
*ippp = control->sinfo_tsn;
ippp++;
*ippp = ((control->sinfo_stream << 16) | control->sinfo_ssn);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_1;
sctp_abort_an_association(stcb->sctp_ep, stcb,
SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return;
}
if (nxt_todel == control->sinfo_ssn) {
/* can be delivered right away? */
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_STR_LOGGING_ENABLE) {
sctp_log_strm_del(control, NULL, SCTP_STR_LOG_FROM_IMMED_DEL);
}
queue_needed = 0;
asoc->size_on_all_streams -= control->length;
sctp_ucount_decr(asoc->cnt_on_all_streams);
strm->last_sequence_delivered++;
sctp_add_to_readq(stcb->sctp_ep, stcb,
control,
&stcb->sctp_socket->so_rcv, 1, SCTP_SO_NOT_LOCKED);
control = TAILQ_FIRST(&strm->inqueue);
while (control != NULL) {
/* all delivered */
nxt_todel = strm->last_sequence_delivered + 1;
if (nxt_todel == control->sinfo_ssn) {
at = TAILQ_NEXT(control, next);
TAILQ_REMOVE(&strm->inqueue, control, next);
asoc->size_on_all_streams -= control->length;
sctp_ucount_decr(asoc->cnt_on_all_streams);
strm->last_sequence_delivered++;
/*
* We ignore the return of deliver_data here
* since we always can hold the chunk on the
* d-queue. And we have a finite number that
* can be delivered from the strq.
*/
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_STR_LOGGING_ENABLE) {
sctp_log_strm_del(control, NULL,
SCTP_STR_LOG_FROM_IMMED_DEL);
}
sctp_add_to_readq(stcb->sctp_ep, stcb,
control,
&stcb->sctp_socket->so_rcv, 1, SCTP_SO_NOT_LOCKED);
control = at;
continue;
}
break;
}
}
if (queue_needed) {
/*
* Ok, we did not deliver this guy, find the correct place
* to put it on the queue.
*/
if ((compare_with_wrap(asoc->cumulative_tsn,
control->sinfo_tsn, MAX_TSN)) ||
(control->sinfo_tsn == asoc->cumulative_tsn)) {
goto protocol_error;
}
if (TAILQ_EMPTY(&strm->inqueue)) {
/* Empty queue */
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_STR_LOGGING_ENABLE) {
sctp_log_strm_del(control, NULL, SCTP_STR_LOG_FROM_INSERT_HD);
}
TAILQ_INSERT_HEAD(&strm->inqueue, control, next);
} else {
TAILQ_FOREACH(at, &strm->inqueue, next) {
if (compare_with_wrap(at->sinfo_ssn,
control->sinfo_ssn, MAX_SEQ)) {
/*
* one in queue is bigger than the
* new one, insert before this one
*/
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_STR_LOGGING_ENABLE) {
sctp_log_strm_del(control, at,
SCTP_STR_LOG_FROM_INSERT_MD);
}
TAILQ_INSERT_BEFORE(at, control, next);
break;
} else if (at->sinfo_ssn == control->sinfo_ssn) {
/*
* Gak, He sent me a duplicate str
* seq number
*/
/*
* foo bar, I guess I will just free
* this new guy, should we abort
* too? FIX ME MAYBE? Or it COULD be
* that the SSN's have wrapped.
* Maybe I should compare to TSN
* somehow... sigh for now just blow
* away the chunk!
*/
if (control->data)
sctp_m_freem(control->data);
control->data = NULL;
asoc->size_on_all_streams -= control->length;
sctp_ucount_decr(asoc->cnt_on_all_streams);
if (control->whoFrom)
sctp_free_remote_addr(control->whoFrom);
control->whoFrom = NULL;
sctp_free_a_readq(stcb, control);
return;
} else {
if (TAILQ_NEXT(at, next) == NULL) {
/*
* We are at the end, insert
* it after this one
*/
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_STR_LOGGING_ENABLE) {
sctp_log_strm_del(control, at,
SCTP_STR_LOG_FROM_INSERT_TL);
}
TAILQ_INSERT_AFTER(&strm->inqueue,
at, control, next);
break;
}
}
}
}
}
}
/*
* Returns two things: You get the total size of the deliverable parts of the
* first fragmented message on the reassembly queue. And you get a 1 back if
* all of the message is ready or a 0 back if the message is still incomplete
*/
static int
sctp_is_all_msg_on_reasm(struct sctp_association *asoc, uint32_t * t_size)
{
struct sctp_tmit_chunk *chk;
uint32_t tsn;
*t_size = 0;
chk = TAILQ_FIRST(&asoc->reasmqueue);
if (chk == NULL) {
/* nothing on the queue */
return (0);
}
if ((chk->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG) == 0) {
/* Not a first on the queue */
return (0);
}
tsn = chk->rec.data.TSN_seq;
while (chk) {
if (tsn != chk->rec.data.TSN_seq) {
return (0);
}
*t_size += chk->send_size;
if (chk->rec.data.rcv_flags & SCTP_DATA_LAST_FRAG) {
return (1);
}
tsn++;
chk = TAILQ_NEXT(chk, sctp_next);
}
return (0);
}
static void
sctp_deliver_reasm_check(struct sctp_tcb *stcb, struct sctp_association *asoc)
{
struct sctp_tmit_chunk *chk;
uint16_t nxt_todel;
uint32_t tsize;
doit_again:
chk = TAILQ_FIRST(&asoc->reasmqueue);
if (chk == NULL) {
/* Huh? */
asoc->size_on_reasm_queue = 0;
asoc->cnt_on_reasm_queue = 0;
return;
}
if (asoc->fragmented_delivery_inprogress == 0) {
nxt_todel =
asoc->strmin[chk->rec.data.stream_number].last_sequence_delivered + 1;
if ((chk->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG) &&
(nxt_todel == chk->rec.data.stream_seq ||
(chk->rec.data.rcv_flags & SCTP_DATA_UNORDERED))) {
/*
* Yep the first one is here and its ok to deliver
* but should we?
*/
if ((sctp_is_all_msg_on_reasm(asoc, &tsize) ||
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
(tsize >= stcb->sctp_ep->partial_delivery_point))) {
/*
* Yes, we setup to start reception, by
* backing down the TSN just in case we
* can't deliver. If we
*/
asoc->fragmented_delivery_inprogress = 1;
asoc->tsn_last_delivered =
chk->rec.data.TSN_seq - 1;
asoc->str_of_pdapi =
chk->rec.data.stream_number;
asoc->ssn_of_pdapi = chk->rec.data.stream_seq;
asoc->pdapi_ppid = chk->rec.data.payloadtype;
asoc->fragment_flags = chk->rec.data.rcv_flags;
sctp_service_reassembly(stcb, asoc);
}
}
} else {
/*
* Service re-assembly will deliver stream data queued at
* the end of fragmented delivery.. but it wont know to go
* back and call itself again... we do that here with the
* got doit_again
*/
sctp_service_reassembly(stcb, asoc);
if (asoc->fragmented_delivery_inprogress == 0) {
/*
* finished our Fragmented delivery, could be more
* waiting?
*/
goto doit_again;
}
}
}
/*
* Dump onto the re-assembly queue, in its proper place. After dumping on the
* queue, see if anthing can be delivered. If so pull it off (or as much as
* we can. If we run out of space then we must dump what we can and set the
* appropriate flag to say we queued what we could.
*/
static void
sctp_queue_data_for_reasm(struct sctp_tcb *stcb, struct sctp_association *asoc,
struct sctp_tmit_chunk *chk, int *abort_flag)
{
struct mbuf *oper;
uint32_t cum_ackp1, last_tsn, prev_tsn, post_tsn;
u_char last_flags;
struct sctp_tmit_chunk *at, *prev, *next;
prev = next = NULL;
cum_ackp1 = asoc->tsn_last_delivered + 1;
if (TAILQ_EMPTY(&asoc->reasmqueue)) {
/* This is the first one on the queue */
TAILQ_INSERT_HEAD(&asoc->reasmqueue, chk, sctp_next);
/*
* we do not check for delivery of anything when only one
* fragment is here
*/
asoc->size_on_reasm_queue = chk->send_size;
sctp_ucount_incr(asoc->cnt_on_reasm_queue);
if (chk->rec.data.TSN_seq == cum_ackp1) {
if (asoc->fragmented_delivery_inprogress == 0 &&
(chk->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG) !=
SCTP_DATA_FIRST_FRAG) {
/*
* An empty queue, no delivery inprogress,
* we hit the next one and it does NOT have
* a FIRST fragment mark.
*/
SCTPDBG(SCTP_DEBUG_INDATA1, "Gak, Evil plot, its not first, no fragmented delivery in progress\n");
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(sizeof(uint32_t) * 3);
ph = mtod(oper, struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length = htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_2);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_2;
sctp_abort_an_association(stcb->sctp_ep, stcb,
SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
} else if (asoc->fragmented_delivery_inprogress &&
(chk->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG) == SCTP_DATA_FIRST_FRAG) {
/*
* We are doing a partial delivery and the
* NEXT chunk MUST be either the LAST or
* MIDDLE fragment NOT a FIRST
*/
SCTPDBG(SCTP_DEBUG_INDATA1, "Gak, Evil plot, it IS a first and fragmented delivery in progress\n");
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper, struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length = htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_3);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_3;
sctp_abort_an_association(stcb->sctp_ep, stcb,
SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
} else if (asoc->fragmented_delivery_inprogress) {
/*
* Here we are ok with a MIDDLE or LAST
* piece
*/
if (chk->rec.data.stream_number !=
asoc->str_of_pdapi) {
/* Got to be the right STR No */
SCTPDBG(SCTP_DEBUG_INDATA1, "Gak, Evil plot, it IS not same stream number %d vs %d\n",
chk->rec.data.stream_number,
asoc->str_of_pdapi);
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(sizeof(uint32_t) * 3);
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_4);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_4;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
} else if ((asoc->fragment_flags & SCTP_DATA_UNORDERED) !=
SCTP_DATA_UNORDERED &&
chk->rec.data.stream_seq !=
asoc->ssn_of_pdapi) {
/* Got to be the right STR Seq */
SCTPDBG(SCTP_DEBUG_INDATA1, "Gak, Evil plot, it IS not same stream seq %d vs %d\n",
chk->rec.data.stream_seq,
asoc->ssn_of_pdapi);
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_5);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_5;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
}
}
}
return;
}
/* Find its place */
TAILQ_FOREACH(at, &asoc->reasmqueue, sctp_next) {
if (compare_with_wrap(at->rec.data.TSN_seq,
chk->rec.data.TSN_seq, MAX_TSN)) {
/*
* one in queue is bigger than the new one, insert
* before this one
*/
/* A check */
asoc->size_on_reasm_queue += chk->send_size;
sctp_ucount_incr(asoc->cnt_on_reasm_queue);
next = at;
TAILQ_INSERT_BEFORE(at, chk, sctp_next);
break;
} else if (at->rec.data.TSN_seq == chk->rec.data.TSN_seq) {
/* Gak, He sent me a duplicate str seq number */
/*
* foo bar, I guess I will just free this new guy,
* should we abort too? FIX ME MAYBE? Or it COULD be
* that the SSN's have wrapped. Maybe I should
* compare to TSN somehow... sigh for now just blow
* away the chunk!
*/
if (chk->data) {
sctp_m_freem(chk->data);
chk->data = NULL;
}
sctp_free_a_chunk(stcb, chk);
return;
} else {
last_flags = at->rec.data.rcv_flags;
last_tsn = at->rec.data.TSN_seq;
prev = at;
if (TAILQ_NEXT(at, sctp_next) == NULL) {
/*
* We are at the end, insert it after this
* one
*/
/* check it first */
asoc->size_on_reasm_queue += chk->send_size;
sctp_ucount_incr(asoc->cnt_on_reasm_queue);
TAILQ_INSERT_AFTER(&asoc->reasmqueue, at, chk, sctp_next);
break;
}
}
}
/* Now the audits */
if (prev) {
prev_tsn = chk->rec.data.TSN_seq - 1;
if (prev_tsn == prev->rec.data.TSN_seq) {
/*
* Ok the one I am dropping onto the end is the
* NEXT. A bit of valdiation here.
*/
if ((prev->rec.data.rcv_flags & SCTP_DATA_FRAG_MASK) ==
SCTP_DATA_FIRST_FRAG ||
(prev->rec.data.rcv_flags & SCTP_DATA_FRAG_MASK) ==
SCTP_DATA_MIDDLE_FRAG) {
/*
* Insert chk MUST be a MIDDLE or LAST
* fragment
*/
if ((chk->rec.data.rcv_flags & SCTP_DATA_FRAG_MASK) ==
SCTP_DATA_FIRST_FRAG) {
SCTPDBG(SCTP_DEBUG_INDATA1, "Prev check - It can be a midlle or last but not a first\n");
SCTPDBG(SCTP_DEBUG_INDATA1, "Gak, Evil plot, it's a FIRST!\n");
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_6);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_6;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return;
}
if (chk->rec.data.stream_number !=
prev->rec.data.stream_number) {
/*
* Huh, need the correct STR here,
* they must be the same.
*/
SCTP_PRINTF("Prev check - Gak, Evil plot, ssn:%d not the same as at:%d\n",
chk->rec.data.stream_number,
prev->rec.data.stream_number);
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_7);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_7;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return;
}
if ((prev->rec.data.rcv_flags & SCTP_DATA_UNORDERED) == 0 &&
chk->rec.data.stream_seq !=
prev->rec.data.stream_seq) {
/*
* Huh, need the correct STR here,
* they must be the same.
*/
SCTPDBG(SCTP_DEBUG_INDATA1, "Prev check - Gak, Evil plot, sseq:%d not the same as at:%d\n",
chk->rec.data.stream_seq,
prev->rec.data.stream_seq);
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_8);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_8;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return;
}
} else if ((prev->rec.data.rcv_flags & SCTP_DATA_FRAG_MASK) ==
SCTP_DATA_LAST_FRAG) {
/* Insert chk MUST be a FIRST */
if ((chk->rec.data.rcv_flags & SCTP_DATA_FRAG_MASK) !=
SCTP_DATA_FIRST_FRAG) {
SCTPDBG(SCTP_DEBUG_INDATA1, "Prev check - Gak, evil plot, its not FIRST and it must be!\n");
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_9);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_9;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return;
}
}
}
}
if (next) {
post_tsn = chk->rec.data.TSN_seq + 1;
if (post_tsn == next->rec.data.TSN_seq) {
/*
* Ok the one I am inserting ahead of is my NEXT
* one. A bit of valdiation here.
*/
if (next->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG) {
/* Insert chk MUST be a last fragment */
if ((chk->rec.data.rcv_flags & SCTP_DATA_FRAG_MASK)
!= SCTP_DATA_LAST_FRAG) {
SCTPDBG(SCTP_DEBUG_INDATA1, "Next chk - Next is FIRST, we must be LAST\n");
SCTPDBG(SCTP_DEBUG_INDATA1, "Gak, Evil plot, its not a last!\n");
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_10);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_10;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return;
}
} else if ((next->rec.data.rcv_flags & SCTP_DATA_FRAG_MASK) ==
SCTP_DATA_MIDDLE_FRAG ||
(next->rec.data.rcv_flags & SCTP_DATA_FRAG_MASK) ==
SCTP_DATA_LAST_FRAG) {
/*
* Insert chk CAN be MIDDLE or FIRST NOT
* LAST
*/
if ((chk->rec.data.rcv_flags & SCTP_DATA_FRAG_MASK) ==
SCTP_DATA_LAST_FRAG) {
SCTPDBG(SCTP_DEBUG_INDATA1, "Next chk - Next is a MIDDLE/LAST\n");
SCTPDBG(SCTP_DEBUG_INDATA1, "Gak, Evil plot, new prev chunk is a LAST\n");
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_11);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_11;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return;
}
if (chk->rec.data.stream_number !=
next->rec.data.stream_number) {
/*
* Huh, need the correct STR here,
* they must be the same.
*/
SCTPDBG(SCTP_DEBUG_INDATA1, "Next chk - Gak, Evil plot, ssn:%d not the same as at:%d\n",
chk->rec.data.stream_number,
next->rec.data.stream_number);
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_12);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_12;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return;
}
if ((next->rec.data.rcv_flags & SCTP_DATA_UNORDERED) == 0 &&
chk->rec.data.stream_seq !=
next->rec.data.stream_seq) {
/*
* Huh, need the correct STR here,
* they must be the same.
*/
SCTPDBG(SCTP_DEBUG_INDATA1, "Next chk - Gak, Evil plot, sseq:%d not the same as at:%d\n",
chk->rec.data.stream_seq,
next->rec.data.stream_seq);
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_13);
ippp++;
*ippp = chk->rec.data.TSN_seq;
ippp++;
*ippp = ((chk->rec.data.stream_number << 16) | chk->rec.data.stream_seq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_13;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return;
}
}
}
}
/* Do we need to do some delivery? check */
sctp_deliver_reasm_check(stcb, asoc);
}
/*
* This is an unfortunate routine. It checks to make sure a evil guy is not
* stuffing us full of bad packet fragments. A broken peer could also do this
* but this is doubtful. It is to bad I must worry about evil crackers sigh
* :< more cycles.
*/
static int
sctp_does_tsn_belong_to_reasm(struct sctp_association *asoc,
uint32_t TSN_seq)
{
struct sctp_tmit_chunk *at;
uint32_t tsn_est;
TAILQ_FOREACH(at, &asoc->reasmqueue, sctp_next) {
if (compare_with_wrap(TSN_seq,
at->rec.data.TSN_seq, MAX_TSN)) {
/* is it one bigger? */
tsn_est = at->rec.data.TSN_seq + 1;
if (tsn_est == TSN_seq) {
/* yep. It better be a last then */
if ((at->rec.data.rcv_flags & SCTP_DATA_FRAG_MASK) !=
SCTP_DATA_LAST_FRAG) {
/*
* Ok this guy belongs next to a guy
* that is NOT last, it should be a
* middle/last, not a complete
* chunk.
*/
return (1);
} else {
/*
* This guy is ok since its a LAST
* and the new chunk is a fully
* self- contained one.
*/
return (0);
}
}
} else if (TSN_seq == at->rec.data.TSN_seq) {
/* Software error since I have a dup? */
return (1);
} else {
/*
* Ok, 'at' is larger than new chunk but does it
* need to be right before it.
*/
tsn_est = TSN_seq + 1;
if (tsn_est == at->rec.data.TSN_seq) {
/* Yep, It better be a first */
if ((at->rec.data.rcv_flags & SCTP_DATA_FRAG_MASK) !=
SCTP_DATA_FIRST_FRAG) {
return (1);
} else {
return (0);
}
}
}
}
return (0);
}
static int
sctp_process_a_data_chunk(struct sctp_tcb *stcb, struct sctp_association *asoc,
struct mbuf **m, int offset, struct sctp_data_chunk *ch, int chk_length,
struct sctp_nets *net, uint32_t * high_tsn, int *abort_flag,
int *break_flag, int last_chunk)
{
/* Process a data chunk */
/* struct sctp_tmit_chunk *chk; */
struct sctp_tmit_chunk *chk;
uint32_t tsn, gap;
struct mbuf *dmbuf;
int indx, the_len;
int need_reasm_check = 0;
uint16_t strmno, strmseq;
struct mbuf *oper;
struct sctp_queued_to_read *control;
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
int ordered;
uint32_t protocol_id;
uint8_t chunk_flags;
struct sctp_stream_reset_list *liste;
chk = NULL;
tsn = ntohl(ch->dp.tsn);
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
chunk_flags = ch->ch.chunk_flags;
if ((chunk_flags & SCTP_DATA_SACK_IMMEDIATELY) == SCTP_DATA_SACK_IMMEDIATELY) {
asoc->send_sack = 1;
}
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
protocol_id = ch->dp.protocol_id;
ordered = ((ch->ch.chunk_flags & SCTP_DATA_UNORDERED) == 0);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
sctp_log_map(tsn, asoc->cumulative_tsn, asoc->highest_tsn_inside_map, SCTP_MAP_TSN_ENTERS);
}
if (stcb == NULL) {
return (0);
}
SCTP_LTRACE_CHK(stcb->sctp_ep, stcb, ch->ch.chunk_type, tsn);
if (compare_with_wrap(asoc->cumulative_tsn, tsn, MAX_TSN) ||
asoc->cumulative_tsn == tsn) {
/* It is a duplicate */
SCTP_STAT_INCR(sctps_recvdupdata);
if (asoc->numduptsns < SCTP_MAX_DUP_TSNS) {
/* Record a dup for the next outbound sack */
asoc->dup_tsns[asoc->numduptsns] = tsn;
asoc->numduptsns++;
}
asoc->send_sack = 1;
return (0);
}
/* Calculate the number of TSN's between the base and this TSN */
if (tsn >= asoc->mapping_array_base_tsn) {
gap = tsn - asoc->mapping_array_base_tsn;
} else {
gap = (MAX_TSN - asoc->mapping_array_base_tsn) + tsn + 1;
}
if (gap >= (SCTP_MAPPING_ARRAY << 3)) {
/* Can't hold the bit in the mapping at max array, toss it */
return (0);
}
if (gap >= (uint32_t) (asoc->mapping_array_size << 3)) {
SCTP_TCB_LOCK_ASSERT(stcb);
if (sctp_expand_mapping_array(asoc, gap)) {
/* Can't expand, drop it */
return (0);
}
}
if (compare_with_wrap(tsn, *high_tsn, MAX_TSN)) {
*high_tsn = tsn;
}
/* See if we have received this one already */
if (SCTP_IS_TSN_PRESENT(asoc->mapping_array, gap)) {
SCTP_STAT_INCR(sctps_recvdupdata);
if (asoc->numduptsns < SCTP_MAX_DUP_TSNS) {
/* Record a dup for the next outbound sack */
asoc->dup_tsns[asoc->numduptsns] = tsn;
asoc->numduptsns++;
}
asoc->send_sack = 1;
return (0);
}
/*
* Check to see about the GONE flag, duplicates would cause a sack
* to be sent up above
*/
if (((stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_SOCKET_GONE) ||
(stcb->sctp_ep->sctp_flags & SCTP_PCB_FLAGS_SOCKET_ALLGONE) ||
(stcb->asoc.state & SCTP_STATE_CLOSED_SOCKET))
) {
/*
* wait a minute, this guy is gone, there is no longer a
* receiver. Send peer an ABORT!
*/
struct mbuf *op_err;
op_err = sctp_generate_invmanparam(SCTP_CAUSE_OUT_OF_RESC);
sctp_abort_an_association(stcb->sctp_ep, stcb, 0, op_err, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return (0);
}
/*
* Now before going further we see if there is room. If NOT then we
* MAY let one through only IF this TSN is the one we are waiting
* for on a partial delivery API.
*/
/* now do the tests */
if (((asoc->cnt_on_all_streams +
asoc->cnt_on_reasm_queue +
asoc->cnt_msg_on_sb) >= SCTP_BASE_SYSCTL(sctp_max_chunks_on_queue)) ||
(((int)asoc->my_rwnd) <= 0)) {
/*
* When we have NO room in the rwnd we check to make sure
* the reader is doing its job...
*/
if (stcb->sctp_socket->so_rcv.sb_cc) {
/* some to read, wake-up */
#if defined (__APPLE__) || defined(SCTP_SO_LOCK_TESTING)
struct socket *so;
so = SCTP_INP_SO(stcb->sctp_ep);
atomic_add_int(&stcb->asoc.refcnt, 1);
SCTP_TCB_UNLOCK(stcb);
SCTP_SOCKET_LOCK(so, 1);
SCTP_TCB_LOCK(stcb);
atomic_subtract_int(&stcb->asoc.refcnt, 1);
if (stcb->asoc.state & SCTP_STATE_CLOSED_SOCKET) {
/* assoc was freed while we were unlocked */
SCTP_SOCKET_UNLOCK(so, 1);
return (0);
}
#endif
sctp_sorwakeup(stcb->sctp_ep, stcb->sctp_socket);
#if defined (__APPLE__) || defined(SCTP_SO_LOCK_TESTING)
SCTP_SOCKET_UNLOCK(so, 1);
#endif
}
/* now is it in the mapping array of what we have accepted? */
if (compare_with_wrap(tsn, asoc->highest_tsn_inside_map, MAX_TSN)) {
/* Nope not in the valid range dump it */
sctp_set_rwnd(stcb, asoc);
if ((asoc->cnt_on_all_streams +
asoc->cnt_on_reasm_queue +
asoc->cnt_msg_on_sb) >= SCTP_BASE_SYSCTL(sctp_max_chunks_on_queue)) {
SCTP_STAT_INCR(sctps_datadropchklmt);
} else {
SCTP_STAT_INCR(sctps_datadroprwnd);
}
indx = *break_flag;
*break_flag = 1;
return (0);
}
}
strmno = ntohs(ch->dp.stream_id);
if (strmno >= asoc->streamincnt) {
struct sctp_paramhdr *phdr;
struct mbuf *mb;
mb = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) * 2),
0, M_DONTWAIT, 1, MT_DATA);
if (mb != NULL) {
/* add some space up front so prepend will work well */
SCTP_BUF_RESV_UF(mb, sizeof(struct sctp_chunkhdr));
phdr = mtod(mb, struct sctp_paramhdr *);
/*
* Error causes are just param's and this one has
* two back to back phdr, one with the error type
* and size, the other with the streamid and a rsvd
*/
SCTP_BUF_LEN(mb) = (sizeof(struct sctp_paramhdr) * 2);
phdr->param_type = htons(SCTP_CAUSE_INVALID_STREAM);
phdr->param_length =
htons(sizeof(struct sctp_paramhdr) * 2);
phdr++;
/* We insert the stream in the type field */
phdr->param_type = ch->dp.stream_id;
/* And set the length to 0 for the rsvd field */
phdr->param_length = 0;
sctp_queue_op_err(stcb, mb);
}
SCTP_STAT_INCR(sctps_badsid);
SCTP_TCB_LOCK_ASSERT(stcb);
SCTP_SET_TSN_PRESENT(asoc->mapping_array, gap);
if (compare_with_wrap(tsn, asoc->highest_tsn_inside_map, MAX_TSN)) {
/* we have a new high score */
asoc->highest_tsn_inside_map = tsn;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
sctp_log_map(0, 2, asoc->highest_tsn_inside_map, SCTP_MAP_SLIDE_RESULT);
}
}
if (tsn == (asoc->cumulative_tsn + 1)) {
/* Update cum-ack */
asoc->cumulative_tsn = tsn;
}
return (0);
}
/*
* Before we continue lets validate that we are not being fooled by
* an evil attacker. We can only have 4k chunks based on our TSN
* spread allowed by the mapping array 512 * 8 bits, so there is no
* way our stream sequence numbers could have wrapped. We of course
* only validate the FIRST fragment so the bit must be set.
