tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
/*
|
|
|
|
* net/tipc/group.c: TIPC group messaging code
|
|
|
|
*
|
|
|
|
* Copyright (c) 2017, Ericsson AB
|
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions are met:
|
|
|
|
*
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
* 3. Neither the names of the copyright holders nor the names of its
|
|
|
|
* contributors may be used to endorse or promote products derived from
|
|
|
|
* this software without specific prior written permission.
|
|
|
|
*
|
|
|
|
* Alternatively, this software may be distributed under the terms of the
|
|
|
|
* GNU General Public License ("GPL") version 2 as published by the Free
|
|
|
|
* Software Foundation.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
|
|
|
|
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
|
|
|
|
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
|
|
|
|
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
|
|
|
|
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
|
|
|
|
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
|
|
|
|
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
|
|
|
|
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
|
|
|
|
* POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "core.h"
|
|
|
|
#include "addr.h"
|
|
|
|
#include "group.h"
|
|
|
|
#include "bcast.h"
|
2018-02-15 17:40:51 +08:00
|
|
|
#include "topsrv.h"
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
#include "msg.h"
|
|
|
|
#include "socket.h"
|
|
|
|
#include "node.h"
|
|
|
|
#include "name_table.h"
|
|
|
|
#include "subscr.h"
|
|
|
|
|
|
|
|
#define ADV_UNIT (((MAX_MSG_SIZE + MAX_H_SIZE) / FLOWCTL_BLK_SZ) + 1)
|
|
|
|
#define ADV_IDLE ADV_UNIT
|
2017-10-13 17:04:26 +08:00
|
|
|
#define ADV_ACTIVE (ADV_UNIT * 12)
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
|
|
|
|
enum mbr_state {
|
|
|
|
MBR_JOINING,
|
|
|
|
MBR_PUBLISHED,
|
|
|
|
MBR_JOINED,
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
MBR_PENDING,
|
|
|
|
MBR_ACTIVE,
|
|
|
|
MBR_RECLAIMING,
|
|
|
|
MBR_REMITTED,
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
MBR_LEAVING
|
|
|
|
};
|
|
|
|
|
|
|
|
struct tipc_member {
|
|
|
|
struct rb_node tree_node;
|
|
|
|
struct list_head list;
|
2018-01-04 22:20:44 +08:00
|
|
|
struct list_head small_win;
|
2017-10-13 17:04:30 +08:00
|
|
|
struct sk_buff_head deferredq;
|
2017-10-13 17:04:26 +08:00
|
|
|
struct tipc_group *group;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
u32 node;
|
|
|
|
u32 port;
|
2017-10-13 17:04:24 +08:00
|
|
|
u32 instance;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
enum mbr_state state;
|
2017-10-13 17:04:26 +08:00
|
|
|
u16 advertised;
|
|
|
|
u16 window;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
u16 bc_rcv_nxt;
|
tipc: guarantee delivery of last broadcast before DOWN event
The following scenario is possible:
- A user sends a broadcast message, and thereafter immediately leaves
the group.
- The LEAVE message, following a different path than the broadcast,
arrives ahead of the broadcast, and the sending member is removed
from the receiver's list.
- The broadcast message arrives, but is dropped because the sender
now is unknown to the receipient.
We fix this by sequence numbering membership events, just like ordinary
unicast messages. Currently, when a JOIN is sent to a peer, it contains
a synchronization point, - the sequence number of the next sent
broadcast, in order to give the receiver a start synchronization point.
We now let even LEAVE messages contain such an "end synchronization"
point, so that the recipient can delay the removal of the sending member
until it knows that all messages have been received.
The received synchronization points are added as sequence numbers to the
generated membership events, making it possible to handle them almost
the same way as regular unicasts in the receiving filter function. In
particular, a DOWN event with a too high sequence number will be kept
in the reordering queue until the missing broadcast(s) arrive and have
been delivered.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:33 +08:00
|
|
|
u16 bc_syncpt;
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
u16 bc_acked;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
struct tipc_group {
|
|
|
|
struct rb_root members;
|
2018-01-04 22:20:44 +08:00
|
|
|
struct list_head small_win;
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
struct list_head pending;
|
|
|
|
struct list_head active;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
struct tipc_nlist dests;
|
|
|
|
struct net *net;
|
|
|
|
int subid;
|
|
|
|
u32 type;
|
|
|
|
u32 instance;
|
|
|
|
u32 scope;
|
|
|
|
u32 portid;
|
|
|
|
u16 member_cnt;
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
u16 active_cnt;
|
|
|
|
u16 max_active;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
u16 bc_snd_nxt;
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
u16 bc_ackers;
|
2018-01-17 23:42:46 +08:00
|
|
|
bool *open;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
bool loopback;
|
2017-10-13 17:04:25 +08:00
|
|
|
bool events;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
static void tipc_group_proto_xmit(struct tipc_group *grp, struct tipc_member *m,
|
|
|
|
int mtyp, struct sk_buff_head *xmitq);
|
|
|
|
|
2018-01-09 04:03:31 +08:00
|
|
|
static void tipc_group_open(struct tipc_member *m, bool *wakeup)
|
|
|
|
{
|
|
|
|
*wakeup = false;
|
|
|
|
if (list_empty(&m->small_win))
|
|
|
|
return;
|
|
|
|
list_del_init(&m->small_win);
|
2018-01-17 23:42:46 +08:00
|
|
|
*m->group->open = true;
|
2018-01-09 04:03:31 +08:00
|
|
|
*wakeup = true;
|
|
|
|
}
|
|
|
|
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
static void tipc_group_decr_active(struct tipc_group *grp,
|
|
|
|
struct tipc_member *m)
|
|
|
|
{
|
2017-12-30 02:48:02 +08:00
|
|
|
if (m->state == MBR_ACTIVE || m->state == MBR_RECLAIMING ||
|
|
|
|
m->state == MBR_REMITTED)
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
