Hi,
What is the preferred way of getting out of an accept in an
multithreaded application? On linux it works that the in-kernel
filedescriptor is closed from the signal handler but that does not seem
to do the trick in FreeBSD 7.1 or 7.2. Is using poll the only option or
preferred anyway?
On Dec 25, 2008, at 10:10 PM, Hartmut Brandt wrote:
In any case it should re-read the kernel table only every 10 minutes
and in the mean time monitor the routing socket to update its copy
of the table. If of course someone is doing a lot of updates on
Why does it have to re-read the
I couldn't get Apache 2.2 ab to work on 7.0 at all. The ab from 2.0
worked fine...
Pete
Andre Oppermann wrote:
Paulo Fragoso wrote:
Hi,
We was using one machine with FreeBSD 6.4-RELEASE running
apache-worker-2.2.3 + mysql, this server can answer high request from
one client using ab:
Patrick Tracanelli wrote:
I have raised the queue lenght a lot, up to 40960, and the behavior
was the same. Ill keep trying and let you know if any success.
No queue depth is going to help you if you receive more data than you
can process.
Pete
Pardon the basic question, but is the current patchset zero copy or
one copy? The paper I saw a link to described a mechanism to eliminate
one of the two copies the traditional bpf approach makes but I haven't
taken a look into the actual code.
Pete
Any chance the recent root zone changes would make it to 7.0?
Pete
___
freebsd-net@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-net
To unsubscribe, send any mail to [EMAIL PROTECTED]
How about routing domain or forwarding domain?
Pete
Julian Elischer wrote:
So, I'm playing with some multiple routing table support..
the first version is a minimal impact version with very limited
functionality.
It's done that way so I can put it in RELENG_6/7 without breaking ABIs
(I
Julian Elischer wrote:
Petri Helenius wrote:
How about routing domain or forwarding domain?
which shortens too
rdom / rd ?
fd would be quite overloaded acronym.
vrf would work for me too. Quite accepted industry term.
Pete
freebsd-net@freebsd.org mailing list
http
Increase sendspace an recvspace depending which way your data is going:
net.inet.tcp.sendspace: 57344
net.inet.tcp.recvspace: 256000
TCP window scaling is enabled by default nowadays if I remember correctly.
Pete
Morgan wrote:
Hi.
I'm trying som file transfers across the globe. The RTT is
Robert Watson wrote:
I tend to agree, but implemented full queueing support for if_em to
make sure I understood to complexity implications of completely
removing queueing from the ifnet side dispatch. I guess an
interesting question for us is how we decide what the right threshold
is to
Brian Candler wrote:
On Sun, Dec 11, 2005 at 03:26:50AM -0800, kamal kc wrote:
i modified the bridge.c file and added a routine to
compress/decompress
ip packet. i put my code in bdg_forward(). And ran the
pc in bridge mode.
The modified kernel is deployed in network where the
datarate
Benjamin Rosenblum wrote:
the em driver in itself is extremly buggy. many people, myself
included, are hitting some major problems with this driver that are
causeing some serious issues. i cant transfer any large files to my
server because the em driver panics and drops the connection for
Steve Shorter wrote:
Howdy!
I was wondering how come the listed media types supported
by sk(4) and em(4) is only 1000BaseSX. Even for single mode
1000BaseLX NICs such as sk 9841 and Intel PWLA8490LX, ifconfig
show the media type as 1000BaseSX. Though the NIC appears to operate in
Mike Tancsa wrote:
I like this idea as well, but you need to control how the routes would
come back after the interface comes back up ? This seems more of the
province of a routing daemon like quagga as opposed to a kernel
feature no ?
The connected interface should try to transmit
Aziz Kezzou wrote:
Hi all,
I worked a bit with netgraph nodes and I find them very amazing and
powerfull... Since my netgraph experience is still quite limited (
they are out of the scope of my project actually) I would like to know
if the following claim is true, I need to be sure because it
Is there a way to configure multiple IPv6 address aliases without
knowing the prefix in advance and just specifying the lower 64 bits on
the ifconfig_ lines on rc.conf?