*/
strmseq = ntohs(ch->dp.stream_sequence);
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
#ifdef SCTP_ASOCLOG_OF_TSNS
SCTP_TCB_LOCK_ASSERT(stcb);
if (asoc->tsn_in_at >= SCTP_TSN_LOG_SIZE) {
asoc->tsn_in_at = 0;
asoc->tsn_in_wrapped = 1;
}
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
asoc->in_tsnlog[asoc->tsn_in_at].tsn = tsn;
asoc->in_tsnlog[asoc->tsn_in_at].strm = strmno;
asoc->in_tsnlog[asoc->tsn_in_at].seq = strmseq;
asoc->in_tsnlog[asoc->tsn_in_at].sz = chk_length;
asoc->in_tsnlog[asoc->tsn_in_at].flgs = chunk_flags;
asoc->in_tsnlog[asoc->tsn_in_at].stcb = (void *)stcb;
asoc->in_tsnlog[asoc->tsn_in_at].in_pos = asoc->tsn_in_at;
asoc->in_tsnlog[asoc->tsn_in_at].in_out = 1;
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
asoc->tsn_in_at++;
#endif
if ((chunk_flags & SCTP_DATA_FIRST_FRAG) &&
- fixed autclose to not allow setting on 1-2-1 model. - bounded cookie-life to 1 second minimum in socket option set. - Delayed_ack_time becomes delayed_ack per new socket api document. - Improve port number selection, we now use low/high bounds and no chance of a endless loop. Only one call to random per bind as well. - fixes so set_peer_primary pre-screens addresses to be valid to this host. - maxseg did not allow setting on an assoc basis. We needed to thus track and use an association value instead of a inp value. - Fixed ep get of HB status to report back properly. - use settings flag to tell if assoc level hb is on off not the timer.. since the timer may still run if unconf address are present. - check for crazy ENABLE/DISABLE conditions. - set and get of pmtud (fixed path mtu) not always taking into account ovh. - Getting PMTU info on stcb only needs to return PMTUD_ENABLED if any net is doing PMTU discovery. - Panic or warning fixed to not do so when a valid ip frag is taking place. - sndrcvinfo appearing in both inp and stcb was full size, instead of the non-pad version. This saves about 92 bytes from each struct by carefully converting to use the smaller version. - one-2-one model get(maxseg) would always get ep value, never the tcb's value. - The delayed ack time could be under a tick, this fixes so it bounds it to at least 1 tick for platforms whos tick is more than a ms. - Fragment interleave level set to wrong default value. - Fragment interleave could not set level 0. - Defered stream reset was broken due to a guard check and ntohl issue. - Found two lock order reversals and fixed. - Tighten up address checking, if the user gives an address the sa_len had better be set properly. - Get asoc by assoc-id would return a locked tcb when it was asked not to if the tcb was in the restart hash. - sysctl to dig down and get more association details Reviewed by: gnn
2007-05-28 11:17:24 +00:00
(TAILQ_EMPTY(&asoc->resetHead)) &&
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
(chunk_flags & SCTP_DATA_UNORDERED) == 0 &&
(compare_with_wrap(asoc->strmin[strmno].last_sequence_delivered,
strmseq, MAX_SEQ) ||
asoc->strmin[strmno].last_sequence_delivered == strmseq)) {
/* The incoming sseq is behind where we last delivered? */
SCTPDBG(SCTP_DEBUG_INDATA1, "EVIL/Broken-Dup S-SEQ:%d delivered:%d from peer, Abort!\n",
strmseq, asoc->strmin[strmno].last_sequence_delivered);
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) = sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper, struct sctp_paramhdr *);
ph->param_type = htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length = htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_14);
ippp++;
*ippp = tsn;
ippp++;
*ippp = ((strmno << 16) | strmseq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_14;
sctp_abort_an_association(stcb->sctp_ep, stcb,
SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return (0);
}
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
/************************************
* From here down we may find ch-> invalid
* so its a good idea NOT to use it.
*************************************/
the_len = (chk_length - sizeof(struct sctp_data_chunk));
if (last_chunk == 0) {
- Macroizes the V6ONLY flag check. - Added a short time wait (not used yet) constant - Corrected the type of the crc32c table (it was unsigned long and really is a uint32_t - Got rid of the user of MHeaders until they are truely needed by lower layers. - Fixed an initialization problem in the readq structure (ordering was off). - Found yet another collision bug when the random number generator returns two numbers on one side (during a collision) that are the same. Also added some tracking of cookies that will go away when we know that we have the last collision bug gone. - Fixed an init bug for book_size_scale, that was causing Early FR code to run when it should not. - Fixed a flight size tracking bug that was associated with Early FR but due to above bug also effected all FR's - Fixed it so Max Burst also will apply to Fast Retransmit. - Fixed a bug in the temporary logging code that allowed a static log array overflow - hashinit_flags is now used. - Two last mcopym's were converted to the macro sctp_m_copym that has always been used by all other places - macro sctp_m_copym was converted to upper case. - We now validate sinfo_flags on input (we did not before). - Fixed a bug that prevented a user from sending data and immediately shuting down with one send operation. - Moved to use hashdestroy instead of free() in our macros. - Fixed an init problem in our timed_wait vtag where we did not fully initialize our time-wait blocks. - Timer stops were re-positioned. - A pcb cleanup method was added, however this probably will not be used in BSD.. unless we make module loadable protocols - I think this fixes the mysterious timer bug.. it was a ordering of locks problem in the way we did timers. It now conforms to the timeout(9) manual (except for the _drain part, we had to do this a different way due to locks). - Fixed error return code so we get either CONNREUSED or CONNRESET depending on where one is in progression - Purged an unused clone macro. - Fixed a read erro code issue where we were NOT getting the proper error when the connection was reset. - Purged an unused clone macro. - Fixed a read erro code issue where we were NOT getting the proper error when the connection was reset. Approved by: gnn
2007-01-15 15:12:10 +00:00
dmbuf = SCTP_M_COPYM(*m,
(offset + sizeof(struct sctp_data_chunk)),
the_len, M_DONTWAIT);
#ifdef SCTP_MBUF_LOGGING
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MBUF_LOGGING_ENABLE) {
struct mbuf *mat;
mat = dmbuf;
while (mat) {
if (SCTP_BUF_IS_EXTENDED(mat)) {
sctp_log_mb(mat, SCTP_MBUF_ICOPY);
}
mat = SCTP_BUF_NEXT(mat);
}
}
#endif
} else {
/* We can steal the last chunk */
int l_len;
dmbuf = *m;
/* lop off the top part */
m_adj(dmbuf, (offset + sizeof(struct sctp_data_chunk)));
if (SCTP_BUF_NEXT(dmbuf) == NULL) {
l_len = SCTP_BUF_LEN(dmbuf);
} else {
/*
* need to count up the size hopefully does not hit
* this to often :-0
*/
struct mbuf *lat;
l_len = 0;
lat = dmbuf;
while (lat) {
l_len += SCTP_BUF_LEN(lat);
lat = SCTP_BUF_NEXT(lat);
}
}
if (l_len > the_len) {
/* Trim the end round bytes off too */
m_adj(dmbuf, -(l_len - the_len));
}
}
if (dmbuf == NULL) {
SCTP_STAT_INCR(sctps_nomem);
return (0);
}
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
if ((chunk_flags & SCTP_DATA_NOT_FRAG) == SCTP_DATA_NOT_FRAG &&
asoc->fragmented_delivery_inprogress == 0 &&
TAILQ_EMPTY(&asoc->resetHead) &&
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
((ordered == 0) ||
((asoc->strmin[strmno].last_sequence_delivered + 1) == strmseq &&
TAILQ_EMPTY(&asoc->strmin[strmno].inqueue)))) {
/* Candidate for express delivery */
/*
* Its not fragmented, No PD-API is up, Nothing in the
* delivery queue, Its un-ordered OR ordered and the next to
* deliver AND nothing else is stuck on the stream queue,
* And there is room for it in the socket buffer. Lets just
* stuff it up the buffer....
*/
/* It would be nice to avoid this copy if we could :< */
sctp_alloc_a_readq(stcb, control);
sctp_build_readq_entry_mac(control, stcb, asoc->context, net, tsn,
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
protocol_id,
stcb->asoc.context,
strmno, strmseq,
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
chunk_flags,
dmbuf);
if (control == NULL) {
goto failed_express_del;
}
sctp_add_to_readq(stcb->sctp_ep, stcb, control, &stcb->sctp_socket->so_rcv, 1, SCTP_SO_NOT_LOCKED);
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
if ((chunk_flags & SCTP_DATA_UNORDERED) == 0) {
/* for ordered, bump what we delivered */
asoc->strmin[strmno].last_sequence_delivered++;
}
SCTP_STAT_INCR(sctps_recvexpress);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_STR_LOGGING_ENABLE) {
sctp_log_strm_del_alt(stcb, tsn, strmseq, strmno,
SCTP_STR_LOG_FROM_EXPRS_DEL);
}
control = NULL;
goto finish_express_del;
}
failed_express_del:
/* If we reach here this is a new chunk */
chk = NULL;
control = NULL;
/* Express for fragmented delivery? */
if ((asoc->fragmented_delivery_inprogress) &&
(stcb->asoc.control_pdapi) &&
(asoc->str_of_pdapi == strmno) &&
(asoc->ssn_of_pdapi == strmseq)
) {
control = stcb->asoc.control_pdapi;
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
if ((chunk_flags & SCTP_DATA_FIRST_FRAG) == SCTP_DATA_FIRST_FRAG) {
/* Can't be another first? */
goto failed_pdapi_express_del;
}
if (tsn == (control->sinfo_tsn + 1)) {
/* Yep, we can add it on */
int end = 0;
uint32_t cumack;
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
if (chunk_flags & SCTP_DATA_LAST_FRAG) {
end = 1;
}
cumack = asoc->cumulative_tsn;
if ((cumack + 1) == tsn)
cumack = tsn;
if (sctp_append_to_readq(stcb->sctp_ep, stcb, control, dmbuf, end,
tsn,
&stcb->sctp_socket->so_rcv)) {
SCTP_PRINTF("Append fails end:%d\n", end);
goto failed_pdapi_express_del;
}
SCTP_STAT_INCR(sctps_recvexpressm);
control->sinfo_tsn = tsn;
asoc->tsn_last_delivered = tsn;
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
asoc->fragment_flags = chunk_flags;
asoc->tsn_of_pdapi_last_delivered = tsn;
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
asoc->last_flags_delivered = chunk_flags;
asoc->last_strm_seq_delivered = strmseq;
asoc->last_strm_no_delivered = strmno;
if (end) {
/* clean up the flags and such */
asoc->fragmented_delivery_inprogress = 0;
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
if ((chunk_flags & SCTP_DATA_UNORDERED) == 0) {
asoc->strmin[strmno].last_sequence_delivered++;
}
stcb->asoc.control_pdapi = NULL;
if (TAILQ_EMPTY(&asoc->reasmqueue) == 0) {
/*
* There could be another message
* ready
*/
need_reasm_check = 1;
}
}
control = NULL;
goto finish_express_del;
}
}
failed_pdapi_express_del:
control = NULL;
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
if ((chunk_flags & SCTP_DATA_NOT_FRAG) != SCTP_DATA_NOT_FRAG) {
sctp_alloc_a_chunk(stcb, chk);
if (chk == NULL) {
/* No memory so we drop the chunk */
SCTP_STAT_INCR(sctps_nomem);
if (last_chunk == 0) {
/* we copied it, free the copy */
sctp_m_freem(dmbuf);
}
return (0);
}
chk->rec.data.TSN_seq = tsn;
chk->no_fr_allowed = 0;
chk->rec.data.stream_seq = strmseq;
chk->rec.data.stream_number = strmno;
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
chk->rec.data.payloadtype = protocol_id;
chk->rec.data.context = stcb->asoc.context;
chk->rec.data.doing_fast_retransmit = 0;
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
chk->rec.data.rcv_flags = chunk_flags;
chk->asoc = asoc;
chk->send_size = the_len;
chk->whoTo = net;
atomic_add_int(&net->ref_count, 1);
chk->data = dmbuf;
} else {
sctp_alloc_a_readq(stcb, control);
sctp_build_readq_entry_mac(control, stcb, asoc->context, net, tsn,
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
protocol_id,
stcb->asoc.context,
strmno, strmseq,
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
chunk_flags,
dmbuf);
if (control == NULL) {
/* No memory so we drop the chunk */
SCTP_STAT_INCR(sctps_nomem);
if (last_chunk == 0) {
/* we copied it, free the copy */
sctp_m_freem(dmbuf);
}
return (0);
}
control->length = the_len;
}
/* Mark it as received */
/* Now queue it where it belongs */
if (control != NULL) {
/* First a sanity check */
if (asoc->fragmented_delivery_inprogress) {
/*
* Ok, we have a fragmented delivery in progress if
* this chunk is next to deliver OR belongs in our
* view to the reassembly, the peer is evil or
* broken.
*/
uint32_t estimate_tsn;
estimate_tsn = asoc->tsn_last_delivered + 1;
if (TAILQ_EMPTY(&asoc->reasmqueue) &&
(estimate_tsn == control->sinfo_tsn)) {
/* Evil/Broke peer */
sctp_m_freem(control->data);
control->data = NULL;
if (control->whoFrom) {
sctp_free_remote_addr(control->whoFrom);
control->whoFrom = NULL;
}
sctp_free_a_readq(stcb, control);
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper, struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length = htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_15);
ippp++;
*ippp = tsn;
ippp++;
*ippp = ((strmno << 16) | strmseq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_15;
sctp_abort_an_association(stcb->sctp_ep, stcb,
SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return (0);
} else {
if (sctp_does_tsn_belong_to_reasm(asoc, control->sinfo_tsn)) {
sctp_m_freem(control->data);
control->data = NULL;
if (control->whoFrom) {
sctp_free_remote_addr(control->whoFrom);
control->whoFrom = NULL;
}
sctp_free_a_readq(stcb, control);
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_16);
ippp++;
*ippp = tsn;
ippp++;
*ippp = ((strmno << 16) | strmseq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_16;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return (0);
}
}
} else {
/* No PDAPI running */
if (!TAILQ_EMPTY(&asoc->reasmqueue)) {
/*
* Reassembly queue is NOT empty validate
* that this tsn does not need to be in
* reasembly queue. If it does then our peer
* is broken or evil.
*/
if (sctp_does_tsn_belong_to_reasm(asoc, control->sinfo_tsn)) {
sctp_m_freem(control->data);
control->data = NULL;
if (control->whoFrom) {
sctp_free_remote_addr(control->whoFrom);
control->whoFrom = NULL;
}
sctp_free_a_readq(stcb, control);
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) =
sizeof(struct sctp_paramhdr) +
(3 * sizeof(uint32_t));
ph = mtod(oper,
struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length =
htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_17);
ippp++;
*ippp = tsn;
ippp++;
*ippp = ((strmno << 16) | strmseq);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_17;
sctp_abort_an_association(stcb->sctp_ep,
stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
*abort_flag = 1;
return (0);
}
}
}
/* ok, if we reach here we have passed the sanity checks */
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
if (chunk_flags & SCTP_DATA_UNORDERED) {
/* queue directly into socket buffer */
sctp_add_to_readq(stcb->sctp_ep, stcb,
control,
&stcb->sctp_socket->so_rcv, 1, SCTP_SO_NOT_LOCKED);
} else {
/*
* Special check for when streams are resetting. We
* could be more smart about this and check the
* actual stream to see if it is not being reset..
* that way we would not create a HOLB when amongst
* streams being reset and those not being reset.
*
* We take complete messages that have a stream reset
* intervening (aka the TSN is after where our
* cum-ack needs to be) off and put them on a
* pending_reply_queue. The reassembly ones we do
* not have to worry about since they are all sorted
* and proceessed by TSN order. It is only the
* singletons I must worry about.
*/
if (((liste = TAILQ_FIRST(&asoc->resetHead)) != NULL) &&
- fixed autclose to not allow setting on 1-2-1 model. - bounded cookie-life to 1 second minimum in socket option set. - Delayed_ack_time becomes delayed_ack per new socket api document. - Improve port number selection, we now use low/high bounds and no chance of a endless loop. Only one call to random per bind as well. - fixes so set_peer_primary pre-screens addresses to be valid to this host. - maxseg did not allow setting on an assoc basis. We needed to thus track and use an association value instead of a inp value. - Fixed ep get of HB status to report back properly. - use settings flag to tell if assoc level hb is on off not the timer.. since the timer may still run if unconf address are present. - check for crazy ENABLE/DISABLE conditions. - set and get of pmtud (fixed path mtu) not always taking into account ovh. - Getting PMTU info on stcb only needs to return PMTUD_ENABLED if any net is doing PMTU discovery. - Panic or warning fixed to not do so when a valid ip frag is taking place. - sndrcvinfo appearing in both inp and stcb was full size, instead of the non-pad version. This saves about 92 bytes from each struct by carefully converting to use the smaller version. - one-2-one model get(maxseg) would always get ep value, never the tcb's value. - The delayed ack time could be under a tick, this fixes so it bounds it to at least 1 tick for platforms whos tick is more than a ms. - Fragment interleave level set to wrong default value. - Fragment interleave could not set level 0. - Defered stream reset was broken due to a guard check and ntohl issue. - Found two lock order reversals and fixed. - Tighten up address checking, if the user gives an address the sa_len had better be set properly. - Get asoc by assoc-id would return a locked tcb when it was asked not to if the tcb was in the restart hash. - sysctl to dig down and get more association details Reviewed by: gnn
2007-05-28 11:17:24 +00:00
((compare_with_wrap(tsn, liste->tsn, MAX_TSN)))
) {
/*
* yep its past where we need to reset... go
* ahead and queue it.
*/
if (TAILQ_EMPTY(&asoc->pending_reply_queue)) {
/* first one on */
TAILQ_INSERT_TAIL(&asoc->pending_reply_queue, control, next);
} else {
struct sctp_queued_to_read *ctlOn;
unsigned char inserted = 0;
ctlOn = TAILQ_FIRST(&asoc->pending_reply_queue);
while (ctlOn) {
if (compare_with_wrap(control->sinfo_tsn,
ctlOn->sinfo_tsn, MAX_TSN)) {
ctlOn = TAILQ_NEXT(ctlOn, next);
} else {
/* found it */
TAILQ_INSERT_BEFORE(ctlOn, control, next);
inserted = 1;
break;
}
}
if (inserted == 0) {
/*
* must be put at end, use
* prevP (all setup from
* loop) to setup nextP.
*/
TAILQ_INSERT_TAIL(&asoc->pending_reply_queue, control, next);
}
}
} else {
sctp_queue_data_to_stream(stcb, asoc, control, abort_flag);
if (*abort_flag) {
return (0);
}
}
}
} else {
/* Into the re-assembly queue */
sctp_queue_data_for_reasm(stcb, asoc, chk, abort_flag);
if (*abort_flag) {
/*
* the assoc is now gone and chk was put onto the
* reasm queue, which has all been freed.
*/
*m = NULL;
return (0);
}
}
finish_express_del:
if (compare_with_wrap(tsn, asoc->highest_tsn_inside_map, MAX_TSN)) {
/* we have a new high score */
asoc->highest_tsn_inside_map = tsn;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
sctp_log_map(0, 2, asoc->highest_tsn_inside_map, SCTP_MAP_SLIDE_RESULT);
}
}
if (tsn == (asoc->cumulative_tsn + 1)) {
/* Update cum-ack */
asoc->cumulative_tsn = tsn;
}
if (last_chunk) {
*m = NULL;
}
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
if (ordered) {
SCTP_STAT_INCR_COUNTER64(sctps_inorderchunks);
} else {
SCTP_STAT_INCR_COUNTER64(sctps_inunorderchunks);
}
SCTP_STAT_INCR(sctps_recvdata);
/* Set it present please */
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_STR_LOGGING_ENABLE) {
sctp_log_strm_del_alt(stcb, tsn, strmseq, strmno, SCTP_STR_LOG_FROM_MARK_TSN);
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
sctp_log_map(asoc->mapping_array_base_tsn, asoc->cumulative_tsn,
asoc->highest_tsn_inside_map, SCTP_MAP_PREPARE_SLIDE);
}
SCTP_TCB_LOCK_ASSERT(stcb);
SCTP_SET_TSN_PRESENT(asoc->mapping_array, gap);
/* check the special flag for stream resets */
if (((liste = TAILQ_FIRST(&asoc->resetHead)) != NULL) &&
- fixed autclose to not allow setting on 1-2-1 model. - bounded cookie-life to 1 second minimum in socket option set. - Delayed_ack_time becomes delayed_ack per new socket api document. - Improve port number selection, we now use low/high bounds and no chance of a endless loop. Only one call to random per bind as well. - fixes so set_peer_primary pre-screens addresses to be valid to this host. - maxseg did not allow setting on an assoc basis. We needed to thus track and use an association value instead of a inp value. - Fixed ep get of HB status to report back properly. - use settings flag to tell if assoc level hb is on off not the timer.. since the timer may still run if unconf address are present. - check for crazy ENABLE/DISABLE conditions. - set and get of pmtud (fixed path mtu) not always taking into account ovh. - Getting PMTU info on stcb only needs to return PMTUD_ENABLED if any net is doing PMTU discovery. - Panic or warning fixed to not do so when a valid ip frag is taking place. - sndrcvinfo appearing in both inp and stcb was full size, instead of the non-pad version. This saves about 92 bytes from each struct by carefully converting to use the smaller version. - one-2-one model get(maxseg) would always get ep value, never the tcb's value. - The delayed ack time could be under a tick, this fixes so it bounds it to at least 1 tick for platforms whos tick is more than a ms. - Fragment interleave level set to wrong default value. - Fragment interleave could not set level 0. - Defered stream reset was broken due to a guard check and ntohl issue. - Found two lock order reversals and fixed. - Tighten up address checking, if the user gives an address the sa_len had better be set properly. - Get asoc by assoc-id would return a locked tcb when it was asked not to if the tcb was in the restart hash. - sysctl to dig down and get more association details Reviewed by: gnn
2007-05-28 11:17:24 +00:00
((compare_with_wrap(asoc->cumulative_tsn, liste->tsn, MAX_TSN)) ||
(asoc->cumulative_tsn == liste->tsn))
) {
/*
* we have finished working through the backlogged TSN's now
* time to reset streams. 1: call reset function. 2: free
* pending_reply space 3: distribute any chunks in
* pending_reply_queue.
*/
struct sctp_queued_to_read *ctl;
sctp_reset_in_stream(stcb, liste->number_entries, liste->req.list_of_streams);
TAILQ_REMOVE(&asoc->resetHead, liste, next_resp);
SCTP_FREE(liste, SCTP_M_STRESET);
/* sa_ignore FREED_MEMORY */
liste = TAILQ_FIRST(&asoc->resetHead);
ctl = TAILQ_FIRST(&asoc->pending_reply_queue);
if (ctl && (liste == NULL)) {
/* All can be removed */
while (ctl) {
TAILQ_REMOVE(&asoc->pending_reply_queue, ctl, next);
sctp_queue_data_to_stream(stcb, asoc, ctl, abort_flag);
if (*abort_flag) {
return (0);
}
ctl = TAILQ_FIRST(&asoc->pending_reply_queue);
}
} else if (ctl) {
/* more than one in queue */
- fixed autclose to not allow setting on 1-2-1 model. - bounded cookie-life to 1 second minimum in socket option set. - Delayed_ack_time becomes delayed_ack per new socket api document. - Improve port number selection, we now use low/high bounds and no chance of a endless loop. Only one call to random per bind as well. - fixes so set_peer_primary pre-screens addresses to be valid to this host. - maxseg did not allow setting on an assoc basis. We needed to thus track and use an association value instead of a inp value. - Fixed ep get of HB status to report back properly. - use settings flag to tell if assoc level hb is on off not the timer.. since the timer may still run if unconf address are present. - check for crazy ENABLE/DISABLE conditions. - set and get of pmtud (fixed path mtu) not always taking into account ovh. - Getting PMTU info on stcb only needs to return PMTUD_ENABLED if any net is doing PMTU discovery. - Panic or warning fixed to not do so when a valid ip frag is taking place. - sndrcvinfo appearing in both inp and stcb was full size, instead of the non-pad version. This saves about 92 bytes from each struct by carefully converting to use the smaller version. - one-2-one model get(maxseg) would always get ep value, never the tcb's value. - The delayed ack time could be under a tick, this fixes so it bounds it to at least 1 tick for platforms whos tick is more than a ms. - Fragment interleave level set to wrong default value. - Fragment interleave could not set level 0. - Defered stream reset was broken due to a guard check and ntohl issue. - Found two lock order reversals and fixed. - Tighten up address checking, if the user gives an address the sa_len had better be set properly. - Get asoc by assoc-id would return a locked tcb when it was asked not to if the tcb was in the restart hash. - sysctl to dig down and get more association details Reviewed by: gnn
2007-05-28 11:17:24 +00:00
while (!compare_with_wrap(ctl->sinfo_tsn, liste->tsn, MAX_TSN)) {
/*
* if ctl->sinfo_tsn is <= liste->tsn we can
* process it which is the NOT of
* ctl->sinfo_tsn > liste->tsn
*/
TAILQ_REMOVE(&asoc->pending_reply_queue, ctl, next);
sctp_queue_data_to_stream(stcb, asoc, ctl, abort_flag);
if (*abort_flag) {
return (0);
}
ctl = TAILQ_FIRST(&asoc->pending_reply_queue);
}
}
/*
* Now service re-assembly to pick up anything that has been
* held on reassembly queue?
