Solaris 10 TCP bug?

I have a palm M505 with a wireless lan jacket. The networking was fine with solaris 7,8 and 9 but fails under Solaris 10.
Every now and again the tcp packets seem to get stuck in a loop. see snoop log. any ideas (It's definately related to Sol 10 as the same operation still works on earlier releases.
using build 69 at moment.
This repeats forever!
Freshfields is the Sol10 box
freshfields -> fresh-11 TCP D=43481 S=3128 Ack=2660256227 Seq=291580582 Len=0 Win=49312
fresh-11 -> freshfields TCP D=3128 S=43481 Ack=291580658 Seq=2660256227 Len=0 Win=804
freshfields -> fresh-11 TCP D=43481 S=3128 Ack=2660256227 Seq=291580582 Len=0 Win=49312
fresh-11 -> freshfields TCP D=3128 S=43481 Ack=291580658 Seq=2660256227 Len=0 Win=804

Similar Messages

Solaris TCP settings

Hi all,
I remember there was a recommendation for the Solaris TCP settings like
tcp_time_wait_interval and tcp_conn_req_max_q. I can't seem to find it now.
Is it still relevant? If so where? I have heard that it is nolonger required
for WLS5.1. Is this correct?
thanks
-lucio

Lucio,
If you are using Solaris - then I strongly recommend this document -
http://www.rvs.uni-hannover.de/people/voeckler/tune/EN/tune.html in
particular section 5.1
regards,
-Rob
Robert Castaneda [email protected]
CustomWare http://www.customware.com
"Lucio Piccoli" <[email protected]> wrote in message
news:[email protected]..
Hi all,
I remember there was a recommendation for the Solaris TCP settings like
tcp_time_wait_interval and tcp_conn_req_max_q. I can't seem to find itnow.
Is it still relevant? If so where? I have heard that it is nolongerrequired
for WLS5.1. Is this correct?
thanks
-lucio

Solaris TCP Tuning

If you are running Weblogic on Solaris you'll need to change some of the
default TCP configuration parameters.
Sun's default TCP settings are not correct for a web server.
We noticed the server was stalling during certain peak periods. netstat
showed that many sockets opened to port 80 were in CLOSE_WAIT or
FIN_WAIT_2.
The following url is a great resource for tuning Solaris TCP/IP
http://www.rvs.uni-hannover.de/people/voeckler/tune/EN/tune.html
Changing the following TCP params solved our stalling problem. Also make
sure you up the ulimit as stated in the weblogic doc.
# to display the TCP values.
ndd /dev/tcp tcp_close_wait_interval tcp_fin_wait_2_flush_interval
tcp_keepalive_interval
# set tcp_close_wait_interval from 2400000 to 60000
ndd -set /dev/tcp tcp_close_wait_interval 60000
# set tcp_fin_wait_2_flush_interval from 675000 to 67500
ndd -set /dev/tcp tcp_fin_wait_2_flush_interval 67500
#set tcp_keepalive_interval from 7200000 to 300000
ndd -set /dev/tcp tcp_keepalive_interval 300000

Hi Victor,
We are facing similar problems as you are while running load testing using Load runner simulating a load of 100 concurrent users.
We have Oracle 8.1.7.4 client running on WebLogic 6.1 SP4 on Solaris 2.6 connecting to Oracle OPS 8.1.7 running on Solaris 2.6. We are getting core dumps while running stress testing using Load Runner for 100 concurrent users. The performance is very good for a while in the beginning of the test with CPU and memory utilization low and all of a sudden we see requests for connections queueing up on the WLS and the server crashes.
If you have found a solution to your problem Please send us details of it. Also it will be very helpful if you can send us the fix applied.
You can send your reply to my email at [email protected]
Regards,
Raj

Solaris TCP/IP

I tried to write program, that will listen to the opening new tcp connections in the Solaris system.
Fot the target I try to overwrite the <code>pr_usrreq()</code> function from the <code>protosw<code>
protocol switch table struct. The <code>pr_usrreq</code> function gets first parameter socket, according to the documentation from the include file /usr/include/sys/protosw.h, the parameter from the <code>struct socket</code> type. I cannot find the definition of the <code>struct socket </code> in the include files
of the Solaris system. I cannot to pass the compilation of the my module.
May be:
Anyone tried to resolve the problem ?
The protocol switch table is not relevanted in the Solaris system?
The first parameter has another type ?
The there is another solution ?
Thanks in advance.
P.S. I found the <code>struct sonode</code> type in the /usr/include/sys/socketvar.h file, that like to the <code>struct socket</code>.