Pete
___
freebsd-net@freebsd.org mailing list
JINMEI Tatuya / wrote:
On Thu, 28 Apr 2005 14:20:07 +0300,
Petri Helenius [EMAIL PROTECTED] said:
Is there a way to configure multiple IPv6 address aliases without
knowing the prefix in advance and just specifying the lower 64 bits on
the ifconfig_ lines on rc.conf
Claudio Jeker wrote:
On Wed, Apr 13, 2005 at 05:14:52PM +0200, Lars Erik Gullerud wrote:
On Thu, 7 Apr 2005, Dag-Erling Smørgrav wrote:
Charles Swiger [EMAIL PROTECTED] writes:
It's dead, I think: Cisco's lawyers started making predatory noises
about their intellectual property.
Claudio Jeker wrote:
On Wed, Apr 13, 2005 at 09:22:48PM +0300, Petri Helenius wrote:
Claudio Jeker wrote:
Did this recently change since looking at /etc/protocols it does not
seem to be the case for most of them anyway?
Most new protocols come from either some company, DARPA
Does somebody have the programming specs for the em chips? Despite of
multiple contacts and promises Intel has been unable to produce them.
Pete
Chris Tracy wrote:
Hi,
I have been attempting to get iperf to generate a line-rate TCP flow
(~989Mbps) across a GigE link but can only get a maximum of
Is there a way to send packets from userland process to a specific altq
defined queue?
Pete
___
freebsd-net@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-net
To unsubscribe, send any mail to [EMAIL PROTECTED]
Danny Braniss wrote:
with tags enabled, iSCSI is much faster, but it also causes a deadlock :-(
this is what i run:
newfs -U /iscsi device
cd /iscsi device
restore rf /home/file.dump
What are you using / what's recommended as iSCSI server?
Pete
___
Alexey Zelkin wrote:
Why not ? Having small and reliable kernel http server able to handle static
content only and limited functionality, would be useful in many cases.
Why?
Pete
___
freebsd-net@freebsd.org mailing list
Julian Elischer wrote:
Petri Helenius wrote:
Alexey Zelkin wrote:
Why not ? Having small and reliable kernel http server able to
handle static
content only and limited functionality, would be useful in many cases.
Why?
able to run without a filesystem, while in single user mode?
Didn't
ukasz Bromirski wrote:
Hi,
Is anyone working on a port of OpenBGPd, or current version of Quagga
(0.97.3)?
openbgpd compiles fairly painlessly on 5.3. Making it work on 5.2.1 was
definetly more work.
Pete
___
[EMAIL PROTECTED] mailing list
Pawel Malachowski wrote:
Hello,
I would like to ask people using mpd about performance on particular hardware
setups. I am interested in the numbers of sessions (probably PPTP with weak
encryption) and total bandwith, that can be achieved with, e.g.:
. 300MHz CPU,
. 1GHz CPU,
. 2GHz CPU.
Bruce M Simpson wrote:
Here is a non-critical patch to bring em(4) into line with other
drivers, by using the sysctl tree created for each device by the
bus framework.
Does anyone here have an idea why some platforms (like Thinkpad X31 or
i875 Supermicros) have trouble rebooting with 5.3-BETA
Garrett Wollman wrote:
On Wed, 06 Oct 2004 17:57:14 +0200, Waldemar Kornewald [EMAIL PROTECTED] said:
Yes, something in that direction, plus: protocols:
IPv4, IPv6, TCP, UDP, ICMP, IPX, etc.
Just about everything as modules.
It is not generally regarded as a good idea to make artificial
Kalin Hristov wrote:
It would be very nice
As they say about the MPLS The feature of the Internet
Did they change the acronym from Mostly Pointless Lamp Switching?