*/
sctp_deliver_reasm_check(stcb, asoc);
need_reasm_check = 0;
}
if (need_reasm_check) {
/* Another one waits ? */
sctp_deliver_reasm_check(stcb, asoc);
}
return (1);
}
int8_t sctp_map_lookup_tab[256] = {
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 3,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 4,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 3,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 5,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 3,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 4,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 3,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 6,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 3,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 4,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 3,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 5,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 3,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 4,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 3,
-1, 0, -1, 1, -1, 0, -1, 2,
-1, 0, -1, 1, -1, 0, -1, 7,
};
void
sctp_sack_check(struct sctp_tcb *stcb, int ok_to_sack, int was_a_gap, int *abort_flag)
{
/*
* Now we also need to check the mapping array in a couple of ways.
* 1) Did we move the cum-ack point?
*/
struct sctp_association *asoc;
int at;
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
int last_all_ones = 0;
int slide_from, slide_end, lgap, distance;
uint32_t old_cumack, old_base, old_highest;
unsigned char aux_array[64];
asoc = &stcb->asoc;
at = 0;
old_cumack = asoc->cumulative_tsn;
old_base = asoc->mapping_array_base_tsn;
old_highest = asoc->highest_tsn_inside_map;
if (asoc->mapping_array_size < 64)
memcpy(aux_array, asoc->mapping_array,
asoc->mapping_array_size);
else
memcpy(aux_array, asoc->mapping_array, 64);
/*
* We could probably improve this a small bit by calculating the
* offset of the current cum-ack as the starting point.
*/
at = 0;
for (slide_from = 0; slide_from < stcb->asoc.mapping_array_size; slide_from++) {
if (asoc->mapping_array[slide_from] == 0xff) {
at += 8;
last_all_ones = 1;
} else {
/* there is a 0 bit */
at += sctp_map_lookup_tab[asoc->mapping_array[slide_from]];
last_all_ones = 0;
break;
}
}
asoc->cumulative_tsn = asoc->mapping_array_base_tsn + (at - last_all_ones);
/* at is one off, since in the table a embedded -1 is present */
at++;
if (compare_with_wrap(asoc->cumulative_tsn,
asoc->highest_tsn_inside_map,
MAX_TSN)) {
#ifdef INVARIANTS
panic("huh, cumack 0x%x greater than high-tsn 0x%x in map",
asoc->cumulative_tsn, asoc->highest_tsn_inside_map);
#else
SCTP_PRINTF("huh, cumack 0x%x greater than high-tsn 0x%x in map - should panic?\n",
asoc->cumulative_tsn, asoc->highest_tsn_inside_map);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
sctp_log_map(0, 6, asoc->highest_tsn_inside_map, SCTP_MAP_SLIDE_RESULT);
}
asoc->highest_tsn_inside_map = asoc->cumulative_tsn;
#endif
}
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
if ((asoc->cumulative_tsn == asoc->highest_tsn_inside_map) && (at >= 8)) {
/* The complete array was completed by a single FR */
/* higest becomes the cum-ack */
int clr;
asoc->cumulative_tsn = asoc->highest_tsn_inside_map;
/* clear the array */
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
clr = (at >> 3) + 1;
if (clr > asoc->mapping_array_size) {
clr = asoc->mapping_array_size;
}
memset(asoc->mapping_array, 0, clr);
/* base becomes one ahead of the cum-ack */
asoc->mapping_array_base_tsn = asoc->cumulative_tsn + 1;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
sctp_log_map(old_base, old_cumack, old_highest,
SCTP_MAP_PREPARE_SLIDE);
sctp_log_map(asoc->mapping_array_base_tsn, asoc->cumulative_tsn,
asoc->highest_tsn_inside_map, SCTP_MAP_SLIDE_CLEARED);
}
} else if (at >= 8) {
/* we can slide the mapping array down */
/* slide_from holds where we hit the first NON 0xff byte */
/*
* now calculate the ceiling of the move using our highest
* TSN value
*/
if (asoc->highest_tsn_inside_map >= asoc->mapping_array_base_tsn) {
lgap = asoc->highest_tsn_inside_map -
asoc->mapping_array_base_tsn;
} else {
lgap = (MAX_TSN - asoc->mapping_array_base_tsn) +
asoc->highest_tsn_inside_map + 1;
}
slide_end = lgap >> 3;
if (slide_end < slide_from) {
#ifdef INVARIANTS
panic("impossible slide");
#else
printf("impossible slide?\n");
return;
#endif
}
if (slide_end > asoc->mapping_array_size) {
#ifdef INVARIANTS
panic("would overrun buffer");
#else
printf("Gak, would have overrun map end:%d slide_end:%d\n",
asoc->mapping_array_size, slide_end);
slide_end = asoc->mapping_array_size;
#endif
}
distance = (slide_end - slide_from) + 1;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
sctp_log_map(old_base, old_cumack, old_highest,
SCTP_MAP_PREPARE_SLIDE);
sctp_log_map((uint32_t) slide_from, (uint32_t) slide_end,
(uint32_t) lgap, SCTP_MAP_SLIDE_FROM);
}
if (distance + slide_from > asoc->mapping_array_size ||
distance < 0) {
/*
* Here we do NOT slide forward the array so that
* hopefully when more data comes in to fill it up
* we will be able to slide it forward. Really I
* don't think this should happen :-0
*/
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
sctp_log_map((uint32_t) distance, (uint32_t) slide_from,
(uint32_t) asoc->mapping_array_size,
SCTP_MAP_SLIDE_NONE);
}
} else {
int ii;
for (ii = 0; ii < distance; ii++) {
asoc->mapping_array[ii] =
asoc->mapping_array[slide_from + ii];
}
for (ii = distance; ii <= slide_end; ii++) {
asoc->mapping_array[ii] = 0;
}
asoc->mapping_array_base_tsn += (slide_from << 3);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
sctp_log_map(asoc->mapping_array_base_tsn,
asoc->cumulative_tsn, asoc->highest_tsn_inside_map,
SCTP_MAP_SLIDE_RESULT);
}
}
}
/*
* Now we need to see if we need to queue a sack or just start the
* timer (if allowed).
*/
if (ok_to_sack) {
if (SCTP_GET_STATE(asoc) == SCTP_STATE_SHUTDOWN_SENT) {
/*
* Ok special case, in SHUTDOWN-SENT case. here we
* maker sure SACK timer is off and instead send a
* SHUTDOWN and a SACK
*/
if (SCTP_OS_TIMER_PENDING(&stcb->asoc.dack_timer.timer)) {
sctp_timer_stop(SCTP_TIMER_TYPE_RECV,
stcb->sctp_ep, stcb, NULL, SCTP_FROM_SCTP_INDATA + SCTP_LOC_18);
}
sctp_send_shutdown(stcb, stcb->asoc.primary_destination);
sctp_send_sack(stcb);
} else {
int is_a_gap;
/* is there a gap now ? */
is_a_gap = compare_with_wrap(stcb->asoc.highest_tsn_inside_map,
stcb->asoc.cumulative_tsn, MAX_TSN);
/*
* CMT DAC algorithm: increase number of packets
* received since last ack
*/
stcb->asoc.cmt_dac_pkts_rcvd++;
if ((stcb->asoc.send_sack == 1) || /* We need to send a
* SACK */
((was_a_gap) && (is_a_gap == 0)) || /* was a gap, but no
* longer is one */
(stcb->asoc.numduptsns) || /* we have dup's */
(is_a_gap) || /* is still a gap */
(stcb->asoc.delayed_ack == 0) || /* Delayed sack disabled */
(stcb->asoc.data_pkts_seen >= stcb->asoc.sack_freq) /* hit limit of pkts */
) {
if ((SCTP_BASE_SYSCTL(sctp_cmt_on_off)) &&
(SCTP_BASE_SYSCTL(sctp_cmt_use_dac)) &&
(stcb->asoc.send_sack == 0) &&
(stcb->asoc.numduptsns == 0) &&
(stcb->asoc.delayed_ack) &&
(!SCTP_OS_TIMER_PENDING(&stcb->asoc.dack_timer.timer))) {
/*
* CMT DAC algorithm: With CMT,
* delay acks even in the face of
*
* reordering. Therefore, if acks that
* do not have to be sent because of
* the above reasons, will be
* delayed. That is, acks that would
* have been sent due to gap reports
* will be delayed with DAC. Start
* the delayed ack timer.
*/
sctp_timer_start(SCTP_TIMER_TYPE_RECV,
stcb->sctp_ep, stcb, NULL);
} else {
/*
* Ok we must build a SACK since the
* timer is pending, we got our
* first packet OR there are gaps or
* duplicates.
*/
(void)SCTP_OS_TIMER_STOP(&stcb->asoc.dack_timer.timer);
sctp_send_sack(stcb);
}
} else {
if (!SCTP_OS_TIMER_PENDING(&stcb->asoc.dack_timer.timer)) {
sctp_timer_start(SCTP_TIMER_TYPE_RECV,
stcb->sctp_ep, stcb, NULL);
}
}
}
}
}
void
sctp_service_queues(struct sctp_tcb *stcb, struct sctp_association *asoc)
{
struct sctp_tmit_chunk *chk;
uint32_t tsize;
uint16_t nxt_todel;
if (asoc->fragmented_delivery_inprogress) {
sctp_service_reassembly(stcb, asoc);
}
/* Can we proceed further, i.e. the PD-API is complete */
if (asoc->fragmented_delivery_inprogress) {
/* no */
return;
}
/*
* Now is there some other chunk I can deliver from the reassembly
* queue.
*/
doit_again:
chk = TAILQ_FIRST(&asoc->reasmqueue);
if (chk == NULL) {
asoc->size_on_reasm_queue = 0;
asoc->cnt_on_reasm_queue = 0;
return;
}
nxt_todel = asoc->strmin[chk->rec.data.stream_number].last_sequence_delivered + 1;
if ((chk->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG) &&
((nxt_todel == chk->rec.data.stream_seq) ||
(chk->rec.data.rcv_flags & SCTP_DATA_UNORDERED))) {
/*
* Yep the first one is here. We setup to start reception,
* by backing down the TSN just in case we can't deliver.
*/
/*
* Before we start though either all of the message should
* be here or 1/4 the socket buffer max or nothing on the
* delivery queue and something can be delivered.
*/
if ((sctp_is_all_msg_on_reasm(asoc, &tsize) ||
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
(tsize >= stcb->sctp_ep->partial_delivery_point))) {
asoc->fragmented_delivery_inprogress = 1;
asoc->tsn_last_delivered = chk->rec.data.TSN_seq - 1;
asoc->str_of_pdapi = chk->rec.data.stream_number;
asoc->ssn_of_pdapi = chk->rec.data.stream_seq;
asoc->pdapi_ppid = chk->rec.data.payloadtype;
asoc->fragment_flags = chk->rec.data.rcv_flags;
sctp_service_reassembly(stcb, asoc);
if (asoc->fragmented_delivery_inprogress == 0) {
goto doit_again;
}
}
}
}
int
sctp_process_data(struct mbuf **mm, int iphlen, int *offset, int length,
struct sctphdr *sh, struct sctp_inpcb *inp, struct sctp_tcb *stcb,
struct sctp_nets *net, uint32_t * high_tsn)
{
struct sctp_data_chunk *ch, chunk_buf;
struct sctp_association *asoc;
int num_chunks = 0; /* number of control chunks processed */
int stop_proc = 0;
int chk_length, break_flag, last_chunk;
int abort_flag = 0, was_a_gap = 0;
struct mbuf *m;
/* set the rwnd */
sctp_set_rwnd(stcb, &stcb->asoc);
m = *mm;
SCTP_TCB_LOCK_ASSERT(stcb);
asoc = &stcb->asoc;
if (compare_with_wrap(stcb->asoc.highest_tsn_inside_map,
stcb->asoc.cumulative_tsn, MAX_TSN)) {
/* there was a gap before this data was processed */
was_a_gap = 1;
}
/*
* setup where we got the last DATA packet from for any SACK that
* may need to go out. Don't bump the net. This is done ONLY when a
* chunk is assigned.
*/
asoc->last_data_chunk_from = net;
/*-
* Now before we proceed we must figure out if this is a wasted
* cluster... i.e. it is a small packet sent in and yet the driver
* underneath allocated a full cluster for it. If so we must copy it
* to a smaller mbuf and free up the cluster mbuf. This will help
* with cluster starvation. Note for __Panda__ we don't do this
* since it has clusters all the way down to 64 bytes.
*/
- Macroizes the V6ONLY flag check. - Added a short time wait (not used yet) constant - Corrected the type of the crc32c table (it was unsigned long and really is a uint32_t - Got rid of the user of MHeaders until they are truely needed by lower layers. - Fixed an initialization problem in the readq structure (ordering was off). - Found yet another collision bug when the random number generator returns two numbers on one side (during a collision) that are the same. Also added some tracking of cookies that will go away when we know that we have the last collision bug gone. - Fixed an init bug for book_size_scale, that was causing Early FR code to run when it should not. - Fixed a flight size tracking bug that was associated with Early FR but due to above bug also effected all FR's - Fixed it so Max Burst also will apply to Fast Retransmit. - Fixed a bug in the temporary logging code that allowed a static log array overflow - hashinit_flags is now used. - Two last mcopym's were converted to the macro sctp_m_copym that has always been used by all other places - macro sctp_m_copym was converted to upper case. - We now validate sinfo_flags on input (we did not before). - Fixed a bug that prevented a user from sending data and immediately shuting down with one send operation. - Moved to use hashdestroy instead of free() in our macros. - Fixed an init problem in our timed_wait vtag where we did not fully initialize our time-wait blocks. - Timer stops were re-positioned. - A pcb cleanup method was added, however this probably will not be used in BSD.. unless we make module loadable protocols - I think this fixes the mysterious timer bug.. it was a ordering of locks problem in the way we did timers. It now conforms to the timeout(9) manual (except for the _drain part, we had to do this a different way due to locks). - Fixed error return code so we get either CONNREUSED or CONNRESET depending on where one is in progression - Purged an unused clone macro. - Fixed a read erro code issue where we were NOT getting the proper error when the connection was reset. - Purged an unused clone macro. - Fixed a read erro code issue where we were NOT getting the proper error when the connection was reset. Approved by: gnn
2007-01-15 15:12:10 +00:00
if (SCTP_BUF_LEN(m) < (long)MLEN && SCTP_BUF_NEXT(m) == NULL) {
/* we only handle mbufs that are singletons.. not chains */
m = sctp_get_mbuf_for_msg(SCTP_BUF_LEN(m), 0, M_DONTWAIT, 1, MT_DATA);
if (m) {
/* ok lets see if we can copy the data up */
caddr_t *from, *to;
/* get the pointers and copy */
to = mtod(m, caddr_t *);
from = mtod((*mm), caddr_t *);
memcpy(to, from, SCTP_BUF_LEN((*mm)));
/* copy the length and free up the old */
SCTP_BUF_LEN(m) = SCTP_BUF_LEN((*mm));
sctp_m_freem(*mm);
/* sucess, back copy */
*mm = m;
} else {
/* We are in trouble in the mbuf world .. yikes */
m = *mm;
}
}
/* get pointer to the first chunk header */
ch = (struct sctp_data_chunk *)sctp_m_getptr(m, *offset,
sizeof(struct sctp_data_chunk), (uint8_t *) & chunk_buf);
if (ch == NULL) {
return (1);
}
/*
* process all DATA chunks...
*/
*high_tsn = asoc->cumulative_tsn;
break_flag = 0;
asoc->data_pkts_seen++;
while (stop_proc == 0) {
/* validate chunk length */
chk_length = ntohs(ch->ch.chunk_length);
if (length - *offset < chk_length) {
/* all done, mutulated chunk */
stop_proc = 1;
break;
}
if (ch->ch.chunk_type == SCTP_DATA) {
if ((size_t)chk_length < sizeof(struct sctp_data_chunk) + 1) {
/*
* Need to send an abort since we had a
* invalid data chunk.
*/
struct mbuf *op_err;
op_err = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 2 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (op_err) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(op_err) = sizeof(struct sctp_paramhdr) +
(2 * sizeof(uint32_t));
ph = mtod(op_err, struct sctp_paramhdr *);
ph->param_type =
htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length = htons(SCTP_BUF_LEN(op_err));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_19);
ippp++;
*ippp = asoc->cumulative_tsn;
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_19;
sctp_abort_association(inp, stcb, m, iphlen, sh,
op_err, 0, net->port);
return (2);
}
#ifdef SCTP_AUDITING_ENABLED
sctp_audit_log(0xB1, 0);
#endif
if (SCTP_SIZE32(chk_length) == (length - *offset)) {
last_chunk = 1;
} else {
last_chunk = 0;
}
if (sctp_process_a_data_chunk(stcb, asoc, mm, *offset, ch,
chk_length, net, high_tsn, &abort_flag, &break_flag,
last_chunk)) {
num_chunks++;
}
if (abort_flag)
return (2);
if (break_flag) {
/*
* Set because of out of rwnd space and no
* drop rep space left.
*/
stop_proc = 1;
break;
}
} else {
/* not a data chunk in the data region */
switch (ch->ch.chunk_type) {
case SCTP_INITIATION:
case SCTP_INITIATION_ACK:
case SCTP_SELECTIVE_ACK:
case SCTP_HEARTBEAT_REQUEST:
case SCTP_HEARTBEAT_ACK:
case SCTP_ABORT_ASSOCIATION:
case SCTP_SHUTDOWN:
case SCTP_SHUTDOWN_ACK:
case SCTP_OPERATION_ERROR:
case SCTP_COOKIE_ECHO:
case SCTP_COOKIE_ACK:
case SCTP_ECN_ECHO:
case SCTP_ECN_CWR:
case SCTP_SHUTDOWN_COMPLETE:
case SCTP_AUTHENTICATION:
case SCTP_ASCONF_ACK:
case SCTP_PACKET_DROPPED:
case SCTP_STREAM_RESET:
case SCTP_FORWARD_CUM_TSN:
case SCTP_ASCONF:
/*
* Now, what do we do with KNOWN chunks that
* are NOT in the right place?
*
* For now, I do nothing but ignore them. We
* may later want to add sysctl stuff to
* switch out and do either an ABORT() or
* possibly process them.
*/
if (SCTP_BASE_SYSCTL(sctp_strict_data_order)) {
struct mbuf *op_err;
op_err = sctp_generate_invmanparam(SCTP_CAUSE_PROTOCOL_VIOLATION);
sctp_abort_association(inp, stcb, m, iphlen, sh, op_err, 0, net->port);
return (2);
}
break;
default:
/* unknown chunk type, use bit rules */
if (ch->ch.chunk_type & 0x40) {
/* Add a error report to the queue */
- fixed autclose to not allow setting on 1-2-1 model. - bounded cookie-life to 1 second minimum in socket option set. - Delayed_ack_time becomes delayed_ack per new socket api document. - Improve port number selection, we now use low/high bounds and no chance of a endless loop. Only one call to random per bind as well. - fixes so set_peer_primary pre-screens addresses to be valid to this host. - maxseg did not allow setting on an assoc basis. We needed to thus track and use an association value instead of a inp value. - Fixed ep get of HB status to report back properly. - use settings flag to tell if assoc level hb is on off not the timer.. since the timer may still run if unconf address are present. - check for crazy ENABLE/DISABLE conditions. - set and get of pmtud (fixed path mtu) not always taking into account ovh. - Getting PMTU info on stcb only needs to return PMTUD_ENABLED if any net is doing PMTU discovery. - Panic or warning fixed to not do so when a valid ip frag is taking place. - sndrcvinfo appearing in both inp and stcb was full size, instead of the non-pad version. This saves about 92 bytes from each struct by carefully converting to use the smaller version. - one-2-one model get(maxseg) would always get ep value, never the tcb's value. - The delayed ack time could be under a tick, this fixes so it bounds it to at least 1 tick for platforms whos tick is more than a ms. - Fragment interleave level set to wrong default value. - Fragment interleave could not set level 0. - Defered stream reset was broken due to a guard check and ntohl issue. - Found two lock order reversals and fixed. - Tighten up address checking, if the user gives an address the sa_len had better be set properly. - Get asoc by assoc-id would return a locked tcb when it was asked not to if the tcb was in the restart hash. - sysctl to dig down and get more association details Reviewed by: gnn
2007-05-28 11:17:24 +00:00
struct mbuf *merr;
struct sctp_paramhdr *phd;
- fixed autclose to not allow setting on 1-2-1 model. - bounded cookie-life to 1 second minimum in socket option set. - Delayed_ack_time becomes delayed_ack per new socket api document. - Improve port number selection, we now use low/high bounds and no chance of a endless loop. Only one call to random per bind as well. - fixes so set_peer_primary pre-screens addresses to be valid to this host. - maxseg did not allow setting on an assoc basis. We needed to thus track and use an association value instead of a inp value. - Fixed ep get of HB status to report back properly. - use settings flag to tell if assoc level hb is on off not the timer.. since the timer may still run if unconf address are present. - check for crazy ENABLE/DISABLE conditions. - set and get of pmtud (fixed path mtu) not always taking into account ovh. - Getting PMTU info on stcb only needs to return PMTUD_ENABLED if any net is doing PMTU discovery. - Panic or warning fixed to not do so when a valid ip frag is taking place. - sndrcvinfo appearing in both inp and stcb was full size, instead of the non-pad version. This saves about 92 bytes from each struct by carefully converting to use the smaller version. - one-2-one model get(maxseg) would always get ep value, never the tcb's value. - The delayed ack time could be under a tick, this fixes so it bounds it to at least 1 tick for platforms whos tick is more than a ms. - Fragment interleave level set to wrong default value. - Fragment interleave could not set level 0. - Defered stream reset was broken due to a guard check and ntohl issue. - Found two lock order reversals and fixed. - Tighten up address checking, if the user gives an address the sa_len had better be set properly. - Get asoc by assoc-id would return a locked tcb when it was asked not to if the tcb was in the restart hash. - sysctl to dig down and get more association details Reviewed by: gnn
2007-05-28 11:17:24 +00:00
merr = sctp_get_mbuf_for_msg(sizeof(*phd), 0, M_DONTWAIT, 1, MT_DATA);
if (merr) {
phd = mtod(merr, struct sctp_paramhdr *);
/*
* We cheat and use param
* type since we did not
* bother to define a error
* cause struct. They are
* the same basic format
* with different names.