grp->active_cnt--;
|
|
|
|
}
|
|
|
|
|
2017-10-13 17:04:26 +08:00
|
|
|
static int tipc_group_rcvbuf_limit(struct tipc_group *grp)
|
|
|
|
{
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
int max_active, active_pool, idle_pool;
|
2017-10-13 17:04:26 +08:00
|
|
|
int mcnt = grp->member_cnt + 1;
|
|
|
|
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
/* Limit simultaneous reception from other members */
|
|
|
|
max_active = min(mcnt / 8, 64);
|
|
|
|
max_active = max(max_active, 16);
|
|
|
|
grp->max_active = max_active;
|
|
|
|
|
|
|
|
/* Reserve blocks for active and idle members */
|
|
|
|
active_pool = max_active * ADV_ACTIVE;
|
|
|
|
idle_pool = (mcnt - max_active) * ADV_IDLE;
|
|
|
|
|
2017-10-13 17:04:26 +08:00
|
|
|
/* Scale to bytes, considering worst-case truesize/msgsize ratio */
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
return (active_pool + idle_pool) * FLOWCTL_BLK_SZ * 4;
|
2017-10-13 17:04:26 +08:00
|
|
|
}
|
|
|
|
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
u16 tipc_group_bc_snd_nxt(struct tipc_group *grp)
|
|
|
|
{
|
|
|
|
return grp->bc_snd_nxt;
|
|
|
|
}
|
|
|
|
|
2018-01-04 22:20:44 +08:00
|
|
|
static bool tipc_group_is_receiver(struct tipc_member *m)
|
2017-10-13 17:04:26 +08:00
|
|
|
{
|
2018-01-09 04:03:25 +08:00
|
|
|
return m && m->state != MBR_JOINING && m->state != MBR_LEAVING;
|
2017-10-13 17:04:26 +08:00
|
|
|
}
|
|
|
|
|
2018-01-04 22:20:44 +08:00
|
|
|
static bool tipc_group_is_sender(struct tipc_member *m)
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
{
|
2018-01-09 04:03:28 +08:00
|
|
|
return m && m->state != MBR_JOINING && m->state != MBR_PUBLISHED;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
}
|
|
|
|
|
2017-10-13 17:04:28 +08:00
|
|
|
u32 tipc_group_exclude(struct tipc_group *grp)
|
|
|
|
{
|
|
|
|
if (!grp->loopback)
|
|
|
|
return grp->portid;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
struct tipc_group *tipc_group_create(struct net *net, u32 portid,
|
2018-01-17 23:42:46 +08:00
|
|
|
struct tipc_group_req *mreq,
|
|
|
|
bool *group_is_open)
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
{
|
2018-01-09 04:03:30 +08:00
|
|
|
u32 filter = TIPC_SUB_PORTS | TIPC_SUB_NO_STATUS;
|
|
|
|
bool global = mreq->scope != TIPC_NODE_SCOPE;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
struct tipc_group *grp;
|
|
|
|
u32 type = mreq->type;
|
|
|
|
|
|
|
|
grp = kzalloc(sizeof(*grp), GFP_ATOMIC);
|
|
|
|
if (!grp)
|
|
|
|
return NULL;
|
|
|
|
tipc_nlist_init(&grp->dests, tipc_own_addr(net));
|
2018-01-04 22:20:44 +08:00
|
|
|
INIT_LIST_HEAD(&grp->small_win);
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
INIT_LIST_HEAD(&grp->active);
|
|
|
|
INIT_LIST_HEAD(&grp->pending);
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
grp->members = RB_ROOT;
|
|
|
|
grp->net = net;
|
|
|
|
grp->portid = portid;
|
|
|
|
grp->type = type;
|
|
|
|
grp->instance = mreq->instance;
|
|
|
|
grp->scope = mreq->scope;
|
|
|
|
grp->loopback = mreq->flags & TIPC_GROUP_LOOPBACK;
|
2017-10-13 17:04:25 +08:00
|
|
|
grp->events = mreq->flags & TIPC_GROUP_MEMBER_EVTS;
|
2018-01-17 23:42:46 +08:00
|
|
|
grp->open = group_is_open;
|
2018-02-27 03:14:04 +08:00
|
|
|
*grp->open = false;
|
2018-01-09 04:03:30 +08:00
|
|
|
filter |= global ? TIPC_SUB_CLUSTER_SCOPE : TIPC_SUB_NODE_SCOPE;
|
|
|
|
if (tipc_topsrv_kern_subscr(net, portid, type, 0, ~0,
|
|
|
|
filter, &grp->subid))
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
return grp;
|
|
|
|
kfree(grp);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2018-01-09 04:03:28 +08:00
|
|
|
void tipc_group_join(struct net *net, struct tipc_group *grp, int *sk_rcvbuf)
|
|
|
|
{
|
|
|
|
struct rb_root *tree = &grp->members;
|
|
|
|
struct tipc_member *m, *tmp;
|
|
|
|
struct sk_buff_head xmitq;
|
|
|
|
|
2019-08-15 22:42:50 +08:00
|
|
|
__skb_queue_head_init(&xmitq);
|
2018-01-09 04:03:28 +08:00
|
|
|
rbtree_postorder_for_each_entry_safe(m, tmp, tree, tree_node) {
|
|
|
|
tipc_group_proto_xmit(grp, m, GRP_JOIN_MSG, &xmitq);
|
|
|
|
tipc_group_update_member(m, 0);
|
|
|
|
}
|
|
|
|
tipc_node_distr_xmit(net, &xmitq);
|
|
|
|
*sk_rcvbuf = tipc_group_rcvbuf_limit(grp);
|
|
|
|
}
|
|
|
|
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
void tipc_group_delete(struct net *net, struct tipc_group *grp)
|
|
|
|
{
|
|
|
|
struct rb_root *tree = &grp->members;
|
|
|
|
struct tipc_member *m, *tmp;
|
|
|
|
struct sk_buff_head xmitq;
|
|
|
|
|
|
|
|
__skb_queue_head_init(&xmitq);
|
|
|
|
|
|
|
|
rbtree_postorder_for_each_entry_safe(m, tmp, tree, tree_node) {
|
|
|
|
tipc_group_proto_xmit(grp, m, GRP_LEAVE_MSG, &xmitq);
|
2019-06-16 17:24:07 +08:00
|
|
|
__skb_queue_purge(&m->deferredq);
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
list_del(&m->list);
|
|
|
|
kfree(m);
|
|
|
|
}
|
|
|
|
tipc_node_distr_xmit(net, &xmitq);
|
|
|
|
tipc_nlist_purge(&grp->dests);
|
|
|
|
tipc_topsrv_kern_unsubscr(net, grp->subid);
|
|
|
|
kfree(grp);
|
|
|
|
}
|
|
|
|
|
2018-07-19 17:16:59 +08:00
|
|
|
static struct tipc_member *tipc_group_find_member(struct tipc_group *grp,
|
|
|
|
u32 node, u32 port)
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
{
|
|
|
|
struct rb_node *n = grp->members.rb_node;
|
|
|
|
u64 nkey, key = (u64)node << 32 | port;
|
|
|
|
struct tipc_member *m;
|
|
|
|
|
|
|
|
while (n) {
|
|
|
|
m = container_of(n, struct tipc_member, tree_node);
|
|
|
|
nkey = (u64)m->node << 32 | m->port;
|
|
|
|
if (key < nkey)
|
|
|
|
n = n->rb_left;
|
|
|
|
else if (key > nkey)
|
|
|
|
n = n->rb_right;
|
|
|
|
else
|
|
|
|
return m;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2017-10-13 17:04:27 +08:00
|
|
|
static struct tipc_member *tipc_group_find_dest(struct tipc_group *grp,
|
|
|
|
u32 node, u32 port)
|
|
|
|
{
|
|
|
|
struct tipc_member *m;
|
|
|
|
|
|
|
|
m = tipc_group_find_member(grp, node, port);
|
2018-01-04 22:20:44 +08:00
|
|
|
if (m && tipc_group_is_receiver(m))
|
2017-10-13 17:04:27 +08:00
|
|
|