Pete
___
[EMAIL PROTECTED] mailing list
Pavel V.Zheltobryukhov wrote:
I send Ctrl+C and boot process was continued. But network card doesn't
work. I check cable - it's good. I can't send any ping packet to other
computers. I tried other transmit modes - I switch NIC to 10Mbit half
duplex and 10Mbit full duplex via ifconfig options.
Andre Oppermann wrote:
BTW: You may be better off using pfil_hooks instead of netgraph for your
tool. You'll save one m_copym and m_freem for each packet.
Is pfil zero copy or one copy by default? If the driver supports it,
does a packet get directly DMA'd in mbufs and passed over the pf
Colin Alston wrote:
What exactly is the point/benefit of such a change?
On related note, it would be nice if the OS bundled dhclient would
report OS version like it does on Windows and Linux. Would make some
operations easier.
Pete
___
[EMAIL
James wrote:
uRPF should not emit an ICMP when it drops a -reject route. Even with
ip unreachables, Cisco won't emit ICMP when uRPF is killing a packet. The source
that triggered uRPF drop condition cannot be trusted as it may have spoofed the
packet.
Where would the ICMP go anyway because
ming fu wrote:
Does this one replace the em driver?
It does not. However as far as I understand the semantics of the chips
aren´t that much different so I wonder why another driver instead of
adding to em.
Pete
___
[EMAIL PROTECTED] mailing list
ming fu wrote:
Is there a suggestion on how to trigger the watchdog to be called. It
is really time consuming to diagnose this as it takes hours or dates
for the em to lockup once.
Some vendors are kind enough to sell motherboards broken enough for em
to never get interrupts. However the
Christian Hiris wrote:
About a year ago i observed strong nfs performance decrease when using
RLT8139A nics. Nfs transfers leaded into high system load, because of an
excessive high packet retransmission rate. Switching over to 3Com nics solved
my problem.
The specific model and it's
Scott Long wrote:
I'm looking a t a similar system right now and it definitely looks
like an interrupt routing problem, not a driver problem. The
interesting thing is that (with 5.2-current as of two days ago)
disabling neither
ACPI nor APIC helps. I guess that we might want to get John
I´m highly confident that this is a case of integrated CSA ethernet
with broken BIOS. I suspect you get an message about that when booting.
Pete
Søren Schmidt wrote:
John Polstra wrote:
On 05-May-2004 Søren Schmidt wrote:
For what its worth I have problems with one em based interface as
well,
Søren Schmidt wrote:
Nope. no messages to that effect, oh and it works in windows(tm)...
The last thing I see if I try to use em0 is:
em0: Link is up 100 Mbps Full Duplex
and then the system locks up hard.
Would you mind posting full dmesg output?
Pete
Luigi Rizzo wrote:
Can someone explain what is the goal ? Reuse a number if an
interface has the same name of a previously existing one and
the index is free ? And does it make sense, anyways, or
we could just simplify that code and just reuse the first
available entry in ifindex_table[] ?
The
Luigi Rizzo wrote:
I do not think the current code supports this, as any free
index at the top of the array is reused. So if you had 'vlan12' there
and it went away, the next interface that comes in will grab that
index and vlan12 will get a new one. On top of this you can rename
interfaces
David Malone wrote:
Mind you, Petri originally asked about evidence for two machines
back-to-back, and 100ms is rather long for that (unless you're at
Steven Low's lab ;-)
Another interesting figure which comes to mind is whether bursty loss
is the usual way a multigigabit optical link loses
Brooks Davis wrote:
For that matter, there are sufficent drops on 10GbE from data errors to
insure that two boxes connected back to back won't achieve line speed on
a single TCP session.
Do you have data to back this up or are you just using broadcom chipsets?