*/
phd->param_type =
htons(SCTP_CAUSE_UNRECOG_CHUNK);
phd->param_length =
htons(chk_length + sizeof(*phd));
- fixed autclose to not allow setting on 1-2-1 model. - bounded cookie-life to 1 second minimum in socket option set. - Delayed_ack_time becomes delayed_ack per new socket api document. - Improve port number selection, we now use low/high bounds and no chance of a endless loop. Only one call to random per bind as well. - fixes so set_peer_primary pre-screens addresses to be valid to this host. - maxseg did not allow setting on an assoc basis. We needed to thus track and use an association value instead of a inp value. - Fixed ep get of HB status to report back properly. - use settings flag to tell if assoc level hb is on off not the timer.. since the timer may still run if unconf address are present. - check for crazy ENABLE/DISABLE conditions. - set and get of pmtud (fixed path mtu) not always taking into account ovh. - Getting PMTU info on stcb only needs to return PMTUD_ENABLED if any net is doing PMTU discovery. - Panic or warning fixed to not do so when a valid ip frag is taking place. - sndrcvinfo appearing in both inp and stcb was full size, instead of the non-pad version. This saves about 92 bytes from each struct by carefully converting to use the smaller version. - one-2-one model get(maxseg) would always get ep value, never the tcb's value. - The delayed ack time could be under a tick, this fixes so it bounds it to at least 1 tick for platforms whos tick is more than a ms. - Fragment interleave level set to wrong default value. - Fragment interleave could not set level 0. - Defered stream reset was broken due to a guard check and ntohl issue. - Found two lock order reversals and fixed. - Tighten up address checking, if the user gives an address the sa_len had better be set properly. - Get asoc by assoc-id would return a locked tcb when it was asked not to if the tcb was in the restart hash. - sysctl to dig down and get more association details Reviewed by: gnn
2007-05-28 11:17:24 +00:00
SCTP_BUF_LEN(merr) = sizeof(*phd);
SCTP_BUF_NEXT(merr) = SCTP_M_COPYM(m, *offset,
SCTP_SIZE32(chk_length),
M_DONTWAIT);
- fixed autclose to not allow setting on 1-2-1 model. - bounded cookie-life to 1 second minimum in socket option set. - Delayed_ack_time becomes delayed_ack per new socket api document. - Improve port number selection, we now use low/high bounds and no chance of a endless loop. Only one call to random per bind as well. - fixes so set_peer_primary pre-screens addresses to be valid to this host. - maxseg did not allow setting on an assoc basis. We needed to thus track and use an association value instead of a inp value. - Fixed ep get of HB status to report back properly. - use settings flag to tell if assoc level hb is on off not the timer.. since the timer may still run if unconf address are present. - check for crazy ENABLE/DISABLE conditions. - set and get of pmtud (fixed path mtu) not always taking into account ovh. - Getting PMTU info on stcb only needs to return PMTUD_ENABLED if any net is doing PMTU discovery. - Panic or warning fixed to not do so when a valid ip frag is taking place. - sndrcvinfo appearing in both inp and stcb was full size, instead of the non-pad version. This saves about 92 bytes from each struct by carefully converting to use the smaller version. - one-2-one model get(maxseg) would always get ep value, never the tcb's value. - The delayed ack time could be under a tick, this fixes so it bounds it to at least 1 tick for platforms whos tick is more than a ms. - Fragment interleave level set to wrong default value. - Fragment interleave could not set level 0. - Defered stream reset was broken due to a guard check and ntohl issue. - Found two lock order reversals and fixed. - Tighten up address checking, if the user gives an address the sa_len had better be set properly. - Get asoc by assoc-id would return a locked tcb when it was asked not to if the tcb was in the restart hash. - sysctl to dig down and get more association details Reviewed by: gnn
2007-05-28 11:17:24 +00:00
if (SCTP_BUF_NEXT(merr)) {
sctp_queue_op_err(stcb, merr);
} else {
- fixed autclose to not allow setting on 1-2-1 model. - bounded cookie-life to 1 second minimum in socket option set. - Delayed_ack_time becomes delayed_ack per new socket api document. - Improve port number selection, we now use low/high bounds and no chance of a endless loop. Only one call to random per bind as well. - fixes so set_peer_primary pre-screens addresses to be valid to this host. - maxseg did not allow setting on an assoc basis. We needed to thus track and use an association value instead of a inp value. - Fixed ep get of HB status to report back properly. - use settings flag to tell if assoc level hb is on off not the timer.. since the timer may still run if unconf address are present. - check for crazy ENABLE/DISABLE conditions. - set and get of pmtud (fixed path mtu) not always taking into account ovh. - Getting PMTU info on stcb only needs to return PMTUD_ENABLED if any net is doing PMTU discovery. - Panic or warning fixed to not do so when a valid ip frag is taking place. - sndrcvinfo appearing in both inp and stcb was full size, instead of the non-pad version. This saves about 92 bytes from each struct by carefully converting to use the smaller version. - one-2-one model get(maxseg) would always get ep value, never the tcb's value. - The delayed ack time could be under a tick, this fixes so it bounds it to at least 1 tick for platforms whos tick is more than a ms. - Fragment interleave level set to wrong default value. - Fragment interleave could not set level 0. - Defered stream reset was broken due to a guard check and ntohl issue. - Found two lock order reversals and fixed. - Tighten up address checking, if the user gives an address the sa_len had better be set properly. - Get asoc by assoc-id would return a locked tcb when it was asked not to if the tcb was in the restart hash. - sysctl to dig down and get more association details Reviewed by: gnn
2007-05-28 11:17:24 +00:00
sctp_m_freem(merr);
}
}
}
if ((ch->ch.chunk_type & 0x80) == 0) {
/* discard the rest of this packet */
stop_proc = 1;
} /* else skip this bad chunk and
* continue... */
break;
}; /* switch of chunk type */
}
*offset += SCTP_SIZE32(chk_length);
if ((*offset >= length) || stop_proc) {
/* no more data left in the mbuf chain */
stop_proc = 1;
continue;
}
ch = (struct sctp_data_chunk *)sctp_m_getptr(m, *offset,
sizeof(struct sctp_data_chunk), (uint8_t *) & chunk_buf);
if (ch == NULL) {
*offset = length;
stop_proc = 1;
break;
}
} /* while */
if (break_flag) {
/*
* we need to report rwnd overrun drops.
*/
sctp_send_packet_dropped(stcb, net, *mm, iphlen, 0);
}
if (num_chunks) {
/*
* Did we get data, if so update the time for auto-close and
* give peer credit for being alive.
*/
SCTP_STAT_INCR(sctps_recvpktwithdata);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_THRESHOLD_LOGGING) {
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
sctp_misc_ints(SCTP_THRESHOLD_CLEAR,
stcb->asoc.overall_error_count,
0,
SCTP_FROM_SCTP_INDATA,
__LINE__);
}
stcb->asoc.overall_error_count = 0;
(void)SCTP_GETTIME_TIMEVAL(&stcb->asoc.time_last_rcvd);
}
/* now service all of the reassm queue if needed */
if (!(TAILQ_EMPTY(&asoc->reasmqueue)))
sctp_service_queues(stcb, asoc);
if (SCTP_GET_STATE(asoc) == SCTP_STATE_SHUTDOWN_SENT) {
/* Assure that we ack right away */
stcb->asoc.send_sack = 1;
}
/* Start a sack timer or QUEUE a SACK for sending */
if ((stcb->asoc.cumulative_tsn == stcb->asoc.highest_tsn_inside_map) &&
(stcb->asoc.mapping_array[0] != 0xff)) {
if ((stcb->asoc.data_pkts_seen >= stcb->asoc.sack_freq) ||
(stcb->asoc.delayed_ack == 0) ||
(stcb->asoc.numduptsns) ||
(stcb->asoc.send_sack == 1)) {
if (SCTP_OS_TIMER_PENDING(&stcb->asoc.dack_timer.timer)) {
(void)SCTP_OS_TIMER_STOP(&stcb->asoc.dack_timer.timer);
}
sctp_send_sack(stcb);
} else {
if (!SCTP_OS_TIMER_PENDING(&stcb->asoc.dack_timer.timer)) {
sctp_timer_start(SCTP_TIMER_TYPE_RECV,
stcb->sctp_ep, stcb, NULL);
}
}
} else {
sctp_sack_check(stcb, 1, was_a_gap, &abort_flag);
}
if (abort_flag)
return (2);
return (0);
}
static void
sctp_handle_segments(struct mbuf *m, int *offset, struct sctp_tcb *stcb, struct sctp_association *asoc,
struct sctp_sack_chunk *ch, uint32_t last_tsn, uint32_t * biggest_tsn_acked,
uint32_t * biggest_newly_acked_tsn, uint32_t * this_sack_lowest_newack,
int num_seg, int *ecn_seg_sums)
{
/************************************************/
/* process fragments and update sendqueue */
/************************************************/
struct sctp_sack *sack;
struct sctp_gap_ack_block *frag, block;
struct sctp_tmit_chunk *tp1;
int i;
unsigned int j;
int num_frs = 0;
uint16_t frag_strt, frag_end, primary_flag_set;
u_long last_frag_high;
/*
* @@@ JRI : TODO: This flag is not used anywhere .. remove?
*/
if (asoc->primary_destination->dest_state & SCTP_ADDR_SWITCH_PRIMARY) {
primary_flag_set = 1;
} else {
primary_flag_set = 0;
}
sack = &ch->sack;
frag = (struct sctp_gap_ack_block *)sctp_m_getptr(m, *offset,
sizeof(struct sctp_gap_ack_block), (uint8_t *) & block);
*offset += sizeof(block);
if (frag == NULL) {
return;
}
tp1 = NULL;
last_frag_high = 0;
for (i = 0; i < num_seg; i++) {
frag_strt = ntohs(frag->start);
frag_end = ntohs(frag->end);
/* some sanity checks on the fargment offsets */
if (frag_strt > frag_end) {
/* this one is malformed, skip */
frag++;
continue;
}
if (compare_with_wrap((frag_end + last_tsn), *biggest_tsn_acked,
MAX_TSN))
*biggest_tsn_acked = frag_end + last_tsn;
/* mark acked dgs and find out the highestTSN being acked */
if (tp1 == NULL) {
tp1 = TAILQ_FIRST(&asoc->sent_queue);
/* save the locations of the last frags */
last_frag_high = frag_end + last_tsn;
} else {
/*
* now lets see if we need to reset the queue due to
* a out-of-order SACK fragment
*/
if (compare_with_wrap(frag_strt + last_tsn,
last_frag_high, MAX_TSN)) {
/*
* if the new frag starts after the last TSN
* frag covered, we are ok and this one is
* beyond the last one
*/
;
} else {
/*
* ok, they have reset us, so we need to
* reset the queue this will cause extra
* hunting but hey, they chose the
* performance hit when they failed to order
* there gaps..
*/
tp1 = TAILQ_FIRST(&asoc->sent_queue);
}
last_frag_high = frag_end + last_tsn;
}
for (j = frag_strt + last_tsn; (compare_with_wrap((frag_end + last_tsn), j, MAX_TSN)); j++) {
while (tp1) {
if (tp1->rec.data.doing_fast_retransmit)
num_frs++;
/*
* CMT: CUCv2 algorithm. For each TSN being
* processed from the sent queue, track the
* next expected pseudo-cumack, or
* rtx_pseudo_cumack, if required. Separate
* cumack trackers for first transmissions,
* and retransmissions.
*/
if ((tp1->whoTo->find_pseudo_cumack == 1) && (tp1->sent < SCTP_DATAGRAM_RESEND) &&
(tp1->snd_count == 1)) {
tp1->whoTo->pseudo_cumack = tp1->rec.data.TSN_seq;
tp1->whoTo->find_pseudo_cumack = 0;
}
if ((tp1->whoTo->find_rtx_pseudo_cumack == 1) && (tp1->sent < SCTP_DATAGRAM_RESEND) &&
(tp1->snd_count > 1)) {
tp1->whoTo->rtx_pseudo_cumack = tp1->rec.data.TSN_seq;
tp1->whoTo->find_rtx_pseudo_cumack = 0;
}
if (tp1->rec.data.TSN_seq == j) {
if (tp1->sent != SCTP_DATAGRAM_UNSENT) {
/*
* must be held until
* cum-ack passes
*/
/*
* ECN Nonce: Add the nonce
* value to the sender's
* nonce sum
*/
if (tp1->sent < SCTP_DATAGRAM_RESEND) {
/*-
* If it is less than RESEND, it is
* now no-longer in flight.
* Higher values may already be set
* via previous Gap Ack Blocks...
* i.e. ACKED or RESEND.
*/
if (compare_with_wrap(tp1->rec.data.TSN_seq,
*biggest_newly_acked_tsn, MAX_TSN)) {
*biggest_newly_acked_tsn = tp1->rec.data.TSN_seq;
}
/*
* CMT: SFR algo
* (and HTNA) - set
* saw_newack to 1
* for dest being
* newly acked.
* update
* this_sack_highest_
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
* newack if
* appropriate.
*/
if (tp1->rec.data.chunk_was_revoked == 0)
tp1->whoTo->saw_newack = 1;
if (compare_with_wrap(tp1->rec.data.TSN_seq,
tp1->whoTo->this_sack_highest_newack,
MAX_TSN)) {
tp1->whoTo->this_sack_highest_newack =
tp1->rec.data.TSN_seq;
}
/*
* CMT DAC algo:
* also update
* this_sack_lowest_n
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
* ewack
*/
if (*this_sack_lowest_newack == 0) {
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_SACK_LOGGING_ENABLE) {
sctp_log_sack(*this_sack_lowest_newack,
last_tsn,
tp1->rec.data.TSN_seq,
0,
0,
SCTP_LOG_TSN_ACKED);
}
*this_sack_lowest_newack = tp1->rec.data.TSN_seq;
}
/*
* CMT: CUCv2
* algorithm. If
* (rtx-)pseudo-cumac
* k for corresp
* dest is being
* acked, then we
* have a new
* (rtx-)pseudo-cumac
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
* k. Set
* new_(rtx_)pseudo_c
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
* umack to TRUE so
* that the cwnd for
* this dest can be
* updated. Also
* trigger search
* for the next
* expected
* (rtx-)pseudo-cumac
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
* k. Separate
* pseudo_cumack
* trackers for
* first
* transmissions and
* retransmissions.
*/
if (tp1->rec.data.TSN_seq == tp1->whoTo->pseudo_cumack) {
if (tp1->rec.data.chunk_was_revoked == 0) {
tp1->whoTo->new_pseudo_cumack = 1;
}
tp1->whoTo->find_pseudo_cumack = 1;
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
sctp_log_cwnd(stcb, tp1->whoTo, tp1->rec.data.TSN_seq, SCTP_CWND_LOG_FROM_SACK);
}
if (tp1->rec.data.TSN_seq == tp1->whoTo->rtx_pseudo_cumack) {
if (tp1->rec.data.chunk_was_revoked == 0) {
tp1->whoTo->new_pseudo_cumack = 1;
}
tp1->whoTo->find_rtx_pseudo_cumack = 1;
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_SACK_LOGGING_ENABLE) {
sctp_log_sack(*biggest_newly_acked_tsn,
last_tsn,
tp1->rec.data.TSN_seq,
frag_strt,
frag_end,
SCTP_LOG_TSN_ACKED);
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FLIGHT_LOGGING_ENABLE) {
sctp_misc_ints(SCTP_FLIGHT_LOG_DOWN_GAP,
tp1->whoTo->flight_size,
tp1->book_size,
(uintptr_t) tp1->whoTo,
tp1->rec.data.TSN_seq);
}
sctp_flight_size_decrease(tp1);
sctp_total_flight_decrease(stcb, tp1);
tp1->whoTo->net_ack += tp1->send_size;
if (tp1->snd_count < 2) {
/*
* True
* non-retran
* smited
* chunk */
tp1->whoTo->net_ack2 += tp1->send_size;
/*
* update RTO
* too ? */
if (tp1->do_rtt) {
tp1->whoTo->RTO =
sctp_calculate_rto(stcb,
asoc,
tp1->whoTo,
&tp1->sent_rcv_time,
sctp_align_safe_nocopy);
tp1->do_rtt = 0;
}
}
}
if (tp1->sent <= SCTP_DATAGRAM_RESEND) {
(*ecn_seg_sums) += tp1->rec.data.ect_nonce;
(*ecn_seg_sums) &= SCTP_SACK_NONCE_SUM;
if (compare_with_wrap(tp1->rec.data.TSN_seq,
asoc->this_sack_highest_gap,
MAX_TSN)) {
asoc->this_sack_highest_gap =
tp1->rec.data.TSN_seq;
}
if (tp1->sent == SCTP_DATAGRAM_RESEND) {
sctp_ucount_decr(asoc->sent_queue_retran_cnt);
#ifdef SCTP_AUDITING_ENABLED
sctp_audit_log(0xB2,
(asoc->sent_queue_retran_cnt & 0x000000ff));
#endif
}
}
/*
* All chunks NOT UNSENT
* fall through here and are
* marked
*/
tp1->sent = SCTP_DATAGRAM_MARKED;
if (tp1->rec.data.chunk_was_revoked) {
/* deflate the cwnd */
tp1->whoTo->cwnd -= tp1->book_size;
tp1->rec.data.chunk_was_revoked = 0;
}
}
break;
} /* if (tp1->TSN_seq == j) */
if (compare_with_wrap(tp1->rec.data.TSN_seq, j,
MAX_TSN))
break;
tp1 = TAILQ_NEXT(tp1, sctp_next);
} /* end while (tp1) */
} /* end for (j = fragStart */
frag = (struct sctp_gap_ack_block *)sctp_m_getptr(m, *offset,
sizeof(struct sctp_gap_ack_block), (uint8_t *) & block);
*offset += sizeof(block);
if (frag == NULL) {
break;
}
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) {
if (num_frs)
sctp_log_fr(*biggest_tsn_acked,
*biggest_newly_acked_tsn,
last_tsn, SCTP_FR_LOG_BIGGEST_TSNS);
}
}
static void
sctp_check_for_revoked(struct sctp_tcb *stcb,
struct sctp_association *asoc, uint32_t cumack,
u_long biggest_tsn_acked)
{
struct sctp_tmit_chunk *tp1;
int tot_revoked = 0;
tp1 = TAILQ_FIRST(&asoc->sent_queue);
while (tp1) {
if (compare_with_wrap(tp1->rec.data.TSN_seq, cumack,
MAX_TSN)) {
/*
* ok this guy is either ACK or MARKED. If it is
* ACKED it has been previously acked but not this
* time i.e. revoked. If it is MARKED it was ACK'ed
* again.
*/
if (compare_with_wrap(tp1->rec.data.TSN_seq, biggest_tsn_acked,
MAX_TSN))
break;
if (tp1->sent == SCTP_DATAGRAM_ACKED) {
/* it has been revoked */
tp1->sent = SCTP_DATAGRAM_SENT;
tp1->rec.data.chunk_was_revoked = 1;
/*
* We must add this stuff back in to assure
* timers and such get started.
*/
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FLIGHT_LOGGING_ENABLE) {
sctp_misc_ints(SCTP_FLIGHT_LOG_UP_REVOKE,
tp1->whoTo->flight_size,
tp1->book_size,
(uintptr_t) tp1->whoTo,
tp1->rec.data.TSN_seq);
}
sctp_flight_size_increase(tp1);
sctp_total_flight_increase(stcb, tp1);
/*
* We inflate the cwnd to compensate for our
* artificial inflation of the flight_size.
*/
tp1->whoTo->cwnd += tp1->book_size;
tot_revoked++;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_SACK_LOGGING_ENABLE) {
sctp_log_sack(asoc->last_acked_seq,
cumack,
tp1->rec.data.TSN_seq,
0,
0,
SCTP_LOG_TSN_REVOKED);
}
} else if (tp1->sent == SCTP_DATAGRAM_MARKED) {
/* it has been re-acked in this SACK */
tp1->sent = SCTP_DATAGRAM_ACKED;
}
}
if (tp1->sent == SCTP_DATAGRAM_UNSENT)
break;
tp1 = TAILQ_NEXT(tp1, sctp_next);
}
if (tot_revoked > 0) {
/*
* Setup the ecn nonce re-sync point. We do this since once
* data is revoked we begin to retransmit things, which do
* NOT have the ECN bits set. This means we are now out of
* sync and must wait until we get back in sync with the
* peer to check ECN bits.
*/
tp1 = TAILQ_FIRST(&asoc->send_queue);
if (tp1 == NULL) {
asoc->nonce_resync_tsn = asoc->sending_seq;
} else {
asoc->nonce_resync_tsn = tp1->rec.data.TSN_seq;
}
asoc->nonce_wait_for_ecne = 0;
asoc->nonce_sum_check = 0;
}
}
static void
sctp_strike_gap_ack_chunks(struct sctp_tcb *stcb, struct sctp_association *asoc,
u_long biggest_tsn_acked, u_long biggest_tsn_newly_acked, u_long this_sack_lowest_newack, int accum_moved)
{
struct sctp_tmit_chunk *tp1;
int strike_flag = 0;
struct timeval now;
int tot_retrans = 0;
uint32_t sending_seq;
struct sctp_nets *net;
int num_dests_sacked = 0;
/*
* select the sending_seq, this is either the next thing ready to be
* sent but not transmitted, OR, the next seq we assign.
*/
tp1 = TAILQ_FIRST(&stcb->asoc.send_queue);
if (tp1 == NULL) {
sending_seq = asoc->sending_seq;
} else {
sending_seq = tp1->rec.data.TSN_seq;
}
/* CMT DAC algo: finding out if SACK is a mixed SACK */
if (SCTP_BASE_SYSCTL(sctp_cmt_on_off) && SCTP_BASE_SYSCTL(sctp_cmt_use_dac)) {
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
if (net->saw_newack)
num_dests_sacked++;
}
}
if (stcb->asoc.peer_supports_prsctp) {
(void)SCTP_GETTIME_TIMEVAL(&now);
}
tp1 = TAILQ_FIRST(&asoc->sent_queue);
while (tp1) {
strike_flag = 0;
if (tp1->no_fr_allowed) {
/* this one had a timeout or something */
tp1 = TAILQ_NEXT(tp1, sctp_next);
continue;
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) {
if (tp1->sent < SCTP_DATAGRAM_RESEND)
sctp_log_fr(biggest_tsn_newly_acked,
tp1->rec.data.TSN_seq,
tp1->sent,
SCTP_FR_LOG_CHECK_STRIKE);
}
if (compare_with_wrap(tp1->rec.data.TSN_seq, biggest_tsn_acked,
MAX_TSN) ||
tp1->sent == SCTP_DATAGRAM_UNSENT) {
/* done */
break;
}
if (stcb->asoc.peer_supports_prsctp) {
if ((PR_SCTP_TTL_ENABLED(tp1->flags)) && tp1->sent < SCTP_DATAGRAM_ACKED) {
/* Is it expired? */
if (
/*
* TODO sctp_constants.h needs alternative
* time macros when _KERNEL is undefined.
*/
(timevalcmp(&now, &tp1->rec.data.timetodrop, >))
) {
/* Yes so drop it */
if (tp1->data != NULL) {
(void)sctp_release_pr_sctp_chunk(stcb, tp1,
(SCTP_RESPONSE_TO_USER_REQ | SCTP_NOTIFY_DATAGRAM_SENT),
&asoc->sent_queue, SCTP_SO_NOT_LOCKED);
}
tp1 = TAILQ_NEXT(tp1, sctp_next);
continue;
}
}
if ((PR_SCTP_RTX_ENABLED(tp1->flags)) && tp1->sent < SCTP_DATAGRAM_ACKED) {
/* Has it been retransmitted tv_sec times? */
if (tp1->snd_count > tp1->rec.data.timetodrop.tv_sec) {
/* Yes, so drop it */
if (tp1->data != NULL) {
(void)sctp_release_pr_sctp_chunk(stcb, tp1,
(SCTP_RESPONSE_TO_USER_REQ | SCTP_NOTIFY_DATAGRAM_SENT),
&asoc->sent_queue, SCTP_SO_NOT_LOCKED);
}
tp1 = TAILQ_NEXT(tp1, sctp_next);
continue;
}
}
}
if (compare_with_wrap(tp1->rec.data.TSN_seq,
asoc->this_sack_highest_gap, MAX_TSN)) {
/* we are beyond the tsn in the sack */
break;
}
if (tp1->sent >= SCTP_DATAGRAM_RESEND) {
/* either a RESEND, ACKED, or MARKED */
/* skip */
tp1 = TAILQ_NEXT(tp1, sctp_next);
continue;
}
/*
* CMT : SFR algo (covers part of DAC and HTNA as well)
*/
if (tp1->whoTo && tp1->whoTo->saw_newack == 0) {
/*
* No new acks were receieved for data sent to this
* dest. Therefore, according to the SFR algo for
* CMT, no data sent to this dest can be marked for
* FR using this SACK.
*/
tp1 = TAILQ_NEXT(tp1, sctp_next);
continue;
} else if (tp1->whoTo && compare_with_wrap(tp1->rec.data.TSN_seq,
tp1->whoTo->this_sack_highest_newack, MAX_TSN)) {
/*
* CMT: New acks were receieved for data sent to
* this dest. But no new acks were seen for data
* sent after tp1. Therefore, according to the SFR
* algo for CMT, tp1 cannot be marked for FR using
* this SACK. This step covers part of the DAC algo
* and the HTNA algo as well.
*/
tp1 = TAILQ_NEXT(tp1, sctp_next);
continue;
}
/*
* Here we check to see if we were have already done a FR
* and if so we see if the biggest TSN we saw in the sack is
* smaller than the recovery point. If so we don't strike
* the tsn... otherwise we CAN strike the TSN.
*/
/*
* @@@ JRI: Check for CMT if (accum_moved &&
* asoc->fast_retran_loss_recovery && (sctp_cmt_on_off ==
* 0)) {
*/
if (accum_moved && asoc->fast_retran_loss_recovery) {
/*
* Strike the TSN if in fast-recovery and cum-ack
* moved.
*/
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) {
sctp_log_fr(biggest_tsn_newly_acked,
tp1->rec.data.TSN_seq,
tp1->sent,
SCTP_FR_LOG_STRIKE_CHUNK);
}
if (tp1->sent < SCTP_DATAGRAM_RESEND) {
tp1->sent++;
}
if (SCTP_BASE_SYSCTL(sctp_cmt_on_off) && SCTP_BASE_SYSCTL(sctp_cmt_use_dac)) {
/*
* CMT DAC algorithm: If SACK flag is set to
* 0, then lowest_newack test will not pass
* because it would have been set to the
* cumack earlier. If not already to be
* rtx'd, If not a mixed sack and if tp1 is
* not between two sacked TSNs, then mark by
* one more. NOTE that we are marking by one
* additional time since the SACK DAC flag
* indicates that two packets have been
* received after this missing TSN.
*/
if ((tp1->sent < SCTP_DATAGRAM_RESEND) && (num_dests_sacked == 1) &&
compare_with_wrap(this_sack_lowest_newack, tp1->rec.data.TSN_seq, MAX_TSN)) {
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) {
sctp_log_fr(16 + num_dests_sacked,
tp1->rec.data.TSN_seq,
tp1->sent,
SCTP_FR_LOG_STRIKE_CHUNK);
}
tp1->sent++;
}
}
} else if ((tp1->rec.data.doing_fast_retransmit) && (SCTP_BASE_SYSCTL(sctp_cmt_on_off) == 0)) {
/*
* For those that have done a FR we must take
* special consideration if we strike. I.e the
* biggest_newly_acked must be higher than the
* sending_seq at the time we did the FR.
