linux.git/net/sctp, branch v2.6.16

[SCTP]: Fix 'fast retransmit' to send a TSN only once.

2006-02-03T00:57:31+00:00

SCTP used to "fast retransmit" a TSN every time we hit the number
of missing reports for the TSN.  However the Implementers Guide
specifies that we should only "fast retransmit" a given TSN once.
Subsequent retransmits should be timeouts only. Also change the
number of missing reports to 3 as per the latest IG(similar to TCP).

Signed-off-by: Vlad Yasevich 
Signed-off-by: Sridhar Samudrala 
Signed-off-by: David S. Miller

[SCTP]: heartbeats exceed maximum retransmssion limit

2006-01-31T00:00:40+00:00

The number of HEARTBEAT chunks that an association may transmit is
limited by Association.Max.Retrans count; however, the code allows
us to send one extra heartbeat.

This patch limits the number of heartbeats to the maximum count.

Signed-off-by: Vlad Yasevich 
Signed-off-by: Sridhar Samudrala 
Signed-off-by: David S. Miller

[SCTP]: correct the number of INIT retransmissions

2006-01-30T23:59:54+00:00

We currently count the initial INIT/COOKIE_ECHO chunk toward the
retransmit count and thus sends a total of sctp_max_retrans_init chunks.
The correct behavior is to retransmit the chunk sctp_max_retrans_init in
addition to sending the original.

Signed-off-by: Vlad Yasevich 
Signed-off-by: Sridhar Samudrala 
Signed-off-by: David S. Miller

[SCTP]: Fix sctp_rcv_ootb() to handle the last chunk of a packet correctly.

2006-01-17T19:57:09+00:00

Signed-off-by: Tsutomu Fujii 
Signed-off-by: Sridhar Samudrala

[SCTP]: Fix couple of races between sctp_peeloff() and sctp_rcv().

2006-01-17T19:56:26+00:00

Validate and update the sk in sctp_rcv() to avoid the race where an
assoc/ep could move to a different socket after we get the sk, but before
the skb is added to the backlog.

Also migrate the skb's in backlog queue to new sk when doing a peeloff.

Signed-off-by: Sridhar Samudrala

[SCTP]: Fix machine check/connection hang on IA64.

2006-01-17T19:55:57+00:00

sctp_unpack_cookie used an on-stack array called digest as a result/out
parameter in the call to crypto_hmac. However, hmac code
(crypto_hmac_final)
assumes that the 'out' argument is in virtual memory (identity mapped
region)
and can use virt_to_page call on it.  This does not work with the on-stack
declared digest.  The problems observed so far have been:
 a) incorrect hmac digest
 b) machine check and hardware reset.

Solution is to define the digest in an identity mapped region by
kmalloc'ing
it.  We can do this once as part of the endpoint structure and re-use it
when
verifying the SCTP cookie.

Signed-off-by: Vlad Yasevich 
Signed-off-by: Sridhar Samudrala

[SCTP]: Fix bad sysctl formatting of SCTP timeout values on 64-bit m/cs.

2006-01-17T19:55:17+00:00

Change all the structure members that hold jiffies to be of type
unsigned long.  This also corrects bad sysctl formating on 64 bit
architectures.

Signed-off-by: Vlad Yasevich 
Signed-off-by: Sridhar Samudrala

[SCTP]: Fix sctp_assoc_seq_show() panics on big-endian systems.

2006-01-17T19:54:06+00:00

This patch corrects the panic by casting the argument to the
pointer of correct size.  On big-endian systems we ended up loading
only 32 bits of data because we are treating the pointer as an int*.
By treating this pointer as loff_t*, we'll load the full 64 bits
and then let regular integer demotion take place which will give us
the correct value.

Signed-off-by: Vlad Yaseivch 
Signed-off-by: Sridhar Samudrala

[SCTP]: sctp doesn't show all associations/endpoints in /proc

2006-01-17T19:53:06+00:00

When creating a very large number of associations (and endpoints),
/proc/assocs and /proc/eps will not show all of them.  As a result
netstat will not show all of the either.  This is particularly evident
when creating 1000+ associations (or endpoints).  As an example with
1500 tcp style associations over loopback, netstat showed 1420 on my
system instead of 3000.

The reason for this is that the seq_operations start method is invoked
multiple times bacause of the amount of data that is provided.  The
start method always increments the position parameter and since we use
the position as the hash bucket id, we end up skipping hash buckets.

This patch corrects this situation and get's rid of the silly hash-1
decrement.

Signed-off-by: Vlad Yasevich 
Signed-off-by: Sridhar Samudrala

[SCTP]: Fix sctp_cookie alignment in the packet.

2006-01-17T19:52:12+00:00

On 64 bit architectures, sctp_cookie sent as part of INIT-ACK is not
aligned on a 64 bit boundry and thus causes unaligned access exceptions.

The layout of the cookie prameter is this:
|<----- Parameter Header --------------------|<--- Cookie DATA --------
-----------------------------------------------------------------------
| param type (16 bits) | param len (16 bits) | sig [32 bytes] | cookie..
-----------------------------------------------------------------------

The cookie data portion contains 64 bit values on 64 bit architechtures
(timeval) that fall on a 32 bit alignment boundry when used as part of
the on-wire format, but align correctly when used in internal
structures.  This patch explicitely pads the on-wire format so that
it is properly aligned.

Signed-off-by: Vlad Yasevich 
Signed-off-by: Sridhar Samudrala