Pete
Brooks Davis wrote:
The problem is that the BER of a typical optical link is high enough
that the link will almost certantly discard at least one packet before
you get out of slow-start and once that happens it, AIMK means it take
hours or even days to get back up to the top even assuming you
Mike Silbersack wrote:
SACK itself really doesn't do much, it's all the new congestion control
schemes (FACK, Rate Halving, etc) that come shipped with most SACK
implementations that do the work and contain most of the complexity.
And all this would be non-issue within normal operational
Andre Oppermann wrote:
What else is missing in FreeBSD?
Cannot resist
MPLS?
1/2 :-)
Pete
___
[EMAIL PROTECTED] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-net
To unsubscribe, send any mail to [EMAIL PROTECTED]
Julian Elischer wrote:
please remeber this next time someone tries to have it deleted from the
system :-)
I tried to google for such a discussion but fortunately couldn´t find
any. Why somebody would want to take away netgraph?
Pete
___
[EMAIL
I´m asking the net list because I came across this when browsing through
the networking / resolver code.
The question is if the code should check for zero value before calling
malloc or is malloc(0) legal if the pointer is never used?
I came across this when using dmalloc library and it
Maxime Henrion wrote:
In C99, malloc(0) is legal. From n869.txt :
%%
If the size of the space requested is zero, the behavior is
implementation-defined: either a null pointer is returned,
or the behavior is as if the size were some nonzero value,
except that the returned pointer shall not
Gareth Bailey wrote:
Does anyone know how to configure FreeBSD to use a wireless USB WLAN adapter. The adapter is a X-Micro WLAN USB adapter.
When i plug it into FreeBSD i get a ugen0 device loaded message. I understand this means that the OS doesn't specifically recognise it as a WIFI adapter,
Guy Helmer wrote:
Emre Bastuz wrote on Thursday, February 12, 2004 3:43 PM
Hi,
for sniffing purposes I have a FreeBSD 4.9-STABLE box running on highend,
state-of-the-art hardware (Xeon something) with all bells and whistles.
The NIC´s an onboard copper em0 with gig-e capabilities.
Bruce M Simpson wrote:
I'd be happy to review other patches for this. My personal feeling though
is that the actual performance increase may not be that great, but it's
a case of someone implementing it and doing the math.
Writing to memory, specially on SMP systems, is very expensive. So
Tanmay Ganacharya wrote:
Hello,
I am currently using FreeBSD 4.8. I am want to record RTT and RTO
values of a tcp connection.
Could anyone please tell me which variables in which files hold the
current RTT and RTO values.
Once known I can put printfs and print these values.
This seems to come
Bruce M Simpson wrote:
The main motivation for committing the port was to bring in features like
this without disrupting the vendor branch of tcpdump/libpcap in the FreeBSD
base system.
Is there a port of libpcap? The system tcpdump seems to be out of synch
with libpcap already
since tcpdump
popsong old wrote:
Ah, my fault. I didn't read your patch carefully and assumed that ip
fastforward do flow caching as ip_flow does. However, I think ip flow
caching is a good thing and maybe implementing it in ip fastforward is
a good idea.
Caching flows has mixed results, it works for you if
(I might sound like a broken record, but...)
Any chance of libpcap vendor import before 5.2 release?
Pete
___
[EMAIL PROTECTED] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-net
To unsubscribe, send any mail to [EMAIL PROTECTED]
Harti Brandt wrote:
On Wed, 1 Oct 2003, Bruce M Simpson wrote:
BMSOn Wed, Oct 01, 2003 at 02:17:59PM -0400, Barney Wolff wrote:
BMS Are you talking about running the phone line directly to the fbsd box
BMS with no dsl modem?
BMS
BMSYes. Also, PPPoA in FreeBSD is currently only implemented if you
Bruce M Simpson wrote:
ports would seem to be an acceptable halfway house, though, for people who
want to use pcap/tcpdump of a more recent vintage, than has been determined
to be suitable for a FreeBSD release. does it not?
