summaryrefslogtreecommitdiff
Commit message (Collapse)AuthorAgeFilesLines
* unicast: Fix RIB stats for congestion avoidanceDimitri Staessens2021-07-101-2/+2
| | | | | | | The upstream/downstream stats were switched. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* lib: Don't initialize process RIB for IPCPsDimitri Staessens2021-07-102-10/+5
| | | | | | | | | | | | | This will skip rib_init() at __init() for IPCPs (or at least, processes that have "ipcpd" in the executable name). The previous code tried to unmount the generic mount and then remount under the ipcp name, but it often failed because fuse_mount() is asynchronous and the mount was not up at the time of the unmount() call. Renaming the mount instead of unmounting failed for the same reason. This is a better fix for now. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* lib: Application RIB with FRCT statisticsDimitri Staessens2021-06-305-17/+173
| | | | | | | | | | Application flows can now be monitored from the RIB, exposing FRCT statistics (window edges, retransmission timeout, rtt estimate, etc). Application RIB requires user permissions to be able to access /dev/fuse. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Fix layer info RIBDimitri Staessens2021-06-301-3/+8
| | | | | | | | Layer info was not converted to parse the full path with the latest RIB change. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* irmd: Fix unused variable in builds without fuseDimitri Staessens2021-06-301-1/+2
| | | | | | | The mode_t mask wasn't guarded by ifdefs. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* lib: Pass full path for RIB entriesDimitri Staessens2021-06-295-18/+42
| | | | | | | | | | | | | | | | | | | | | | | | | | | The read functions for the RIB will now receive the full path, instead of only the entry name. For IPCPs, we organized the RIB in an /<ipcp>/<component>/entries structure with a directory per component, so we don't need the full path at this point. For process flow information, it's a lot more convenient to organize it the following way /<pid>/<fd>/stat We can then register/unregister the flow descriptor when the frct instance is created, and for getting the stats, we'd know the flow descriptor from the fuse file path. If we would create a file per flow instead of a directory per flow, something like /<pid>/flows/<fd> we'd need to do additional bookkeeping to list the contents of that directory (we would need to track all flows with an active FRCT instance), that fuse knows because it tracks the directories. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* irmd: Fix setting mode on RIB directoryDimitri Staessens2021-06-291-0/+5
| | | | | | | | The irmd was creating the directory with the wrong mode set because the default mask (022) was not cleared. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Fix unregistering fa and dt RIBsDimitri Staessens2021-06-282-10/+13
| | | | | | | These RIBs were not properly unregistered on shutdown. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* lib: Remove struct stat from RIB APIDimitri Staessens2021-06-286-90/+73
| | | | | | | | | | The RIB API had a struct stat in the getattr() function, which made all components that exposed variables via the RIB dependent on <sys/stat.h>. The rib now has its own struct rib_attr to set attributes such as size and last modified time. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* build: Fix compilation with fuse (RIB) on FreeBSDDimitri Staessens2021-06-285-7/+21
| | | | | | | | | | | Compilation failed on FreeBSD 14 with fuse enabled because of some missing definitions. __XSI_VISIBLE must be set before including <ouroboros/rib.h> for some definitions in <sys/stat.h>. FreeBSD doesn't know the MSG_CONFIRM flag to sendto() or CLOCK_REALTIME_COARSE, which are Linux-specific. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Fix missing return statement0.18.2Dimitri Staessens2021-06-232-1/+3
| | | | | | | | | The fa_handle_packet function loop is non-void but didn't have a return statement. Only got picked up if I build from AUR, which is weird. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* lib, ipcpd, irmd: Wrap pthread unlocks for cleanupDimitri Staessens2021-06-2327-123/+136
| | | | | | | | | | | | This add an ouroboros/pthread.h header that wraps the pthread_..._unlock() functions for cleanup using pthread_cleanup_push() as this casting is not safe (and there were definitely bad casts in the code). The close() function is now also wrapped for cleanup in ouroboros/sockets.h. This allows enabling more compiler checks. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Add missing default for switch in shim-dataDimitri Staessens2021-06-231-0/+3
| | | | | | | | There was a switch-case without a default, causing some compiler warnings. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* lib: Bypass assertion in shm_rdrbuff0.18.1Dimitri Staessens2021-06-212-2/+2
| | | | | | | | | This assert() causes ipcpd and subsequent irmd abort() when shutting down debug builds. Should be fixed some day when other components are more robust (frct retransmissions and routing). Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Disable core lock by defaultDimitri Staessens2021-06-211-1/+1