*/
if (
#ifdef SCTP_FR_TO_ALTERNATE
/*
* If FR's go to new networks, then we must only do
* this for singly homed asoc's. However if the FR's
* go to the same network (Armando's work) then its
* ok to FR multiple times.
*/
(asoc->numnets < 2)
#else
(1)
#endif
) {
if ((compare_with_wrap(biggest_tsn_newly_acked,
tp1->rec.data.fast_retran_tsn, MAX_TSN)) ||
(biggest_tsn_newly_acked ==
tp1->rec.data.fast_retran_tsn)) {
/*
* Strike the TSN, since this ack is
* beyond where things were when we
* did a FR.
*/
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) {
sctp_log_fr(biggest_tsn_newly_acked,
tp1->rec.data.TSN_seq,
tp1->sent,
SCTP_FR_LOG_STRIKE_CHUNK);
}
if (tp1->sent < SCTP_DATAGRAM_RESEND) {
tp1->sent++;
}
strike_flag = 1;
if (SCTP_BASE_SYSCTL(sctp_cmt_on_off) && SCTP_BASE_SYSCTL(sctp_cmt_use_dac)) {
/*
* CMT DAC algorithm: If
* SACK flag is set to 0,
* then lowest_newack test
* will not pass because it
* would have been set to
* the cumack earlier. If
* not already to be rtx'd,
* If not a mixed sack and
* if tp1 is not between two
* sacked TSNs, then mark by
* one more. NOTE that we
* are marking by one
* additional time since the
* SACK DAC flag indicates
* that two packets have
* been received after this
* missing TSN.
*/
if ((tp1->sent < SCTP_DATAGRAM_RESEND) &&
(num_dests_sacked == 1) &&
compare_with_wrap(this_sack_lowest_newack,
tp1->rec.data.TSN_seq, MAX_TSN)) {
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) {
sctp_log_fr(32 + num_dests_sacked,
tp1->rec.data.TSN_seq,
tp1->sent,
SCTP_FR_LOG_STRIKE_CHUNK);
}
if (tp1->sent < SCTP_DATAGRAM_RESEND) {
tp1->sent++;
}
}
}
}
}
/*
* JRI: TODO: remove code for HTNA algo. CMT's SFR
* algo covers HTNA.
*/
} else if (compare_with_wrap(tp1->rec.data.TSN_seq,
biggest_tsn_newly_acked, MAX_TSN)) {
/*
* We don't strike these: This is the HTNA
* algorithm i.e. we don't strike If our TSN is
* larger than the Highest TSN Newly Acked.
*/
;
} else {
/* Strike the TSN */
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) {
sctp_log_fr(biggest_tsn_newly_acked,
tp1->rec.data.TSN_seq,
tp1->sent,
SCTP_FR_LOG_STRIKE_CHUNK);
}
if (tp1->sent < SCTP_DATAGRAM_RESEND) {
tp1->sent++;
}
if (SCTP_BASE_SYSCTL(sctp_cmt_on_off) && SCTP_BASE_SYSCTL(sctp_cmt_use_dac)) {
/*
* CMT DAC algorithm: If SACK flag is set to
* 0, then lowest_newack test will not pass
* because it would have been set to the
* cumack earlier. If not already to be
* rtx'd, If not a mixed sack and if tp1 is
* not between two sacked TSNs, then mark by
* one more. NOTE that we are marking by one
* additional time since the SACK DAC flag
* indicates that two packets have been
* received after this missing TSN.
*/
if ((tp1->sent < SCTP_DATAGRAM_RESEND) && (num_dests_sacked == 1) &&
compare_with_wrap(this_sack_lowest_newack, tp1->rec.data.TSN_seq, MAX_TSN)) {
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) {
sctp_log_fr(48 + num_dests_sacked,
tp1->rec.data.TSN_seq,
tp1->sent,
SCTP_FR_LOG_STRIKE_CHUNK);
}
tp1->sent++;
}
}
}
if (tp1->sent == SCTP_DATAGRAM_RESEND) {
/* Increment the count to resend */
struct sctp_nets *alt;
/* printf("OK, we are now ready to FR this guy\n"); */
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FR_LOGGING_ENABLE) {
sctp_log_fr(tp1->rec.data.TSN_seq, tp1->snd_count,
0, SCTP_FR_MARKED);
}
if (strike_flag) {
/* This is a subsequent FR */
SCTP_STAT_INCR(sctps_sendmultfastretrans);
}
sctp_ucount_incr(stcb->asoc.sent_queue_retran_cnt);
if (SCTP_BASE_SYSCTL(sctp_cmt_on_off)) {
/*
* CMT: Using RTX_SSTHRESH policy for CMT.
* If CMT is being used, then pick dest with
* largest ssthresh for any retransmission.
*/
tp1->no_fr_allowed = 1;
alt = tp1->whoTo;
/* sa_ignore NO_NULL_CHK */
if (SCTP_BASE_SYSCTL(sctp_cmt_on_off) && SCTP_BASE_SYSCTL(sctp_cmt_pf)) {
/*
* JRS 5/18/07 - If CMT PF is on,
* use the PF version of
* find_alt_net()
*/
alt = sctp_find_alternate_net(stcb, alt, 2);
} else {
/*
* JRS 5/18/07 - If only CMT is on,
* use the CMT version of
* find_alt_net()
*/
/* sa_ignore NO_NULL_CHK */
alt = sctp_find_alternate_net(stcb, alt, 1);
}
if (alt == NULL) {
alt = tp1->whoTo;
}
/*
* CUCv2: If a different dest is picked for
* the retransmission, then new
* (rtx-)pseudo_cumack needs to be tracked
* for orig dest. Let CUCv2 track new (rtx-)
* pseudo-cumack always.
*/
if (tp1->whoTo) {
tp1->whoTo->find_pseudo_cumack = 1;
tp1->whoTo->find_rtx_pseudo_cumack = 1;
}
} else {/* CMT is OFF */
#ifdef SCTP_FR_TO_ALTERNATE
/* Can we find an alternate? */
alt = sctp_find_alternate_net(stcb, tp1->whoTo, 0);
#else
/*
* default behavior is to NOT retransmit
* FR's to an alternate. Armando Caro's
* paper details why.
*/
alt = tp1->whoTo;
#endif
}
tp1->rec.data.doing_fast_retransmit = 1;
tot_retrans++;
/* mark the sending seq for possible subsequent FR's */
/*
* printf("Marking TSN for FR new value %x\n",
* (uint32_t)tpi->rec.data.TSN_seq);
*/
if (TAILQ_EMPTY(&asoc->send_queue)) {
/*
* If the queue of send is empty then its
* the next sequence number that will be
* assigned so we subtract one from this to
* get the one we last sent.
*/
tp1->rec.data.fast_retran_tsn = sending_seq;
} else {
/*
* If there are chunks on the send queue
* (unsent data that has made it from the
* stream queues but not out the door, we
* take the first one (which will have the
* lowest TSN) and subtract one to get the
* one we last sent.
*/
struct sctp_tmit_chunk *ttt;
ttt = TAILQ_FIRST(&asoc->send_queue);
tp1->rec.data.fast_retran_tsn =
ttt->rec.data.TSN_seq;
}
if (tp1->do_rtt) {
/*
* this guy had a RTO calculation pending on
* it, cancel it
*/
tp1->do_rtt = 0;
}
/* fix counts and things */
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FLIGHT_LOGGING_ENABLE) {
sctp_misc_ints(SCTP_FLIGHT_LOG_DOWN_RSND,
(tp1->whoTo ? (tp1->whoTo->flight_size) : 0),
tp1->book_size,
(uintptr_t) tp1->whoTo,
tp1->rec.data.TSN_seq);
}
if (tp1->whoTo) {
tp1->whoTo->net_ack++;
sctp_flight_size_decrease(tp1);
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_RWND_ENABLE) {
sctp_log_rwnd(SCTP_INCREASE_PEER_RWND,
asoc->peers_rwnd, tp1->send_size, SCTP_BASE_SYSCTL(sctp_peer_chunk_oh));
}
/* add back to the rwnd */
asoc->peers_rwnd += (tp1->send_size + SCTP_BASE_SYSCTL(sctp_peer_chunk_oh));
/* remove from the total flight */
sctp_total_flight_decrease(stcb, tp1);
if (alt != tp1->whoTo) {
/* yes, there is an alternate. */
sctp_free_remote_addr(tp1->whoTo);
/* sa_ignore FREED_MEMORY */
tp1->whoTo = alt;
atomic_add_int(&alt->ref_count, 1);
}
}
tp1 = TAILQ_NEXT(tp1, sctp_next);
} /* while (tp1) */
if (tot_retrans > 0) {
/*
* Setup the ecn nonce re-sync point. We do this since once
* we go to FR something we introduce a Karn's rule scenario
* and won't know the totals for the ECN bits.
*/
asoc->nonce_resync_tsn = sending_seq;
asoc->nonce_wait_for_ecne = 0;
asoc->nonce_sum_check = 0;
}
}
struct sctp_tmit_chunk *
sctp_try_advance_peer_ack_point(struct sctp_tcb *stcb,
struct sctp_association *asoc)
{
struct sctp_tmit_chunk *tp1, *tp2, *a_adv = NULL;
struct timeval now;
int now_filled = 0;
if (asoc->peer_supports_prsctp == 0) {
return (NULL);
}
tp1 = TAILQ_FIRST(&asoc->sent_queue);
while (tp1) {
if (tp1->sent != SCTP_FORWARD_TSN_SKIP &&
tp1->sent != SCTP_DATAGRAM_RESEND) {
/* no chance to advance, out of here */
break;
}
if (!PR_SCTP_ENABLED(tp1->flags)) {
/*
* We can't fwd-tsn past any that are reliable aka
* retransmitted until the asoc fails.
*/
break;
}
if (!now_filled) {
(void)SCTP_GETTIME_TIMEVAL(&now);
now_filled = 1;
}
tp2 = TAILQ_NEXT(tp1, sctp_next);
/*
* now we got a chunk which is marked for another
* retransmission to a PR-stream but has run out its chances
* already maybe OR has been marked to skip now. Can we skip
* it if its a resend?
*/
if (tp1->sent == SCTP_DATAGRAM_RESEND &&
(PR_SCTP_TTL_ENABLED(tp1->flags))) {
/*
* Now is this one marked for resend and its time is
* now up?
*/
if (timevalcmp(&now, &tp1->rec.data.timetodrop, >)) {
/* Yes so drop it */
if (tp1->data) {
(void)sctp_release_pr_sctp_chunk(stcb, tp1,
(SCTP_RESPONSE_TO_USER_REQ | SCTP_NOTIFY_DATAGRAM_SENT),
&asoc->sent_queue, SCTP_SO_NOT_LOCKED);
}
} else {
/*
* No, we are done when hit one for resend
* whos time as not expired.
*/
break;
}
}
/*
* Ok now if this chunk is marked to drop it we can clean up
* the chunk, advance our peer ack point and we can check
* the next chunk.
*/
if (tp1->sent == SCTP_FORWARD_TSN_SKIP) {
/* advance PeerAckPoint goes forward */
asoc->advanced_peer_ack_point = tp1->rec.data.TSN_seq;
a_adv = tp1;
/*
* we don't want to de-queue it here. Just wait for
* the next peer SACK to come with a new cumTSN and
* then the chunk will be droped in the normal
* fashion.
*/
if (tp1->data) {
sctp_free_bufspace(stcb, asoc, tp1, 1);
/*
* Maybe there should be another
* notification type
*/
sctp_ulp_notify(SCTP_NOTIFY_DG_FAIL, stcb,
(SCTP_RESPONSE_TO_USER_REQ | SCTP_NOTIFY_DATAGRAM_SENT),
tp1, SCTP_SO_NOT_LOCKED);
sctp_m_freem(tp1->data);
tp1->data = NULL;
if (stcb->sctp_socket) {
#if defined (__APPLE__) || defined(SCTP_SO_LOCK_TESTING)
struct socket *so;
so = SCTP_INP_SO(stcb->sctp_ep);
atomic_add_int(&stcb->asoc.refcnt, 1);
SCTP_TCB_UNLOCK(stcb);
SCTP_SOCKET_LOCK(so, 1);
SCTP_TCB_LOCK(stcb);
atomic_subtract_int(&stcb->asoc.refcnt, 1);
if (stcb->asoc.state & SCTP_STATE_CLOSED_SOCKET) {
/*
* assoc was freed while we
* were unlocked
*/
SCTP_SOCKET_UNLOCK(so, 1);
return (NULL);
}
#endif
sctp_sowwakeup(stcb->sctp_ep, stcb->sctp_socket);
#if defined (__APPLE__) || defined(SCTP_SO_LOCK_TESTING)
SCTP_SOCKET_UNLOCK(so, 1);
#endif
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_WAKE_LOGGING_ENABLE) {
sctp_wakeup_log(stcb, tp1->rec.data.TSN_seq, 1, SCTP_WAKESND_FROM_FWDTSN);
}
}
}
} else {
/*
* If it is still in RESEND we can advance no
* further
*/
break;
}
/*
* If we hit here we just dumped tp1, move to next tsn on
* sent queue.
*/
tp1 = tp2;
}
return (a_adv);
}
static void
sctp_fs_audit(struct sctp_association *asoc)
{
struct sctp_tmit_chunk *chk;
int inflight = 0, resend = 0, inbetween = 0, acked = 0, above = 0;
TAILQ_FOREACH(chk, &asoc->sent_queue, sctp_next) {
if (chk->sent < SCTP_DATAGRAM_RESEND) {
inflight++;
} else if (chk->sent == SCTP_DATAGRAM_RESEND) {
resend++;
} else if (chk->sent < SCTP_DATAGRAM_ACKED) {
inbetween++;
} else if (chk->sent > SCTP_DATAGRAM_ACKED) {
above++;
} else {
acked++;
}
}
if ((inflight > 0) || (inbetween > 0)) {
#ifdef INVARIANTS
panic("Flight size-express incorrect? \n");
#else
SCTP_PRINTF("Flight size-express incorrect inflight:%d inbetween:%d\n",
inflight, inbetween);
#endif
}
}
static void
sctp_window_probe_recovery(struct sctp_tcb *stcb,
struct sctp_association *asoc,
struct sctp_nets *net,
struct sctp_tmit_chunk *tp1)
{
struct sctp_tmit_chunk *chk;
/* First setup this one and get it moved back */
tp1->sent = SCTP_DATAGRAM_UNSENT;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FLIGHT_LOGGING_ENABLE) {
sctp_misc_ints(SCTP_FLIGHT_LOG_DOWN_WP,
tp1->whoTo->flight_size,
tp1->book_size,
(uintptr_t) tp1->whoTo,
tp1->rec.data.TSN_seq);
}
sctp_flight_size_decrease(tp1);
sctp_total_flight_decrease(stcb, tp1);
TAILQ_REMOVE(&asoc->sent_queue, tp1, sctp_next);
TAILQ_INSERT_HEAD(&asoc->send_queue, tp1, sctp_next);
asoc->sent_queue_cnt--;
asoc->send_queue_cnt++;
/*
* Now all guys marked for RESEND on the sent_queue must be moved
* back too.
*/
TAILQ_FOREACH(chk, &asoc->sent_queue, sctp_next) {
if (chk->sent == SCTP_DATAGRAM_RESEND) {
/* Another chunk to move */
chk->sent = SCTP_DATAGRAM_UNSENT;
/* It should not be in flight */
TAILQ_REMOVE(&asoc->sent_queue, chk, sctp_next);
TAILQ_INSERT_AFTER(&asoc->send_queue, tp1, chk, sctp_next);
asoc->sent_queue_cnt--;
asoc->send_queue_cnt++;
sctp_ucount_decr(asoc->sent_queue_retran_cnt);
}
}
}
void
sctp_express_handle_sack(struct sctp_tcb *stcb, uint32_t cumack,
uint32_t rwnd, int nonce_sum_flag, int *abort_now)
{
struct sctp_nets *net;
struct sctp_association *asoc;
struct sctp_tmit_chunk *tp1, *tp2;
uint32_t old_rwnd;
int win_probe_recovery = 0;
int win_probe_recovered = 0;
int j, done_once = 0;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_SACK_ARRIVALS_ENABLE) {
sctp_misc_ints(SCTP_SACK_LOG_EXPRESS, cumack,
rwnd, stcb->asoc.last_acked_seq, stcb->asoc.peers_rwnd);
}
SCTP_TCB_LOCK_ASSERT(stcb);
#ifdef SCTP_ASOCLOG_OF_TSNS
stcb->asoc.cumack_log[stcb->asoc.cumack_log_at] = cumack;
stcb->asoc.cumack_log_at++;
if (stcb->asoc.cumack_log_at > SCTP_TSN_LOG_SIZE) {
stcb->asoc.cumack_log_at = 0;
}
#endif
asoc = &stcb->asoc;
old_rwnd = asoc->peers_rwnd;
if (compare_with_wrap(asoc->last_acked_seq, cumack, MAX_TSN)) {
/* old ack */
return;
} else if (asoc->last_acked_seq == cumack) {
/* Window update sack */
asoc->peers_rwnd = sctp_sbspace_sub(rwnd,
(uint32_t) (asoc->total_flight + (asoc->sent_queue_cnt * SCTP_BASE_SYSCTL(sctp_peer_chunk_oh))));
if (asoc->peers_rwnd < stcb->sctp_ep->sctp_ep.sctp_sws_sender) {
/* SWS sender side engages */
asoc->peers_rwnd = 0;
}
if (asoc->peers_rwnd > old_rwnd) {
goto again;
}
return;
}
/* First setup for CC stuff */
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
net->prev_cwnd = net->cwnd;
net->net_ack = 0;
net->net_ack2 = 0;
/*
* CMT: Reset CUC and Fast recovery algo variables before
* SACK processing
*/
net->new_pseudo_cumack = 0;
net->will_exit_fast_recovery = 0;
}
if (SCTP_BASE_SYSCTL(sctp_strict_sacks)) {
uint32_t send_s;
if (!TAILQ_EMPTY(&asoc->sent_queue)) {
tp1 = TAILQ_LAST(&asoc->sent_queue,
sctpchunk_listhead);
send_s = tp1->rec.data.TSN_seq + 1;
} else {
send_s = asoc->sending_seq;
}
if ((cumack == send_s) ||
compare_with_wrap(cumack, send_s, MAX_TSN)) {
#ifndef INVARIANTS
struct mbuf *oper;
#endif
#ifdef INVARIANTS
panic("Impossible sack 1");
#else
*abort_now = 1;
/* XXX */
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) = sizeof(struct sctp_paramhdr) +
sizeof(uint32_t);
ph = mtod(oper, struct sctp_paramhdr *);
ph->param_type = htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length = htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_25);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_25;
sctp_abort_an_association(stcb->sctp_ep, stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
return;
#endif
}
}
asoc->this_sack_highest_gap = cumack;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_THRESHOLD_LOGGING) {
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
sctp_misc_ints(SCTP_THRESHOLD_CLEAR,
stcb->asoc.overall_error_count,
0,
SCTP_FROM_SCTP_INDATA,
__LINE__);
}
stcb->asoc.overall_error_count = 0;
if (compare_with_wrap(cumack, asoc->last_acked_seq, MAX_TSN)) {
/* process the new consecutive TSN first */
tp1 = TAILQ_FIRST(&asoc->sent_queue);
while (tp1) {
tp2 = TAILQ_NEXT(tp1, sctp_next);
if (compare_with_wrap(cumack, tp1->rec.data.TSN_seq,
MAX_TSN) ||
cumack == tp1->rec.data.TSN_seq) {
if (tp1->sent == SCTP_DATAGRAM_UNSENT) {
printf("Warning, an unsent is now acked?\n");
}
/*
* ECN Nonce: Add the nonce to the sender's
* nonce sum
*/
asoc->nonce_sum_expect_base += tp1->rec.data.ect_nonce;
if (tp1->sent < SCTP_DATAGRAM_ACKED) {
/*
* If it is less than ACKED, it is
* now no-longer in flight. Higher
* values may occur during marking
*/
if (tp1->sent < SCTP_DATAGRAM_RESEND) {
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FLIGHT_LOGGING_ENABLE) {
sctp_misc_ints(SCTP_FLIGHT_LOG_DOWN_CA,
tp1->whoTo->flight_size,
tp1->book_size,
(uintptr_t) tp1->whoTo,
tp1->rec.data.TSN_seq);
}
sctp_flight_size_decrease(tp1);
/* sa_ignore NO_NULL_CHK */
sctp_total_flight_decrease(stcb, tp1);
}
tp1->whoTo->net_ack += tp1->send_size;
if (tp1->snd_count < 2) {
/*
* True non-retransmited
* chunk
*/
tp1->whoTo->net_ack2 +=
tp1->send_size;
/* update RTO too? */
if (tp1->do_rtt) {
tp1->whoTo->RTO =
/*
* sa_ignore
* NO_NULL_CHK
*/
sctp_calculate_rto(stcb,
asoc, tp1->whoTo,
&tp1->sent_rcv_time,
sctp_align_safe_nocopy);
tp1->do_rtt = 0;
}
}
/*
* CMT: CUCv2 algorithm. From the
* cumack'd TSNs, for each TSN being
* acked for the first time, set the
* following variables for the
* corresp destination.
* new_pseudo_cumack will trigger a
* cwnd update.
* find_(rtx_)pseudo_cumack will
* trigger search for the next
* expected (rtx-)pseudo-cumack.
*/
tp1->whoTo->new_pseudo_cumack = 1;
tp1->whoTo->find_pseudo_cumack = 1;
tp1->whoTo->find_rtx_pseudo_cumack = 1;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
/* sa_ignore NO_NULL_CHK */
sctp_log_cwnd(stcb, tp1->whoTo, tp1->rec.data.TSN_seq, SCTP_CWND_LOG_FROM_SACK);
}
}
if (tp1->sent == SCTP_DATAGRAM_RESEND) {
sctp_ucount_decr(asoc->sent_queue_retran_cnt);
}
if (tp1->rec.data.chunk_was_revoked) {
/* deflate the cwnd */
tp1->whoTo->cwnd -= tp1->book_size;
tp1->rec.data.chunk_was_revoked = 0;
}
tp1->sent = SCTP_DATAGRAM_ACKED;
TAILQ_REMOVE(&asoc->sent_queue, tp1, sctp_next);
if (tp1->data) {
/* sa_ignore NO_NULL_CHK */
sctp_free_bufspace(stcb, asoc, tp1, 1);
sctp_m_freem(tp1->data);
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_SACK_LOGGING_ENABLE) {
sctp_log_sack(asoc->last_acked_seq,
cumack,
tp1->rec.data.TSN_seq,
0,
0,
SCTP_LOG_FREE_SENT);
}
tp1->data = NULL;
asoc->sent_queue_cnt--;
sctp_free_a_chunk(stcb, tp1);
tp1 = tp2;
} else {
break;
}
}
}
/* sa_ignore NO_NULL_CHK */
if (stcb->sctp_socket) {
#if defined (__APPLE__) || defined(SCTP_SO_LOCK_TESTING)
struct socket *so;
#endif
SOCKBUF_LOCK(&stcb->sctp_socket->so_snd);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_WAKE_LOGGING_ENABLE) {
/* sa_ignore NO_NULL_CHK */
sctp_wakeup_log(stcb, cumack, 1, SCTP_WAKESND_FROM_SACK);
}
#if defined (__APPLE__) || defined(SCTP_SO_LOCK_TESTING)
so = SCTP_INP_SO(stcb->sctp_ep);
atomic_add_int(&stcb->asoc.refcnt, 1);
SCTP_TCB_UNLOCK(stcb);
SCTP_SOCKET_LOCK(so, 1);
SCTP_TCB_LOCK(stcb);
atomic_subtract_int(&stcb->asoc.refcnt, 1);
if (stcb->asoc.state & SCTP_STATE_CLOSED_SOCKET) {
/* assoc was freed while we were unlocked */
SCTP_SOCKET_UNLOCK(so, 1);
return;
}
#endif
sctp_sowwakeup_locked(stcb->sctp_ep, stcb->sctp_socket);
#if defined (__APPLE__) || defined(SCTP_SO_LOCK_TESTING)
SCTP_SOCKET_UNLOCK(so, 1);
#endif
} else {
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_WAKE_LOGGING_ENABLE) {
sctp_wakeup_log(stcb, cumack, 1, SCTP_NOWAKE_FROM_SACK);
}
}
/* JRS - Use the congestion control given in the CC module */
if (asoc->last_acked_seq != cumack)
asoc->cc_functions.sctp_cwnd_update_after_sack(stcb, asoc, 1, 0, 0);
asoc->last_acked_seq = cumack;
if (TAILQ_EMPTY(&asoc->sent_queue)) {
/* nothing left in-flight */
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
net->flight_size = 0;
net->partial_bytes_acked = 0;
}
asoc->total_flight = 0;
asoc->total_flight_count = 0;
}
/* Fix up the a-p-a-p for future PR-SCTP sends */
if (compare_with_wrap(cumack, asoc->advanced_peer_ack_point, MAX_TSN)) {
asoc->advanced_peer_ack_point = cumack;
}
/* ECN Nonce updates */
if (asoc->ecn_nonce_allowed) {
if (asoc->nonce_sum_check) {
if (nonce_sum_flag != ((asoc->nonce_sum_expect_base) & SCTP_SACK_NONCE_SUM)) {
if (asoc->nonce_wait_for_ecne == 0) {
struct sctp_tmit_chunk *lchk;
lchk = TAILQ_FIRST(&asoc->send_queue);
asoc->nonce_wait_for_ecne = 1;
if (lchk) {
asoc->nonce_wait_tsn = lchk->rec.data.TSN_seq;
} else {
asoc->nonce_wait_tsn = asoc->sending_seq;
}
} else {
if (compare_with_wrap(asoc->last_acked_seq, asoc->nonce_wait_tsn, MAX_TSN) ||
(asoc->last_acked_seq == asoc->nonce_wait_tsn)) {
/*
* Misbehaving peer. We need
* to react to this guy
*/
asoc->ecn_allowed = 0;
asoc->ecn_nonce_allowed = 0;
}
}
}
} else {
/* See if Resynchronization Possible */
if (compare_with_wrap(asoc->last_acked_seq, asoc->nonce_resync_tsn, MAX_TSN)) {
asoc->nonce_sum_check = 1;
/*
* now we must calculate what the base is.