On general case, I agree with you. However libpcap does not change
What is the right way to set the TRAFFIC CLASS field in
ipv6 packets sent from an application? I´m looking for
function similar to setsockopt IP_TOS with IPv4.
Pete
___
[EMAIL PROTECTED] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-net
Hajimu UMEMOTO wrote:
TCLASS is not supported in current FreeBSD. It will merged into
FreeBSD during next KAME merge.
Is this planned before or after 5.2?
Pete
___
[EMAIL PROTECTED] mailing list
Bruce M Simpson wrote:
This is coming up more often. Perhaps we should consider net/libpcap-devel
and net/tcpdump-devel ports for people who wish to track CVS and/or
snapshots of these tools? This might relieve some of the pressure on Bill
to update the vendor branch so often.
How often is so
Edwin Groothuis wrote:
On Wed, Sep 17, 2003 at 06:31:03PM -0700, Josh Brooks wrote:
Whenever I run:
tcpdump -vvv
when I am finished, I am surprised to see:
27441 packets received by filter
7866 packets dropped by kernel
That's because the buffer of captures-but-not-yet-processed
Bruce M Simpson wrote:
We should avoid applying patches on the import if we possibly can, it's
Not Right.
I know and I agree with that. That´s why I would like to have the tree
right to avoid patching
it locally. Same problem, different leg.
I'd suggest submitting a patch via Sourceforge. I
Bruce M Simpson wrote:
Er, if you check this URL:
http://www.freebsd.org/cgi/cvsweb.cgi/src/contrib/tcpdump/CHANGES
Shurely you mean tcpdump 3.7.2, which is already imported (by fenner, with
additional hacks)?
I mean libpcap, which also tcpdump uses, if I´m not mistaken. Look in
Don Bowman wrote:
I found that increasing the bpf buffer size in libpcap
to 256K from the default of 4K made a tremendous difference.
We generally use sizes from 512k to 8M depending on network interface and
hardware configuration. Used to do larger but run into some issues with KVM
jakae wrote:
Hello,
I am wondering if there is a way for rejecting some system (shell
account) users to access some networks and hosts.
I wonder if the network virtualization patch would do it, if I remember
correctly, it has
separate interfaces for processes. Look in the list archives and
Since it seems that tcpdump people are more or less gone, is it still
too much to ask
that the broken code in src/contrib/libpcap/pcap-bpf.c starting from
line 236
which breaks the bufsize would be removed from the FreeBSD repository?
Obviously the option is to fix the code but that´s much more
You probably need to install a service to the internet background
radiation ports.
Look at the samba port, it´ll receive packets on the ports most
virused microsoft
clients send to random destination addresses.
Pete
Colin Watson wrote:
Hi, I've got a rather strange issue with UDP loss (at
Julian Elischer wrote:
On Fri, 5 Sep 2003, Konstantin KABASSANOV wrote:
Hi,
I'm actually working on a teredo server/relay implementation as a
netgraph node. Can you tell me, please, who I need to contact for an
eventual integration of my code into the netgraph source.
contact me..
ipfw seems to have developed a bug lately on 5-CURRENT;
# ipfw add 2042 allow tcp from 0.0.0.0/0 to me
42
02042 allow tcp from me to me dst-port 42
It used to work that 0.0.0.0/0 was any instead of me. Last I checked
the notation is also widely used in
Maxim Konovalov wrote:
On Thu, 28 Aug 2003, 23:01+0300, Petri Helenius wrote:
ipfw seems to have developed a bug lately on 5-CURRENT;
# ipfw add 2042 allow tcp from 0.0.0.0/0 to me
42
02042 allow tcp from me to me dst-port 42
It used to work that 0.0.0.0/0 was any instead of me. Last I
My D-link USB DUB-E100 adapter seems to work only for a while
on recent and older 5.1-CURRENT. After that it seems to lose
sight of it's PHY;
Also on boots it complains about MII without any PHY! randomly,
pulling and re-plugging it will eventually get it running and
it works for 5-10 minutes
This would add additional delays to the code path for both ingress
and egress. In a world where gigabit ethernet is becoming the norm,
every nanosecond counts. I don't think the benefits of your proposal
would justify the performance loss. At the very least, I'd want the
extra calls to
I´ve seen lost interrupts with 5.1 with em driver and the ata driver seems to suffer
from this also when doing detach/attach. Getting a old kernel module from 5.0 and
recompiling
it for 5.1 does not help with the issue so I suspect a more generic issue with
interrupts.