| | | | | | | | This was beneficial on dual-XEON CPU systems, but on most hardware, this kills performance when running multiple IPCPs on a single CPU. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Fix buffer overflow in flow allocator RIB.Dimitri Staessens2021-06-211-3/+3
| | | | | | | | Max value of UINT64 can be 20 characters when printed, need an extra byte for sprintf trailing '\0'. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Fix congestion window scalingDimitri Staessens2021-06-211-18/+41
| | | | | | | This fixes scaling of the congestion window, which was a buggy mess. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Move RIB initialization to common groundDimitri Staessens2021-06-218-32/+167
| | | | | | | | | | | | | | This moves Resource Information Base (RIB) initialization into the ipcp_init() function, so all IPCPs initialize a RIB. The RIB not shows some common IPCP information, such as the IPCP name, IPCP state and the layer name if the IPCP is part of a layer. The initialization of the hash algorithm and layer name was moved out of the common ipcp source because IPCPs may only know this information after enrollment. Some IPCPs were not even storing this information. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* tools: Support broadcast IPCP in irm ipcp listDimitri Staessens2021-06-181-5/+10
| | | | | | | | The broadcast IPCP was missing in the known types, causing the type to show as UNKNOWN. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* build: Fix version in CMakeLists.txtDimitri Staessens2021-06-181-2/+2
| | | | | | | | The tag was set to 0.18, but the version was still at 17.5 in CMake. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Fix free of uninitialized variable in graph.cDimitri Staessens2021-05-282-5/+1
| | | | | | | | | The new GCC 11.1 compiler discovered s_dist would be uninitialized with an unknown policy, so it doesn't need to be free'd. Also removes some unneeded includes in broadcast dt.c that I had pending. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* irmd: Remove unused variable in prog_tableDimitri Staessens2021-04-133-41/+3
| | | | | | | This removes a program name variable that was not used anymore. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* build: Move connmgr and enrolment to common groundDimitri Staessens2021-03-2821-1890/+944
| | | | | | | | | | The connection manager and enrolment components of the unicast and broadcast IPCP have a lot in common, as conjectured in the paper. The initial implementation of the broadcast IPCP just duplicated the code. This moves the shared functionality to common ground. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* build: Remove raptor IPCPDimitri Staessens2021-03-2811-1208/+8
| | | | | | | | | | This removes the raptor IPCP. The code hasn't been updated for a while, and wouldn't compile. Raptor served its purpose as a PoC for Ouroboros-over-Ethernet-Layer-1, but giving the extreme niche hardware needed to run it, it's not worth maintaining this anymore. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Don't include enroll.h in connmgrDimitri Staessens2021-02-251-1/+0
| | | | | | | It's not needed to include enroll.h. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Single UDP port for the ipcpd-udp0.18.0Dimitri Staessens2021-01-036-180/+85
| | | | | | | | | | | | | | | | | | The UDP layer will now use a single (configurable) UDP port, default 3435. This makes it easer to allocate flows as a client from behind a NAT firewall without having to configure port forwarding rules. So basically, from now on Ouroboros traffic is transported over a bidirectional <src><port>:<dst><port> UDP tunnel. The reason for not using/allowing different client/server ports is that it would require reading from different sockets using select() or something similar, but since we need the EID anyway (mgmt packets arrive on the same server UDP port), there's not a lot of benefit in doing it. Now the operation is similar to the ipcpd-eth, with the port somewhat functioning as a "layer name", where in UDP, the Ethertype functions as a "layer name". Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* build: Update email addressesDimitri Staessens2021-01-03206-412/+412
| | | | | | | | The ugent email addresses are shut down, updated to Ouroboros mail addresses. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* build: Update copyright to 2021Dimitri Staessens2021-01-03206-206/+206
| | | | | | | Happy New Year, Ouroboros! Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Pass qoscube to ECN marking functionDimitri Staessens2020-12-208-3/+13
| | | | | | | | The ECN marking function should be able to use the packet QoS to allow prioritizing traffic under congestion. Not yet implemented in MB-ECN. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Remove unused variable in MB-ECN policyDimitri Staessens2020-12-121-4/+0
| | | | | | | | The t_sent variable is a remnant from the first version and isn't needed anymore. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Pass previous ECN value in congestion APIDimitri Staessens2020-12-128-18/+30