I tried to write program, that will listen to the
opening new tcp connections in the Solaris system.
Fot the target I try to overwrite the
<code>pr_usrreq()</code> function from the
<code>protosw<code>
protocol switch table struct. The
<code>pr_usrreq</code> function gets first parameter
socket, according to the documentation from the
include file /usr/include/sys/protosw.h, the parameter
from the <code>struct socket</code> type. I cannot
find the definition of the <code>struct socket </code>
in the include files
of the Solaris system. I cannot to pass the
compilation of the my module.Solaris TCP/IP doesn't use pr_usrreq() or protosw struct. Can you describe in more detail
what are you trying to achieve?

Solaris 9 Bug with Installer Scripts?

The installer for my product, which works fine on Solaris versions from 2.5 through 2.9 Early Release, does not work on the released Solaris 9. The problem is visible right away in my "request" file, which early in the file uses the standard shell variable '$0'.
The result I get is something like:
"/tmp/installBOayFd/checkinstallDOayFd", rather than the expected
"/myhome/mypkg/install/request". I don't know why files are being referenced from "/tmp", and the garbage characters look suspiciously like a string terminator is being left out someplace.
Does anybody know anything about this? If so, is there an expected fix or a workaround?

Here is the output from the command:
bash-2.05$ /usr/jdk/latest/bin/java -version:"1.5+" -version
java version "1.5.0_15"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.5.0_15-b04)
Java HotSpot(TM) Client VM (build 1.5.0_15-b04, mixed mode, sharing)
It looks like the java version should be okay, but it doesn't seem to include the definition for TransportAdapterException.
Thanks for pointing out that this cron job doesn't actually do what I was hoping that it did, which was to run a sort of "smpatch update" automatically each night to handle the downloading and/or installation of "safe" patches. I'll probably just drop it.

What RFC:s are implemented in Solaris�s TCP stack?

I am trying to find out wich of the following TCP related RFC:s are implemented in the Solaris TCP stack:
RFC0793
RFC1122
RFC2581
RFC1323
RFC2081
RFC1191
RFC1981
RFC2414
RFC2481
RFC2884

Hi Jeny,
refer this URL:
http://www.rvs.uni-hannover.de/people/voeckler/tune/EN/tune.html
Thanks.
regards,
senthilkumar
SUN - DTS

Solaris 8 Floppy Disk Drive Problem

Hi to all!, I need help with Solaris 8 10/00 Intel. I�ve installed it and everything works fine, but I cant make the system to recognize the floppy disk drive, everytime I do a "volcheck", it just ignores it. Inside CDE happens the same thing, when I do check the floppy, nothing happens. Any clue?. I�m using an AMD Athlon 700, with VIA KT133 chipset. Thank a lot!.

Try "ftp" + your email address. (but "anonymous" works, too).
Perhaps a typo or there was some other temporary problem with the
web site.
The K7M boards have different bios release numbers, it seems the
first bios release had a revision number of 1007:
ftp://ftp.asus.com.tw/pub/ASUS/mb/slota/k7m
Do you have to boot using the DCA floppy for some specific reason?
(for example to install using the >32GB ide-drive patch; network install
and no local cdrom drive; El-Torito CDROM boot does not work at all)?
Or do you have the system already installed by booting from CDROM,
and you just want to get usable floppy/serial/parallel ports under
solaris 8? In that case have a look at http://www.tools.de/solaris/via/. I've written a small driver that attaches to the VIA
686A SouthBridge PCI/ISA device and it re-enables the Super-I/O
controller on that chip. The Super-I/O chip is disabled due to the
solaris ACPI bug and is the reason all these on-board devices do
not work under solaris 8. According to
http://www.asus.com.tw/products/Motherboard/SlotA/k7m/spec.html, the K7M uses the same SouthBridge chip, so my driver should work for
this board, too.

Solaris 10 (x86) + XDMCP + Java = No Num keypad

Hi all,
I have a Solaris 10 Server running on Sun Fire hardware. Users login via XDMCP and use a WYSE Linux Based terminal to connect to the Sun Server. From inside the terminal numlock works as you would imagine it should. However when running a Java Applet, If the user presses Numlock, the Backspace and Delete keys stop working and the numeric keys do not activate.
This only affects Java Applet based applications and does not effect apps like Star Office.
Running JRE/JDK 1.5.0_06 with Java Desktop 3
Windows users running the same applet combo do not exhibit this problem nor is it native on the console. It only exists through things like CygwinX, ReflectionX, KEAX or any hardware based XTERM.
I have read http://www.freedesktop.org/wiki/CygwinXTroubleShooting which suggests this is an existing historical Solaris problem

