sliver-openvswitch.git
13 years agodpif: Use sset instead of svec in dpif interface.
Ben Pfaff [Fri, 25 Mar 2011 20:00:13 +0000 (13:00 -0700)]
dpif: Use sset instead of svec in dpif interface.

13 years agoConvert shash users that don't use the 'data' value to sset instead.
Ben Pfaff [Fri, 25 Mar 2011 22:26:30 +0000 (15:26 -0700)]
Convert shash users that don't use the 'data' value to sset instead.

In each of the cases converted here, an shash was used simply to maintain
a set of strings, with the shash_nodes' 'data' values set to NULL.  This
commit converts them to use sset instead.

13 years agosset: New data type for a set of strings.
Ben Pfaff [Wed, 30 Mar 2011 20:44:10 +0000 (13:44 -0700)]
sset: New data type for a set of strings.

Many uses of "shash" or "svec" data structures really call for a "set of
strings" data type.  This commit introduces such a data structure.  Later
commits convert inappropriate uses of shash and svec to use sset instead.

13 years agolearning-switch: Remove dead assignment.
Ethan Jackson [Tue, 29 Mar 2011 19:37:49 +0000 (12:37 -0700)]
learning-switch: Remove dead assignment.

13 years agoovs-ofctl: Remove dead assignment.
Ethan Jackson [Thu, 31 Mar 2011 23:38:56 +0000 (16:38 -0700)]
ovs-ofctl: Remove dead assignment.

13 years agonetdev-linux: Remove dead assignments.
Ethan Jackson [Tue, 29 Mar 2011 19:25:17 +0000 (12:25 -0700)]
netdev-linux: Remove dead assignments.

13 years agoofproto: Use new timer library.
Ethan Jackson [Thu, 31 Mar 2011 20:55:23 +0000 (13:55 -0700)]
ofproto: Use new timer library.

13 years agocfm: Use new timer library.
Ethan Jackson [Thu, 31 Mar 2011 20:54:44 +0000 (13:54 -0700)]
cfm: Use new timer library.

13 years agolacp: Use new timer library.
Ethan Jackson [Thu, 31 Mar 2011 20:54:15 +0000 (13:54 -0700)]
lacp: Use new timer library.

13 years agolib: Create new timer library.
Ethan Jackson [Thu, 31 Mar 2011 20:46:04 +0000 (13:46 -0700)]
lib: Create new timer library.

Scattered throughout the code base we use long integers to
implement timers.  When the result of timer_msec() is greater than
the time stored, we preform some action.

This commit creates a new timer library intended to replace these
manually managed timers.  Code using the timer library will be more
obviously correct, and more consistent with other code using the
library.

13 years agocfm: Fix appctl negative report.
Ethan Jackson [Thu, 31 Mar 2011 23:12:01 +0000 (16:12 -0700)]
cfm: Fix appctl negative report.

When the cfm module has never received a bad CCM message, it would
report a negative time.

13 years agobridge: Destroy ofproto before deleting dpif.
Ben Pfaff [Thu, 31 Mar 2011 21:12:57 +0000 (14:12 -0700)]
bridge: Destroy ofproto before deleting dpif.

Otherwise the ofproto's attempt to flush flows from the dpif will fail with
an error, causing a spurious log message.

13 years agoconnmgr: Fix wild pointer dereference in connmgr_broadcast().
Ben Pfaff [Thu, 31 Mar 2011 21:13:36 +0000 (14:13 -0700)]
connmgr: Fix wild pointer dereference in connmgr_broadcast().

Fixes a segfault when fail-open goes into effect.

13 years agoofproto: Fix order of destruction in ofproto_destroy().
Ben Pfaff [Thu, 31 Mar 2011 21:11:57 +0000 (14:11 -0700)]
ofproto: Fix order of destruction in ofproto_destroy().

ofproto_flush_flows() calls into the connmgr (via connmgr_flushed()) so
it must be called before destroying the connmgr to avoid a use-after-free
error.

Bug #5231.
Reported-by: Krishna Miriyala <krishna@nicira.com>
13 years agodatapath: Update for changes in 2.6.39-rc1
Simon Horman [Thu, 31 Mar 2011 07:32:07 +0000 (16:32 +0900)]
datapath: Update for changes in 2.6.39-rc1

Update for flowi4 and ip_route_output_flow() changes
in 2.6.39-rc1.

Signed-off-by: Simon Horman <horms@verge.net.au>
[Jesse: drop redundant unlikely() from IS_ERR()]
Signed-off-by: Jesse Gross <jesse@nicira.com>
13 years agoofpbuf: Make ofpbufs initialized with ofpbuf_use_stack() not expandable.
Ben Pfaff [Wed, 2 Mar 2011 21:39:59 +0000 (13:39 -0800)]
ofpbuf: Make ofpbufs initialized with ofpbuf_use_stack() not expandable.

My original intent for ofpbufs initialized with ofpbuf_use_stack() was that
the caller was providing enough space on the stack for the common case,
with dynamic allocation as a fallback.  But in practice, none of the
clients actually do this.  Instead, all of them actually know that the
stack-allocated buffer is big enough and, since they don't want to bother
with having to call ofpbuf_delete(), they instead assert that the buffer
wasn't reallocated.