return m;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
static struct tipc_member *tipc_group_find_node(struct tipc_group *grp,
|
|
|
|
u32 node)
|
|
|
|
{
|
|
|
|
struct tipc_member *m;
|
|
|
|
struct rb_node *n;
|
|
|
|
|
|
|
|
for (n = rb_first(&grp->members); n; n = rb_next(n)) {
|
|
|
|
m = container_of(n, struct tipc_member, tree_node);
|
|
|
|
if (m->node == node)
|
|
|
|
return m;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void tipc_group_add_to_tree(struct tipc_group *grp,
|
|
|
|
struct tipc_member *m)
|
|
|
|
{
|
|
|
|
u64 nkey, key = (u64)m->node << 32 | m->port;
|
|
|
|
struct rb_node **n, *parent = NULL;
|
|
|
|
struct tipc_member *tmp;
|
|
|
|
|
|
|
|
n = &grp->members.rb_node;
|
|
|
|
while (*n) {
|
|
|
|
tmp = container_of(*n, struct tipc_member, tree_node);
|
|
|
|
parent = *n;
|
|
|
|
tmp = container_of(parent, struct tipc_member, tree_node);
|
|
|
|
nkey = (u64)tmp->node << 32 | tmp->port;
|
|
|
|
if (key < nkey)
|
|
|
|
n = &(*n)->rb_left;
|
|
|
|
else if (key > nkey)
|
|
|
|
n = &(*n)->rb_right;
|
|
|
|
else
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
rb_link_node(&m->tree_node, parent, n);
|
|
|
|
rb_insert_color(&m->tree_node, &grp->members);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct tipc_member *tipc_group_create_member(struct tipc_group *grp,
|
|
|
|
u32 node, u32 port,
|
2018-01-09 04:03:28 +08:00
|
|
|
u32 instance, int state)
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
{
|
|
|
|
struct tipc_member *m;
|
|
|
|
|
|
|
|
m = kzalloc(sizeof(*m), GFP_ATOMIC);
|
|
|
|
if (!m)
|
|
|
|
return NULL;
|
|
|
|
INIT_LIST_HEAD(&m->list);
|
2018-01-04 22:20:44 +08:00
|
|
|
INIT_LIST_HEAD(&m->small_win);
|
2017-10-13 17:04:30 +08:00
|
|
|
__skb_queue_head_init(&m->deferredq);
|
2017-10-13 17:04:26 +08:00
|
|
|
m->group = grp;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
m->node = node;
|
|
|
|
m->port = port;
|
2018-01-09 04:03:28 +08:00
|
|
|
m->instance = instance;
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
m->bc_acked = grp->bc_snd_nxt - 1;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
grp->member_cnt++;
|
|
|
|
tipc_group_add_to_tree(grp, m);
|
|
|
|
tipc_nlist_add(&grp->dests, m->node);
|
|
|
|
m->state = state;
|
|
|
|
return m;
|
|
|
|
}
|
|
|
|
|
2018-01-09 04:03:28 +08:00
|
|
|
void tipc_group_add_member(struct tipc_group *grp, u32 node,
|
|
|
|
u32 port, u32 instance)
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
{
|
2018-01-09 04:03:28 +08:00
|
|
|
tipc_group_create_member(grp, node, port, instance, MBR_PUBLISHED);
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void tipc_group_delete_member(struct tipc_group *grp,
|
|
|
|
struct tipc_member *m)
|
|
|
|
{
|
|
|
|
rb_erase(&m->tree_node, &grp->members);
|
|
|
|
grp->member_cnt--;
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
|
|
|
|
/* Check if we were waiting for replicast ack from this member */
|
|
|
|
if (grp->bc_ackers && less(m->bc_acked, grp->bc_snd_nxt - 1))
|
|
|
|
grp->bc_ackers--;
|
|
|
|
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
list_del_init(&m->list);
|
2018-01-04 22:20:44 +08:00
|
|
|
list_del_init(&m->small_win);
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
tipc_group_decr_active(grp, m);
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
|
|
|
|
/* If last member on a node, remove node from dest list */
|
|
|
|
if (!tipc_group_find_node(grp, m->node))
|
|
|
|
tipc_nlist_del(&grp->dests, m->node);
|
|
|
|
|
|
|
|
kfree(m);
|
|
|
|
}
|
|
|
|
|
|
|
|
struct tipc_nlist *tipc_group_dests(struct tipc_group *grp)
|
|
|
|
{
|
|
|
|
return &grp->dests;
|
|
|
|
}
|
|
|
|
|
|
|
|
void tipc_group_self(struct tipc_group *grp, struct tipc_name_seq *seq,
|
|
|
|
int *scope)
|
|
|
|
{
|
|
|
|
seq->type = grp->type;
|
|
|
|
seq->lower = grp->instance;
|
|
|
|
seq->upper = grp->instance;
|
|
|
|
*scope = grp->scope;
|
|
|
|
}
|
|
|
|
|
2017-10-13 17:04:26 +08:00
|
|
|
void tipc_group_update_member(struct tipc_member *m, int len)
|
|
|
|
{
|
|
|
|
struct tipc_group *grp = m->group;
|
|
|
|
struct tipc_member *_m, *tmp;
|
|
|
|
|
2018-01-04 22:20:44 +08:00
|
|
|
if (!tipc_group_is_receiver(m))
|
2017-10-13 17:04:26 +08:00
|
|
|
return;
|
|
|
|
|
|
|
|
m->window -= len;
|
|
|
|
|
|
|
|
if (m->window >= ADV_IDLE)
|
|
|
|
return;
|
|
|
|
|
2018-01-04 22:20:44 +08:00
|
|
|
list_del_init(&m->small_win);
|
2017-10-13 17:04:26 +08:00
|
|
|
|
2018-01-04 22:20:44 +08:00
|
|
|
/* Sort member into small_window members' list */
|
|
|
|
list_for_each_entry_safe(_m, tmp, &grp->small_win, small_win) {
|
2018-01-04 22:20:45 +08:00
|
|
|
if (_m->window > m->window)
|
|
|
|
break;
|
2017-10-13 17:04:26 +08:00
|
|
|
}
|
2018-01-04 22:20:45 +08:00
|
|
|
list_add_tail(&m->small_win, &_m->small_win);
|
2017-10-13 17:04:26 +08:00
|
|
|
}
|
|
|
|
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
void tipc_group_update_bc_members(struct tipc_group *grp, int len, bool ack)
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
{
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
u16 prev = grp->bc_snd_nxt - 1;
|
2017-10-13 17:04:26 +08:00
|
|
|
struct tipc_member *m;
|
|
|
|
struct rb_node *n;
|
2017-12-21 20:07:11 +08:00
|
|
|
u16 ackers = 0;
|
2017-10-13 17:04:26 +08:00
|
|
|
|
|
|
|
for (n = rb_first(&grp->members); n; n = rb_next(n)) {
|
|
|
|
m = container_of(n, struct tipc_member, tree_node);
|
2018-01-04 22:20:44 +08:00
|
|
|
if (tipc_group_is_receiver(m)) {
|
2017-10-13 17:04:26 +08:00
|
|
|
tipc_group_update_member(m, len);
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
m->bc_acked = prev;
|
2017-12-21 20:07:11 +08:00
|
|
|
ackers++;
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
}
|
2017-10-13 17:04:26 +08:00
|
|
|
}
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
|
|
|
|
/* Mark number of acknowledges to expect, if any */
|
|
|
|
if (ack)
|
2017-12-21 20:07:11 +08:00
|
|
|
grp->bc_ackers = ackers;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
grp->bc_snd_nxt++;
|
|
|
|
}
|
|
|
|
|