This bug also surfaces in FreeBSD's port of JDK 1.5.0. With Num Lock switched on, Backspace and delete keys don't work, nor for that matter does Tab, (or Shift-Tab, or the cursor keys, Home, End, Page Up or Page Down... In fact, any non-meta key that happens to be coloured grey on my keyboard. To delete, I have use to Ctrl-H, which is nice if I want to be reminded of teletype days, but not nice these days.
Switch Num Lock off, and you lose use of the Number pad for numbers, but the usual functionality for the grey keys on the keyboard returns.
I spotted it the other day on a Solaris 10 x86 machine, and couldn't believe that it surfaces there as well, so it's obviously a core bug that both the Solaris build and FreeBSD build is affected by.

Solaris 9: USCSICMD ioctl on volume management devices fails

Upto and including Solaris 8, it was possible to use the USCSICMD
ioctl on devices managed by the vold daemon (e.g. the CDROM device,
/vol/dev/aliases/cdrom0) without the need for root permission.
It seems this does not work any more with Solaris 9, it requires
root permission now.
An example program is appended below. On Solaris 8 with a uid != 0
it prints the CDROM drive's manufacturer and product id:
% uname -r
5.8
% id
uid=109(jk) gid=20(usr)
% vol
TOSHIBA DVD-ROM SD-M1612
It seems that on Solaris 9, the USCSICMD ioctl fails with an EPERM
error. Can anyone reproduce this on Solaris 9?
Is this a Solaris 9 bug, or a Solaris 9 feature?
/* compile with: cc -o vol vol.c */
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/types.h>
#include <sys/ioctl.h>
#include <sys/scsi/generic/commands.h>
#include <sys/scsi/impl/uscsi.h>
#define     CD_DEVICE     "/vol/dev/aliases/cdrom0"
/* execute a SCSI INQUIRY command on device |fd| */
int inquiry(int fd, void *inq_data, int inq_len)
struct uscsi_cmd sc;
union scsi_cdb inq_cdb;
memset(&inq_cdb, 0, sizeof(inq_cdb));
inq_cdb.scc_cmd = SCMD_INQUIRY;
inq_cdb.g0_count0 = inq_len;
memset(&sc, 0, sizeof(sc));
sc.uscsi_cdb = (caddr_t)&inq_cdb;
sc.uscsi_cdblen = 6;
sc.uscsi_bufaddr = inq_data;
sc.uscsi_buflen = inq_len;
sc.uscsi_flags = USCSI_ISOLATE|USCSI_READ;
if (ioctl(fd, USCSICMD, &sc)) {
     perror("USCSICMD ioctl, inquiry");
     exit(1);
if (sc.uscsi_status) {
     fprintf(stderr, "bad status for INQUIRY command (%d)\n",
          sc.uscsi_status);
     return -1;
return 0;
int main(int argc, char **argv)
struct scsi_inquiry inq_data;
int fd;
char *cd_device = CD_DEVICE;
if ((fd = open(cd_device, O_RDONLY)) < 0) {
     perror(cd_device);
     exit(1);
if (inquiry(fd, &inq_data, sizeof(inq_data))) {
     fprintf(stderr, "scsi inquiry command failed\n");
     exit(1);
/* print vendor/product from scsi inquiry data */
printf("%.*s %.*s\n",
     sizeof(inq_data.inq_vid), inq_data.inq_vid,
     sizeof(inq_data.inq_pid), inq_data.inq_pid);
exit(0);

According to a usenet thread (where I posted the same question)...
http://groups.google.de/groups?hl=de&lr=&ie=UTF-8&oe=UTF-8&selm=wyvg45125q.fsf%40tools.de
... it's a security enhancement. Or in other words: It's a feature, not a bug.
It's possible to work around the problem in your own programs using
the "-lsmedia" library, but this is not (yet?) a supported interface, it seems:
#include <sys/smedia.h>
int fd = open("/vol/dev/...", ...);
struct uscsi_cmd scsi_cmd;
smedia_handle_t handle = smedia_get_handle(fd);
/* fill in scsi_cmd structure ... */
smedia_uscsi_cmd(handle, &scsi_cmd);
smedia_release_handle(handle);