Since this is a bit of a pain, this commit changes the semantics of
ofpbuf_use_stack() to be that the stack-allocated buffer cannot be
reallocated at all.  This is more convenient for the existing clients.

13 years agoodp-util: Replace ODPUTIL_FLOW_KEY_U32S by new struct odputil_keybuf.
Ben Pfaff [Wed, 2 Mar 2011 21:25:10 +0000 (13:25 -0800)]
odp-util: Replace ODPUTIL_FLOW_KEY_U32S by new struct odputil_keybuf.

This seems to me to better encapsulate the inherent ugliness.

13 years agodatapath: Fix mysterious GRE-over-IPSEC problems.
Ben Pfaff [Wed, 30 Mar 2011 21:54:26 +0000 (14:54 -0700)]
datapath: Fix mysterious GRE-over-IPSEC problems.

We've noticed that packets that go up to userspace and then back down to
the kernel and then enter an GRE tunnel that is then ESP encapsulated
by IPSEC end up with a bad ESP "next header" value: it ends up as zero
instead of 0x2f (IPPROTO_GRE).  Just putting packets from userspace into
a freshly allocated skb fixes the problem.

The underlying problem that this works around is still unknown.

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Bug #4769.

13 years agomac-learning: Fix mac_entry_is_grat_arp_locked().
Ben Pfaff [Wed, 30 Mar 2011 20:01:36 +0000 (13:01 -0700)]
mac-learning: Fix mac_entry_is_grat_arp_locked().

The lock is asserted if its expiration time has not arrived yet, not the
reverse.

13 years agoofproto: Get rid of send_port_status() trivial wrapper function.
Ben Pfaff [Tue, 15 Mar 2011 23:51:46 +0000 (16:51 -0700)]
ofproto: Get rid of send_port_status() trivial wrapper function.

13 years agoofp-util: Remove flow_stats_iterator, flows_stats_first(), flow_stats_next()
Ben Pfaff [Tue, 15 Mar 2011 21:56:09 +0000 (14:56 -0700)]
ofp-util: Remove flow_stats_iterator, flows_stats_first(), flow_stats_next()

Nothing uses these anymore.  ofputil_decode_flow_stats_reply() is a better
alternative.

13 years agotimeval: Only log poll intervals longer than 50 ms.
Ben Pfaff [Mon, 28 Mar 2011 20:43:32 +0000 (13:43 -0700)]
timeval: Only log poll intervals longer than 50 ms.

When poll interval-based logging was introduced a long time, we were
actively interested in looking at almost every long poll interval.  But
these days, with OVS working rather well, with pretty good latency, most
of the messages are red herrings that bother some administrators and
provoke false reports.  So this commit suppresses all but the most
egregious long poll intervals that may in fact be worth looking at.

NIC-366.

13 years agoAUTHORS: Add Gaetano Catalli <gaetano.catalli@gmail.com>.
Ben Pfaff [Wed, 30 Mar 2011 18:21:28 +0000 (11:21 -0700)]
AUTHORS: Add Gaetano Catalli <gaetano.catalli@gmail.com>.

13 years agoFix compilation of openvswitch-1.1.0pre2 on FreeBSD-8.1
Gaetano Catalli [Wed, 30 Mar 2011 15:25:36 +0000 (17:25 +0200)]
Fix compilation of openvswitch-1.1.0pre2 on FreeBSD-8.1

13 years agoofproto: Factor OpenFlow connection management into new "connmgr".
Ben Pfaff [Tue, 29 Mar 2011 19:24:28 +0000 (12:24 -0700)]
ofproto: Factor OpenFlow connection management into new "connmgr".

This removes a lot of code from ofproto.c and makes the ofproto code
easier to understand.

13 years agoofproto: New function ofconn_pktbuf_retrieve().
Ben Pfaff [Thu, 10 Feb 2011 21:16:56 +0000 (13:16 -0800)]
ofproto: New function ofconn_pktbuf_retrieve().

This helps to increase the level of abstraction of "struct ofconn",
in preparation for moving it from ofproto.c into a new file.

13 years agoofproto: Add functions to get and set ofconn type and role.
Ben Pfaff [Thu, 10 Feb 2011 21:11:07 +0000 (13:11 -0800)]
ofproto: Add functions to get and set ofconn type and role.

This helps to increase the level of abstraction of "struct ofconn",
in preparation for moving it from ofproto.c into a new file.

13 years agoofproto: Add ofconn_get_miss_send_len(), ofconn_set_miss_send_len().
Ben Pfaff [Thu, 10 Feb 2011 20:37:02 +0000 (12:37 -0800)]
ofproto: Add ofconn_get_miss_send_len(), ofconn_set_miss_send_len().

This helps to increase the level of abstraction of "struct ofconn",
in preparation for moving it from ofproto.c into a new file.

