Re: [PATCH] Fix bug in at2 7bit encoding with offset

Hi,

On Thu, 11 Dec 2008 17:04:09 +0100
Alexander Malysh <[email protected]> wrote:

> could you please explain in more detail why this patch is needed?
> Unfortunately I'm unable to reproduce or understand the issue...
> some examples maybe?

Since the PDU/7bit -bug seems not to be so obvious even after 5 months,
some explaining with example could be in order.

Lets send a message with the lenght of 161 + n*8 to number 0123456789
using the HTTP API: curl 'http://localhost:13013/cgi-bin/sendsms?
username=*&password=*&to=0123456789&text=abcdefghijklmnopqrstuvwxyzABC
DEFGHIJKLMNOPQRSTUVWXYZ0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJK
LMNOPQRSTUVWXYZ0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRS'


Current Kannel version does this according to the bearerbox.log:

# part 2
AT+CMGS=33^M
0051000A8110325476980000A7170500030202028845E31199542E994DE7131A954E

# part 1
AT+CMGS=153^M
0051000A8110325476980000A7A0050003020201C2E231B96C3EA3D3EA35BB
ED7EC3E3F239BD6EBFE3F3FAA070482C1A8FC8A472C96C3A9FD0A8744AAD5A
AFD8AC161693CD6835DB0D970B8BC7E4B2F98C4EABD7ECB6FB0D8FCBE7F4BA
FD8ECFEB83C221B1683C2293CA25B3E97C42A3D229B56ABD62B35A584C36A3
D56C375C2E2C1E93CBE6333AAD5EB3DBEE373C2E9FD3EBF63B3EAF0F0A87


The first part (that is actually the big latter one, why does Kannel
sends those messages in 'wrong' order?) is ok, but the second part is
one octet too short. It should be 34 octets but it's only 33.

To be sure, lets decode the data manually (by blindly trusting the
specs here: http://dreamfabric.com/sms/). 

# part 1
00 - use default SMSC
51 - SMS-SUBMIT: TP-UDHI (User data has header) |
     TP-VPF (validity period is in realtive format == one octet) |
     TP-MTI (SMS-SUBMIT)
00 - auto msg ref number
0A - length of the phone number (10 digits)
81 - phone address type (unknown type of number, telephone numbering
plan) 1032547698 - the phone number (read by swapping the nibbles, 10
32 54.. -> 01 23 45..)
00 - short message type 0
00 - data is coded in 7 bits (important!)
A7 - validity period is 24 hours
A0 - length of the msg in septets == 160 7bit chars. That would be
     exactly 140 (160*7/8) bytes.
050004020201...
   - remaining 140 bytes of data, everything is ok. The
     data has some kind of another header within the message, i think
     it at least defines the timestamp to the message. However, that
     part is in the data section in PDU, so we are not interested.


# part 2
# Most of the header is the same than in part 1:
00 51 00 0A 81 1032547698 00 00 A7

# Now the interesting part:
17 - the data length, 23 septets == 161 bits. That needs 21 bytes ...
0500030202028845E31199542E994DE7131A954E
   - ... but the rest of the data is only 20 bytes (40 nibbles), so
     we just lost one last bit of the data.


So now to my patch. destRemain was calculated incorrectly, because it
didn't take into account the offset value (in bits) in function
Octstr *at2_encode7bituncompressed(Octstr *source, int offset):

-int destRemain = (int)ceil ((octstr_len(source) * 7.0) / 8.0);
+int destRemain = (int)ceil ((octstr_len(source) * 7.0 + offset) / 8.0);


With my patch, the PDU is generated one byte longer with the second
part, and the data now has the last bit there correctly:

# part 2, with patch
AT+CMGS=34^M
0051000A8140658790740000A7170500032C02028845E31199542E994DE7131A954E01

(the timestamp is different, that's why some bytes in the beginning
are different to the old one)


The patch has been in production use since July 2008, and without this
Kannel makes Nokia 30 GSM Modem to crash. Kannel restart was required
to recover from that state.


-- 
Riku Palomäki
Hypercube Software