Tcp_reass_timer in Solaris 11

Hi,
in Solaris 11 TCP code there is a timer for the reassembly queue (tcp_reass_timer). If out-of-order data sits in the reassembly queue for 100*SECONDS (default timeout value for tcp_reass_timeout) and no data is read from the reassembly queue during this period, all the out-of-order data and related SACK information is deleted (data reneging) using the tcp_reass_timer() function defined in common/inet/tcp/tcp_timers.c. Such a timer is not specified in TCP RFC and to my knowledge there is no other OS that implements such a timer. Could you please clarify why Solaris TCP stack have a reassembly queue timer? How does it help to the OS to delete successfully received out-of-order data?
Regards,
nasif ekiz

tcp_reass_timeout was implemented as result of Bugfix for 6759500 which addressed TCP security vunerabilties

WebStack 1.5 - Apache stops responding - Solaris 10x86 Zone

I've got a Solaris 10x86 installation with multiple zones. Webstack is installed in the global but only one zone is using the Apache as a generic web server. Other zones are quite happy and fast (could mail servers, simple ftp server).
The web server however is driving me nuts. It simply stops responding for no apparent reason. Nothing in the apache logs, error, virtual hosts, nothing in the system logs. It just sits there when you are using one of the websites until you stop and restart it.
I thought perhaps this was a memory issue, but we added more memory to the box and it acts the same. I would expect to see some sort of error but none are displayed.
I currently have a script that connects to a site on a regular basis to see if it is still responding and if not, restarts the Apache server (includes MySQL as well), but I am at a loss why this is doing this.
The previous box was very old and used Xampp instead (very very old) and never had this problem.
If I can't get this to stabilize, I will have to put something else on it (Xampp again perhaps?)..
Anyone else see this before or have any suggestions. BTW Truss on the master pid of the Apache doesn't really show anything interesting.

Seems I found the reason....
http://sysadminsjourney.com/content/2009/09/01/apache-2212-2213-and-solaris-10-bug-nastiness
Don't know for sure as the article doesn't state if it was Sparc or x86, but it sure sounds like it. This is older software so getting a fix appears to not be favorable. I really can't rebuild the system until after the first of the year.
No idea if the person mentioned (Jeff Trawick ), would still have the work around .so file.
Sad that many of the Sun links are gone or redirected into Oracle neverland...

Binding a Second IP to an Interface Card

Hey All,
I need to bind a second IP address to my single network card on a Solaris 7 server. I've done this in Linux by adding an additional ifcfg-eth0 file to /etc/sysconfig/network-scripts, specifying the new IP address. With Solaris though, the new landscape has me a bit flustered. I'd very much appreciate a guide through the woods here, my flashlight is going dim...
Much Appreciated,
Chris

Hi Chris,
Here's a write up on this question:
Virtual Interfaces (One interface, different IPs):
Solaris TCP/IP allows multiple logical interfaces to be associated with a physical network interface. This allows a single machine to be
assigned multiple IP addresses, even though it may have only one
network interface. (see the ifconfig man page)
You can define up to 255 "logical addresses" for a given network interface on Solaris 2.x.
There is no supported method to do so on SunOS 4.1.X (Solaris 1.x)
In Solaris 2.6 you can go greater than 256 ( 0-255 ) which is configurable with NDD. The command "ndd -set /dev/ip ip_addrs_per_if 1-8192" will permit you to configure as many as your addressing structure will permit
(up to 8191 virtual hosts.)
1. Create an entry in the /etc/hosts file ( or nis host map) for each of the virtual interfaces with the hostname and IP address you want to use. Don't forget to add the hostname to /etc/hosts NIS, NIS+, and/or DNS database.
2. Create a /etc/hostname.<interface:#> file for each of the interfaces. (/etc/hostname.le0:1, hostname.le0:2, hostname.le0:3...le0:255.)
The file can contain a single line with either the ip number or hostname reference. For example, if you are using the le0 interfaces, you would create the following files:
/etc/hostname.le0:1 (do NOT use a le0:0 - same as le0 )
/etc/hostname.le0:2
and so on
This can also be done for an hme interface. You can have up to 1024 virtual interfaces under 2.5.1 This file should include the hostname or IP address for the corresponding interface.
3. If subnetting is being used, there should be an entry in the
/etc/netmasks file in the following format:
network_address netmask
The network_address should be the network address before
subnetting. For example, a class B network that is subnetted to
a Class C would have an entry such as:
157.145.0.0 255.255.255.0
4. Reboot the system
5. Verify changes with an ifconfig -a
Note: there are some bugs with third party applications/daemons that do not operate well with many virtual interfaces.
Additional Note: For Security and network isolation. You can set ip_forwarding off and ip_strict_dst_multihoming on, if you are trying to prevent access to the other interfaces.
"ndd /dev/ip ip_forwarding" Determines if the workstation will route packets.
"ndd /dev/ip ip_strict_dst_multihoming" Determines whether to use Strict Destination Multihoming. If this variable is set to True, and ip_forwarding is turned off, then the machine will not except packets destined for a different interface. RFC112
If duplicate additional local routes are added (by ifconfig) when the virtual interfaces are on same subnet, then you may need to "route delete net" on these.
A /etc/rc2.d/S99vif file can be added to ifconfig and add/delete routes, etc.
On 2.6 it may be neccessary to turn off interface groups.
To disable this feature (which is on by default) the following line
must be added to /etc/rc2.d/S69inet
ndd -set /dev/ip ip_enable_group_ifs 0
Thanks
Sujeet (Sun-DTS)