13 years agoofproto: Add ofconn_get_flow_format(), ofconn_set_flow_format().
Ben Pfaff [Thu, 10 Feb 2011 19:26:16 +0000 (11:26 -0800)]
ofproto: Add ofconn_get_flow_format(), ofconn_set_flow_format().

This helps to increase the level of abstraction of "struct ofconn",
in preparation for moving it from ofproto.c into a new file.

13 years agoofproto: Use ofconn_send_reply() for sending OpenFlow replies.
Ben Pfaff [Thu, 10 Feb 2011 19:22:03 +0000 (11:22 -0800)]
ofproto: Use ofconn_send_reply() for sending OpenFlow replies.

This helps to increase the level of abstraction of "struct ofconn",
in preparation for moving it from ofproto.c into a new file.

13 years agoofproto: Introduce ofconn_get_ofproto() accessor function.
Ben Pfaff [Tue, 22 Mar 2011 16:14:59 +0000 (09:14 -0700)]
ofproto: Introduce ofconn_get_ofproto() accessor function.

This helps to increase the level of abstraction of "struct ofconn",
in preparation for moving it from ofproto.c into a new file.

13 years agoofproto: Move hton_ofp_phy_port() to ofp-util.
Ben Pfaff [Tue, 15 Mar 2011 17:08:54 +0000 (10:08 -0700)]
ofproto: Move hton_ofp_phy_port() to ofp-util.

This removes some code from ofproto.c that doesn't really seem to
belong there to begin with.

13 years agoofproto: Break packet_in encoding out of ofproto into ofp-util.
Ben Pfaff [Tue, 22 Mar 2011 16:13:02 +0000 (09:13 -0700)]
ofproto: Break packet_in encoding out of ofproto into ofp-util.

This removes some code from ofproto.c.

13 years agoofproto: Factor building of flow_removed messages out into ofp-util.
Ben Pfaff [Tue, 15 Mar 2011 16:49:14 +0000 (09:49 -0700)]
ofproto: Factor building of flow_removed messages out into ofp-util.

This removes some code from ofproto.c.

13 years agoovs-vsctl: Remove dead assignment.
Ethan Jackson [Tue, 29 Mar 2011 17:58:45 +0000 (10:58 -0700)]
ovs-vsctl: Remove dead assignment.

Noticed this last night while playing around with the clang static
analyzer.

13 years agoovs-vsctl: Remove dead code.
Ethan Jackson [Tue, 29 Mar 2011 00:46:43 +0000 (17:46 -0700)]
ovs-vsctl: Remove dead code.

Coverity #10710.

13 years agobridge: Always wait for MAC learning table and ports.
Ben Pfaff [Tue, 29 Mar 2011 17:08:16 +0000 (10:08 -0700)]
bridge: Always wait for MAC learning table and ports.

The test ofproto_has_primary_controller() is meaningless, since OFPP_NORMAL
can cause the MAC learning table and port bonding to be in use even when
there is a controller.

I see that this bug has been here since early 2009, when the OFPP_NORMAL
feature was introduced in the bridge.  (Obviously it's not a severe
problem.)

13 years agoxenserver: Wait for ovs-xapi-sync to exit in "stop" command.
Ben Pfaff [Mon, 28 Mar 2011 23:22:59 +0000 (16:22 -0700)]
xenserver: Wait for ovs-xapi-sync to exit in "stop" command.

It seems possible that "restart" or a quick application of "stop" then
"start" could kill ovs-xapi-sync without starting it again, if
ovs-xapi-sync takes a little while to die, long enough for the next
instance of it to see that its pidfile is still open and locked.

I hope that this fixes some odd races that we've noticed in the "restart"
command.

Signed-off-by: Ben Pfaff <blp@nicira.com>
13 years agodaemon: Avoid redundant code in already_running().
Ben Pfaff [Tue, 29 Mar 2011 16:44:55 +0000 (09:44 -0700)]
daemon: Avoid redundant code in already_running().

This function substantially duplicated read_pidfile(), so reuse that
code instead.

13 years agodaemon: Write "already running" message to log also.
Ben Pfaff [Tue, 29 Mar 2011 17:09:23 +0000 (10:09 -0700)]
daemon: Write "already running" message to log also.

Otherwise it's hard to diagnose later if the daemon failed to start because
it thinks that it is already running.

13 years agocfm: No longer keep track of bad remote MPs and MAIDS.
Ethan Jackson [Mon, 28 Mar 2011 20:10:12 +0000 (13:10 -0700)]
cfm: No longer keep track of bad remote MPs and MAIDS.

Ben pointed out that an attacker could cause OVS to use infinite
memory by sending a series of CCMs with different MAIDs.  Each
message would cause a remote_maid to be allocated and stored for
several seconds.

Since Commit 1c2e2d2fc8 (cfm: Don't report unexpected remote
endpoints) no longer reports unexpected remote MAIDS and MPs in the
database, the only reason to keep track of this information is for
debugging purposes.  In my judgment, it provides negligible useful
debugging information at the expense of significantly increased
code complexity.  This commit rips it out entirely.

