| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This adds flow liveness monitoring for flows, with a fixed timeout of
120s. I will make it configurable at flow allocation later on (timeout
needs to be communicated to the peer). If one peer dies, or doesn't
call any IPC calls (flow_write/flow_read/fevent) it will stop sending
keepalives and the other peer's read/writes will error on an
-EFLOWDOWN after the timeout expires.
Packets without a payload (0 length packets) are interpreted as
keepalive packets for the flow. They can be sent from any application,
but they will not trigger a message read at the receiver side (0 as a
return value on flow_read indicates a previous partial read has
completed at exactly the buffer size).
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
The flow_set will now keep a list of the flows in the set, this makes
it more efficient to iterate over the flows. Extending the public API
for fset_t with an iterator will also be useful.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
When handling management frames, there was a cancellation point after
the unlock, which would cause the cleanup handler to attempt a double
unlock if the thread was cancelled at that point.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
The blocking read from the rbuff was not correctly handling flow down
states, returning a valid index. The attempt to fetch the header then
failed on an assertion. The blocking read will now return -EFLOWDOWN
if the flow is marked down by the IPCP.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
Small refactor taking the wait for the flow allocation to complete
out.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
This refactors the single long function that handles incoming packets
destined for the flow allocator.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
The pft hash function assumed mem_hash allocates memory, but it does
not. There was also a memcpy with potentially overlapping memory
regions, which is undefined behaviour.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
The dht_del function was called under lock in dht_unreg, and then
tried to take the lock again, a 100% deadlock. Also fix uninitialized
value in dht_retrieve.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
Less code, and less chance of a collision.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
The count value could be IPCP_UDP_BUF_SIZE, overflowing buf.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
Probably a leftover from previous shutdown logic.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
Better to check the error code than the out parameter.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
Off-by-one error in the bounds check.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Unlocking the flows while iterating could cause a modification during
the iteration. Added pthread_cleanup handlers as the thread could get
cancelled while holding a lock.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
The width of the Kademlia hash function (dht->b) was set only after
the ID was created. This should have failed miserably, but the bytes
after were fine as they were just a randomized ID in the Kademlia
network. Nasty.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The argvdup function didn't handle the case where argc is 0.
Small refactor that also handles this case correctly.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
The free of the buffer in the failure path of the readdir RIB
functions was taking the wrong pointer in a couple of places. The FRCT
RIB readdir was missing error handling for malloc and strdup.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The fccntl call FRCTSFLAGS was using a pointer to a flags so set
flags, which should just be a regular uint16_t.
For instance, the FRCTLINGER flags can now be turned off using
fccntl(fd, FRCTSFLAGS, FRCTFRESCNTL | FRCTFRTX)
leaving only resource control (flow control, FRCTFRESCNTL) and
retransmission enabled. Note that retransmission (FRCTFRTX) can't be
enabled or disabled on a live flow, it will be set on flow allocation.
Updates the man page for fccntl to add these FRCT options.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The multipath pff entry was modified (rotated) under a read lock,
which is now changed to a write lock.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
To allow merging large network layers, a situation will arise where
multiple directories need to coexist within the layer. This reverts
commit 9422e6be94ac1007e8115a920379fd545055e531.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
It was taking a write lock when a read lock was sufficient.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
This is a fix to wait for outstanding retransmissions when a flow is
deallocated. Instead of waiting the full timeout, it will now wait in
the same tic increments used within FRCT. Bit of a stopgap at the
moment, FRCT and the flows are in need of a serious refactor.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
There was a missing unlock in FRCT. Also fixes some indentation.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
The timeout variable was not correctly passed to the IPCP, causing
flow IDs to be reused immediately instead of waiting for the full
Delta-t to expire. This caused all kinds of havoc with retransmissions
in reliable flows.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
If the timeout is already expired, the wait variable would be negative
and return a negative value for the __frcti_dealloc function, thinking
that the timeout was not expired causing an unnecessary wait even if
all packets are acknowledged.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
| |
The initial sender right window edge (indicating acknowledged packet
sequence number) was initialized to seqno - 1. This should be the same
as seqno, since we acknowledge with the next expected sequence number.
It also indicates that a flow without traffic has no outstanding
acknowledgements.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
The dt component had init/start commands somewhat outside of the
overall flow of startup of the unicast IPCP. This was probably some
old code and wasn't needed.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
Each policy folder will now have a pol.h file, so that (in most cases)
adding a new policy only requires changes inside the policy folder.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The DHT is now a proper directory policy instead of a unicast IPCP
component.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
This makes the DHT a single directory implementation and moves it to
the stack (init/fini instead of create/destroy). This is a step
towards making it a directory policy, in line with our other policy
implementations.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The policies were all in a single folder pol/, and have been moved to
a folder per component/mechanism to keep things a bit more orderly.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
There was some leftover code in dev.c wrt to the process RIB that is
not needed anymore.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
Some cases were not guarded by explicit fallthrough where needed.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Arithmetic with NULL pointers is undefined behaviour. Caught by clang
13. Fixed by using uintptr_t, which is guaranteed to be the size of a
pointer.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
On more recent Mac OS X (>= Catalina) the system volume is read-only,
even with System Integrity Protection disabled. This requires an
additional remount step to install something into /usr. Probably
better to play by their rules and default the install to /usr/local.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
The last couple of commits fix compilation on OS X. So bumping the
patchlevel makes sense to fix compilation for the master branch as
well.
Signed-off-by: Sander Vrijders <[email protected]>
Signed-off-by: Dimitri Staessens <[email protected]>
|
|
|
|
|
|
|
|
| |
The print output for MB-ECN had a size_t conversion to uint64_t that
some compilers complain about.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
prctl.h is linux only.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Apparently that function isn't implemented on some versions of OS
X. On these systems, we can just use sigwait, but now the IPCP will
also accept signals not coming from the IRMd.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The logic for additive increase was botched. It was adding to the
current window limit, and to avoid a count-to-infinity when sending
below the limit, I added a check that the application was trying to
send more than the current limit. This fails in congestion avoidance
mode when the IPCP is throttling traffic below the limit; causing the
app to never increase the congestion window (and even worse, to keep
decreasing in some cases).
The Additive Increase will now always add bandwidth to the latest
sending rate instead of the window bandwidth limit, avoiding all the
problems.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The u_snd and u_rcv variables were not guarded by ifdefs when
updating.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The RIB of the flow allocator will now export the number of flow
updates sent/received per flow.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
| |
This will reduce the linux high resolution slack timer in IPCPs. Linux
default for userspace processes is 50us. It is configurable at build
using IPCP_LINUX_SLACKTIMER_NS. Default is now 1us.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
| |
The rate was supposed to be 1 update per 8 data packets, but the
calculation was doing 1 update per 4 data packets.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
| |
The upstream/downstream stats were switched.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
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]>
|
|
|
|
|
|
|
|
|
|
| |
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]>
|
|
|
|
|
|
|
|
| |
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]>
|
|
|
|
|
|
|
| |
The mode_t mask wasn't guarded by ifdefs.
Signed-off-by: Dimitri Staessens <[email protected]>
Signed-off-by: Sander Vrijders <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
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]>
|