| Age | Commit message (Collapse) | Author |
|---|
|
Since Linux kernel 2.6.29, ctnetlink reports the changes that have
been done using ctnetlink. With this patch, conntrackd can recognize
who is the origin of the event messages. For example, this is
interesting to avoid a messy implicit bulk send during the commit
of entries.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patch uses the appropriate printf format to display the
commit time taken (it was using %llu instead of %lu).
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
During the commit phase, the entries in the external cache entries
are inserted in the kernel conntrack table. Currently, we use a
fixed timeout that is specified in the config file. With this patch,
if you don't specify the fixed timeout value via CommitTimeout, the
daemon calculates the real timeout value during the commit phase.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
The lifetime feature is used by all working modes, it is useful to
know how long it has been an entry living in the cache. This patch
moves the lifetime feature to the main caching code.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
With this patch, the `-t' option adds an alarm that will flush the
cache after CONFIG(purge_timeout) seconds specified in the config
file. This looks much cleaner and more performance that looping
on the entire conntrack table to set the new timeout of every single
entry.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patch removes the clone conntrack objects created before
calling nl_*_conntrack functions since they are not required
anymore (the previous patch guarantees that objects passed as
parameter are not modified).
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patch moves the timeout mangling inside nl_*_conntrack().
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
The current commit code updates an entry it still exists in the
kernel. With this patch, we delete the entry and create a new one
to make sure that we start a clean session.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patch adds the status field to the cache object. This avoids
the (ab)use of the alarm to check if an entry is active or dead.
This is the first step to possibly move the alarm to the cache_extra
memory space of the ftfw (which is the only use by now).
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
Currently, the caching system is implemented in a two layer
architecture: hashtable (inner layer) and cache (upper layer).
This patch reworks the hash-cache infrastructure to solve some
initial design problems to make it more flexible, the main strong
points of this patch are:
* Memory handling is done in the cache layer, not in the inner
hashtable layer. This removes one of the main dependencies between
the hashtable and the cache classes.
* Remove excessive encapsulation: the former cache used to hide a
lot of details of the inner hashtable implementation.
* Fix over-hashing of some operations: lookup-delete-add required
three hash calculations. Similarly, the update-or-add operation
required two hash calculations. Now, we calculate the hash once
and re-use the value how many times as we need.
This patch simplifies the caching system. As a result, we save ~130
lines of code. Small code means and less complexity means less
chance to have bugs.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patch reports to the logfile the time taken to commit the
entries. The output is expressed in seconds.microseconds.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patch adds cache statistics that you can check via
`conntrackd -s cache'. This information is useful for
trouble-shooting.
This patch replaces several log messages that can be triggered in
runtime. The idea behind this patch is to avoid log message flooding
under errors.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patch uses a clone object that includes the original tuple and
the new timer to be set. This fixes EINVAL and EBUSY errors reporting
while trying to update the timer of some conntrack entries.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patch fixes a race condition that triggers EILSEQ errors
(wrong sequence message). The problems is triggered when the child
process resets the timers at the same time that the parent process
requests a resync. Since both the child and the parent process use
the same descriptors, the sequence tracking code in libnfnetlink
gets confused as it considers that it is receiving out of sequence
netlink messages.
This patch introduces internal handlers to commit and reset timers
so that the parent and the child do not use the same descriptors
to operate with the kernel.
This patch changes the prototype of all nf_*_conntrack() functions.
Now, the nfct handler is passed as first parameter, this change is
required to fix this problem. The rest of the changes on the API
is done for consistency.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patches fixes two problems:
- If we failt to update an entry, we remove it and try again. This
happens when we still have an entry in a final state like TIME_WAIT
while we see a new connection (SYN_SENT) with the same tuple. In
this particular case, we fail to update since some status bits are
only settable, but not unsettable.
- If we hit ETIME in an update, we have to go over the creation
patch, otherwise we hit ENOENT in the next run.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
Commit master entries before related ones to avoid ENOENT errors.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patch adds the clause PurgeTimeout that sets the new timer
when conntrackd -t is called. This command is particularly useful
when the sysadmin triggers hand-overs between several nodes without
rebooting as it reduces the timers of the remaining entries in
the kernel. Thus, avoiding clashes between new and old entries that
may trigger INVALID packets.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
Do not report ENOENT to log files, this may confuse users. There's a
race condition when shortening the timers and handling the destroy
messages. However, this problem is not serious as the point of the
shortening is to reduce the lifetime of the conntracks. If the conntrack
is dying, there's no point to shorten their lifetime anymore :)
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patch adds the new option `-t' for conntrackd. This option shortens
the value of the timeout for the cached entries that lives in the
kernel. This option is particularly useful to remove the zombie
established entries that remain in kernel if the user tests the platform
by forcing the takeover from one to another node several times.
We currently use the value of CommitTimeout which is sane for it. Adding
a new option does not seem to add more flexibility IMO.
Once we get the patches to notify user changes via ctnetlink and the
netlink flag NLM_F_ECHO works, we may directly invoke a massive purge of
the entries, however, such solution would still need evaluation.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
This patch hides information that may confuse users while they are
diagnosing problems in their setup. For example, we hide entries
that are schedule to expire - from the user side, they are already
destroyed entries; and we show in the counters the real active entries,
not all that are stored in the caches.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
Some users are reporting ETIME errors in the update. This happens
when you try to update a conntrack that is expiring. To avoid this
problem, we retry once at least. We do similar for ENOMEM errors,
although only users in virtual machines have reported this AFAIK.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
|
|
|
|
only which is incompatible AFAIK
|
|
has been Max Kellermann and has no issues with relicensing their contributions.
|
|
Simplify logging infrastructure
|
|
import only required C headers and put local headers on top to check
|
|
Fix tons of gcc warnings
|
|
add missing function prototypes
|
|
o minor irrelevant fixes for uncommon error paths and fix several typos
o use LOG_INFO for connection logging, use LOG_NOTICE for other information
o minor error handling updates
|
|
feature is particularly useful for active-active setup without connection persistency, ie. you cannot know which firewall would filter a packet that belongs to a connection.
|
|
|
|
|
|
|
|
- add len field to nethdr
- implement buffered send/recv to batch messages
- stop using netlink format for network messages: use similar TLV-based format
- reduce synchronization messages size up to 60%
- introduce periodic alive messages for sync-nack protocol
- timeslice alarm implementation: remove alarm pthread, remove locking
- simplify debugging functions: use nfct_snprintf instead
- remove major use of libnfnetlink functions: use libnetfilter_conntrack API
- deprecate conntrackd -F, use conntrack -F instead
- major rework of the network infrastructure: much simple, less messy
|
|
|
|
- remove several debug calls
- create a child to dispatch dump requests: this will help to simplify the
current locking schema. Later.
|
|
|
|
o remove reminiscents of delay destroy message and relax transitions
o remove confusing StripNAT parameter: NAT support enabled by default
o relax event tracking: *_update callbacks use cache_update_force
o use wraparound-aware functions after/before/between
o lots of cleanups
|
|
- empty debug_ct function if DEBUG_CT is not set
- revisit overrun handler: this is a hard battle, just try to do our best here, call Patrick :)
- explicit warning message when netlink_buffer_max_growth is reached
- fix silly bug in stats-mode when dumping in XML format
- fix UDP handler for conntrack
|
|
- flush nack queue in the conntrackd -f path
- do not increase add_fail counter for EEXIST errors
- cleanup sync-nack code
- improve mcast_recv_netmsg: sanity check before checksumming!
|
|
chain
|