13 years agoovsdb-tool: Fix cut-and-paste error in manpage.
Ben Pfaff [Mon, 28 Mar 2011 20:35:32 +0000 (13:35 -0700)]
ovsdb-tool: Fix cut-and-paste error in manpage.

Reported-by: Paul Ingram <paul@nicira.com>
13 years agoovs-brcompatd: Delete write-only variable.
Ben Pfaff [Fri, 25 Mar 2011 22:22:10 +0000 (15:22 -0700)]
ovs-brcompatd: Delete write-only variable.

13 years agocfm: Create new cfm/show appctl command.
Ethan Jackson [Fri, 25 Mar 2011 22:30:33 +0000 (15:30 -0700)]
cfm: Create new cfm/show appctl command.

This will be useful for debugging CFM problems in the future.

13 years agobridge: Write CFM updates more quickly.
Ethan Jackson [Fri, 25 Mar 2011 20:36:43 +0000 (13:36 -0700)]
bridge: Write CFM updates more quickly.

This commit causes updates to CFM status to be written immediately.
A rate limit of 1 second is introduced to avoid performance
problems.

13 years agocfm: Reduce missed CCM detection time.
Ethan Jackson [Fri, 25 Mar 2011 21:26:53 +0000 (14:26 -0700)]
cfm: Reduce missed CCM detection time.

The specification says that a fault should be signaled when 3.5 *
ccm_interval milliseconds have passed.  This commit respects that
requirement, possibly increasing the responsiveness of fault
detection slightly.

13 years agocfm: Don't report unexpected remote endpoints.
Ethan Jackson [Fri, 25 Mar 2011 01:36:56 +0000 (18:36 -0700)]
cfm: Don't report unexpected remote endpoints.

Before this patch, CFM would report unexpected remote maintenance
points in the database.  This commit no longer exposes this
information.

Information about precisely why a link is faulty is more interesting
to a system administrator debugging a problem than a controller
which will generally only care about whether or not a link is
faulty.  For simplicity sake, this commit removes this information
from the database where it was somewhat awkwardly placed.  In the
future it may be valuable to report the information through
ovs-appctl commands for debugging purposes.

13 years agocfm: Immediately signal fault on bad CCM reception.
Ethan Jackson [Fri, 25 Mar 2011 02:04:21 +0000 (19:04 -0700)]
cfm: Immediately signal fault on bad CCM reception.

Commit af5739857a (cfm: Immediately signal a fault upon receiving
an unexpected MPID.) caused the CFM library to immediately signal a
fault upon reception of an unexpected remote MPID.  This commit
does the same for MAIDs, and remote maintenance points with invalid
CCM intervals.

13 years agopackets: Move CFM related packet information to cfm header file.
Ethan Jackson [Fri, 25 Mar 2011 20:57:21 +0000 (13:57 -0700)]
packets: Move CFM related packet information to cfm header file.

13 years agoschema: Monitor's remote_mps is not ephemeral.
Ethan Jackson [Fri, 25 Mar 2011 00:53:16 +0000 (17:53 -0700)]
schema: Monitor's remote_mps is not ephemeral.

The remote_mps column of the Monitor table is a configuration
parameter and should not be ephemeral.

13 years agodatapath: Avoid memory leak in odp_packet_cmd_execute().
Ben Pfaff [Fri, 25 Mar 2011 00:06:58 +0000 (17:06 -0700)]
datapath: Avoid memory leak in odp_packet_cmd_execute().

The error path needs to free 'packet'.

Signed-off-by: Ben Pfaff <blp@nicira.com>
13 years agocfm: cfm_run() return ccm instead of packet.
Ethan Jackson [Tue, 22 Mar 2011 23:57:43 +0000 (16:57 -0700)]
cfm: cfm_run() return ccm instead of packet.

It doesn't really make sense for the CFM code to be composing
packets.  Its caller is better placed to compose the appropriate
L2 header.  This commit pulls that logic out of the CFM library.

13 years agopackets: Create new compose_packet() function.
Ethan Jackson [Wed, 23 Mar 2011 19:59:40 +0000 (12:59 -0700)]
packets: Create new compose_packet() function.

This commit generalizes compose_lacp_packet() into new
compose_packet() function.  This new function will be used to send
CCM messages in future patches.

13 years agocfm: Move destination address to cfm.h
Ethan Jackson [Tue, 22 Mar 2011 21:28:22 +0000 (14:28 -0700)]
cfm: Move destination address to cfm.h

Future patches will require the use of this information globally.

13 years agodatapath: Add compatibility with sk_buff's vlan_tci before 2.6.33.
Ben Pfaff [Wed, 23 Mar 2011 17:34:37 +0000 (10:34 -0700)]
datapath: Add compatibility with sk_buff's vlan_tci before 2.6.33.

Between 2.6.27 and 2.6.32, the vlan_tci member of struct sk_buff was the
raw value of the 802.1Q header's TCI field, without the CFI bit being set.
In 2.6.33 and later, the CFI bit is always set if an 802.1Q header is
present, correcting a corner case.