JES 2 (2004/Q2 and 118263-14 - Portal 6.3 PC 14)

Hi all,
Just wanted to 'ping' the crowd, has anyone else tested JES 2 and the current public release 118263-14 (portal 6.3 and patch cluster 14) ?
My PC12 (plus patch from 13 to fix a bug regression) is rock solid, but doing some testing with PC14, I find that once 1 single user logs in, and retrieves the authenticated portal desktop, you can refresh the screen (F5) and you'll start to see the 'webservd' threads escalate, and after a short while, it continues past "999" and will soon core dump the webservd process (a single user can cause this within 30 seconds)
Has anyone else experienced this ?
I'm not rocking the boat, and only have 115766-01 (Access manager 6.2 PC 1 installed)
SUN support is on this, but wondering if anyone else has experienced this?
Thanks
Dave

Hey There,
If you are using 6.0.. stop, pull out! eject ! Do not pass Go. at a minimum java client versions will give you grief. Yep the docs are a tad... errr.. lacking...
I'd suggest you at a min use 2005/Q1, it seems to be relatively stable, or patch 2004/Q2 to at least PC_9 or PC_10, and you should have a stable platform.
- We stopped using Netfile (the built-in samba share) due to it's poor performance, and reverted to a GPL version:
http://www.dragonsdawn.net/~gordon/smb-network/
- it's awesome with SRA, as it's incrediably fast, only drawback is it doesn't do deletes, whereas netfile gives you this.
- as far as the original thread, yes waaay back, I saw this error, and it had to do with some of the solaris TCP variables missing on a base install.
specifically "domainname" (type that at prompt and you should not see a "blank" return. (defaultdomain variable). Check all your other TCP settings as well and ensure everything is fully working,
Thanks
dave

"Failed to Send the HTTP Message..." - HTTPSupport problem

Hello,
We're having a few difficulties when the HTTPServer in UDS5.0.3 running on Solaris 8. Basically, we have a pair of MS .NET servers using some XML Webservices that we reside on the UDS server.
The symptoms of the problem are that a lot of "Failed to send the HTTP message..." messages are being seen in the log for the partition that contains instances of the HTTPServer. The reason given is a UsageException on the ExternalConnection - stating that the "Object is not open for I/O so it cannot be read or written" (ExternalConnection::Write at 1). This only occurs when a HTTP Response is to be sent back to the .NET server.
Unfortunately, this exception appears to be occurring somewhere within the code held in the HTTPSupport library, so we can't easily trace the error. Worse still, we don't actually know what's causing the error.
Does the error indicate a problem with comms between our UDS HTTPServer and the .NET servers? We've also noticed a large number of socket connections being left in a state of CLOSE_WAIT when this error occurs (which eventually block the port altogether resulting in the HTTPServer instances becoming unresponsive).
As far as I can tell, the Solaris TCP/IP settings are all at the OS default.
Many thanks for any assistance,
Jason Bridgen.

Hi Gordon,
The problem was solved, the customer changed his ip adress on the server and it's OK now!
Thank you !!
Hilla.

118872-03 ksh patch breaks set construct wildcard

Not sure where to report solaris 10 bug, anyone know? So I will post it here.... Solaris 10 sparc patch 118872-03 breaks set construct wildcard. For example:
% ls [A-Z]*
will return filename of both upper and lowercase..
back out 118872-03 fix the problem. 118872-02 works..
-- Frogdeep

OK... not so sure the problem is with 118872-03 now. I look carefully even with the 118872-02 patch I still have the problem. So it is some other patch that is causing the problem. Let's say my system is fully patched.
On system we only have recommendated cluster patched:
% touch A aa B C D ee
% ls [a-z]
[a-z]: No such file or directory
On fully patched box:
% touch A aa B C D ee
% ls [a-z]
A B C D
Doh! Help! Something is not right :*(
-- Frogdeep

Solaris 10 TCP bug?

Similar Messages

Maybe you are looking for