2017-10-13 17:04:27 +08:00
|
|
|
bool tipc_group_cong(struct tipc_group *grp, u32 dnode, u32 dport,
|
|
|
|
int len, struct tipc_member **mbr)
|
2017-10-13 17:04:26 +08:00
|
|
|
{
|
2017-10-13 17:04:27 +08:00
|
|
|
struct sk_buff_head xmitq;
|
2017-10-13 17:04:26 +08:00
|
|
|
struct tipc_member *m;
|
2017-10-13 17:04:27 +08:00
|
|
|
int adv, state;
|
|
|
|
|
|
|
|
m = tipc_group_find_dest(grp, dnode, dport);
|
2018-01-09 04:03:31 +08:00
|
|
|
if (!tipc_group_is_receiver(m)) {
|
|
|
|
*mbr = NULL;
|
2017-10-13 17:04:27 +08:00
|
|
|
return false;
|
2018-01-09 04:03:31 +08:00
|
|
|
}
|
|
|
|
*mbr = m;
|
|
|
|
|
2017-10-13 17:04:27 +08:00
|
|
|
if (m->window >= len)
|
|
|
|
return false;
|
2018-01-09 04:03:31 +08:00
|
|
|
|
2018-01-17 23:42:46 +08:00
|
|
|
*grp->open = false;
|
2017-10-13 17:04:27 +08:00
|
|
|
|
|
|
|
/* If not fully advertised, do it now to prevent mutual blocking */
|
|
|
|
adv = m->advertised;
|
|
|
|
state = m->state;
|
|
|
|
if (state == MBR_JOINED && adv == ADV_IDLE)
|
|
|
|
return true;
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
if (state == MBR_ACTIVE && adv == ADV_ACTIVE)
|
|
|
|
return true;
|
|
|
|
if (state == MBR_PENDING && adv == ADV_IDLE)
|
|
|
|
return true;
|
2019-08-15 22:42:50 +08:00
|
|
|
__skb_queue_head_init(&xmitq);
|
2017-10-13 17:04:27 +08:00
|
|
|
tipc_group_proto_xmit(grp, m, GRP_ADV_MSG, &xmitq);
|
|
|
|
tipc_node_distr_xmit(grp->net, &xmitq);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool tipc_group_bc_cong(struct tipc_group *grp, int len)
|
|
|
|
{
|
|
|
|
struct tipc_member *m = NULL;
|
2017-10-13 17:04:26 +08:00
|
|
|
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
/* If prev bcast was replicast, reject until all receivers have acked */
|
2018-01-09 04:03:31 +08:00
|
|
|
if (grp->bc_ackers) {
|
2018-01-17 23:42:46 +08:00
|
|
|
*grp->open = false;
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
return true;
|
2018-01-09 04:03:31 +08:00
|
|
|
}
|
2018-01-04 22:20:44 +08:00
|
|
|
if (list_empty(&grp->small_win))
|
2017-10-13 17:04:26 +08:00
|
|
|
return false;
|
|
|
|
|
2018-01-04 22:20:44 +08:00
|
|
|
m = list_first_entry(&grp->small_win, struct tipc_member, small_win);
|
2017-10-13 17:04:26 +08:00
|
|
|
if (m->window >= len)
|
|
|
|
return false;
|
|
|
|
|
2017-10-13 17:04:27 +08:00
|
|
|
return tipc_group_cong(grp, m->node, m->port, len, &m);
|
2017-10-13 17:04:26 +08:00
|
|
|
}
|
|
|
|
|
2017-10-13 17:04:30 +08:00
|
|
|
/* tipc_group_sort_msg() - sort msg into queue by bcast sequence number
|
|
|
|
*/
|
|
|
|
static void tipc_group_sort_msg(struct sk_buff *skb, struct sk_buff_head *defq)
|
|
|
|
{
|
|
|
|
struct tipc_msg *_hdr, *hdr = buf_msg(skb);
|
|
|
|
u16 bc_seqno = msg_grp_bc_seqno(hdr);
|
|
|
|
struct sk_buff *_skb, *tmp;
|
|
|
|
int mtyp = msg_type(hdr);
|
|
|
|
|
tipc: guarantee delivery of last broadcast before DOWN event
The following scenario is possible:
- A user sends a broadcast message, and thereafter immediately leaves
the group.
- The LEAVE message, following a different path than the broadcast,
arrives ahead of the broadcast, and the sending member is removed
from the receiver's list.
- The broadcast message arrives, but is dropped because the sender
now is unknown to the receipient.
We fix this by sequence numbering membership events, just like ordinary
unicast messages. Currently, when a JOIN is sent to a peer, it contains
a synchronization point, - the sequence number of the next sent
broadcast, in order to give the receiver a start synchronization point.
We now let even LEAVE messages contain such an "end synchronization"
point, so that the recipient can delay the removal of the sending member
until it knows that all messages have been received.
The received synchronization points are added as sequence numbers to the
generated membership events, making it possible to handle them almost
the same way as regular unicasts in the receiving filter function. In
particular, a DOWN event with a too high sequence number will be kept
in the reordering queue until the missing broadcast(s) arrive and have
been delivered.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:33 +08:00
|
|
|
/* Bcast/mcast may be bypassed by ucast or other bcast, - sort it in */
|
2017-10-13 17:04:30 +08:00
|
|
|
if (mtyp == TIPC_GRP_BCAST_MSG || mtyp == TIPC_GRP_MCAST_MSG) {
|
|
|
|
skb_queue_walk_safe(defq, _skb, tmp) {
|
|
|
|
_hdr = buf_msg(_skb);
|
|
|
|
if (!less(bc_seqno, msg_grp_bc_seqno(_hdr)))
|
|
|
|
continue;
|
|
|
|
__skb_queue_before(defq, _skb, skb);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
/* Bcast was not bypassed, - add to tail */
|
|
|
|
}
|
|
|
|
/* Unicasts are never bypassed, - always add to tail */
|
|
|
|
__skb_queue_tail(defq, skb);
|
|
|
|
}
|
|
|
|
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
/* tipc_group_filter_msg() - determine if we should accept arriving message
|
|
|
|
*/
|
|
|
|
void tipc_group_filter_msg(struct tipc_group *grp, struct sk_buff_head *inputq,
|
|
|
|
struct sk_buff_head *xmitq)
|
|
|
|
{
|
|
|
|
struct sk_buff *skb = __skb_dequeue(inputq);
|
tipc: guarantee delivery of last broadcast before DOWN event
The following scenario is possible:
- A user sends a broadcast message, and thereafter immediately leaves
the group.
- The LEAVE message, following a different path than the broadcast,
arrives ahead of the broadcast, and the sending member is removed
from the receiver's list.
- The broadcast message arrives, but is dropped because the sender
now is unknown to the receipient.
We fix this by sequence numbering membership events, just like ordinary
unicast messages. Currently, when a JOIN is sent to a peer, it contains
a synchronization point, - the sequence number of the next sent
broadcast, in order to give the receiver a start synchronization point.
We now let even LEAVE messages contain such an "end synchronization"
point, so that the recipient can delay the removal of the sending member
until it knows that all messages have been received.