Until now, OVS has not consistently dealt with this.  If a packet arrived
at a datapath from a network device directly, or if it was set with an
ODP_ACTION_ATTR_SET_DL_TCI action, then the CFI bit would not be set in
vlan_tci.  In flow_extract(), OVS copies vlan_tci directly to dl_tci in the
flow structure (via vlan_get_tci()), so the CFI bit would also not be set
in dl_tci.  But if OVS had to send a packet up to userspace (converting the
vlan_tci back to an 802.1Q header along the way) and got it back, then it
would set the VLAN CFI bit in dl_tci when it parsed the 802.1Q header in
parse_vlan().  This had the effect that a flow set up by userspace (with
the CFI bit set) would never be matched by a packet arriving from a network
device, because they would have different dl_tci values.

This fixes the problem, by making the vlan_get_tci() and vlan_set_tci()
interface consistent across kernel versions.  Now, they always accept or
return a value where the VLAN CFI bit is set if an 802.1Q header is
present.

Build-tested only.

Problem isolated by Ethan Jackson <ethan@nicira.com>

Signed-off-by: Ben Pfaff <blp@nicira.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Reported-by: Ram Jothikumar <rjothikumar@nicira.com>
Bug #4915.

13 years agoofproto: Allow tunnel id to be in decimal or hex for trace command.
Justin Pettit [Wed, 23 Mar 2011 00:06:07 +0000 (17:06 -0700)]
ofproto: Allow tunnel id to be in decimal or hex for trace command.

Suggested-by: Pankaj Thakkar <thakkar@nicira.com>
13 years agoodp-util: Print tunnel ids in host-order.
Justin Pettit [Tue, 22 Mar 2011 23:45:03 +0000 (16:45 -0700)]
odp-util: Print tunnel ids in host-order.

Suggested-by: Pankaj Thakkar <thakkar@nicira.com>
13 years agogitignore: Added ovs-vlan-bug-workaround and tests
Ethan Jackson [Wed, 23 Mar 2011 02:11:50 +0000 (19:11 -0700)]
gitignore: Added ovs-vlan-bug-workaround and tests

13 years agoutilities: Add ovs-vlan-bug-workaround to make distclean
Ethan Jackson [Wed, 23 Mar 2011 01:55:53 +0000 (18:55 -0700)]
utilities: Add ovs-vlan-bug-workaround to make distclean

This was causing and error during make distcheck for me.

13 years agovswitchd: Properly calculate output port.
Ethan Jackson [Tue, 22 Mar 2011 22:31:59 +0000 (15:31 -0700)]
vswitchd: Properly calculate output port.

This was causing segfaults on my system.

13 years agoxenserver: Collect /proc/slabinfo in a Server Status Report
Sajjad Lateef [Tue, 22 Mar 2011 18:17:33 +0000 (11:17 -0700)]
xenserver: Collect /proc/slabinfo in a Server Status Report

/proc/slabinfo gives information about memory usage
as slab level and can be used to diagnose memory issues
on memory-limited systems

13 years agovswitchd: Destroy lacp in port_destroy().
Ethan Jackson [Tue, 22 Mar 2011 20:30:15 +0000 (13:30 -0700)]
vswitchd: Destroy lacp in port_destroy().

Port destruction could cause dangling lacp objects to live in the
lacp module's 'all_lacps' list.  This could cause bogus output for
the lacp/show appctl command.

Bug #5088.

13 years agobridge: Memory leak in port_destroy()
Ethan Jackson [Tue, 22 Mar 2011 20:16:24 +0000 (13:16 -0700)]
bridge: Memory leak in port_destroy()

13 years agobridge: Change bridge's 'ports' member from array to hmap.
Ben Pfaff [Mon, 21 Mar 2011 17:07:39 +0000 (10:07 -0700)]
bridge: Change bridge's 'ports' member from array to hmap.

In my opinion, this makes the code more obviously correct in many places,
because there are generally fewer variables.  One must generally keep two
variables in sync for iterating through an array: the array index and the
contents of the array at that index.  For iterating through an hmap, only
the map element is necessary.

A linked list would also be a reasonable choice for the bridge's collection
of ports.  I chose to use an hmap because we already had an index by name
and it seemed OK to use only one index.  I decided not to keep the shash
because they are less convenient for iteration than an hmap.

13 years agobridge: Avoid flushing entire MAC learning table for common operations.
Ben Pfaff [Fri, 18 Mar 2011 22:42:29 +0000 (15:42 -0700)]
bridge: Avoid flushing entire MAC learning table for common operations.

Adding and removing ports is fairly common in a virtual environment,
because it happens whenever a VM boots or shuts down.  It is best to
avoid flushing the whole MAC learning table when that happens, because
that means that, briefly, every packet will get flooded, wasting CPU cycles
and network bandwidth.

This commit breaks flushing the MAC table out of bridge_flush().  Instead,
each caller is now responsible for flushing the MAC table if it is
necessary.  In a few cases, no flushing was necessary, so those callers
were not modified.  In the case of removing a port or modifying its VLAN
assignments, it is necessary to expire all of the MAC learning entries
associated with that port, so this commit does that.  Finally, some
operations do require a MAC learning flush but they are rare enough that
in my opinion it's not worth taking care to avoid a MAC table flush.