* We do this based on two things, we know
* the total's for all the segments
* gap-acked in the SACK (none), We also
* know the SACK's nonce sum, its in
* nonce_sum_flag. So we can build a truth
* table to back-calculate the new value of
* asoc->nonce_sum_expect_base:
*
* SACK-flag-Value Seg-Sums Base 0 0 0
* 1 0 1 0 1 1 1
* 1 0
*/
asoc->nonce_sum_expect_base = (0 ^ nonce_sum_flag) & SCTP_SACK_NONCE_SUM;
}
}
}
/* RWND update */
asoc->peers_rwnd = sctp_sbspace_sub(rwnd,
(uint32_t) (asoc->total_flight + (asoc->sent_queue_cnt * SCTP_BASE_SYSCTL(sctp_peer_chunk_oh))));
if (asoc->peers_rwnd < stcb->sctp_ep->sctp_ep.sctp_sws_sender) {
/* SWS sender side engages */
asoc->peers_rwnd = 0;
}
if (asoc->peers_rwnd > old_rwnd) {
win_probe_recovery = 1;
}
/* Now assure a timer where data is queued at */
again:
j = 0;
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
if (win_probe_recovery && (net->window_probe)) {
net->window_probe = 0;
win_probe_recovered = 1;
/*
* Find first chunk that was used with window probe
* and clear the sent
*/
/* sa_ignore FREED_MEMORY */
TAILQ_FOREACH(tp1, &asoc->sent_queue, sctp_next) {
if (tp1->window_probe) {
/* move back to data send queue */
sctp_window_probe_recovery(stcb, asoc, net, tp1);
break;
}
}
}
if (net->flight_size) {
int to_ticks;
if (net->RTO == 0) {
to_ticks = MSEC_TO_TICKS(stcb->asoc.initial_rto);
} else {
to_ticks = MSEC_TO_TICKS(net->RTO);
}
j++;
(void)SCTP_OS_TIMER_START(&net->rxt_timer.timer, to_ticks,
sctp_timeout_handler, &net->rxt_timer);
} else {
if (SCTP_OS_TIMER_PENDING(&net->rxt_timer.timer)) {
sctp_timer_stop(SCTP_TIMER_TYPE_SEND, stcb->sctp_ep,
stcb, net,
SCTP_FROM_SCTP_INDATA + SCTP_LOC_22);
}
if (SCTP_BASE_SYSCTL(sctp_early_fr)) {
if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
SCTP_STAT_INCR(sctps_earlyfrstpidsck4);
sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
SCTP_FROM_SCTP_INDATA + SCTP_LOC_23);
}
}
}
}
if ((j == 0) &&
(!TAILQ_EMPTY(&asoc->sent_queue)) &&
(asoc->sent_queue_retran_cnt == 0) &&
(win_probe_recovered == 0) &&
(done_once == 0)) {
/* huh, this should not happen */
sctp_fs_audit(asoc);
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
net->flight_size = 0;
}
asoc->total_flight = 0;
asoc->total_flight_count = 0;
asoc->sent_queue_retran_cnt = 0;
TAILQ_FOREACH(tp1, &asoc->sent_queue, sctp_next) {
if (tp1->sent < SCTP_DATAGRAM_RESEND) {
sctp_flight_size_increase(tp1);
sctp_total_flight_increase(stcb, tp1);
} else if (tp1->sent == SCTP_DATAGRAM_RESEND) {
asoc->sent_queue_retran_cnt++;
}
}
done_once = 1;
goto again;
}
/**********************************/
/* Now what about shutdown issues */
/**********************************/
if (TAILQ_EMPTY(&asoc->send_queue) && TAILQ_EMPTY(&asoc->sent_queue)) {
/* nothing left on sendqueue.. consider done */
/* clean up */
if ((asoc->stream_queue_cnt == 1) &&
((asoc->state & SCTP_STATE_SHUTDOWN_PENDING) ||
(asoc->state & SCTP_STATE_SHUTDOWN_RECEIVED)) &&
(asoc->locked_on_sending)
) {
struct sctp_stream_queue_pending *sp;
/*
* I may be in a state where we got all across.. but
* cannot write more due to a shutdown... we abort
* since the user did not indicate EOR in this case.
* The sp will be cleaned during free of the asoc.
*/
sp = TAILQ_LAST(&((asoc->locked_on_sending)->outqueue),
sctp_streamhead);
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
if ((sp) && (sp->length == 0)) {
/* Let cleanup code purge it */
if (sp->msg_is_complete) {
asoc->stream_queue_cnt--;
} else {
asoc->state |= SCTP_STATE_PARTIAL_MSG_LEFT;
asoc->locked_on_sending = NULL;
asoc->stream_queue_cnt--;
}
}
}
if ((asoc->state & SCTP_STATE_SHUTDOWN_PENDING) &&
(asoc->stream_queue_cnt == 0)) {
if (asoc->state & SCTP_STATE_PARTIAL_MSG_LEFT) {
/* Need to abort here */
struct mbuf *oper;
abort_out_now:
*abort_now = 1;
/* XXX */
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) = sizeof(struct sctp_paramhdr) +
sizeof(uint32_t);
ph = mtod(oper, struct sctp_paramhdr *);
ph->param_type = htons(SCTP_CAUSE_USER_INITIATED_ABT);
ph->param_length = htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_24);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_24;
sctp_abort_an_association(stcb->sctp_ep, stcb, SCTP_RESPONSE_TO_USER_REQ, oper, SCTP_SO_NOT_LOCKED);
} else {
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
if ((SCTP_GET_STATE(asoc) == SCTP_STATE_OPEN) ||
(SCTP_GET_STATE(asoc) == SCTP_STATE_SHUTDOWN_RECEIVED)) {
SCTP_STAT_DECR_GAUGE32(sctps_currestab);
}
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
SCTP_SET_STATE(asoc, SCTP_STATE_SHUTDOWN_SENT);
SCTP_CLEAR_SUBSTATE(asoc, SCTP_STATE_SHUTDOWN_PENDING);
sctp_stop_timers_for_shutdown(stcb);
sctp_send_shutdown(stcb,
stcb->asoc.primary_destination);
sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWN,
stcb->sctp_ep, stcb, asoc->primary_destination);
sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNGUARD,
stcb->sctp_ep, stcb, asoc->primary_destination);
}
} else if ((SCTP_GET_STATE(asoc) == SCTP_STATE_SHUTDOWN_RECEIVED) &&
(asoc->stream_queue_cnt == 0)) {
if (asoc->state & SCTP_STATE_PARTIAL_MSG_LEFT) {
goto abort_out_now;
}
SCTP_STAT_DECR_GAUGE32(sctps_currestab);
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
SCTP_SET_STATE(asoc, SCTP_STATE_SHUTDOWN_ACK_SENT);
SCTP_CLEAR_SUBSTATE(asoc, SCTP_STATE_SHUTDOWN_PENDING);
sctp_send_shutdown_ack(stcb,
stcb->asoc.primary_destination);
sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNACK,
stcb->sctp_ep, stcb, asoc->primary_destination);
}
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_SACK_RWND_LOGGING_ENABLE) {
sctp_misc_ints(SCTP_SACK_RWND_UPDATE,
rwnd,
stcb->asoc.peers_rwnd,
stcb->asoc.total_flight,
stcb->asoc.total_output_queue_size);
}
}
void
sctp_handle_sack(struct mbuf *m, int offset,
struct sctp_sack_chunk *ch, struct sctp_tcb *stcb,
struct sctp_nets *net_from, int *abort_now, int sack_len, uint32_t rwnd)
{
struct sctp_association *asoc;
struct sctp_sack *sack;
struct sctp_tmit_chunk *tp1, *tp2;
uint32_t cum_ack, last_tsn, biggest_tsn_acked, biggest_tsn_newly_acked,
this_sack_lowest_newack;
uint32_t sav_cum_ack;
uint16_t num_seg, num_dup;
uint16_t wake_him = 0;
unsigned int sack_length;
uint32_t send_s = 0;
long j;
int accum_moved = 0;
int will_exit_fast_recovery = 0;
uint32_t a_rwnd, old_rwnd;
int win_probe_recovery = 0;
int win_probe_recovered = 0;
struct sctp_nets *net = NULL;
int nonce_sum_flag, ecn_seg_sums = 0;
int done_once;
uint8_t reneged_all = 0;
uint8_t cmt_dac_flag;
/*
* we take any chance we can to service our queues since we cannot
* get awoken when the socket is read from :<
*/
/*
* Now perform the actual SACK handling: 1) Verify that it is not an
* old sack, if so discard. 2) If there is nothing left in the send
* queue (cum-ack is equal to last acked) then you have a duplicate
* too, update any rwnd change and verify no timers are running.
* then return. 3) Process any new consequtive data i.e. cum-ack
* moved process these first and note that it moved. 4) Process any
* sack blocks. 5) Drop any acked from the queue. 6) Check for any
* revoked blocks and mark. 7) Update the cwnd. 8) Nothing left,
* sync up flightsizes and things, stop all timers and also check
* for shutdown_pending state. If so then go ahead and send off the
* shutdown. If in shutdown recv, send off the shutdown-ack and
* start that timer, Ret. 9) Strike any non-acked things and do FR
* procedure if needed being sure to set the FR flag. 10) Do pr-sctp
* procedures. 11) Apply any FR penalties. 12) Assure we will SACK
* if in shutdown_recv state.
*/
SCTP_TCB_LOCK_ASSERT(stcb);
sack = &ch->sack;
/* CMT DAC algo */
this_sack_lowest_newack = 0;
j = 0;
sack_length = (unsigned int)sack_len;
/* ECN Nonce */
SCTP_STAT_INCR(sctps_slowpath_sack);
nonce_sum_flag = ch->ch.chunk_flags & SCTP_SACK_NONCE_SUM;
cum_ack = last_tsn = ntohl(sack->cum_tsn_ack);
#ifdef SCTP_ASOCLOG_OF_TSNS
stcb->asoc.cumack_log[stcb->asoc.cumack_log_at] = cum_ack;
stcb->asoc.cumack_log_at++;
if (stcb->asoc.cumack_log_at > SCTP_TSN_LOG_SIZE) {
stcb->asoc.cumack_log_at = 0;
}
#endif
num_seg = ntohs(sack->num_gap_ack_blks);
a_rwnd = rwnd;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_SACK_ARRIVALS_ENABLE) {
sctp_misc_ints(SCTP_SACK_LOG_NORMAL, cum_ack,
rwnd, stcb->asoc.last_acked_seq, stcb->asoc.peers_rwnd);
}
/* CMT DAC algo */
cmt_dac_flag = ch->ch.chunk_flags & SCTP_SACK_CMT_DAC;
num_dup = ntohs(sack->num_dup_tsns);
old_rwnd = stcb->asoc.peers_rwnd;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_THRESHOLD_LOGGING) {
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
sctp_misc_ints(SCTP_THRESHOLD_CLEAR,
stcb->asoc.overall_error_count,
0,
SCTP_FROM_SCTP_INDATA,
__LINE__);
}
stcb->asoc.overall_error_count = 0;
asoc = &stcb->asoc;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_SACK_LOGGING_ENABLE) {
sctp_log_sack(asoc->last_acked_seq,
cum_ack,
0,
num_seg,
num_dup,
SCTP_LOG_NEW_SACK);
}
if ((num_dup) && (SCTP_BASE_SYSCTL(sctp_logging_level) & (SCTP_FR_LOGGING_ENABLE | SCTP_EARLYFR_LOGGING_ENABLE))) {
int off_to_dup, iii;
uint32_t *dupdata, dblock;
off_to_dup = (num_seg * sizeof(struct sctp_gap_ack_block)) + sizeof(struct sctp_sack_chunk);
if ((off_to_dup + (num_dup * sizeof(uint32_t))) <= sack_length) {
dupdata = (uint32_t *) sctp_m_getptr(m, off_to_dup,
sizeof(uint32_t), (uint8_t *) & dblock);
off_to_dup += sizeof(uint32_t);
if (dupdata) {
for (iii = 0; iii < num_dup; iii++) {
sctp_log_fr(*dupdata, 0, 0, SCTP_FR_DUPED);
dupdata = (uint32_t *) sctp_m_getptr(m, off_to_dup,
sizeof(uint32_t), (uint8_t *) & dblock);
if (dupdata == NULL)
break;
off_to_dup += sizeof(uint32_t);
}
}
} else {
SCTP_PRINTF("Size invalid offset to dups:%d number dups:%d sack_len:%d num gaps:%d\n",
off_to_dup, num_dup, sack_length, num_seg);
}
}
if (SCTP_BASE_SYSCTL(sctp_strict_sacks)) {
/* reality check */
if (!TAILQ_EMPTY(&asoc->sent_queue)) {
tp1 = TAILQ_LAST(&asoc->sent_queue,
sctpchunk_listhead);
send_s = tp1->rec.data.TSN_seq + 1;
} else {
send_s = asoc->sending_seq;
}
if (cum_ack == send_s ||
compare_with_wrap(cum_ack, send_s, MAX_TSN)) {
#ifndef INVARIANTS
struct mbuf *oper;
#endif
#ifdef INVARIANTS
hopeless_peer:
panic("Impossible sack 1");
#else
/*
* no way, we have not even sent this TSN out yet.
* Peer is hopelessly messed up with us.
*/
hopeless_peer:
*abort_now = 1;
/* XXX */
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) = sizeof(struct sctp_paramhdr) +
sizeof(uint32_t);
ph = mtod(oper, struct sctp_paramhdr *);
ph->param_type = htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length = htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_25);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_25;
sctp_abort_an_association(stcb->sctp_ep, stcb, SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
return;
#endif
}
}
/**********************/
/* 1) check the range */
/**********************/
if (compare_with_wrap(asoc->last_acked_seq, last_tsn, MAX_TSN)) {
/* acking something behind */
return;
}
sav_cum_ack = asoc->last_acked_seq;
/* update the Rwnd of the peer */
if (TAILQ_EMPTY(&asoc->sent_queue) &&
TAILQ_EMPTY(&asoc->send_queue) &&
(asoc->stream_queue_cnt == 0)
) {
/* nothing left on send/sent and strmq */
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_RWND_ENABLE) {
sctp_log_rwnd_set(SCTP_SET_PEER_RWND_VIA_SACK,
asoc->peers_rwnd, 0, 0, a_rwnd);
}
asoc->peers_rwnd = a_rwnd;
if (asoc->sent_queue_retran_cnt) {
asoc->sent_queue_retran_cnt = 0;
}
if (asoc->peers_rwnd < stcb->sctp_ep->sctp_ep.sctp_sws_sender) {
/* SWS sender side engages */
asoc->peers_rwnd = 0;
}
/* stop any timers */
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
sctp_timer_stop(SCTP_TIMER_TYPE_SEND, stcb->sctp_ep,
stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_26);
if (SCTP_BASE_SYSCTL(sctp_early_fr)) {
if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
SCTP_STAT_INCR(sctps_earlyfrstpidsck1);
sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
SCTP_FROM_SCTP_INDATA + SCTP_LOC_26);
}
}
net->partial_bytes_acked = 0;
net->flight_size = 0;
}
asoc->total_flight = 0;
asoc->total_flight_count = 0;
return;
}
/*
* We init netAckSz and netAckSz2 to 0. These are used to track 2
* things. The total byte count acked is tracked in netAckSz AND
* netAck2 is used to track the total bytes acked that are un-
* amibguious and were never retransmitted. We track these on a per
* destination address basis.
*/
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
net->prev_cwnd = net->cwnd;
net->net_ack = 0;
net->net_ack2 = 0;
/*
* CMT: Reset CUC and Fast recovery algo variables before
* SACK processing
*/
net->new_pseudo_cumack = 0;
net->will_exit_fast_recovery = 0;
}
/* process the new consecutive TSN first */
tp1 = TAILQ_FIRST(&asoc->sent_queue);
while (tp1) {
if (compare_with_wrap(last_tsn, tp1->rec.data.TSN_seq,
MAX_TSN) ||
last_tsn == tp1->rec.data.TSN_seq) {
if (tp1->sent != SCTP_DATAGRAM_UNSENT) {
/*
* ECN Nonce: Add the nonce to the sender's
* nonce sum
*/
asoc->nonce_sum_expect_base += tp1->rec.data.ect_nonce;
accum_moved = 1;
if (tp1->sent < SCTP_DATAGRAM_ACKED) {
/*
* If it is less than ACKED, it is
* now no-longer in flight. Higher
* values may occur during marking
*/
if ((tp1->whoTo->dest_state &
SCTP_ADDR_UNCONFIRMED) &&
(tp1->snd_count < 2)) {
/*
* If there was no retran
* and the address is
* un-confirmed and we sent
* there and are now
* sacked.. its confirmed,
* mark it so.
*/
tp1->whoTo->dest_state &=
~SCTP_ADDR_UNCONFIRMED;
}
if (tp1->sent < SCTP_DATAGRAM_RESEND) {
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FLIGHT_LOGGING_ENABLE) {
sctp_misc_ints(SCTP_FLIGHT_LOG_DOWN_CA,
tp1->whoTo->flight_size,
tp1->book_size,
(uintptr_t) tp1->whoTo,
tp1->rec.data.TSN_seq);
}
sctp_flight_size_decrease(tp1);
sctp_total_flight_decrease(stcb, tp1);
}
tp1->whoTo->net_ack += tp1->send_size;
/* CMT SFR and DAC algos */
this_sack_lowest_newack = tp1->rec.data.TSN_seq;
tp1->whoTo->saw_newack = 1;
if (tp1->snd_count < 2) {
/*
* True non-retransmited
* chunk
*/
tp1->whoTo->net_ack2 +=
tp1->send_size;
/* update RTO too? */
if (tp1->do_rtt) {
tp1->whoTo->RTO =
sctp_calculate_rto(stcb,
asoc, tp1->whoTo,
&tp1->sent_rcv_time,
sctp_align_safe_nocopy);
tp1->do_rtt = 0;
}
}
/*
* CMT: CUCv2 algorithm. From the
* cumack'd TSNs, for each TSN being
* acked for the first time, set the
* following variables for the
* corresp destination.
* new_pseudo_cumack will trigger a
* cwnd update.
* find_(rtx_)pseudo_cumack will
* trigger search for the next
* expected (rtx-)pseudo-cumack.
*/
tp1->whoTo->new_pseudo_cumack = 1;
tp1->whoTo->find_pseudo_cumack = 1;
tp1->whoTo->find_rtx_pseudo_cumack = 1;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_SACK_LOGGING_ENABLE) {
sctp_log_sack(asoc->last_acked_seq,
cum_ack,
tp1->rec.data.TSN_seq,
0,
0,
SCTP_LOG_TSN_ACKED);
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
sctp_log_cwnd(stcb, tp1->whoTo, tp1->rec.data.TSN_seq, SCTP_CWND_LOG_FROM_SACK);
}
}
if (tp1->sent == SCTP_DATAGRAM_RESEND) {
sctp_ucount_decr(asoc->sent_queue_retran_cnt);
#ifdef SCTP_AUDITING_ENABLED
sctp_audit_log(0xB3,
(asoc->sent_queue_retran_cnt & 0x000000ff));
#endif
}
if (tp1->rec.data.chunk_was_revoked) {
/* deflate the cwnd */
tp1->whoTo->cwnd -= tp1->book_size;
tp1->rec.data.chunk_was_revoked = 0;
}
tp1->sent = SCTP_DATAGRAM_ACKED;
}
} else {
break;
}
tp1 = TAILQ_NEXT(tp1, sctp_next);
}
biggest_tsn_newly_acked = biggest_tsn_acked = last_tsn;
/* always set this up to cum-ack */
asoc->this_sack_highest_gap = last_tsn;
/* Move offset up to point to gaps/dups */
offset += sizeof(struct sctp_sack_chunk);
if (((num_seg * (sizeof(struct sctp_gap_ack_block))) + sizeof(struct sctp_sack_chunk)) > sack_length) {
/* skip corrupt segments */
goto skip_segments;
}
if (num_seg > 0) {
/*
* CMT: SFR algo (and HTNA) - this_sack_highest_newack has
* to be greater than the cumack. Also reset saw_newack to 0
* for all dests.
*/
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
net->saw_newack = 0;
net->this_sack_highest_newack = last_tsn;
}
/*
* thisSackHighestGap will increase while handling NEW
* segments this_sack_highest_newack will increase while
* handling NEWLY ACKED chunks. this_sack_lowest_newack is
* used for CMT DAC algo. saw_newack will also change.
*/
sctp_handle_segments(m, &offset, stcb, asoc, ch, last_tsn,
&biggest_tsn_acked, &biggest_tsn_newly_acked, &this_sack_lowest_newack,
num_seg, &ecn_seg_sums);
if (SCTP_BASE_SYSCTL(sctp_strict_sacks)) {
/*
* validate the biggest_tsn_acked in the gap acks if
* strict adherence is wanted.
*/
if ((biggest_tsn_acked == send_s) ||
(compare_with_wrap(biggest_tsn_acked, send_s, MAX_TSN))) {
/*
* peer is either confused or we are under
* attack. We must abort.
*/
goto hopeless_peer;
}
}
}
skip_segments:
/*******************************************/
/* cancel ALL T3-send timer if accum moved */
/*******************************************/
if (SCTP_BASE_SYSCTL(sctp_cmt_on_off)) {
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
if (net->new_pseudo_cumack)
sctp_timer_stop(SCTP_TIMER_TYPE_SEND, stcb->sctp_ep,
stcb, net,
SCTP_FROM_SCTP_INDATA + SCTP_LOC_27);
}
} else {
if (accum_moved) {
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
sctp_timer_stop(SCTP_TIMER_TYPE_SEND, stcb->sctp_ep,
stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_28);
}
}
}
/********************************************/
/* drop the acked chunks from the sendqueue */
/********************************************/
asoc->last_acked_seq = cum_ack;
tp1 = TAILQ_FIRST(&asoc->sent_queue);
if (tp1 == NULL)
goto done_with_it;
do {
if (compare_with_wrap(tp1->rec.data.TSN_seq, cum_ack,
MAX_TSN)) {
break;
}
if (tp1->sent == SCTP_DATAGRAM_UNSENT) {
/* no more sent on list */
printf("Warning, tp1->sent == %d and its now acked?\n",
tp1->sent);
}
tp2 = TAILQ_NEXT(tp1, sctp_next);
TAILQ_REMOVE(&asoc->sent_queue, tp1, sctp_next);
if (tp1->pr_sctp_on) {
if (asoc->pr_sctp_cnt != 0)
asoc->pr_sctp_cnt--;
}
if ((TAILQ_FIRST(&asoc->sent_queue) == NULL) &&
(asoc->total_flight > 0)) {
#ifdef INVARIANTS
panic("Warning flight size is postive and should be 0");
#else
SCTP_PRINTF("Warning flight size incorrect should be 0 is %d\n",
asoc->total_flight);
#endif
asoc->total_flight = 0;
}
if (tp1->data) {
/* sa_ignore NO_NULL_CHK */
sctp_free_bufspace(stcb, asoc, tp1, 1);
sctp_m_freem(tp1->data);
if (PR_SCTP_BUF_ENABLED(tp1->flags)) {
asoc->sent_queue_cnt_removeable--;
}
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_SACK_LOGGING_ENABLE) {
sctp_log_sack(asoc->last_acked_seq,
cum_ack,
tp1->rec.data.TSN_seq,
0,
0,
SCTP_LOG_FREE_SENT);
}
tp1->data = NULL;
asoc->sent_queue_cnt--;
sctp_free_a_chunk(stcb, tp1);
wake_him++;
tp1 = tp2;
} while (tp1 != NULL);
done_with_it:
/* sa_ignore NO_NULL_CHK */
if ((wake_him) && (stcb->sctp_socket)) {
#if defined (__APPLE__) || defined(SCTP_SO_LOCK_TESTING)
struct socket *so;
#endif
SOCKBUF_LOCK(&stcb->sctp_socket->so_snd);
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_WAKE_LOGGING_ENABLE) {
sctp_wakeup_log(stcb, cum_ack, wake_him, SCTP_WAKESND_FROM_SACK);
}
#if defined (__APPLE__) || defined(SCTP_SO_LOCK_TESTING)
so = SCTP_INP_SO(stcb->sctp_ep);
atomic_add_int(&stcb->asoc.refcnt, 1);
SCTP_TCB_UNLOCK(stcb);
SCTP_SOCKET_LOCK(so, 1);
SCTP_TCB_LOCK(stcb);
atomic_subtract_int(&stcb->asoc.refcnt, 1);
if (stcb->asoc.state & SCTP_STATE_CLOSED_SOCKET) {
/* assoc was freed while we were unlocked */
SCTP_SOCKET_UNLOCK(so, 1);
return;
}
#endif
sctp_sowwakeup_locked(stcb->sctp_ep, stcb->sctp_socket);
#if defined (__APPLE__) || defined(SCTP_SO_LOCK_TESTING)
SCTP_SOCKET_UNLOCK(so, 1);
#endif
} else {
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_WAKE_LOGGING_ENABLE) {
sctp_wakeup_log(stcb, cum_ack, wake_him, SCTP_NOWAKE_FROM_SACK);
}
}
if (asoc->fast_retran_loss_recovery && accum_moved) {
if (compare_with_wrap(asoc->last_acked_seq,
asoc->fast_recovery_tsn, MAX_TSN) ||
asoc->last_acked_seq == asoc->fast_recovery_tsn) {
/* Setup so we will exit RFC2582 fast recovery */
will_exit_fast_recovery = 1;
}
}
/*
* Check for revoked fragments:
*
* if Previous sack - Had no frags then we can't have any revoked if
* Previous sack - Had frag's then - If we now have frags aka
* num_seg > 0 call sctp_check_for_revoked() to tell if peer revoked
* some of them. else - The peer revoked all ACKED fragments, since
* we had some before and now we have NONE.
