Recent changes to bugs

how to enable seeding option in code

Anonymous — Mon, 10 Aug 2009 13:16:22 -0000

I am trying to write a torrent client using libbt .It works fine and it seeding at 2kBs.If I want to seed more data what option should i enable?

help please

Segfault

Anonymous — Mon, 16 Jul 2007 18:58:44 -0000

btget 1.05 is getting a Segmentation Fault running under NetBSD 2.0.2 on sparc64

ulimit -n 1024 is needed to avoid running out of files. Some blocks download ok but then a segfault kills the program.

Contact steven_grunza@ieee.org for more details if necessary.

SYSDIE when block is single dir

Leon Kos — Tue, 27 Dec 2005 23:52:50 -0000

When single block contains only directory description
(without any file) btget fails check and dumps core.

Misallocation in `recv_peermsg' function resulting in crash

Anonymous — Tue, 06 Sep 2005 00:00:08 -0000

Dear developers,

I think i've found a small bug in the `recv_peermsg'
function in peer.c (1.04 release).

If i understand the code correctly a message is send
with a 4 byte length in front of a block of data. If
the data length was less than 80 a stack reserved
memory block is used to avoid over-malloc'ing for small
pieces. This block was declared 80 bytes long... so it
would overflow if messages of aprox. 79 bytes were sent
for the length bytes were not accounted for in the
allocation.

To fix, change the allocation "char msg[80]" to at
least "char msg[84]" or more around line 474 of peer.c

With this small fix the repeated crashes that i
experienced are gone.

Hope to be of help,

Reinoud Zandijk,
reinoud@netbsd.org

Check return codes everywhere

Markus Elfring — Sun, 17 Jul 2005 15:41:35 -0000

Some checks for return codes are missing.

Example:
Please add more error handling for null pointers from
"malloc" in the function "net_accept".
http://cvs.sourceforge.net/viewcvs.py/libbt/libbt/src/net.c?rev=1.8&view=markup

Assertion failures in peer.c

Stephen Henson — Wed, 12 Jan 2005 19:54:45 -0000

I'm getting repeated assertion failures in peer.c in the
function recv_peermsg(). This seems to be a client
sending data libbt doesn't like.

One problem is the code:

len = ntohl(nbo_len);
if (len <= 80) {
nmsg = msg;
} else if (len <= MAXMESSAGE) {
nmsg = btmalloc(len+sizeof(int32_t));
} else { /* Too big a packet, kill the peer. */
peer->ios.error = BTERR_LARGEPACKET;
return -1;
}
#if 0
printf("A/%d: looking for %d bytes %d available (addr
%d)\n", peer->ios.fd, len+4, kStream_iqlen( &peer-
>ios), kStream_in_addr(&peer->ios));
#endif
if (len == 0) {
/* keep alive */
err = kStream_fread( &peer->ios, (char *)
&nbo_len, sizeof(nbo_len));
DIE_UNLESS(err == sizeof(nbo_len));
return 0;
}
DIE_UNLESS(len <= MAXMESSAGE && len >= 0);

when len < 0.

Also later if there is another message waiting with
invalid len:

if (res == 0) {
/* check if there is another message waiting */
err = kStream_fpeek( &peer->ios, (char *)
&nbo_len, sizeof(nbo_len));
if (err == sizeof(nbo_len)) {
int tlen;
tlen = ntohl(nbo_len) + sizeof(nbo_len);
#if 0
printf("B/looking for %d bytes %d available
(addr %d)\n", tlen, kStream_iqlen( &peer->ios),
kStream_in_addr(&peer->ios));
#endif
DIE_UNLESS(tlen <= MAXMESSAGE && tlen
>= 0);
if (kStream_iqlen( &peer->ios) >= tlen) res
= 1;
}
}

bug in btString_cmp()

Stephen Henson — Sun, 12 Dec 2004 16:44:46 -0000

In btString_cmp() when two strings are of unequal
length but otherwise match the longer of the two should
be considered the greater.

Swapping the along=1 and along=-1 lines fixes this.

This is a problem because in some torrents it will reorder
a Dictionary type and produce an invalid hash.

Fixes for big endian.

Anonymous — Wed, 08 Dec 2004 19:12:07 -0000

I've been attempting to get libbt going on an nslu2
device which uses an XScale processor in big endian mode.

A couple of problems were apparent.

1. In benc.c benc_put_int() the call to bts_printf()
uses an int format for the 64 bit field 'ival'

2. There are several places in protocol.c where
misaligned 32 bit access can occur. For example:

int block = ntohl (*(int *)(msg + 1));

I've attached a quick fix to these. You might want
something cleaner for #2.

returned parameter variable size too small on Sun

Jim Burke — Fri, 24 Sep 2004 22:55:28 -0000

The call to getsockop() uses: char error;

This fails on Sun Solaris SPARC. getsockopt returns -1.
errno is set to 150, EINVARG.
It should be: int error;
to reserve more space. Or better yet: socklen_t error;

There are a few changes needed to get libbt to build
on Sun Solaris SPARC.
1. Use GNU tools, not the Sun tools.
2. Added some include files to the source and
autoconfig files.
3. Fix the getsockopt() error bug noted above.

Since the lib is targeted at Linux I haven't detailed them.
But anyone desiring the Sun port info can reach me at
jburke(at)sdf.lonestar.org

integer format specifier used for long long type

Jim Burke — Mon, 06 Sep 2004 16:42:10 -0000

In benc.c, routine benc_put_int():
int benc_put_int( struct btStream *bts, btInteger *i) {
int res;

res = bts_printf(bts, "i%de", i->ival);
return res;
}

The format specifier %d is used for ival, but ival is a
64 bit 'long long'. It should be %lld.

I think it happens to work ok on little-endian
machines, but it didn't work on my big-endian SPARC
machine.
The big endian 64-bit value was stored as
00 00 00 00 12 34 56 78
and the 32-bit format specifier read from the start at
the first 00.
On an little endian machine it would be stored as:
78 56 34 12 00 00 00 00, and could be read correctly as
a 32-bit value.