Bug #891.

13 years agomac-learning: Expose function for expiring a single MAC learning entry.
Ben Pfaff [Fri, 18 Mar 2011 22:28:21 +0000 (15:28 -0700)]
mac-learning: Expose function for expiring a single MAC learning entry.

The bridge will soon use this for expiring only the MAC learning entries
associated with a given port at port deletion time.

13 years agobridge: Use pointer to "struct port", not "port_idx", for MAC learning.
Ben Pfaff [Tue, 22 Mar 2011 16:57:06 +0000 (09:57 -0700)]
bridge: Use pointer to "struct port", not "port_idx", for MAC learning.

This takes a step toward changing the data structure used for keeping
track of ports from an array to a more appropriate data structure.

13 years agomac-learning: Change 'port' member to a union.
Ben Pfaff [Fri, 18 Mar 2011 22:03:24 +0000 (15:03 -0700)]
mac-learning: Change 'port' member to a union.

This allow the client a little more flexibility.  The next commit shows
how this can be useful.

13 years agomac-learning: Refactor to increase generality.
Ben Pfaff [Tue, 22 Mar 2011 16:47:02 +0000 (09:47 -0700)]
mac-learning: Refactor to increase generality.

In an upcoming commit I want to store a pointer in MAC learning entries
in the bridge, instead of an integer port number.  The MAC learning library
has other clients, and the others do not gracefully fit this new model, so
in fact the data will have to become a union.  However, this does not fit
well with the current mac_learning API, since mac_learning_learn()
currently initializes and compares the data.  It seems better to break up
the API so that only the client has to know the data's format and how to
initialize it or compare it.  This commit makes this possible.

This commit doesn't change the type of the data stored in a MAC learning
entry yet.

As a side effect this commit has the benefit that clients that don't need
gratuitous ARP locking don't have to specify any policy for it at all.

13 years agobridge: Change port's 'ifaces' member from array to list.
Ben Pfaff [Mon, 21 Mar 2011 17:24:32 +0000 (10:24 -0700)]
bridge: Change port's 'ifaces' member from array to list.

In my opinion, this makes the code more obviously correct in many places,
because there are generally fewer variables.  One must generally keep two
variables in sync for iterating through an array: the array index and the
contents of the array at that index.  For iterating through a list, only
the list element is necessary.

13 years agobridge: Get rid of "port_ifidx" in struct iface. Fix bonding hash.
Ben Pfaff [Mon, 21 Mar 2011 20:25:02 +0000 (13:25 -0700)]
bridge: Get rid of "port_ifidx" in struct iface.  Fix bonding hash.

This is a first step toward changing the array of ifaces in struct port
to a more suitable data structure.

As a side effect this fixes a bonding problem that I noticed via code
inspection.  Before this commit, each bond_entry specified an interface
via index.  If an iface was deleted, however, this shifted all of the
iface indexes, and the code didn't compensate for that.  This commit fixes
the problem by using pointers to ifaces instead, which don't shift around.

13 years agobridge: Expire bond slave assignments when their loads decay to 0.
Ben Pfaff [Fri, 18 Mar 2011 00:12:40 +0000 (17:12 -0700)]
bridge: Expire bond slave assignments when their loads decay to 0.

Until now, if a given MAC ever transmitted, then it would always show up
in bond information output.  There's no benefit to that if the MAC has
gone away permanently.  This commit causes them to be deleted when their
load has gone to 0.  This takes a fairly long time: if a MAC has sent, say,
one million bytes and then stops transmitting entirely, then it will take
about 20 rebalancing intervals (200 seconds) before it decays to 0 and
gets deleted.

Bug #2157.

13 years agobridge: Improve name and comments for bond_entry's "iface_tag" member.
Ben Pfaff [Thu, 17 Mar 2011 23:29:55 +0000 (16:29 -0700)]
bridge: Improve name and comments for bond_entry's "iface_tag" member.

The iface_tag name and comment implied that it was really just a copy of
the 'tag' member of struct iface, but in fact it has a completely different
purpose: it represents the binding of a bond_entry to a particular iface.
It is invalidated when the bond_entry has to be redirected to a different
iface, not when the iface itself changes.  I hope that this commit helps
to clarify.

13 years agobridge: Change struct port's active_iface member from index to pointer.
Ben Pfaff [Thu, 17 Mar 2011 22:57:20 +0000 (15:57 -0700)]
bridge: Change struct port's active_iface member from index to pointer.

This makes the code easier to understand.

As a historical note, the "bridge" code was originally written in an
almighty hurry, and so some design decisions were made on the basis of
being unlikely to cause serious bugs instead of on the basis of being
easy to understand.  That's why there are so many array indexes sprinkled
around the bridge data structures, and so much range checking of their
values, when it would be better to just have pointers that can be followed
directly.  I figured that getting the wrong index would at least do
something half-reasonable in most cases, whereas dereferencing a freed
pointer was likely to segfault sooner or later and cause immediate failure.