| | | | | | | | The previous value of the ECN field should be passed to the congestion notification function. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* lib: Fix flow_accept without opensslDimitri Staessens2020-12-122-5/+7
| | | | | | | | DH key creation was returning -ECRYPT if opennssl is not installed, instead of success (0). Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* lib: Fix return value in function returning voidDimitri Staessens2020-12-121-1/+1
| | | | | | | This causes builds to fail on systems where OpenSSL is not available. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Use 64-bit flow endpoint IDs for DTDimitri Staessens2020-12-076-54/+117
| | | | | | | | | | | | | The EIDs are now 64-bit. This makes it a tad harder to guess them (think of port scanning). The implementation has only the most significant 32 bits random to quickly map EIDs to N+1 flows. While this is equivalent to a random cookie as a check on flows, the rationale is that valid endpoint IDs should be pretty hard to guess (and thus be 64-bit random at least). Ideally one would use content-addressable memory for this kind of mapping. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Fix slow start in multi-bit (F)ECN policyDimitri Staessens2020-12-071-8/+11
| | | | | | | | | | | There is a check not to rapidly double the window to astronomical sizes when there is no congestion experienced for long periods of time, but the if-else logic was botched and it still grew to astronomical sizes (albeit linear instead of exponential). I also lowered the ECN threshold a bit. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Remove DT-FA bypass on receiver sideDimitri Staessens2020-12-075-63/+88
| | | | | | | | | | | | | | | | | | | | | | | | | | The DT will now post all packets for N+1 flows through the flow allocator component. This means that N+1 flows can be monitored through the flow allocator stats, and N-1 flows through the DT stats. The DT component still keeps stats for the local components (FA and DHT), but this can be removed once the DHT has its own RIB output. The flow allocator show statistics for Sent packets: total packets that were presented for sending on this specific flow Send failed: packets that were unable to be sent Received packets: total packets that were presented by the DT component on this specific flow Received failed: packets that were unable to be delivered These stats are presented as both packet counts and byte counts. To know how many were successful, the values for failed need to be subtracted from the values for total. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Fix off-by-one in Multi-bit (F)ECN policyDimitri Staessens2020-12-051-2/+1
| | | | | | | | Noticed an off-by-one in the packet counter because it was incremented before and the byte counter after the flow update. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Add RIB statistics for flow allocatorDimitri Staessens2020-12-058-12/+255
| | | | | | | | | | The RIB will now show some stats for the flow allocator, including congestion avoidance statistics. This is needed before decoupling the data transfer component and the flow allocator as some current stats show in DT will move to FA. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Simplify multi-bit (F)ECN policyDimitri Staessens2020-12-021-54/+69
| | | | | | | | | | | | The mb-ecn policy has a couple of divisions in the math, which I wanted to avoid. Now it measures the number of bytes sent in a window, and updates the next window with AIMD logic. If the number of bytes in the window is reached, the call blocks. To avoid long packet bursts, the window size continually scales to contain between CA_MINPS (8) and CA_MAXPS (64) packets. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Don't update for deallocated flowsDimitri Staessens2020-12-021-0/+5
| | | | | | | | | | | The dt component bypasses the flow allocator on the receiver side, and may try to update congestion context when the flow has already been deallocated by the receiver. I will fix this bypass and always pass through the flow allocator sometime soon; for now, I added a check in the flow allocator call to avoid the SEGV. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Fix enrollment for congestion avoidanceDimitri Staessens2020-12-021-0/+3
| | | | | | | | The enrollment procedure was not passing the policy for congestion avoidance. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* ipcpd: Add congestion avoidance policiesDimitri Staessens2020-12-0221-140/+957
