| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
|
|
|
|
|
|
|
|
|
| |
This recalculates the PFF upon adding a new neighbor, so that the
network reconverges faster. It is safe to do so since it is only done
by the two IPCPs that established the new flow. The PFF is also
recalculated every 4 seconds.
Signed-off-by: Sander Vrijders <[email protected]>
Signed-off-by: Dimitri Staessens <[email protected]>
|
|
|
|
|
|
|
|
|
| |
This will prevent quick calls to ipcp conn for dt flows to create
multiple join threads in parallel, instead the subsequent calls will
wait for the pending join to finish.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
This makes the buffer size used by the UNIX sockets configurable. In
case of a lot of IPCPs in the system it might become too small with
the default value, resulting in irm command failures. The user can now
easily configure it with an adequate value.
Signed-off-by: Sander Vrijders <[email protected]>
Signed-off-by: Dimitri Staessens <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
The DHT performed a lookup upon a new DT connection when it was
running for efficiency reasons. This lookup somehow cause SEGV under
stress (rapid bootstrapping of networks). This will be solved with a
full revision of the DHT implementation.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
The list of commands was not cleaned up on shutdown.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
The lookup struct uses the cookies to track pending request messages,
but they were not removed when a response is processed. Upon reuse of
a cookie for the next message, it could update the wrong lookup. This
removes the cookie for a lookup when it is looked for.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
The DHT has a thread that continued to run during shutdown. The
shutdown state is now checked for. It will not try to send messages at
shutdown.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The connmgr requires the notifier, but the notifier was created after
and destroyed before the connmgr. This is now fixed.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
If a program exits, it cleans its read buffer. However, another
process could still write a packet in that buffer, which would cause
the IPCP or IRMd to run into an assertion failure on shutdown. Setting
the rbuff to ACL_FLOWDOWN prevents this.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
This will shut down IPCPs before destroying IRMd internals. This fixes
warnings from IPCPs trying to send messages and send packets upon IRM
shutdown.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
This will unmount a stale FUSE mountpoint (an IPCP with the same pid
exited ungracefully on the system) before trying to create it.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
A previous commit tried to stat the shared memory file in case it was
already there, but stat does not seem to work for shared memory
files. This simply omits the O_EXCL attribute upon creation of the
shared memory file, since pids are unique anyway.
Signed-off-by: Sander Vrijders <[email protected]>
Signed-off-by: Dimitri Staessens <[email protected]>
|
|
|
|
|
|
|
|
|
| |
This adds a check in the irm ipcp list command to see if
irm_list_ipcps returned an error or not. Before it was only checking
if there were zero IPCPs in the system.
Signed-off-by: Sander Vrijders <[email protected]>
Signed-off-by: Dimitri Staessens <[email protected]>
|
|
|
|
|
|
|
|
|
| |
This unlinks previously allocated flow set shared memory. It could be
lingering in case an application with the same pid was SIGKILLED, and
didn't clean up its flow set shared memory.
Signed-off-by: Sander Vrijders <[email protected]>
Signed-off-by: Dimitri Staessens <[email protected]>
|
|
|
|
|
|
|
|
| |
The type check failed incorrectly if the type was specified because
the specified type was not set.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The new command 'irm ipcp list' was not being printed upon printing
the usage of 'irm ipcp'. It also fixes an unchecked return value.
Signed-off-by: Sander Vrijders <[email protected]>
Signed-off-by: Dimitri Staessens <[email protected]>
|
|
|
|
|
|
|
|
| |
The truncate was setting the length to the frame length, instead of
the actual payload length to be delivered to the N+1.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
An ethertype check was missing for the DIX ethernet IPCP, causing
crashes if there is other traffic on the network.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
This writes into the Ethernet device directly from the rdrbuff to
avoid a copy on the write side in the Ethernet IPCPs. This does not
work for the netmap device.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
This reads from the Ethernet device directly into the rdrbuff to avoid
a copy on the read side in the Ethernet IPCPs. This does not work for
the netmap device.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The signal handler is completely embedded in the source file. There
was no more need to call it from elsewhere.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
This adds a function that locks a thread to a random core. This
greatly improves performance on multi-cpu systems. There is no
portable way to do this, this only implements it for GNU/Linux.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
A bad check caused failure to set the hash algorithm for IPCPs.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The packets were being sent and read into a buffer that had the
payload length instead of the frame length.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
The unreachable statement was missed by some compilers.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
| |
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
| |
The ocbr server had a 1500 byte read buffer, which caused it to
perform partial reads on larger SDUs, slowing it down considerably
when sendin large packets. The buffer has been increased to 512KB,
partial reads disabled, and the client will give an error when larger
frames are sent. It will also warn if the size overflows (avoiding a
SEGV).