But now I think it's time to improve the code.  The code is mature enough
now that we should be able to thoroughly understand the data lifetime
issues.

13 years agobridge: Get rid of redundant bridge_get_local_iface() function.
Ben Pfaff [Thu, 17 Mar 2011 22:38:08 +0000 (15:38 -0700)]
bridge: Get rid of redundant bridge_get_local_iface() function.

Not only was this function redundant, it managed to spend O(n) time doing
a task that should take only O(1)!

13 years agobridge: Drop 'active_iface_tag' from "struct port".
Ben Pfaff [Thu, 17 Mar 2011 22:26:43 +0000 (15:26 -0700)]
bridge: Drop 'active_iface_tag' from "struct port".

I cannot see a reason to have an active_iface_tag that is different from
the tag on the active_iface.  As far as I can see, we can just get rid of
it.

13 years agobridge: Don't bother revalidating in combination with bridge_flush().
Ben Pfaff [Thu, 17 Mar 2011 22:14:46 +0000 (15:14 -0700)]
bridge: Don't bother revalidating in combination with bridge_flush().

iface_destroy() always calls bridge_flush(), which causes every flow to be
revalidated, so there is no point in revalidating individual tags also.

13 years agoxenserver: Don't require other-config:disable-in-band to exist.
Ben Pfaff [Mon, 21 Mar 2011 21:39:45 +0000 (14:39 -0700)]
xenserver: Don't require other-config:disable-in-band to exist.

other-config:disable-in-band is optional but the ovs-vsctl "get" command
by default requires its target to exist.  Adding --if-exists allows it
to be missing.

Reported-by: Henrik Amren <henrik@nicira.com>
Bug #5056.

13 years agolist: Allow list_front(), list_back() to take 'const' arguments.
Ben Pfaff [Mon, 21 Mar 2011 21:06:35 +0000 (14:06 -0700)]
list: Allow list_front(), list_back() to take 'const' arguments.

It totally makes sense to pass a const struct list * to one of these
functions.  Ideally the return type would be the same as the argument
type but C can't handle that, so this is the best second choice.

13 years agolist: Fix indentation.
Ben Pfaff [Mon, 21 Mar 2011 21:03:35 +0000 (14:03 -0700)]
list: Fix indentation.

13 years agobridge: Always delete the correct interfaces in port_del_ifaces().
Ben Pfaff [Fri, 18 Mar 2011 17:56:28 +0000 (10:56 -0700)]
bridge: Always delete the correct interfaces in port_del_ifaces().

This code was checking a name from cfg->interfaces[i] but using the result
to decide whether to delete port->ifaces[i].  The two arrays aren't
necessarily in the same order so this code could delete the wrong iface.

Found by inspection.  This is probably not a common problem in practice
because I imagine that deleting an interface from a bond, without deleting
the bond itself, is a rare operation.  The problem would also be
intermittent--after a few switch reconfiguration cycles I imagine that
the problem would resolve.

13 years agoofproto: Take responsibility for connectivity fault management (CFM).
Ben Pfaff [Fri, 18 Mar 2011 23:56:31 +0000 (16:56 -0700)]
ofproto: Take responsibility for connectivity fault management (CFM).

This takes one step toward the larger goal of getting rid of all of the
hooks from ofproto back into vswitchd, by eliminating one of the reasons
that they are required.

13 years agoofproto: Specialize ofproto_send_packet() for uses the callers really want.
Ben Pfaff [Wed, 16 Mar 2011 20:54:10 +0000 (13:54 -0700)]
ofproto: Specialize ofproto_send_packet() for uses the callers really want.

The callers of ofproto_send_packet() actually just want to send a packet
out on a port, possibly tagged with a VLAN, but the interface forced them
to compose a set of OpenFlow actions, which made it harder to use than
necessary.  This commit specializes the interface for the purposes that
the callers really wanted, making it easier to use.

handle_miss_upcall() can now take advantage of this function, too.

13 years agocfm: Immediately signal a fault upon receiving an unexpected MPID.
Ben Pfaff [Wed, 16 Mar 2011 21:32:38 +0000 (14:32 -0700)]
cfm: Immediately signal a fault upon receiving an unexpected MPID.

An unexpected MPID is always a fault, but the CFM code didn't signal the
fault until the next time cfm_run() was called.  In one experiment I
saw a visible lag in the database (although I wasn't able to reproduce it
again within a few tries).

13 years agobridge: Consistently use miimon status if miimon is configured.
Ben Pfaff [Thu, 17 Mar 2011 17:48:05 +0000 (10:48 -0700)]
bridge: Consistently use miimon status if miimon is configured.

A port can be configured to use miimon reporting as the criterion for
enabling or disabling an interface, but in some cases (such as for reading
the initial link status) the code was reading the carrier status instead.
This commit fixes the problem.

This changes the meaning of the link_status column in the Interface table.
I don't think that the old meaning was useful to the controller in the
case of a bond configured for miimon monitoring, because the controller
could not use it to detect which interfaces the bond considered to be up
or down.