Pete
- Original
Is anyone working on a driver for the Intel 10Ge card (I think they're
the only one actually shipping...)? I was looking to give one a try on
something other than linux :)
Since intel provided the 1Ge driver, I would suspect them to come up with
the 10Ge one. The 10Ge part does not seem
Charlie wrote:
On Wednesday 09 July 2003 08:09, Gleb Smirnoff wrote:
On Wed, Jul 09, 2003 at 01:18:06PM +0800, Eugene Grosbein wrote:
E Does FreeBSD 4.8-STABLE keep per-interface summary for received
E unicast octets? More precisely, I need to know number of unicast
E octets received by my
how come no-one knows about netgraph.. the framework designed to do
exactly this? :-)
It's only been in use for 6 years..
We do this and a lot more with netgraph and love the architechture,
thanks goes for the people who did the architechture.
So this qualifies as somebody knows :)
Pete
Eugene Grosbein wrote:
Luigi Rizzo wrote:
How could the scheduler decide when to drain the queue ?
It should move packets from zero-bw WFQ pipe the interface FIFO
as soon as possible but should consider weights
(100 packets from one queue then 1 from another and so on).
That would
Most common acronym would be Digital Video Broadcasting,
a set of standards defining audio, video, text, teletext, data, etc.
encoding, modulation, encapsulation, etc. over Satellite, Cable
or Terrestial medium.
Only reason why I have a linux-box is to run DVB-S stuff.
Not sure what it would take
specification so applications like vdr, dvbstream, etc. would work
out of the box.
Pete
Eugene Vedistchev
Petri Helenius wrote:
Most common acronym would be Digital Video Broadcasting,
a set of standards defining audio, video, text, teletext, data, etc.
encoding, modulation
Are you aware of the work done in various forums to address this?
For one of the more recent presentations look at
http://www.nanog.org/mtg-0306/ward.html
Pete
- Original Message -
From: Evgeny Dolgopiat [EMAIL PROTECTED]
To: [EMAIL PROTECTED]
Cc: Rogier R. Mulhuijzen [EMAIL
Is there planned support for the 82597EX (10 GbE) chipset and if there is a plan,
is that incremental development on the em driver or completely separate piece of
code? At least the linux driver seems to be separate, although the chip semantics
seem very similar of the later 8254X (1 GbE) chips.
There...
http://www.freebsd.org/cgi/query-pr.cgi?pr=i386/52835
Pete
- Original Message -
From: Robert Watson [EMAIL PROTECTED]
To: Petri Helenius [EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]; [EMAIL PROTECTED]
Sent: Thursday, May 15, 2003 9:54 PM
Subject: Re: 5.1-BETA em
Could you file
You are strongly advised to use IP addresses instead of hostnames in firewall
rulesets, to avoid DNS spoofing attacks subverting your firewall. Ideally, your
firewall should function without depending on any external network resources.
I know that, I control the domains and additionally
If your firewall needs to perform *any* DNS queries, what happens if the DNS
server(s) are down or unreachable when the firewall tries to restart? Does it
fail in a way that you are happy with?
That´s an another defect in ipfw client utility, it stops processing rules if
it fails to lookup
Changing both sides to full-duplex removes to collisions.
However: Changing only one side _always_ results in packet-loss!
It´s only when both sides transmit at once. Which is not always.
It happens almost always though.