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This adds congestion avoidance policies to the unicast IPCP. The default policy is a multi-bit explicit congestion avoidance algorithm based on data-center TCP congestion avoidance (DCTCP) to relay information about the maximum queue depth that packets experienced to the receiver. There's also a "nop" policy to disable congestion avoidance for testing and benchmarking purposes. The (initial) API for congestion avoidance policies is: void * (* ctx_create)(void); void (* ctx_destroy)(void * ctx); These calls create / and or destroy a context for congestion control for a specific flow. Thread-safety of the context is the responsability of the flow allocator (operations on the ctx should be performed under a lock). ca_wnd_t (* ctx_update_snd)(void * ctx, size_t len); This is the sender call to update the context, and should be called for every packet that is sent on the flow. The len parameter in this API is the packet length, which allows calculating the bandwidth. It returns an opaque union type that is used for the call to check/wait if the congestion window is open or closed (and allowing to release locks before waiting). bool (* ctx_update_rcv)(void * ctx, size_t len, uint8_t ecn, uint16_t * ece); This is the call to update the flow congestion context on the receiver side. It should be called for every received packet. It gets the ecn value from the packet and its length, and returns the ECE (explicit congestion experienced) value to be sent to the sender in case of congestion. The boolean returned signals whether or not a congestion update needs to be sent. void (* ctx_update_ece)(void * ctx, uint16_t ece); This is the call for the sending side top update the context when it receives an ECE update from the receiver. void (* wnd_wait)(ca_wnd_t wnd); This is a (blocking) call that waits for the congestion window to clear. It should be stateless (to avoid waiting under locks). This may change later on if passing the context is needed for different algorithms. uint8_t (* calc_ecn)(int fd, size_t len); This is the call that intermediate IPCPs(routers) should use to update the ECN field on passing packets. The multi-bit ECN policy bases the value for the ECN field on the depth of the rbuff queue packets will be sent on. I created another call to grab the queue depth as fccntl is write-locking the application. We can further optimize this to avoid most locking on the rbuff. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* tools: Default ocbr to sleep and add --spin optionDimitri Staessens2020-12-021-1/+4
| | | | | | | | | | The ocbr client was spinning the CPU by default, which made sense on lab servers with dual xeons, but not so much for average users. Now sleeping becomes the default. Busy waiting can be enabled using --spin if needed. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* tools: Use read timeouts in ocbr serverDimitri Staessens2020-11-251-1/+3
| | | | | | | | | The ocbr server was using non-blocking reads (probably because we didn't have read timeouts when we wrote it) and was using a whole CPU core per thread. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* lib: Reduce timerwheel CPU consumptionDimitri Staessens2020-11-252-1/+10
| | | | | | | | | | | The timerwheel is checked during IPC calls (fevent, flow_read), causing huge load on CPU consumption in IPCPs, since they have a lot of fevent() threads for QoS. The timerwheel will need further optimization), but for now I reduced the default tick time to 5 ms and added a boolean to check that the wheel is actually used. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* irmd: Fix data race in flow allocationDimitri Staessens2020-11-251-73/+100
| | | | | | | | | | | The flow information in the main loop is passed as a direct pointer to an irm_flow object in the flow database. This was (probably) not really an issue due to how the flow allocation operations work, but the thread sanitizer was barfing a lot of (correct) data race errors when running bigger tests, so now makes a safe copy of the data. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* lib: Don't default to lockless rbuffDimitri Staessens2020-11-221-1/+1
| | | | | | | | I mistakenly set the default to the (buggy) lockless rbuff implementation instead of the pthread one in commit 3aec660e. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* Revert "lib: Unmount stale RIB directories"Sander Vrijders2020-11-111-10/+1
| | | | | | | | This reverts commit 978266fe4beba21292daad2d341fe5ff22e08aba. We were incorrectly unmounting the directory under normal conditions. Signed-off-by: Sander Vrijders <[email protected]> Signed-off-by: Dimitri Staessens <[email protected]>
* ipcpd: Refactor DT componentDimitri Staessens2020-11-111-76/+38
| | | | | | | The flow stats had quite a lot of duplication. Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
* lib: Add Rendez-Vous mechanism for flow controlDimitri Staessens2020-10-113-31/+130
| | | | | | | | | | | | | This adds the rendez-vous mechanism to handle the case where the sending window is closed and window updates get lost. If the sending window is closed, the sender side will send an RDVS every DELT_RDV time (100ms), and give up after MAX_RDV time (1 second). Upon reception of a RDVS packet, a window update is sent immediately. We can make this much more configurable later on (build options for defaults, fccntl for runtime tuning). Signed-off-by: Dimitri Staessens <[email protected]> Signed-off-by: Sander Vrijders <[email protected]>
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-15 07:45:17 +0000