Issue:
For consecutive messages such as
MAC1 -> VTEP1 add
MAC1 -> VTEP2 add
MAC1 -> VTEP1 add
Final state, i.e. (MAC1 -> VTEP1 add) should be sent via FPM.
But, with current code, FPM will send (MAC1 -> VTEP2 add)
RCA:
When FPM receives (MAC1, VTEP1), it stores it in the FPM processing queue and
hash table.
When FPM receives (MAC1, VTEP2), this entry is stored as another node as hash
table key is (mac, vtep and vni)
IF FPM again receives (MAC1, VTEP1), we fetch this node in the hash table
which is already enqueued.
When the FPM queue is processed, we will send FPM message for (MAC1, VTEP1)
first and then for (MAC1, VTEP2)
This sequencing issue happened because the key of the table is (MAC, VTEP, VNI)
Fix:
Change the key of the hash table to (MAC, VNI)
So, every time we receive a new update for (MAC1, VNI1), we will find a node in
the processing queue corresponding to MAC1 if present.
We will update this same node for every operation related to (MAC1, VNI1)
Thus, at the time when FPM processes this node, it will have latest MAC1 info.
Signed-off-by: Ameya Dharkar <adharkar@vmware.com>
It's been a year search and destroy.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>
When we shut down zebra, we were not doing anything to shut
down the FPM. Perform the necessary occult rituals and
stop the threads from running during early shutdown.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
This change addresses the following :
1. Ensures zlog_debug should be under DEBUG macro check
2. Ensures zlog_err and zlog_warn wherever applicable.
3. Removed few posivite logs from fpm handling, whose frequency is high.
Signed-off-by: vishaldhingra <vdhingra@vmware.com>
We agreed on this several weeks ago on the weekly call, I just forgot to
actually put it in a PR...
A call for any Protobuf FPM users to raise their hand came up empty on
both the mailing list as well as Slack. Let's see if this gets any
response. If not, it'll be time to remove Protobuf FPM.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
- When the connection with the FPM socket is established, iterate through all the
L3VNIs and send all the RMACs for FPM processing zfpm_conn_up_thread_cb"
- We have already handled connection down even in previous commits. When the FPM
connection goes down, empty mac_q and FPM mac info hash table
"zfpm_conn_down_thread_cb"
Signed-off-by: Ameya Dharkar <adharkar@vmware.com>
- FPM write thread calls "zfpm_build_updates()" to process mac_q and dest_q and
to write update buffer over the FPM socket.
- "zfpm_build_updates()" processes all the update queues one by one in a while
loop. It will break the while loop and return if Queue processing function
returns "FPM_WRITE_STOP" OR FPM write buffer is full OR all the queues are
empty (no more update to process).
- "zfpm_build_route_updates()" dequeues and processes route nodes from "dest_q".
- "zfpm_build_mac_updates()" dequeues and processes MAC nodes from "mac_q"
- These queue processing functions return with "FPM_WRITE_STOP" if the write
buffer is full. Return value is "FPM_GOTO_NEXT_Q" if enough updates are
processed from this queue and we want to move on to the next queue.
- In each call, a queue processing function will process max
"FPM_QUEUE_PROCESS_LIMIT (10000)" updates to avoid starvation of other queues.
Signed-off-by: Ameya Dharkar <adharkar@vmware.com>
- Define a hook "zebra_mac_update" which can be registered by multiple
data plane components (e.g. FPM, dplane).
DEFINE_HOOK(zebra_rmac_update, (zebra_mac_t *rmac, zebra_l3vni_t *zl3vni, bool
delete, const char *reason), (rmac, zl3vni, delete, reason))
- While performing RMAC add/delete for an L3VNI, call "zebra_mac_update" hook.
- This hook call triggers "zfpm_trigger_rmac_update". In this function, we do a
lookup for the RMAC in fpm_mac_info_table. If already present, this node is
updated with the latest RMAC info. Else, a new fpm_mac_info_t node is created
and inserted in the queue and hash data structures.
Signed-off-by: Ameya Dharkar <adharkar@vmware.com>
- FPM MAC structure: This data structure will contain all the information
required for FPM message generation for an RMAC.
struct fpm_mac_info_t {
struct ethaddr macaddr;
uint32_t zebra_flags; /* Could be used to build FPM messages */
vni_t vni;
ifindex_t vxlan_if;
ifindex_t svi_if; /* L2 or L3 Bridge interface */
struct in_addr r_vtep_ip; /* Remote VTEP IP */
/* Linkage to put MAC on the FPM processing queue. */
TAILQ_ENTRY(fpm_mac_info_t) fpm_mac_q_entries;
uint8_t fpm_flags;
};
- Queue structure for FPM processing:
For FPM processing, we build a queue of "fpm_mac_info_t". When RMAC is
added or deleted from zebra, fpm_mac_info_t node is enqueued in this queue
for the corresponding operation. FPM thread will dequeue these nodes one by
one to generate a netlink message.