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
The list_ipcps call had a memleak in the failure case. Also fixes a
compiler warning for a possible uninitialized variable and renumbers
the gpb ipcpd message fields.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The reg/unreg API is simplified to registering and unregistering a
single name with a single IPCP. The functionality associated with
registering names was moved from the IRMd to the irm tool. The
function to list IPCPs was simplified to return all IPCPs in the
system with their basic properties needed for management.
The above changes led to some needed changes in the irm tool and the
management functions that were depending on the previous behaviour of
list_ipcps.
Command line functionality to list IPCPs in the system is also added
to the irm tool.
Some older code was refactored.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The IRMd kept a mapping from layer names to registered names, but this
is obsolete since the introduction of the query functionality.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This allows disabling partial reads. It adds a flag FLOWFRNOPART that
disables partial reads. Partial read is different from partial
delivery (FRCTFPARTIAL), which allows delivery of fragments of an
incomplete packet and thus potentially corrupted data. FLOWFRNOPART
will never deliver corrupted data (unless FRCTFPARTIAL is also set).
If FLOWFRNOPART is set and the buffer provided to flow_read is too
small for the SDU, that SDU will be discarded and -EMSGSIZE is
returned;
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The oecho tool still used its old "echo-app" name in the usage() output.
The destination name is also changed to oecho.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
| |
The read buffer had the exact length of a link-state message. With the
partial read implemented, we should then do another read() to check if
there are more parts of the message (which will return 0). To avoid
the additional read() call every time, the buffer was extended by 1
byte.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
This simplifies some functions in the rdrbuff.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
| |
The oping and ocbr servers were using non-blocking read/writes. This
caused writes to fail on high-performance tests if the buffer got
full, instead of waiting for a slot in the buffer. The write failure
caused the server to quit. This fixes the tools by setting the I/O to
blocking write and non-blocking read.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The UNIX sockets were using SOCK_STREAM, which does not preserve the
message boundaries. This switches to SOCK_SEQPACKET.
Signed-off-by: Sander Vrijders <[email protected]>
Signed-off-by: Dimitri Staessens <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This implements partial read of packets if the buffer supplied to
flow_read() is smaller than the packet in the buffer. If the number of
bytes returned by flow_read equals the size of the buffer, the next
read() will deliver the next bytes of the packet (or 0 if the packet
was exactly the size of the buffer on the previous read).
Implements #7.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
This completes the implementation of the SNDTIMEO for a blocking
write.
Fixes #6.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
This will check if the Ethertype value is a valid Ethertype in the irm
tool and the eth-dix IPCPd.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Some NICs added padding to the Ethernet II frames causing bad frame
lengths and GPB unpack fails. This adds a 2-byte length field to the
DIX frame to circumvent this.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This adds an IPC Process that uses DIX Ethernet with an Ethertype that
is configurable at bootstrap. This allows parallel DIX layers over the
same Ethernet network with different Ethertypes (and one LLC
layer). It allows jumbo frames in the future, and should avoid the
problems we have with some routers not handling LLC traffic very
well. The destination endpoint ID is sent as a 16 bit integer, so the
maximum payload is 1498 bytes in standard Ethernet, and 8998 bytes
when Jumbo frames are used.
The implementation is very similar to the Ethernet LLC IPCP, so it is
implemented using preprocessor macros in the single source instead of
duplicating code.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
| |
The rbuff uses the PTHREAD_COND_CLOCK for its condition variables, but
the flow_read was passing a time it got from the CLOCK_REALTIME_COARSE
clock. This causes the blocking reads not to timeout correctly.
The oping was updated to detect server timeouts and finish gracefully.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The number of bytes sent was not counting the data transfer PCI. This
is now fixed.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The reader thread was cancelled as soon as the writer was finished,
which resulted in missed responses and misreported packet loss.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
This adds the FLOWGRXQLEN and FLOWGTXQLEN operations to fccntl to get
the number of packets that are in the receive and transmit buffers
respectively. The flow statistics are updated to show these queue
lengths.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
There was a wrong rcv_bytes where it should have been snd_bytes.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
This fixes the access to the RIB for non-root users.
Thanks to Sitri and Dagger from the ##fuse channel for their
assistance.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
This reduces the default buffer size to 4096 blocks (about 16MB). This
will avoid problems when users try a default build on smaller machines.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
This fixes the access to an uninitialized dt_pci struct when updating
flow statistics in the case there is no next hop, which often resulted
in a segmentation fault.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The dht_get_state() function should be used to get the state of the
DHT. This fixes bug #4.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|