The received synchronization points are added as sequence numbers to the
generated membership events, making it possible to handle them almost
the same way as regular unicasts in the receiving filter function. In
particular, a DOWN event with a too high sequence number will be kept
in the reordering queue until the missing broadcast(s) arrive and have
been delivered.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:33 +08:00
|
|
|
bool ack, deliver, update, leave = false;
|
2017-10-13 17:04:30 +08:00
|
|
|
struct sk_buff_head *defq;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
struct tipc_member *m;
|
|
|
|
struct tipc_msg *hdr;
|
|
|
|
u32 node, port;
|
2017-10-13 17:04:30 +08:00
|
|
|
int mtyp, blks;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
|
|
|
|
if (!skb)
|
|
|
|
return;
|
|
|
|
|
|
|
|
hdr = buf_msg(skb);
|
|
|
|
node = msg_orignode(hdr);
|
|
|
|
port = msg_origport(hdr);
|
|
|
|
|
|
|
|
if (!msg_in_group(hdr))
|
|
|
|
goto drop;
|
|
|
|
|
|
|
|
m = tipc_group_find_member(grp, node, port);
|
2018-01-04 22:20:44 +08:00
|
|
|
if (!tipc_group_is_sender(m))
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
goto drop;
|
|
|
|
|
2017-10-13 17:04:30 +08:00
|
|
|
if (less(msg_grp_bc_seqno(hdr), m->bc_rcv_nxt))
|
|
|
|
goto drop;
|
2017-10-13 17:04:29 +08:00
|
|
|
|
2017-10-13 17:04:30 +08:00
|
|
|
TIPC_SKB_CB(skb)->orig_member = m->instance;
|
|
|
|
defq = &m->deferredq;
|
|
|
|
tipc_group_sort_msg(skb, defq);
|
|
|
|
|
|
|
|
while ((skb = skb_peek(defq))) {
|
|
|
|
hdr = buf_msg(skb);
|
|
|
|
mtyp = msg_type(hdr);
|
2017-11-28 03:13:39 +08:00
|
|
|
blks = msg_blocks(hdr);
|
2017-10-13 17:04:30 +08:00
|
|
|
deliver = true;
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
ack = false;
|
2017-10-13 17:04:30 +08:00
|
|
|
update = false;
|
|
|
|
|
|
|
|
if (more(msg_grp_bc_seqno(hdr), m->bc_rcv_nxt))
|
|
|
|
break;
|
|
|
|
|
|
|
|
/* Decide what to do with message */
|
|
|
|
switch (mtyp) {
|
|
|
|
case TIPC_GRP_MCAST_MSG:
|
|
|
|
if (msg_nameinst(hdr) != grp->instance) {
|
|
|
|
update = true;
|
|
|
|
deliver = false;
|
|
|
|
}
|
2020-08-24 06:36:59 +08:00
|
|
|
fallthrough;
|
2017-10-13 17:04:30 +08:00
|
|
|
case TIPC_GRP_BCAST_MSG:
|
|
|
|
m->bc_rcv_nxt++;
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
ack = msg_grp_bc_ack_req(hdr);
|
2017-10-13 17:04:30 +08:00
|
|
|
break;
|
|
|
|
case TIPC_GRP_UCAST_MSG:
|
|
|
|
break;
|
tipc: guarantee delivery of last broadcast before DOWN event
The following scenario is possible:
- A user sends a broadcast message, and thereafter immediately leaves
the group.
- The LEAVE message, following a different path than the broadcast,
arrives ahead of the broadcast, and the sending member is removed
from the receiver's list.
- The broadcast message arrives, but is dropped because the sender
now is unknown to the receipient.
We fix this by sequence numbering membership events, just like ordinary
unicast messages. Currently, when a JOIN is sent to a peer, it contains
a synchronization point, - the sequence number of the next sent
broadcast, in order to give the receiver a start synchronization point.
We now let even LEAVE messages contain such an "end synchronization"
point, so that the recipient can delay the removal of the sending member
until it knows that all messages have been received.
The received synchronization points are added as sequence numbers to the
generated membership events, making it possible to handle them almost
the same way as regular unicasts in the receiving filter function. In
particular, a DOWN event with a too high sequence number will be kept
in the reordering queue until the missing broadcast(s) arrive and have
been delivered.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:33 +08:00
|
|
|
case TIPC_GRP_MEMBER_EVT:
|
|
|
|
if (m->state == MBR_LEAVING)
|
|
|
|
leave = true;
|
|
|
|
if (!grp->events)
|
|
|
|
deliver = false;
|
|
|
|
break;
|
2017-10-13 17:04:30 +08:00
|
|
|
default:
|
|
|
|
break;
|
2017-10-13 17:04:29 +08:00
|
|
|
}
|
|
|
|
|
2017-10-13 17:04:30 +08:00
|
|
|
/* Execute decisions */
|
|
|
|
__skb_dequeue(defq);
|
|
|
|
if (deliver)
|
|
|
|
__skb_queue_tail(inputq, skb);
|
|
|
|
else
|
|
|
|
kfree_skb(skb);
|
|
|
|
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
if (ack)
|
|
|
|
tipc_group_proto_xmit(grp, m, GRP_ACK_MSG, xmitq);
|
|
|
|
|
tipc: guarantee delivery of last broadcast before DOWN event
The following scenario is possible:
- A user sends a broadcast message, and thereafter immediately leaves
the group.
- The LEAVE message, following a different path than the broadcast,
arrives ahead of the broadcast, and the sending member is removed
from the receiver's list.
- The broadcast message arrives, but is dropped because the sender
now is unknown to the receipient.
We fix this by sequence numbering membership events, just like ordinary
unicast messages. Currently, when a JOIN is sent to a peer, it contains
a synchronization point, - the sequence number of the next sent
broadcast, in order to give the receiver a start synchronization point.
We now let even LEAVE messages contain such an "end synchronization"
point, so that the recipient can delay the removal of the sending member
until it knows that all messages have been received.
The received synchronization points are added as sequence numbers to the
generated membership events, making it possible to handle them almost
the same way as regular unicasts in the receiving filter function. In
particular, a DOWN event with a too high sequence number will be kept
in the reordering queue until the missing broadcast(s) arrive and have
been delivered.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:33 +08:00
|
|
|
if (leave) {
|
|
|
|
__skb_queue_purge(defq);
|
2017-11-21 04:43:03 +08:00
|
|
|
tipc_group_delete_member(grp, m);
|
tipc: guarantee delivery of last broadcast before DOWN event
The following scenario is possible:
- A user sends a broadcast message, and thereafter immediately leaves
the group.
- The LEAVE message, following a different path than the broadcast,
arrives ahead of the broadcast, and the sending member is removed
from the receiver's list.
- The broadcast message arrives, but is dropped because the sender
now is unknown to the receipient.
We fix this by sequence numbering membership events, just like ordinary
unicast messages. Currently, when a JOIN is sent to a peer, it contains
a synchronization point, - the sequence number of the next sent
broadcast, in order to give the receiver a start synchronization point.
We now let even LEAVE messages contain such an "end synchronization"
point, so that the recipient can delay the removal of the sending member
until it knows that all messages have been received.