*/
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
if (num_seg)
sctp_check_for_revoked(stcb, asoc, cum_ack, biggest_tsn_acked);
else if (asoc->saw_sack_with_frags) {
int cnt_revoked = 0;
tp1 = TAILQ_FIRST(&asoc->sent_queue);
if (tp1 != NULL) {
/* Peer revoked all dg's marked or acked */
TAILQ_FOREACH(tp1, &asoc->sent_queue, sctp_next) {
if ((tp1->sent > SCTP_DATAGRAM_RESEND) &&
(tp1->sent < SCTP_FORWARD_TSN_SKIP)) {
tp1->sent = SCTP_DATAGRAM_SENT;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_FLIGHT_LOGGING_ENABLE) {
sctp_misc_ints(SCTP_FLIGHT_LOG_UP_REVOKE,
tp1->whoTo->flight_size,
tp1->book_size,
(uintptr_t) tp1->whoTo,
tp1->rec.data.TSN_seq);
}
sctp_flight_size_increase(tp1);
sctp_total_flight_increase(stcb, tp1);
tp1->rec.data.chunk_was_revoked = 1;
/*
* To ensure that this increase in
* flightsize, which is artificial,
* does not throttle the sender, we
* also increase the cwnd
* artificially.
*/
tp1->whoTo->cwnd += tp1->book_size;
cnt_revoked++;
}
}
if (cnt_revoked) {
reneged_all = 1;
}
}
asoc->saw_sack_with_frags = 0;
}
if (num_seg)
asoc->saw_sack_with_frags = 1;
else
asoc->saw_sack_with_frags = 0;
/* JRS - Use the congestion control given in the CC module */
asoc->cc_functions.sctp_cwnd_update_after_sack(stcb, asoc, accum_moved, reneged_all, will_exit_fast_recovery);
if (TAILQ_EMPTY(&asoc->sent_queue)) {
/* nothing left in-flight */
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
/* stop all timers */
if (SCTP_BASE_SYSCTL(sctp_early_fr)) {
if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
SCTP_STAT_INCR(sctps_earlyfrstpidsck4);
sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
SCTP_FROM_SCTP_INDATA + SCTP_LOC_29);
}
}
sctp_timer_stop(SCTP_TIMER_TYPE_SEND, stcb->sctp_ep,
stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_30);
net->flight_size = 0;
net->partial_bytes_acked = 0;
}
asoc->total_flight = 0;
asoc->total_flight_count = 0;
}
/**********************************/
/* Now what about shutdown issues */
/**********************************/
if (TAILQ_EMPTY(&asoc->send_queue) && TAILQ_EMPTY(&asoc->sent_queue)) {
/* nothing left on sendqueue.. consider done */
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_RWND_ENABLE) {
sctp_log_rwnd_set(SCTP_SET_PEER_RWND_VIA_SACK,
asoc->peers_rwnd, 0, 0, a_rwnd);
}
asoc->peers_rwnd = a_rwnd;
if (asoc->peers_rwnd < stcb->sctp_ep->sctp_ep.sctp_sws_sender) {
/* SWS sender side engages */
asoc->peers_rwnd = 0;
}
/* clean up */
if ((asoc->stream_queue_cnt == 1) &&
((asoc->state & SCTP_STATE_SHUTDOWN_PENDING) ||
(asoc->state & SCTP_STATE_SHUTDOWN_RECEIVED)) &&
(asoc->locked_on_sending)
) {
struct sctp_stream_queue_pending *sp;
/*
* I may be in a state where we got all across.. but
* cannot write more due to a shutdown... we abort
* since the user did not indicate EOR in this case.
*/
sp = TAILQ_LAST(&((asoc->locked_on_sending)->outqueue),
sctp_streamhead);
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
if ((sp) && (sp->length == 0)) {
asoc->locked_on_sending = NULL;
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
if (sp->msg_is_complete) {
asoc->stream_queue_cnt--;
} else {
asoc->state |= SCTP_STATE_PARTIAL_MSG_LEFT;
asoc->stream_queue_cnt--;
}
}
}
if ((asoc->state & SCTP_STATE_SHUTDOWN_PENDING) &&
(asoc->stream_queue_cnt == 0)) {
if (asoc->state & SCTP_STATE_PARTIAL_MSG_LEFT) {
/* Need to abort here */
struct mbuf *oper;
abort_out_now:
*abort_now = 1;
/* XXX */
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) = sizeof(struct sctp_paramhdr) +
sizeof(uint32_t);
ph = mtod(oper, struct sctp_paramhdr *);
ph->param_type = htons(SCTP_CAUSE_USER_INITIATED_ABT);
ph->param_length = htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_31);
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_31;
sctp_abort_an_association(stcb->sctp_ep, stcb, SCTP_RESPONSE_TO_USER_REQ, oper, SCTP_SO_NOT_LOCKED);
return;
} else {
- Copyright updates (aka 2007) - ZONE get now also take a type cast so it does the cast like mtod does. - New macro SCTP_LIST_EMPTY, which in bsd is just LIST_EMPTY - Removal of const in some of the static hmac functions (not needed) - Store length changes to allow for new fields in auth - Auth code updated to current draft (this should be the RFC version we think). - use uint8_t instead of u_char in LOOPBACK address comparison - Some u_int32_t converted to uint32_t (in crc code) - A bug was found in the mib counts for ordered/unordered count, this was fixed (was referencing a freed mbuf). - SCTP_ASOCLOG_OF_TSNS added (code will probably disappear after my testing completes. It allows us to keep a small log on each assoc of the last 40 TSN's in/out and stream assignment. It is NOT in options and so is only good for private builds. - Some CMT changes in prep for Jana fixing his problem with reneging when CMT is enabled (Concurrent Multipath Transfer = CMT). - Some missing mib stats added. - Correction to number of open assoc's count in mib - Correction to os_bsd.h to get right sha2 macros - Add of special AUTH_04 flags so you can compile the code with the old format (in case the peer does not yet support the latest auth code). - Nonce sum was incorrectly being set in when ecn_nonce was NOT on. - LOR in listen with implicit bind found and fixed. - Moved away from using mbuf's for socket options to using just data pointers. The mbufs were used to harmonize NetBSD code since both Net and Open used this method. We have decided to move away from that and more conform to FreeBSD style (which makes more sense). - Very very nasty bug found in some of my "debug" code. The cookie_how collision case tracking had an endless loop in it if you got a second retransmission of a cookie collision case. This would lock up a CPU .. ugly.. - auth function goes to using size_t instead of int which conforms to socketapi better - Found the nasty bug that happens after 9 days of testing.. you get the data chunk, deliver it and due to the reference to a ch-> that every now and then has been deleted (depending on the postion in the mbuf) you have an invalid ch->ch.flags.. and thus you don't advance the stream sequence number.. so you block the stream permanently. The fix is to make local variables of these guys and set them up before you have any chance of trimming the mbuf. - style fix in sctp_util.h, not sure how this got bad maybe in the last patch? (aka it may not be in the real source). - Found interesting bug when using the extended snd/rcv info where we would get an error on receiving with this. Thats because it was NOT padded to the same size as the snd_rcv info. We increase (add the pad) so the two structs are the same size in sctp_uio.h - In sctp_usrreq.c one of the most common things we did for socket options was to cast the pointer and validate the size. This as been macro-ized to help make the code more readable. - in sctputil.c two things, the socketapi class found a missing flag type (the next msg is a notification) and a missing scope recovery was also fixed. Reviewed by: gnn
2007-02-12 23:24:31 +00:00
if ((SCTP_GET_STATE(asoc) == SCTP_STATE_OPEN) ||
(SCTP_GET_STATE(asoc) == SCTP_STATE_SHUTDOWN_RECEIVED)) {
SCTP_STAT_DECR_GAUGE32(sctps_currestab);
}
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
SCTP_SET_STATE(asoc, SCTP_STATE_SHUTDOWN_SENT);
SCTP_CLEAR_SUBSTATE(asoc, SCTP_STATE_SHUTDOWN_PENDING);
sctp_stop_timers_for_shutdown(stcb);
sctp_send_shutdown(stcb,
stcb->asoc.primary_destination);
sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWN,
stcb->sctp_ep, stcb, asoc->primary_destination);
sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNGUARD,
stcb->sctp_ep, stcb, asoc->primary_destination);
}
return;
} else if ((SCTP_GET_STATE(asoc) == SCTP_STATE_SHUTDOWN_RECEIVED) &&
(asoc->stream_queue_cnt == 0)) {
if (asoc->state & SCTP_STATE_PARTIAL_MSG_LEFT) {
goto abort_out_now;
}
SCTP_STAT_DECR_GAUGE32(sctps_currestab);
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
SCTP_SET_STATE(asoc, SCTP_STATE_SHUTDOWN_ACK_SENT);
SCTP_CLEAR_SUBSTATE(asoc, SCTP_STATE_SHUTDOWN_PENDING);
sctp_send_shutdown_ack(stcb,
stcb->asoc.primary_destination);
sctp_timer_start(SCTP_TIMER_TYPE_SHUTDOWNACK,
stcb->sctp_ep, stcb, asoc->primary_destination);
return;
}
}
/*
* Now here we are going to recycle net_ack for a different use...
* HEADS UP.
*/
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
net->net_ack = 0;
}
/*
* CMT DAC algorithm: If SACK DAC flag was 0, then no extra marking
* to be done. Setting this_sack_lowest_newack to the cum_ack will
* automatically ensure that.
*/
if (SCTP_BASE_SYSCTL(sctp_cmt_on_off) && SCTP_BASE_SYSCTL(sctp_cmt_use_dac) && (cmt_dac_flag == 0)) {
this_sack_lowest_newack = cum_ack;
}
if (num_seg > 0) {
sctp_strike_gap_ack_chunks(stcb, asoc, biggest_tsn_acked,
biggest_tsn_newly_acked, this_sack_lowest_newack, accum_moved);
}
/*********************************************/
/* Here we perform PR-SCTP procedures */
/* (section 4.2) */
/*********************************************/
/* C1. update advancedPeerAckPoint */
if (compare_with_wrap(cum_ack, asoc->advanced_peer_ack_point, MAX_TSN)) {
asoc->advanced_peer_ack_point = cum_ack;
}
/* C2. try to further move advancedPeerAckPoint ahead */
if ((asoc->peer_supports_prsctp) && (asoc->pr_sctp_cnt > 0)) {
struct sctp_tmit_chunk *lchk;
lchk = sctp_try_advance_peer_ack_point(stcb, asoc);
/* C3. See if we need to send a Fwd-TSN */
if (compare_with_wrap(asoc->advanced_peer_ack_point, cum_ack,
MAX_TSN)) {
/*
* ISSUE with ECN, see FWD-TSN processing for notes
* on issues that will occur when the ECN NONCE
* stuff is put into SCTP for cross checking.
*/
send_forward_tsn(stcb, asoc);
/*
* ECN Nonce: Disable Nonce Sum check when FWD TSN
* is sent and store resync tsn
*/
asoc->nonce_sum_check = 0;
asoc->nonce_resync_tsn = asoc->advanced_peer_ack_point;
if (lchk) {
/* Assure a timer is up */
sctp_timer_start(SCTP_TIMER_TYPE_SEND,
stcb->sctp_ep, stcb, lchk->whoTo);
}
}
}
/* JRS - Use the congestion control given in the CC module */
asoc->cc_functions.sctp_cwnd_update_after_fr(stcb, asoc);
/******************************************************************
* Here we do the stuff with ECN Nonce checking.
* We basically check to see if the nonce sum flag was incorrect
* or if resynchronization needs to be done. Also if we catch a
* misbehaving receiver we give him the kick.
******************************************************************/
if (asoc->ecn_nonce_allowed) {
if (asoc->nonce_sum_check) {
if (nonce_sum_flag != ((asoc->nonce_sum_expect_base + ecn_seg_sums) & SCTP_SACK_NONCE_SUM)) {
if (asoc->nonce_wait_for_ecne == 0) {
struct sctp_tmit_chunk *lchk;
lchk = TAILQ_FIRST(&asoc->send_queue);
asoc->nonce_wait_for_ecne = 1;
if (lchk) {
asoc->nonce_wait_tsn = lchk->rec.data.TSN_seq;
} else {
asoc->nonce_wait_tsn = asoc->sending_seq;
}
} else {
if (compare_with_wrap(asoc->last_acked_seq, asoc->nonce_wait_tsn, MAX_TSN) ||
(asoc->last_acked_seq == asoc->nonce_wait_tsn)) {
/*
* Misbehaving peer. We need
* to react to this guy
*/
asoc->ecn_allowed = 0;
asoc->ecn_nonce_allowed = 0;
}
}
}
} else {
/* See if Resynchronization Possible */
if (compare_with_wrap(asoc->last_acked_seq, asoc->nonce_resync_tsn, MAX_TSN)) {
asoc->nonce_sum_check = 1;
/*
* now we must calculate what the base is.
* We do this based on two things, we know
* the total's for all the segments
* gap-acked in the SACK, its stored in
* ecn_seg_sums. We also know the SACK's
* nonce sum, its in nonce_sum_flag. So we
* can build a truth table to back-calculate
* the new value of
* asoc->nonce_sum_expect_base:
*
* SACK-flag-Value Seg-Sums Base 0 0 0
* 1 0 1 0 1 1 1
* 1 0
*/
asoc->nonce_sum_expect_base = (ecn_seg_sums ^ nonce_sum_flag) & SCTP_SACK_NONCE_SUM;
}
}
}
/* Now are we exiting loss recovery ? */
if (will_exit_fast_recovery) {
/* Ok, we must exit fast recovery */
asoc->fast_retran_loss_recovery = 0;
}
if ((asoc->sat_t3_loss_recovery) &&
((compare_with_wrap(asoc->last_acked_seq, asoc->sat_t3_recovery_tsn,
MAX_TSN) ||
(asoc->last_acked_seq == asoc->sat_t3_recovery_tsn)))) {
/* end satellite t3 loss recovery */
asoc->sat_t3_loss_recovery = 0;
}
/*
* CMT Fast recovery
*/
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
if (net->will_exit_fast_recovery) {
/* Ok, we must exit fast recovery */
net->fast_retran_loss_recovery = 0;
}
}
/* Adjust and set the new rwnd value */
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_LOG_RWND_ENABLE) {
sctp_log_rwnd_set(SCTP_SET_PEER_RWND_VIA_SACK,
asoc->peers_rwnd, asoc->total_flight, (asoc->sent_queue_cnt * SCTP_BASE_SYSCTL(sctp_peer_chunk_oh)), a_rwnd);
}
asoc->peers_rwnd = sctp_sbspace_sub(a_rwnd,
(uint32_t) (asoc->total_flight + (asoc->sent_queue_cnt * SCTP_BASE_SYSCTL(sctp_peer_chunk_oh))));
if (asoc->peers_rwnd < stcb->sctp_ep->sctp_ep.sctp_sws_sender) {
/* SWS sender side engages */
asoc->peers_rwnd = 0;
}
if (asoc->peers_rwnd > old_rwnd) {
win_probe_recovery = 1;
}
/*
* Now we must setup so we have a timer up for anyone with
* outstanding data.
*/
done_once = 0;
again:
j = 0;
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
if (win_probe_recovery && (net->window_probe)) {
net->window_probe = 0;
win_probe_recovered = 1;
/*-
* Find first chunk that was used with
* window probe and clear the event. Put
* it back into the send queue as if has
* not been sent.
*/
TAILQ_FOREACH(tp1, &asoc->sent_queue, sctp_next) {
if (tp1->window_probe) {
sctp_window_probe_recovery(stcb, asoc, net, tp1);
break;
}
}
}
if (net->flight_size) {
j++;
sctp_timer_start(SCTP_TIMER_TYPE_SEND,
stcb->sctp_ep, stcb, net);
} else {
if (SCTP_OS_TIMER_PENDING(&net->rxt_timer.timer)) {
sctp_timer_stop(SCTP_TIMER_TYPE_SEND, stcb->sctp_ep,
stcb, net,
SCTP_FROM_SCTP_INDATA + SCTP_LOC_22);
}
if (SCTP_BASE_SYSCTL(sctp_early_fr)) {
if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
SCTP_STAT_INCR(sctps_earlyfrstpidsck4);
sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
SCTP_FROM_SCTP_INDATA + SCTP_LOC_23);
}
}
}
}
if ((j == 0) &&
(!TAILQ_EMPTY(&asoc->sent_queue)) &&
(asoc->sent_queue_retran_cnt == 0) &&
(win_probe_recovered == 0) &&
(done_once == 0)) {
/* huh, this should not happen */
sctp_fs_audit(asoc);
TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
net->flight_size = 0;
}
asoc->total_flight = 0;
asoc->total_flight_count = 0;
asoc->sent_queue_retran_cnt = 0;
TAILQ_FOREACH(tp1, &asoc->sent_queue, sctp_next) {
if (tp1->sent < SCTP_DATAGRAM_RESEND) {
sctp_flight_size_increase(tp1);
sctp_total_flight_increase(stcb, tp1);
} else if (tp1->sent == SCTP_DATAGRAM_RESEND) {
asoc->sent_queue_retran_cnt++;
}
}
done_once = 1;
goto again;
}
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_SACK_RWND_LOGGING_ENABLE) {
sctp_misc_ints(SCTP_SACK_RWND_UPDATE,
a_rwnd,
stcb->asoc.peers_rwnd,
stcb->asoc.total_flight,
stcb->asoc.total_output_queue_size);
}
}
void
sctp_update_acked(struct sctp_tcb *stcb, struct sctp_shutdown_chunk *cp,
struct sctp_nets *netp, int *abort_flag)
{
/* Copy cum-ack */
uint32_t cum_ack, a_rwnd;
cum_ack = ntohl(cp->cumulative_tsn_ack);
/* Arrange so a_rwnd does NOT change */
a_rwnd = stcb->asoc.peers_rwnd + stcb->asoc.total_flight;
/* Now call the express sack handling */
sctp_express_handle_sack(stcb, cum_ack, a_rwnd, 0, abort_flag);
}
static void
sctp_kick_prsctp_reorder_queue(struct sctp_tcb *stcb,
struct sctp_stream_in *strmin)
{
struct sctp_queued_to_read *ctl, *nctl;
struct sctp_association *asoc;
int tt;
asoc = &stcb->asoc;
tt = strmin->last_sequence_delivered;
/*
* First deliver anything prior to and including the stream no that
* came in
*/
ctl = TAILQ_FIRST(&strmin->inqueue);
while (ctl) {
nctl = TAILQ_NEXT(ctl, next);
if (compare_with_wrap(tt, ctl->sinfo_ssn, MAX_SEQ) ||
(tt == ctl->sinfo_ssn)) {
/* this is deliverable now */
TAILQ_REMOVE(&strmin->inqueue, ctl, next);
/* subtract pending on streams */
asoc->size_on_all_streams -= ctl->length;
sctp_ucount_decr(asoc->cnt_on_all_streams);
/* deliver it to at least the delivery-q */
if (stcb->sctp_socket) {
sctp_add_to_readq(stcb->sctp_ep, stcb,
ctl,
&stcb->sctp_socket->so_rcv, 1, SCTP_SO_NOT_LOCKED);
}
} else {
/* no more delivery now. */
break;
}
ctl = nctl;
}
/*
* now we must deliver things in queue the normal way if any are
* now ready.
*/
tt = strmin->last_sequence_delivered + 1;
ctl = TAILQ_FIRST(&strmin->inqueue);
while (ctl) {
nctl = TAILQ_NEXT(ctl, next);
if (tt == ctl->sinfo_ssn) {
/* this is deliverable now */
TAILQ_REMOVE(&strmin->inqueue, ctl, next);
/* subtract pending on streams */
asoc->size_on_all_streams -= ctl->length;
sctp_ucount_decr(asoc->cnt_on_all_streams);
/* deliver it to at least the delivery-q */
strmin->last_sequence_delivered = ctl->sinfo_ssn;
if (stcb->sctp_socket) {
sctp_add_to_readq(stcb->sctp_ep, stcb,
ctl,
&stcb->sctp_socket->so_rcv, 1, SCTP_SO_NOT_LOCKED);
}
tt = strmin->last_sequence_delivered + 1;
} else {
break;
}
ctl = nctl;
}
}
void
sctp_handle_forward_tsn(struct sctp_tcb *stcb,
struct sctp_forward_tsn_chunk *fwd, int *abort_flag, struct mbuf *m, int offset)
{
/*
* ISSUES that MUST be fixed for ECN! When we are the sender of the
* forward TSN, when the SACK comes back that acknowledges the
* FWD-TSN we must reset the NONCE sum to match correctly. This will
* get quite tricky since we may have sent more data interveneing
* and must carefully account for what the SACK says on the nonce
* and any gaps that are reported. This work will NOT be done here,
* but I note it here since it is really related to PR-SCTP and
* FWD-TSN's
*/
/* The pr-sctp fwd tsn */
/*
* here we will perform all the data receiver side steps for
* processing FwdTSN, as required in by pr-sctp draft:
*
* Assume we get FwdTSN(x):
*
* 1) update local cumTSN to x 2) try to further advance cumTSN to x +
* others we have 3) examine and update re-ordering queue on
* pr-in-streams 4) clean up re-assembly queue 5) Send a sack to
* report where we are.
*/
struct sctp_association *asoc;
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
uint32_t new_cum_tsn, gap;
unsigned int i, cnt_gone, fwd_sz, cumack_set_flag, m_size;
struct sctp_stream_in *strm;
struct sctp_tmit_chunk *chk, *at;
cumack_set_flag = 0;
asoc = &stcb->asoc;
cnt_gone = 0;
if ((fwd_sz = ntohs(fwd->ch.chunk_length)) < sizeof(struct sctp_forward_tsn_chunk)) {
SCTPDBG(SCTP_DEBUG_INDATA1,
"Bad size too small/big fwd-tsn\n");
return;
}
m_size = (stcb->asoc.mapping_array_size << 3);
/*************************************************************/
/* 1. Here we update local cumTSN and shift the bitmap array */
/*************************************************************/
new_cum_tsn = ntohl(fwd->new_cumulative_tsn);
if (compare_with_wrap(asoc->cumulative_tsn, new_cum_tsn, MAX_TSN) ||
asoc->cumulative_tsn == new_cum_tsn) {
/* Already got there ... */
return;
}
if (compare_with_wrap(new_cum_tsn, asoc->highest_tsn_inside_map,
MAX_TSN)) {
asoc->highest_tsn_inside_map = new_cum_tsn;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
sctp_log_map(0, 0, asoc->highest_tsn_inside_map, SCTP_MAP_SLIDE_RESULT);
}
}
/*
* now we know the new TSN is more advanced, let's find the actual
* gap
*/
if ((compare_with_wrap(new_cum_tsn, asoc->mapping_array_base_tsn,
MAX_TSN)) ||
(new_cum_tsn == asoc->mapping_array_base_tsn)) {
gap = new_cum_tsn - asoc->mapping_array_base_tsn;
} else {
/* try to prevent underflow here */
gap = new_cum_tsn + (MAX_TSN - asoc->mapping_array_base_tsn) + 1;
}
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
if (gap >= m_size) {
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
- Fix address add handling to clear cached routes and source addresses when peer acks the add in case the routing table changes. - Fix sctp_lower_sosend to send shutdown chunk for mbuf send case when sndlen = 0 and sinfoflag = SCTP_EOF - Fix sctp_lower_sosend for SCTP_ABORT mbuf send case with null data, So that it does not send the "null" data mbuf out and cause it to get freed twice. - Fix so auto-asconf sysctl actually effect the socket's asconf state. - Do not allow SCTP_AUTO_ASCONF option to be used on subset bound sockets. - Memset bug in sctp_output.c (arguments were reversed) submitted found and reported by Dave Jones (davej@codemonkey.org.uk). - PD-API point needs to be invoked >= not just > to conform to socket api draft this fixes sctp_indata.c in the two places need to be >=. - move M_NOTIFICATION to use M_PROTO5. - PEER_ADDR_PARAMS did not fail properly if you specify an address that is not in the association with a valid assoc_id. This meant you got or set the stcb level values instead of the destination you thought you were going to get/set. Now validate if the stcb is non-null and the net is NULL that the sa_family is set and the address is unspecified otherwise return an error. - The thread based iterator could crash if associations were freed at the exact time it was running. rework the worker thread to use the increment/decrement to prevent this and no longer use the markers that the timer based iterator uses. - Fix the memleak in sctp_add_addr_to_vrf() for the case when it is detected that ifa is already pointing to a ifn. - Fix it so that if someone is so insane that they drop the send window below the minimal add mark, they still can send. - Changed all state for associations to use mask safe macro. - During front states in association freeing in sctp_inpcbfree, we had a locking problem where locks were not in place where they should have been. - Free association calls were not testing the return value in sctp_inpcb_free() properly... others should be cast void returns where we don't care about the return value. - If a reference count is held on an assoc, even from the "force free" we should not do the actual free.. but instead let the timer free it. - When we enter sctp_input(), if the SCTP_ASOC_ABOUT_TO_BE_FREED flag is set, we must NOT process the packet but handle it like ootb. This is because while freeing an assoc we release the locks to get all the higher order locks so we can purge all the hash tables. This leaves a hole if a packet comes in just at that point. Now sctp_common_input_processing() will call the ootb code in such a case. - Change MBUF M_NOTIFICATION to use M_PROTO5 (per Sam L). This makes it so we don't have a conflict (I think this is a covertity change). We made this change AFTER some conversation and looking to make sure that M_PROTO5 does not have a problem between SCTP and the 802.11 stuff (which is the only other place its used). - Fixed lock order reversal and missing atomic protection around locked_tcb during association lookup and the 1-2-1 model. - Added debug to source address selection. - V6 output must always do checksum even for loopback. - Remove more locks around inp that are not needed for an atomically added/subtracted ref count. - slight optimization in the way we zero the array in sctp_sack_check() - It was possible to respond to a ABORT() with bad checksum with a PKT-DROP. This lead to a PKT-DROP/ABORT war. Add code to NOT send a PKT-DROP to any ABORT(). - Add an option for local logging (useful for macintosh or when you need better performing during debugging). Note no commands are here to get the log info, you must just use kgdb. - The timer code needs to be aware of if it needs to call sctp_sack_check() to slide the maps and adjust the cum-ack. This is because it may be out of sync cum-ack wise. - Added threshold managment logging. - If the user picked just the right size, that just filled the send window minus one mtu, we would enter a forever loop not copying and at the same time not blocking. Change from < to <= solves this. - Sysctl added to control the fragment interleave level which defaults to 1. - My rwnd control was not being used to control the rwnd properly (we did not add and subtract to it :-() this is now fixed so we handle small messages (1 byte etc) better to bring our rwnd down more slowly. Approved by: re@freebsd.org (Bruce Mah)
2007-08-24 00:53:53 +00:00
sctp_log_map(0, 0, asoc->highest_tsn_inside_map, SCTP_MAP_SLIDE_RESULT);
}
if ((long)gap > sctp_sbspace(&stcb->asoc, &stcb->sctp_socket->so_rcv)) {
struct mbuf *oper;
/*
* out of range (of single byte chunks in the rwnd I
* give out). This must be an attacker.