Pete
___
[EMAIL PROTECTED]
Well, I don't see the problem.
My math says that that's .03% collision rate, which is so deep in the
noise as to be practically zero. What do you _think_ it should be?
Even Mr. Inventor of the ethernet himself regrets calling them collisions because
that term has a bad ring people
I haven't looked that deep into why, but em is quite slow on coming up
compared to
fxp for example. Probably something to do with hardware re-initialization.
Pete
Ruslan Ermilov wrote:
On Wed, May 28, 2003 at 07:57:07PM -0400, [EMAIL PROTECTED] wrote:
Hi all,
Is there a known issue with
i have yet to see a cisco ios image supporting ipv6 that was usable
in production environment. and i have tried hard.
This is getting OT but on the subject of repelling users, they´re probably
trying hard to repel their users to the vendor J boxen.
but i will admit to not having seen apollo
Any ideas if netgraph code is accounted for the swiN: net kernel process
or to the interrupt virtual process?
Also ideas what is the usual bottleneck in SMP Xeon system are appreciated,
600Mbps internet traffic seems to generate about 60% (on one of the CPUs) .
This is on -CURRENT.
The number
Any comments on the high cpu consumption of mb_free? Or any other places
where I should look to improve performance?
What do you mean high cpu consumption? The common case of mb_free()
is this:
According to profiling mb_free takes 18.9% of all time consumed in kernel and is
almost
Yes, it's normal. The commit log clearly states that the new
watermarks do nothing for now. I have a patch that changes that but I
haven't committed it yet because I left for vacation last Sunday and I
only returned early this Monday. Since then, I've been too busy to
clean up
Suggestions what would it take to make libpcap included in the FreeBSD
distribution
stop tweaking BPF buffer size by default?
tcpdump.org people have been nonresponsive about changing it there, so I
would suggest
it should be patched in FreeBSD to allow applications to control buffer
size.
Richard A Steenbergen wrote:
On Fri, Dec 13, 2002 at 12:25:42PM +0200, Petri Helenius wrote:
Suggestions what would it take to make libpcap included in the FreeBSD
distribution
stop tweaking BPF buffer size by default?
tcpdump.org people have been nonresponsive about changing it there, so I
Guy Helmer wrote:
I use sysctl debug.dbf_bufsize=131072 on my appliances to increase the
BPF buffer size to something more reasonable without having to directly
modify libpcap.
Hope you're not disappointed to find out that modifying that parameter has
no effect when using applications
I think the same applies for promiscuous mode, if the interface undergoes
configuration changes while it's in promiscuous mode and the hardware
gets reinitialized it forgets about being in that mode. This happens
at least with em.
Pete
Luigi Rizzo wrote:
[Bcc to re@ because it would be good to
David Schultz wrote:
Most kernel memory is not pageable, so swap probably won't help
you. Your `kmem_map too small' error message should report to you
the size of the attempted allocation and the size of kmem_map.
If the map really isn't full, I'm not sure why you would get this
panic, unless
David Schultz wrote:
Thus spake Petri Helenius [EMAIL PROTECTED]:
I seem to get kmem_map too small panics when using large buffers with
bpf. Is there a tunable I should be increasing?
Yes, increase KVA_PAGES in your kernel config.
I put in KVA_PAGES=1024
with following results
Read LINT (or NOTES) carefully. You can't set KVA_PAGES to 1024,
because then your kernel would take up the entire 4 GB virtual
address space. Since the kernel must fit into 4 GB alongside
every user process, that leaves you no room for programs. Try a
more reasonable value like 512 (2
With about 150M in use and KVA_PAGES undefined in config (default),
both 4.7-STABLE and 5.0-CURRENT panic (1G installed memory).
Yes, the default is 256, IIRC. That corresponds to 1 GB of KVA,
and you have only 1 GB of physical memory to back it. I take it
this is a very busy machine.
1 - 100 of 124 matches
Mail list logo