The received synchronization points are added as sequence numbers to the
generated membership events, making it possible to handle them almost
the same way as regular unicasts in the receiving filter function. In
particular, a DOWN event with a too high sequence number will be kept
in the reordering queue until the missing broadcast(s) arrive and have
been delivered.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:33 +08:00
|
|
|
break;
|
|
|
|
}
|
2017-10-13 17:04:30 +08:00
|
|
|
if (!update)
|
|
|
|
continue;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
|
2017-10-13 17:04:30 +08:00
|
|
|
tipc_group_update_rcv_win(grp, blks, node, port, xmitq);
|
|
|
|
}
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
return;
|
|
|
|
drop:
|
|
|
|
kfree_skb(skb);
|
|
|
|
}
|
|
|
|
|
2017-10-13 17:04:26 +08:00
|
|
|
void tipc_group_update_rcv_win(struct tipc_group *grp, int blks, u32 node,
|
|
|
|
u32 port, struct sk_buff_head *xmitq)
|
|
|
|
{
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
struct list_head *active = &grp->active;
|
|
|
|
int max_active = grp->max_active;
|
|
|
|
int reclaim_limit = max_active * 3 / 4;
|
|
|
|
int active_cnt = grp->active_cnt;
|
2017-12-30 02:48:02 +08:00
|
|
|
struct tipc_member *m, *rm, *pm;
|
2017-10-13 17:04:26 +08:00
|
|
|
|
|
|
|
m = tipc_group_find_member(grp, node, port);
|
|
|
|
if (!m)
|
|
|
|
return;
|
|
|
|
|
|
|
|
m->advertised -= blks;
|
|
|
|
|
|
|
|
switch (m->state) {
|
|
|
|
case MBR_JOINED:
|
2018-01-09 04:03:24 +08:00
|
|
|
/* First, decide if member can go active */
|
|
|
|
if (active_cnt <= max_active) {
|
|
|
|
m->state = MBR_ACTIVE;
|
|
|
|
list_add_tail(&m->list, active);
|
|
|
|
grp->active_cnt++;
|
|
|
|
tipc_group_proto_xmit(grp, m, GRP_ADV_MSG, xmitq);
|
|
|
|
} else {
|
|
|
|
m->state = MBR_PENDING;
|
|
|
|
list_add_tail(&m->list, &grp->pending);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (active_cnt < reclaim_limit)
|
|
|
|
break;
|
|
|
|
|
|
|
|
/* Reclaim from oldest active member, if possible */
|
|
|
|
if (!list_empty(active)) {
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
rm = list_first_entry(active, struct tipc_member, list);
|
|
|
|
rm->state = MBR_RECLAIMING;
|
2018-01-09 04:03:23 +08:00
|
|
|
list_del_init(&rm->list);
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
tipc_group_proto_xmit(grp, rm, GRP_RECLAIM_MSG, xmitq);
|
|
|
|
break;
|
|
|
|
}
|
2018-01-09 04:03:24 +08:00
|
|
|
/* Nobody to reclaim from; - revert oldest pending to JOINED */
|
|
|
|
pm = list_first_entry(&grp->pending, struct tipc_member, list);
|
|
|
|
list_del_init(&pm->list);
|
|
|
|
pm->state = MBR_JOINED;
|
|
|
|
tipc_group_proto_xmit(grp, pm, GRP_ADV_MSG, xmitq);
|
|
|
|
break;
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
case MBR_ACTIVE:
|
|
|
|
if (!list_is_last(&m->list, &grp->active))
|
|
|
|
list_move_tail(&m->list, &grp->active);
|
|
|
|
if (m->advertised > (ADV_ACTIVE * 3 / 4))
|
|
|
|
break;
|
|
|
|
tipc_group_proto_xmit(grp, m, GRP_ADV_MSG, xmitq);
|
|
|
|
break;
|
|
|
|
case MBR_REMITTED:
|
|
|
|
if (m->advertised > ADV_IDLE)
|
|
|
|
break;
|
|
|
|
m->state = MBR_JOINED;
|
2018-01-09 04:03:23 +08:00
|
|
|
grp->active_cnt--;
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
if (m->advertised < ADV_IDLE) {
|
|
|
|
pr_warn_ratelimited("Rcv unexpected msg after REMIT\n");
|
2017-10-13 17:04:26 +08:00
|
|
|
tipc_group_proto_xmit(grp, m, GRP_ADV_MSG, xmitq);
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
}
|
2018-01-09 04:03:23 +08:00
|
|
|
|
2017-12-30 02:48:02 +08:00
|
|
|
if (list_empty(&grp->pending))
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* Set oldest pending member to active and advertise */
|
|
|
|
pm = list_first_entry(&grp->pending, struct tipc_member, list);
|
|
|
|
pm->state = MBR_ACTIVE;
|
|
|
|
list_move_tail(&pm->list, &grp->active);
|
|
|
|
grp->active_cnt++;
|
|
|
|
tipc_group_proto_xmit(grp, pm, GRP_ADV_MSG, xmitq);
|
2017-10-13 17:04:26 +08:00
|
|
|
break;
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
case MBR_RECLAIMING:
|
2017-10-13 17:04:26 +08:00
|
|
|
case MBR_JOINING:
|
|
|
|
case MBR_LEAVING:
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-01-09 04:03:26 +08:00
|
|
|
static void tipc_group_create_event(struct tipc_group *grp,
|
|
|
|
struct tipc_member *m,
|
|
|
|
u32 event, u16 seqno,
|
|
|
|
struct sk_buff_head *inputq)
|
|
|
|
{ u32 dnode = tipc_own_addr(grp->net);
|
|
|
|
struct tipc_event evt;
|
|
|
|
struct sk_buff *skb;
|
|
|
|
struct tipc_msg *hdr;
|
|
|
|
|
2018-10-18 23:38:29 +08:00
|
|
|
memset(&evt, 0, sizeof(evt));
|
2018-01-09 04:03:26 +08:00
|
|
|
evt.event = event;
|
|
|
|
evt.found_lower = m->instance;
|
|
|
|
evt.found_upper = m->instance;
|
|
|
|
evt.port.ref = m->port;
|
|
|
|
evt.port.node = m->node;
|
|
|
|
evt.s.seq.type = grp->type;
|
|
|
|
evt.s.seq.lower = m->instance;
|
|
|
|
evt.s.seq.upper = m->instance;
|
|
|
|
|
|
|
|
skb = tipc_msg_create(TIPC_CRITICAL_IMPORTANCE, TIPC_GRP_MEMBER_EVT,
|
|
|
|
GROUP_H_SIZE, sizeof(evt), dnode, m->node,
|
|
|
|
grp->portid, m->port, 0);
|
|
|
|
if (!skb)
|
|
|
|
return;
|
|
|
|
|
|
|
|
hdr = buf_msg(skb);
|
|
|
|
msg_set_nametype(hdr, grp->type);
|
|
|
|
msg_set_grp_evt(hdr, event);
|
|
|
|
msg_set_dest_droppable(hdr, true);
|
|
|
|
msg_set_grp_bc_seqno(hdr, seqno);
|
|
|
|
memcpy(msg_data(hdr), &evt, sizeof(evt));
|
|
|
|
TIPC_SKB_CB(skb)->orig_member = m->instance;
|
|
|
|
__skb_queue_tail(inputq, skb);
|
|
|
|
}
|
|
|
|
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
static void tipc_group_proto_xmit(struct tipc_group *grp, struct tipc_member *m,
|
|
|
|
int mtyp, struct sk_buff_head *xmitq)
|
|
|
|
{
|
|
|
|
struct tipc_msg *hdr;
|
|
|
|
struct sk_buff *skb;
|
2017-10-13 17:04:26 +08:00
|
|
|
int adv = 0;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
|
|
|
|
skb = tipc_msg_create(GROUP_PROTOCOL, mtyp, INT_H_SIZE, 0,
|
|
|
|
m->node, tipc_own_addr(grp->net),
|
|
|
|
m->port, grp->portid, 0);
|
|
|
|
if (!skb)
|
|
|
|
return;
|
|
|
|
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
if (m->state == MBR_ACTIVE)
|
2017-10-13 17:04:26 +08:00
|
|
|
adv = ADV_ACTIVE - m->advertised;
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
else if (m->state == MBR_JOINED || m->state == MBR_PENDING)
|
|
|
|
adv = ADV_IDLE - m->advertised;
|
2017-10-13 17:04:26 +08:00
|
|
|
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
hdr = buf_msg(skb);
|
2017-10-13 17:04:26 +08:00
|
|
|
|
|
|
|
if (mtyp == GRP_JOIN_MSG) {
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
msg_set_grp_bc_syncpt(hdr, grp->bc_snd_nxt);
|
2017-10-13 17:04:26 +08:00
|
|
|
msg_set_adv_win(hdr, adv);
|
|
|
|
m->advertised += adv;
|
tipc: guarantee delivery of last broadcast before DOWN event
The following scenario is possible:
- A user sends a broadcast message, and thereafter immediately leaves
the group.