13 years agodatapath: Use ipv4_dst_hoplimit() instead of direct access.
Jesse Gross [Fri, 18 Mar 2011 21:37:33 +0000 (14:37 -0700)]
datapath: Use ipv4_dst_hoplimit() instead of direct access.

In 2.6.38 dst metrics became shared and use copy-on-write instead
of using private versions.  This means that it is no longer permissible
to directly access the metric and we need to use the helper function
instead.

Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
13 years agodatapath: Backport ip4_dst_hoplimit() from 2.6.38.
Jesse Gross [Fri, 18 Mar 2011 21:43:51 +0000 (14:43 -0700)]
datapath: Backport ip4_dst_hoplimit() from 2.6.38.

Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>
13 years agoovs-ofctl: Make mod-flows and del-flows commands able read files.
Ben Pfaff [Fri, 11 Mar 2011 21:19:12 +0000 (13:19 -0800)]
ovs-ofctl: Make mod-flows and del-flows commands able read files.

This makes it possible to modify or delete multiple flows in one run,
which is sometimes convenient.

Requested-by: Paul Ingram <paul@nicira.com>
13 years agoofp-parse: Generalize parse_ofp_add_flow_file() to parse_ofp_flow_mod_file().
Ben Pfaff [Fri, 18 Mar 2011 21:41:22 +0000 (14:41 -0700)]
ofp-parse: Generalize parse_ofp_add_flow_file() to parse_ofp_flow_mod_file().

An upcoming commit will want to pass a different command.

13 years agoovs-ofctl: New commands "replace-flows" and "diff-flows".
Ben Pfaff [Fri, 11 Mar 2011 20:02:27 +0000 (12:02 -0800)]
ovs-ofctl: New commands "replace-flows" and "diff-flows".

Requested-by: Paul Ingram <paul@nicira.com>
13 years agoovs-ofctl: Use ovs_be32 for big-endian variable.
Ben Pfaff [Fri, 11 Mar 2011 19:56:25 +0000 (11:56 -0800)]
ovs-ofctl: Use ovs_be32 for big-endian variable.

13 years agoofpbuf: New function ofpbuf_steal_data().
Ben Pfaff [Fri, 11 Mar 2011 19:52:49 +0000 (11:52 -0800)]
ofpbuf: New function ofpbuf_steal_data().

This will have its first use in an upcoming commit.

13 years agovlog: Add comment describing the purpose of each log level.
Ben Pfaff [Fri, 18 Mar 2011 17:10:20 +0000 (10:10 -0700)]
vlog: Add comment describing the purpose of each log level.

Based on descriptions from internal wiki.

13 years agolacp: Enable "fast" lacp timing mode.
Ethan Jackson [Wed, 9 Mar 2011 00:34:20 +0000 (16:34 -0800)]
lacp: Enable "fast" lacp timing mode.

13 years agolacp: Transmit more judiciously.
Ethan Jackson [Wed, 9 Mar 2011 02:32:58 +0000 (18:32 -0800)]
lacp: Transmit more judiciously.

Only transmit when the LACP partner needs to be updated instead of
whenever it might need to be updated.

13 years agovswitchd: Modularize LACP.
Ethan Jackson [Mon, 28 Feb 2011 22:48:06 +0000 (14:48 -0800)]
vswitchd: Modularize LACP.

This commit pulls LACP code out of the bridge into its own LACP
module.  Currently this module is only used by the existing bonding
infrastructure.

13 years agovswitchd: Update lacp before bonding.
Ethan Jackson [Wed, 9 Mar 2011 01:20:01 +0000 (17:20 -0800)]
vswitchd: Update lacp before bonding.

Bonding requires lacp attached status to make decisions about which
links are enabled.  However, LACP does not require any bonding
related information (other than which links are aggregated) to
perform.  Having them reversed causes some subtle bugs in the
bonding code's responsiveness to LACP changes.

13 years agopackets: Rename LACP portid and sysid to port_id and sys_id.
Ethan Jackson [Thu, 3 Mar 2011 18:34:22 +0000 (10:34 -0800)]
packets: Rename LACP portid and sysid to port_id and sys_id.

This seems more stylistically consistent with the rest of the
code base and the sys_priority and port_priority in the same
struct.

13 years agopackets: create new compose_lacp_pdu() function.
Ethan Jackson [Tue, 1 Mar 2011 21:27:23 +0000 (13:27 -0800)]
packets: create new compose_lacp_pdu() function.

Creates new compose_lacp_pdu() from the old compose_lacp_packet()
function.  This will allow a LACP PDU to be created without
necessarily knowing the Ethernet Source Address required for
generating the packet.  Future patches rely on this functionality.

13 years agobridge: Pull monitoring out of bonding logic.
Ethan Jackson [Tue, 22 Feb 2011 22:19:18 +0000 (14:19 -0800)]
bridge: Pull monitoring out of bonding logic.

The monitoring logic and bonding code are unrelated.  This commit
pulls the monitoring logic out.  As a result all interfaces, not
just those participating in bonds, are monitored.  This will be
required to run LACP on non-bonded ports.

Also removes the miimon flag from the port structure.