*/
*abort_flag = 1;
oper = sctp_get_mbuf_for_msg((sizeof(struct sctp_paramhdr) + 3 * sizeof(uint32_t)),
0, M_DONTWAIT, 1, MT_DATA);
if (oper) {
struct sctp_paramhdr *ph;
uint32_t *ippp;
SCTP_BUF_LEN(oper) = sizeof(struct sctp_paramhdr) +
(sizeof(uint32_t) * 3);
ph = mtod(oper, struct sctp_paramhdr *);
ph->param_type = htons(SCTP_CAUSE_PROTOCOL_VIOLATION);
ph->param_length = htons(SCTP_BUF_LEN(oper));
ippp = (uint32_t *) (ph + 1);
*ippp = htonl(SCTP_FROM_SCTP_INDATA + SCTP_LOC_33);
ippp++;
*ippp = asoc->highest_tsn_inside_map;
ippp++;
*ippp = new_cum_tsn;
}
stcb->sctp_ep->last_abort_code = SCTP_FROM_SCTP_INDATA + SCTP_LOC_33;
sctp_abort_an_association(stcb->sctp_ep, stcb,
SCTP_PEER_FAULTY, oper, SCTP_SO_NOT_LOCKED);
return;
}
SCTP_STAT_INCR(sctps_fwdtsn_map_over);
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
slide_out:
memset(stcb->asoc.mapping_array, 0, stcb->asoc.mapping_array_size);
cumack_set_flag = 1;
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
asoc->mapping_array_base_tsn = new_cum_tsn + 1;
asoc->cumulative_tsn = asoc->highest_tsn_inside_map = new_cum_tsn;
if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_MAP_LOGGING_ENABLE) {
sctp_log_map(0, 3, asoc->highest_tsn_inside_map, SCTP_MAP_SLIDE_RESULT);
}
asoc->last_echo_tsn = asoc->highest_tsn_inside_map;
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
} else {
SCTP_TCB_LOCK_ASSERT(stcb);
if ((compare_with_wrap(((uint32_t) asoc->cumulative_tsn + gap), asoc->highest_tsn_inside_map, MAX_TSN)) ||
(((uint32_t) asoc->cumulative_tsn + gap) == asoc->highest_tsn_inside_map)) {
goto slide_out;
} else {
for (i = 0; i <= gap; i++) {
SCTP_SET_TSN_PRESENT(asoc->mapping_array, i);
}
}
/*
* Now after marking all, slide thing forward but no sack
* please.
*/
sctp_sack_check(stcb, 0, 0, abort_flag);
if (*abort_flag)
return;
}
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
/*************************************************************/
/* 2. Clear up re-assembly queue */
/*************************************************************/
/*
* First service it if pd-api is up, just in case we can progress it
* forward
*/
if (asoc->fragmented_delivery_inprogress) {
sctp_service_reassembly(stcb, asoc);
}
if (!TAILQ_EMPTY(&asoc->reasmqueue)) {
/* For each one on here see if we need to toss it */
/*
* For now large messages held on the reasmqueue that are
* complete will be tossed too. We could in theory do more
* work to spin through and stop after dumping one msg aka
* seeing the start of a new msg at the head, and call the
* delivery function... to see if it can be delivered... But
* for now we just dump everything on the queue.
*/
chk = TAILQ_FIRST(&asoc->reasmqueue);
while (chk) {
at = TAILQ_NEXT(chk, sctp_next);
if (compare_with_wrap(asoc->cumulative_tsn,
chk->rec.data.TSN_seq, MAX_TSN) ||
asoc->cumulative_tsn == chk->rec.data.TSN_seq) {
/* It needs to be tossed */
TAILQ_REMOVE(&asoc->reasmqueue, chk, sctp_next);
if (compare_with_wrap(chk->rec.data.TSN_seq,
asoc->tsn_last_delivered, MAX_TSN)) {
asoc->tsn_last_delivered =
chk->rec.data.TSN_seq;
asoc->str_of_pdapi =
chk->rec.data.stream_number;
asoc->ssn_of_pdapi =
chk->rec.data.stream_seq;
asoc->fragment_flags =
chk->rec.data.rcv_flags;
}
asoc->size_on_reasm_queue -= chk->send_size;
sctp_ucount_decr(asoc->cnt_on_reasm_queue);
cnt_gone++;
/* Clear up any stream problem */
if ((chk->rec.data.rcv_flags & SCTP_DATA_UNORDERED) !=
SCTP_DATA_UNORDERED &&
(compare_with_wrap(chk->rec.data.stream_seq,
asoc->strmin[chk->rec.data.stream_number].last_sequence_delivered,
MAX_SEQ))) {
/*
* We must dump forward this streams
* sequence number if the chunk is
* not unordered that is being
* skipped. There is a chance that
* if the peer does not include the
* last fragment in its FWD-TSN we
* WILL have a problem here since
* you would have a partial chunk in
* queue that may not be
* deliverable. Also if a Partial
* delivery API as started the user
* may get a partial chunk. The next
* read returning a new chunk...
* really ugly but I see no way
* around it! Maybe a notify??
*/
asoc->strmin[chk->rec.data.stream_number].last_sequence_delivered =
chk->rec.data.stream_seq;
}
if (chk->data) {
sctp_m_freem(chk->data);
chk->data = NULL;
}
sctp_free_a_chunk(stcb, chk);
} else {
/*
* Ok we have gone beyond the end of the
* fwd-tsn's mark. Some checks...
*/
if ((asoc->fragmented_delivery_inprogress) &&
(chk->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG)) {
uint32_t str_seq;
/*
* Special case PD-API is up and
* what we fwd-tsn' over includes
* one that had the LAST_FRAG. We no
* longer need to do the PD-API.
*/
asoc->fragmented_delivery_inprogress = 0;
str_seq = (asoc->str_of_pdapi << 16) | asoc->ssn_of_pdapi;
sctp_ulp_notify(SCTP_NOTIFY_PARTIAL_DELVIERY_INDICATION,
stcb, SCTP_PARTIAL_DELIVERY_ABORTED, (void *)&str_seq, SCTP_SO_NOT_LOCKED);
}
break;
}
chk = at;
}
}
if (asoc->fragmented_delivery_inprogress) {
/*
* Ok we removed cnt_gone chunks in the PD-API queue that
* were being delivered. So now we must turn off the flag.
*/
uint32_t str_seq;
str_seq = (asoc->str_of_pdapi << 16) | asoc->ssn_of_pdapi;
sctp_ulp_notify(SCTP_NOTIFY_PARTIAL_DELVIERY_INDICATION,
stcb, SCTP_PARTIAL_DELIVERY_ABORTED, (void *)&str_seq, SCTP_SO_NOT_LOCKED);
asoc->fragmented_delivery_inprogress = 0;
}
/*************************************************************/
/* 3. Update the PR-stream re-ordering queues */
/*************************************************************/
fwd_sz -= sizeof(*fwd);
if (m && fwd_sz) {
/* New method. */
- fixed autclose to not allow setting on 1-2-1 model. - bounded cookie-life to 1 second minimum in socket option set. - Delayed_ack_time becomes delayed_ack per new socket api document. - Improve port number selection, we now use low/high bounds and no chance of a endless loop. Only one call to random per bind as well. - fixes so set_peer_primary pre-screens addresses to be valid to this host. - maxseg did not allow setting on an assoc basis. We needed to thus track and use an association value instead of a inp value. - Fixed ep get of HB status to report back properly. - use settings flag to tell if assoc level hb is on off not the timer.. since the timer may still run if unconf address are present. - check for crazy ENABLE/DISABLE conditions. - set and get of pmtud (fixed path mtu) not always taking into account ovh. - Getting PMTU info on stcb only needs to return PMTUD_ENABLED if any net is doing PMTU discovery. - Panic or warning fixed to not do so when a valid ip frag is taking place. - sndrcvinfo appearing in both inp and stcb was full size, instead of the non-pad version. This saves about 92 bytes from each struct by carefully converting to use the smaller version. - one-2-one model get(maxseg) would always get ep value, never the tcb's value. - The delayed ack time could be under a tick, this fixes so it bounds it to at least 1 tick for platforms whos tick is more than a ms. - Fragment interleave level set to wrong default value. - Fragment interleave could not set level 0. - Defered stream reset was broken due to a guard check and ntohl issue. - Found two lock order reversals and fixed. - Tighten up address checking, if the user gives an address the sa_len had better be set properly. - Get asoc by assoc-id would return a locked tcb when it was asked not to if the tcb was in the restart hash. - sysctl to dig down and get more association details Reviewed by: gnn
2007-05-28 11:17:24 +00:00
unsigned int num_str;
struct sctp_strseq *stseq, strseqbuf;
offset += sizeof(*fwd);
num_str = fwd_sz / sizeof(struct sctp_strseq);
for (i = 0; i < num_str; i++) {
uint16_t st;
stseq = (struct sctp_strseq *)sctp_m_getptr(m, offset,
sizeof(struct sctp_strseq),
(uint8_t *) & strseqbuf);
offset += sizeof(struct sctp_strseq);
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
if (stseq == NULL) {
break;
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
}
/* Convert */
- send call has a reference to uio->uio_resid in the recent send code, but uio may be NULL on sendfile calls. Change to use sndlen variable. - EMSGSIZE is not being returned in non-blocking mode and needs a small tweak to look if the msg would ever fit when returning EWOULDBLOCK. - FWD-TSN has a bug in stream processing which could cause a panic. This is a follow on to the codenomicon fix. - PDAPI level 1 and 2 do not work unless the reader gets his returned buffer full. Fix so we can break out when at level 1 or 2. - Fix fast-handoff features to copy across properly on accepted sockets - Fix sctp_peeloff() system call when no true system call exists to screen arguments for errors. In cases where a real system call exists the system call itself does this. - Fix raddr leak in recent add-ip code change for bundled asconfs (even when non-bundled asconfs are received) - Make sure ipi_addr lock is held when walking global addr list. Need to change this lock type to a rwlock(). - Add don't wake flag on both input and output when the socket is closing. - When deleting an address verify the interface is correct before allowing the delete to process. This protects panda and unnumbered. - Clean up old sysctl stuff and get rid of the old Open/Net BSD structures. - Add a function to watch the ranges in the sysctl sets. - When appending in the reassembly queue, validate that the assoc has not gone to about to be freed. If so (in the middle) abort out. Note this especially effects MAC I think due to the lock/unlock they do (or with LOCK testing in place). - Netstat patch to get rid of warnings. - Make sure that no data gets queued to inactive/unconfirmed destinations. This especially effect CMT but also makes a impact on regular SCTP as well. - During init collision when we detect seq number out of sync we need to treat it like Case C and discard the cookie (no invarient needed here). - Atomic access to the random store. - When we declare a vtag good, we need to shove it into the time wait hash to prevent further use. When the tag is put into the assoc hash, we need to remove it from the twait hash (where it will surely be). This prevents duplicate tag assignments. - Move decr-ref count to better protect sysctl out of data. - ltrace error corrections in sctp6_usrreq.c - Add hook for interface up/down to be sent to us. - Make sysctl() exported structures independent of processor architecture. - Fix route and src addr cache clearing for delete address case. - Make sure address marked SCTP_DEL_IP_ADDRESS is never selected as src addr. - in icmp handling fixed so we actually look at the icmp codes to figure out what to do. - Modified mobility code. Reception of DELETE IP ADDRESS for a primary destination and SET PRIMARY for a new primary destination is used for retransmission trigger to the new primary destination. Also, in this case, destination of chunks in send_queue are changed to the new primary destination. - Fix so that we disallow sending by mbuf to ever have EEOR mode set upon it. Approved by: re@freebsd.org (B Mah)
2007-09-08 17:48:46 +00:00
st = ntohs(stseq->stream);
stseq->stream = st;
st = ntohs(stseq->sequence);
stseq->sequence = st;
/* now process */
- send call has a reference to uio->uio_resid in the recent send code, but uio may be NULL on sendfile calls. Change to use sndlen variable. - EMSGSIZE is not being returned in non-blocking mode and needs a small tweak to look if the msg would ever fit when returning EWOULDBLOCK. - FWD-TSN has a bug in stream processing which could cause a panic. This is a follow on to the codenomicon fix. - PDAPI level 1 and 2 do not work unless the reader gets his returned buffer full. Fix so we can break out when at level 1 or 2. - Fix fast-handoff features to copy across properly on accepted sockets - Fix sctp_peeloff() system call when no true system call exists to screen arguments for errors. In cases where a real system call exists the system call itself does this. - Fix raddr leak in recent add-ip code change for bundled asconfs (even when non-bundled asconfs are received) - Make sure ipi_addr lock is held when walking global addr list. Need to change this lock type to a rwlock(). - Add don't wake flag on both input and output when the socket is closing. - When deleting an address verify the interface is correct before allowing the delete to process. This protects panda and unnumbered. - Clean up old sysctl stuff and get rid of the old Open/Net BSD structures. - Add a function to watch the ranges in the sysctl sets. - When appending in the reassembly queue, validate that the assoc has not gone to about to be freed. If so (in the middle) abort out. Note this especially effects MAC I think due to the lock/unlock they do (or with LOCK testing in place). - Netstat patch to get rid of warnings. - Make sure that no data gets queued to inactive/unconfirmed destinations. This especially effect CMT but also makes a impact on regular SCTP as well. - During init collision when we detect seq number out of sync we need to treat it like Case C and discard the cookie (no invarient needed here). - Atomic access to the random store. - When we declare a vtag good, we need to shove it into the time wait hash to prevent further use. When the tag is put into the assoc hash, we need to remove it from the twait hash (where it will surely be). This prevents duplicate tag assignments. - Move decr-ref count to better protect sysctl out of data. - ltrace error corrections in sctp6_usrreq.c - Add hook for interface up/down to be sent to us. - Make sysctl() exported structures independent of processor architecture. - Fix route and src addr cache clearing for delete address case. - Make sure address marked SCTP_DEL_IP_ADDRESS is never selected as src addr. - in icmp handling fixed so we actually look at the icmp codes to figure out what to do. - Modified mobility code. Reception of DELETE IP ADDRESS for a primary destination and SET PRIMARY for a new primary destination is used for retransmission trigger to the new primary destination. Also, in this case, destination of chunks in send_queue are changed to the new primary destination. - Fix so that we disallow sending by mbuf to ever have EEOR mode set upon it. Approved by: re@freebsd.org (B Mah)
2007-09-08 17:48:46 +00:00
if (stseq->stream >= asoc->streamincnt) {
- During shutdown pending, when the last sack came in and the last message on the send stream was "null" but still there, a state we allow, we could get hung and not clean it up and wait for the shutdown guard timer to clear the association without a graceful close. Fix this so that that we properly clean up. - Added support for Multiple ASCONF per new RFC. We only (so far) accept input of these and cannot yet generate a multi-asconf. - Sysctl'd support for experimental Fast Handover feature. Always disabled unless sysctl or socket option changes to enable. - Error case in add-ip where the peer supports AUTH and ADD-IP but does NOT require AUTH of ASCONF/ASCONF-ACK. We need to ABORT in this case. - According to the Kyoto summit of socket api developers (Solaris, Linux, BSD). We need to have: o non-eeor mode messages be atomic - Fixed o Allow implicit setup of an assoc in 1-2-1 model if using the sctp_**() send calls - Fixed o Get rid of HAVE_XXX declarations - Done o add a sctp_pr_policy in hole in sndrcvinfo structure - Done o add a PR_SCTP_POLICY_VALID type flag - yet to-do in a future patch! - Optimize sctp6 calls to reuse code in sctp_usrreq. Also optimize when we close sending out the data and disabling Nagle. - Change key concatenation order to match the auth RFC - When sending OOTB shutdown_complete always do csum. - Don't send PKT-DROP to a PKT-DROP - For abort chunks just always checksums same for shutdown-complete. - inpcb_free front state had a bug where in queue data could wedge an assoc. We need to just abandon ones in front states (free_assoc). - If a peer sends us a 64k abort, we would try to assemble a response packet which may be larger than 64k. This then would be dropped by IP. Instead make a "minimum" size for us 64k-2k (we want at least 2k for our initack). If we receive such an init discard it early without all the processing. - When we peel off we must increment the tcb ref count to keep it from being freed from underneath us. - handling fwd-tsn had bugs that caused memory overwrites when given faulty data, fixed so can't happen and we also stop at the first bad stream no. - Fixed so comm-up generates the adaption indication. - peeloff did not get the hmac params copied. - fix it so we lock the addr list when doing src-addr selection (in future we need to use a multi-reader/one writer lock here) - During lowlevel output, we could end up with a _l_addr set to null if the iterator is calling the output routine. This means we would possibly crash when we gather the MTU info. Fix so we only do the gather where we have a src address cached. - we need to be sure to set abort flag on conn state when we receive an abort. - peeloff could leak a socket. Moved code so the close will find the socket if the peeloff fails (uipc_syscalls.c) Approved by: re@freebsd.org(Ken Smith)
2007-08-27 05:19:48 +00:00
/* screwed up streams, stop! */
break;
}
- send call has a reference to uio->uio_resid in the recent send code, but uio may be NULL on sendfile calls. Change to use sndlen variable. - EMSGSIZE is not being returned in non-blocking mode and needs a small tweak to look if the msg would ever fit when returning EWOULDBLOCK. - FWD-TSN has a bug in stream processing which could cause a panic. This is a follow on to the codenomicon fix. - PDAPI level 1 and 2 do not work unless the reader gets his returned buffer full. Fix so we can break out when at level 1 or 2. - Fix fast-handoff features to copy across properly on accepted sockets - Fix sctp_peeloff() system call when no true system call exists to screen arguments for errors. In cases where a real system call exists the system call itself does this. - Fix raddr leak in recent add-ip code change for bundled asconfs (even when non-bundled asconfs are received) - Make sure ipi_addr lock is held when walking global addr list. Need to change this lock type to a rwlock(). - Add don't wake flag on both input and output when the socket is closing. - When deleting an address verify the interface is correct before allowing the delete to process. This protects panda and unnumbered. - Clean up old sysctl stuff and get rid of the old Open/Net BSD structures. - Add a function to watch the ranges in the sysctl sets. - When appending in the reassembly queue, validate that the assoc has not gone to about to be freed. If so (in the middle) abort out. Note this especially effects MAC I think due to the lock/unlock they do (or with LOCK testing in place). - Netstat patch to get rid of warnings. - Make sure that no data gets queued to inactive/unconfirmed destinations. This especially effect CMT but also makes a impact on regular SCTP as well. - During init collision when we detect seq number out of sync we need to treat it like Case C and discard the cookie (no invarient needed here). - Atomic access to the random store. - When we declare a vtag good, we need to shove it into the time wait hash to prevent further use. When the tag is put into the assoc hash, we need to remove it from the twait hash (where it will surely be). This prevents duplicate tag assignments. - Move decr-ref count to better protect sysctl out of data. - ltrace error corrections in sctp6_usrreq.c - Add hook for interface up/down to be sent to us. - Make sysctl() exported structures independent of processor architecture. - Fix route and src addr cache clearing for delete address case. - Make sure address marked SCTP_DEL_IP_ADDRESS is never selected as src addr. - in icmp handling fixed so we actually look at the icmp codes to figure out what to do. - Modified mobility code. Reception of DELETE IP ADDRESS for a primary destination and SET PRIMARY for a new primary destination is used for retransmission trigger to the new primary destination. Also, in this case, destination of chunks in send_queue are changed to the new primary destination. - Fix so that we disallow sending by mbuf to ever have EEOR mode set upon it. Approved by: re@freebsd.org (B Mah)
2007-09-08 17:48:46 +00:00
strm = &asoc->strmin[stseq->stream];
if (compare_with_wrap(stseq->sequence,
strm->last_sequence_delivered, MAX_SEQ)) {
/* Update the sequence number */
strm->last_sequence_delivered =
- send call has a reference to uio->uio_resid in the recent send code, but uio may be NULL on sendfile calls. Change to use sndlen variable. - EMSGSIZE is not being returned in non-blocking mode and needs a small tweak to look if the msg would ever fit when returning EWOULDBLOCK. - FWD-TSN has a bug in stream processing which could cause a panic. This is a follow on to the codenomicon fix. - PDAPI level 1 and 2 do not work unless the reader gets his returned buffer full. Fix so we can break out when at level 1 or 2. - Fix fast-handoff features to copy across properly on accepted sockets - Fix sctp_peeloff() system call when no true system call exists to screen arguments for errors. In cases where a real system call exists the system call itself does this. - Fix raddr leak in recent add-ip code change for bundled asconfs (even when non-bundled asconfs are received) - Make sure ipi_addr lock is held when walking global addr list. Need to change this lock type to a rwlock(). - Add don't wake flag on both input and output when the socket is closing. - When deleting an address verify the interface is correct before allowing the delete to process. This protects panda and unnumbered. - Clean up old sysctl stuff and get rid of the old Open/Net BSD structures. - Add a function to watch the ranges in the sysctl sets. - When appending in the reassembly queue, validate that the assoc has not gone to about to be freed. If so (in the middle) abort out. Note this especially effects MAC I think due to the lock/unlock they do (or with LOCK testing in place). - Netstat patch to get rid of warnings. - Make sure that no data gets queued to inactive/unconfirmed destinations. This especially effect CMT but also makes a impact on regular SCTP as well. - During init collision when we detect seq number out of sync we need to treat it like Case C and discard the cookie (no invarient needed here). - Atomic access to the random store. - When we declare a vtag good, we need to shove it into the time wait hash to prevent further use. When the tag is put into the assoc hash, we need to remove it from the twait hash (where it will surely be). This prevents duplicate tag assignments. - Move decr-ref count to better protect sysctl out of data. - ltrace error corrections in sctp6_usrreq.c - Add hook for interface up/down to be sent to us. - Make sysctl() exported structures independent of processor architecture. - Fix route and src addr cache clearing for delete address case. - Make sure address marked SCTP_DEL_IP_ADDRESS is never selected as src addr. - in icmp handling fixed so we actually look at the icmp codes to figure out what to do. - Modified mobility code. Reception of DELETE IP ADDRESS for a primary destination and SET PRIMARY for a new primary destination is used for retransmission trigger to the new primary destination. Also, in this case, destination of chunks in send_queue are changed to the new primary destination. - Fix so that we disallow sending by mbuf to ever have EEOR mode set upon it. Approved by: re@freebsd.org (B Mah)
2007-09-08 17:48:46 +00:00
stseq->sequence;
}
/* now kick the stream the new way */
/* sa_ignore NO_NULL_CHK */
sctp_kick_prsctp_reorder_queue(stcb, strm);
}
}
if (TAILQ_FIRST(&asoc->reasmqueue)) {
/* now lets kick out and check for more fragmented delivery */
/* sa_ignore NO_NULL_CHK */
sctp_deliver_reasm_check(stcb, &stcb->asoc);
}
}