- The LEAVE message, following a different path than the broadcast,
arrives ahead of the broadcast, and the sending member is removed
from the receiver's list.
- The broadcast message arrives, but is dropped because the sender
now is unknown to the receipient.
We fix this by sequence numbering membership events, just like ordinary
unicast messages. Currently, when a JOIN is sent to a peer, it contains
a synchronization point, - the sequence number of the next sent
broadcast, in order to give the receiver a start synchronization point.
We now let even LEAVE messages contain such an "end synchronization"
point, so that the recipient can delay the removal of the sending member
until it knows that all messages have been received.
The received synchronization points are added as sequence numbers to the
generated membership events, making it possible to handle them almost
the same way as regular unicasts in the receiving filter function. In
particular, a DOWN event with a too high sequence number will be kept
in the reordering queue until the missing broadcast(s) arrive and have
been delivered.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:33 +08:00
|
|
|
} else if (mtyp == GRP_LEAVE_MSG) {
|
|
|
|
msg_set_grp_bc_syncpt(hdr, grp->bc_snd_nxt);
|
2017-10-13 17:04:26 +08:00
|
|
|
} else if (mtyp == GRP_ADV_MSG) {
|
|
|
|
msg_set_adv_win(hdr, adv);
|
|
|
|
m->advertised += adv;
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
} else if (mtyp == GRP_ACK_MSG) {
|
|
|
|
msg_set_grp_bc_acked(hdr, m->bc_rcv_nxt);
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
} else if (mtyp == GRP_REMIT_MSG) {
|
|
|
|
msg_set_grp_remitted(hdr, m->window);
|
2017-10-13 17:04:26 +08:00
|
|
|
}
|
2017-12-19 00:34:16 +08:00
|
|
|
msg_set_dest_droppable(hdr, true);
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
__skb_queue_tail(xmitq, skb);
|
|
|
|
}
|
|
|
|
|
2017-10-13 17:04:26 +08:00
|
|
|
void tipc_group_proto_rcv(struct tipc_group *grp, bool *usr_wakeup,
|
|
|
|
struct tipc_msg *hdr, struct sk_buff_head *inputq,
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
struct sk_buff_head *xmitq)
|
|
|
|
{
|
|
|
|
u32 node = msg_orignode(hdr);
|
|
|
|
u32 port = msg_origport(hdr);
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
struct tipc_member *m, *pm;
|
|
|
|
u16 remitted, in_flight;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
|
|
|
|
if (!grp)
|
|
|
|
return;
|
|
|
|
|
2018-01-09 04:03:30 +08:00
|
|
|
if (grp->scope == TIPC_NODE_SCOPE && node != tipc_own_addr(grp->net))
|
|
|
|
return;
|
|
|
|
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
m = tipc_group_find_member(grp, node, port);
|
|
|
|
|
|
|
|
switch (msg_type(hdr)) {
|
|
|
|
case GRP_JOIN_MSG:
|
|
|
|
if (!m)
|
|
|
|
m = tipc_group_create_member(grp, node, port,
|
2018-01-09 04:03:28 +08:00
|
|
|
0, MBR_JOINING);
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
if (!m)
|
|
|
|
return;
|
tipc: guarantee delivery of last broadcast before DOWN event
The following scenario is possible:
- A user sends a broadcast message, and thereafter immediately leaves
the group.
- The LEAVE message, following a different path than the broadcast,
arrives ahead of the broadcast, and the sending member is removed
from the receiver's list.
- The broadcast message arrives, but is dropped because the sender
now is unknown to the receipient.
We fix this by sequence numbering membership events, just like ordinary
unicast messages. Currently, when a JOIN is sent to a peer, it contains
a synchronization point, - the sequence number of the next sent
broadcast, in order to give the receiver a start synchronization point.
We now let even LEAVE messages contain such an "end synchronization"
point, so that the recipient can delay the removal of the sending member
until it knows that all messages have been received.
The received synchronization points are added as sequence numbers to the
generated membership events, making it possible to handle them almost
the same way as regular unicasts in the receiving filter function. In
particular, a DOWN event with a too high sequence number will be kept
in the reordering queue until the missing broadcast(s) arrive and have
been delivered.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:33 +08:00
|
|
|
m->bc_syncpt = msg_grp_bc_syncpt(hdr);
|
|
|
|
m->bc_rcv_nxt = m->bc_syncpt;
|
2017-10-13 17:04:26 +08:00
|
|
|
m->window += msg_adv_win(hdr);
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
|
2018-01-09 04:03:28 +08:00
|
|
|
/* Wait until PUBLISH event is received if necessary */
|
|
|
|
if (m->state != MBR_PUBLISHED)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* Member can be taken into service */
|
|
|
|
m->state = MBR_JOINED;
|
2018-01-09 04:03:31 +08:00
|
|
|
tipc_group_open(m, usr_wakeup);
|
2017-12-20 18:03:15 +08:00
|
|
|
tipc_group_update_member(m, 0);
|
2018-01-09 04:03:28 +08:00
|
|
|
tipc_group_proto_xmit(grp, m, GRP_ADV_MSG, xmitq);
|
|
|
|
tipc_group_create_event(grp, m, TIPC_PUBLISHED,
|
|
|
|
m->bc_syncpt, inputq);
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
return;
|
|
|
|
case GRP_LEAVE_MSG:
|
|
|
|
if (!m)
|
|
|
|
return;
|
tipc: guarantee delivery of last broadcast before DOWN event
The following scenario is possible:
- A user sends a broadcast message, and thereafter immediately leaves
the group.
- The LEAVE message, following a different path than the broadcast,
arrives ahead of the broadcast, and the sending member is removed
from the receiver's list.
- The broadcast message arrives, but is dropped because the sender
now is unknown to the receipient.
We fix this by sequence numbering membership events, just like ordinary
unicast messages. Currently, when a JOIN is sent to a peer, it contains
a synchronization point, - the sequence number of the next sent
broadcast, in order to give the receiver a start synchronization point.
We now let even LEAVE messages contain such an "end synchronization"
point, so that the recipient can delay the removal of the sending member
until it knows that all messages have been received.