TAILQ_HEAD(zfpm_mac_q, fpm_mac_info_t) mac_q;
- Hash table for "fpm_mac_info_t"
When zebra tries to enqueue fpm_mac_info_t for a new RMAC add/delete
operation, it is possible that this RMAC is already present in the queue. So,
to avoid multiple messages for duplicate RMAC nodes, insert fpm_mac_info_t
into a hash table.
struct hash *fpm_mac_info_table;
- Before enqueueing any MAC, try to fetch the fpm_mac_info_t from the hash
table first.
- Entry is deleted from the hash table when the node is dequeued.
- For hash table key generation, parameters used are "mac adress" and "vni"
This will provide a fairly unique key for a MAC(fpm_mac_info_hash_keymake).
- Compare function uses "mac address", "RVTEP address" and "VNI" as the key
which is sufficient to distinguish any two RMACs. This compare function is
used for fpm_mac_info_t lookup (zfpm_mac_info_cmp).
Signed-off-by: Ameya Dharkar <adharkar@vmware.com>
When we receive a read failure in handling a FPM read
let's add a bit more information to what we think has
gone wrong, in a hope that debugging will be a bit easier.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
The SELECTED_FIB flag was placed upon the entry that we
have inserted into the kernel. Remove this flag and replace
with a `rib_dest_t` *selected_fib. Just keep track of the
selected_fib as we modify it. This removes allot of
FOREACH_RE loops as that we do not need to find the
entry anymore.
At this point in time I think this is a very minor performance
boost. Most `rib_dest_t` structures do not typically carry
more than 1 route_entry, but the minute you start having more
than one entry you can and will start having significant processing
time spent finding the selected_fib.
A future commit may re-order the route entries and possibly
keep more pointers on `rib_dest_t` to avoid lookup. This
is a bit tricky because of the FIB_OVERRIDE code.
Signed-off-by Donald Sharp <sharpd@cumulusnetworks.com>
Some of the deprecated stream.h macros see such little use that we may
as well just remove them and use the non-deprecated macros.
Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>
This fixes the broken indentation of several foreach loops throughout
the code.
From clang's documentation[1]:
ForEachMacros: A vector of macros that should be interpreted as foreach
loops instead of as function calls.
[1] http://clang.llvm.org/docs/ClangFormatStyleOptions.html
Signed-off-by: Renato Westphal <renato@opensourcerouting.org>
it's just an alias for a millisecond timer used in exactly nine places
and serves only to complicate
Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>
The 'struct rib' data structure is missnamed. It really
is a 'struct route_entry' as part of the 'struct route_node'.
We have 1 'struct route_entry' per route src. As such
1 route node can have multiple route entries if multiple
protocols attempt to install the same route.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
The FSF's address changed, and we had a mixture of comment styles for
the GPL file header. (The style with * at the beginning won out with
580 to 141 in existing files.)
Note: I've intentionally left intact other "variations" of the copyright
header, e.g. whether it says "Zebra", "Quagga", "FRR", or nothing.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Pass pointer to pointer instead of assigning by return value. See
previous commit message.
To ensure that the behavior stays functionally correct, any assignments
with the result of a thread_add* function have been transformed to set
the pointer to null before passing it. These can be removed wherever the
pointer is known to already be null.
Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>
The way thread.c is written, a caller who wishes to be able to cancel a
thread or avoid scheduling it twice must keep a reference to the thread.
Typically this is done with a long lived pointer whose value is checked
for null in order to know if the thread is currently scheduled. The
check-and-schedule idiom is so common that several wrapper macros in
thread.h existed solely to provide it.
This patch removes those macros and adds a new parameter to all
thread_add_* functions which is a pointer to the struct thread * to
store the result of a scheduling call. If the value passed is non-null,
the thread will only be scheduled if the value is null. This helps with
consistency.
A Coccinelle spatch has been used to transform code of the form:
if (t == NULL)
t = thread_add_* (...)
to the form
thread_add_* (..., &t)
The THREAD_ON macros have also been transformed to the underlying
thread.c calls.
Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>
The next patch will bulk-replace these, and monotime() has time_t as
return value, not an error indicator.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
There's no need to duplicate the 'vrf_id' and 'name' fields from the 'vrf'
structure into the 'zebra_vrf' structure. Instead of that, add a back
pointer in 'zebra_vrf' that should point to the associated 'vrf' structure.
Additionally, modify the vrf callbacks to pass the whole vrf structure
as a parameter. This allow us to make further simplifications in the code.
Signed-off-by: Renato Westphal <renato@opensourcerouting.org>