The received synchronization points are added as sequence numbers to the
generated membership events, making it possible to handle them almost
the same way as regular unicasts in the receiving filter function. In
particular, a DOWN event with a too high sequence number will be kept
in the reordering queue until the missing broadcast(s) arrive and have
been delivered.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:33 +08:00
|
|
|
m->bc_syncpt = msg_grp_bc_syncpt(hdr);
|
2017-12-19 01:13:34 +08:00
|
|
|
list_del_init(&m->list);
|
2018-01-09 04:03:31 +08:00
|
|
|
tipc_group_open(m, usr_wakeup);
|
2018-01-09 04:03:27 +08:00
|
|
|
tipc_group_decr_active(grp, m);
|
|
|
|
m->state = MBR_LEAVING;
|
2018-01-09 04:03:26 +08:00
|
|
|
tipc_group_create_event(grp, m, TIPC_WITHDRAWN,
|
|
|
|
m->bc_syncpt, inputq);
|
2017-10-13 17:04:26 +08:00
|
|
|
return;
|
|
|
|
case GRP_ADV_MSG:
|
|
|
|
if (!m)
|
|
|
|
return;
|
|
|
|
m->window += msg_adv_win(hdr);
|
2018-01-09 04:03:31 +08:00
|
|
|
tipc_group_open(m, usr_wakeup);
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
return;
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
case GRP_ACK_MSG:
|
|
|
|
if (!m)
|
|
|
|
return;
|
|
|
|
m->bc_acked = msg_grp_bc_acked(hdr);
|
|
|
|
if (--grp->bc_ackers)
|
2018-01-09 04:03:31 +08:00
|
|
|
return;
|
|
|
|
list_del_init(&m->small_win);
|
2018-01-17 23:42:46 +08:00
|
|
|
*m->group->open = true;
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
*usr_wakeup = true;
|
2018-01-09 04:03:31 +08:00
|
|
|
tipc_group_update_member(m, 0);
|
tipc: guarantee that group broadcast doesn't bypass group unicast
We need a mechanism guaranteeing that group unicasts sent out from a
socket are not bypassed by later sent broadcasts from the same socket.
We do this as follows:
- Each time a unicast is sent, we set a the broadcast method for the
socket to "replicast" and "mandatory". This forces the first
subsequent broadcast message to follow the same network and data path
as the preceding unicast to a destination, hence preventing it from
overtaking the latter.
- In order to make the 'same data path' statement above true, we let
group unicasts pass through the multicast link input queue, instead
of as previously through the unicast link input queue.
- In the first broadcast following a unicast, we set a new header flag,
requiring all recipients to immediately acknowledge its reception.
- During the period before all the expected acknowledges are received,
the socket refuses to accept any more broadcast attempts, i.e., by
blocking or returning EAGAIN. This period should typically not be
longer than a few microseconds.
- When all acknowledges have been received, the sending socket will
open up for subsequent broadcasts, this time giving the link layer
freedom to itself select the best transmission method.
- The forced and/or abrupt transmission method changes described above
may lead to broadcasts arriving out of order to the recipients. We
remedy this by introducing code that checks and if necessary
re-orders such messages at the receiving end.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:31 +08:00
|
|
|
return;
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
case GRP_RECLAIM_MSG:
|
|
|
|
if (!m)
|
|
|
|
return;
|
|
|
|
tipc_group_proto_xmit(grp, m, GRP_REMIT_MSG, xmitq);
|
|
|
|
m->window = ADV_IDLE;
|
2018-01-09 04:03:31 +08:00
|
|
|
tipc_group_open(m, usr_wakeup);
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
return;
|
|
|
|
case GRP_REMIT_MSG:
|
|
|
|
if (!m || m->state != MBR_RECLAIMING)
|
|
|
|
return;
|
|
|
|
|
|
|
|
remitted = msg_grp_remitted(hdr);
|
|
|
|
|
|
|
|
/* Messages preceding the REMIT still in receive queue */
|
|
|
|
if (m->advertised > remitted) {
|
|
|
|
m->state = MBR_REMITTED;
|
|
|
|
in_flight = m->advertised - remitted;
|
2017-12-30 02:48:02 +08:00
|
|
|
m->advertised = ADV_IDLE + in_flight;
|
|
|
|
return;
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
}
|
2018-01-09 04:03:23 +08:00
|
|
|
/* This should never happen */
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
if (m->advertised < remitted)
|
2018-01-09 04:03:23 +08:00
|
|
|
pr_warn_ratelimited("Unexpected REMIT msg\n");
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
|
2018-01-09 04:03:23 +08:00
|
|
|
/* All messages preceding the REMIT have been read */
|
|
|
|
m->state = MBR_JOINED;
|
2017-12-30 02:48:02 +08:00
|
|
|
grp->active_cnt--;
|
2018-01-09 04:03:23 +08:00
|
|
|
m->advertised = ADV_IDLE;
|
tipc: add multipoint-to-point flow control
We already have point-to-multipoint flow control within a group. But
we even need the opposite; -a scheme which can handle that potentially
hundreds of sources may try to send messages to the same destination
simultaneously without causing buffer overflow at the recipient. This
commit adds such a mechanism.
The algorithm works as follows:
- When a member detects a new, joining member, it initially set its
state to JOINED and advertises a minimum window to the new member.
This window is chosen so that the new member can send exactly one
maximum sized message, or several smaller ones, to the recipient
before it must stop and wait for an additional advertisement. This
minimum window ADV_IDLE is set to 65 1kB blocks.
- When a member receives the first data message from a JOINED member,
it changes the state of the latter to ACTIVE, and advertises a larger
window ADV_ACTIVE = 12 x ADV_IDLE blocks to the sender, so it can
continue sending with minimal disturbances to the data flow.
- The active members are kept in a dedicated linked list. Each time a
message is received from an active member, it will be moved to the
tail of that list. This way, we keep a record of which members have
been most (tail) and least (head) recently active.
- There is a maximum number (16) of permitted simultaneous active
senders per receiver. When this limit is reached, the receiver will
not advertise anything immediately to a new sender, but instead put
it in a PENDING state, and add it to a corresponding queue. At the
same time, it will pick the least recently active member, send it an
advertisement RECLAIM message, and set this member to state
RECLAIMING.
- The reclaimee member has to respond with a REMIT message, meaning that
it goes back to a send window of ADV_IDLE, and returns its unused
advertised blocks beyond that value to the reclaiming member.
- When the reclaiming member receives the REMIT message, it unlinks
the reclaimee from its active list, resets its state to JOINED, and
notes that it is now back at ADV_IDLE advertised blocks to that
member. If there are still unread data messages sent out by
reclaimee before the REMIT, the member goes into an intermediate
state REMITTED, where it stays until the said messages have been
consumed.
- The returned advertised blocks can now be re-advertised to the
pending member, which is now set to state ACTIVE and added to
the active member list.
- To be proactive, i.e., to minimize the risk that any member will
end up in the pending queue, we start reclaiming resources already
when the number of active members exceeds 3/4 of the permitted
maximum.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:34 +08:00
|
|
|
|
|
|
|
/* Set oldest pending member to active and advertise */
|
|
|
|
if (list_empty(&grp->pending))
|
|
|
|
return;
|
|
|
|
pm = list_first_entry(&grp->pending, struct tipc_member, list);
|
|
|
|
pm->state = MBR_ACTIVE;
|
|
|
|
list_move_tail(&pm->list, &grp->active);
|
|
|
|
grp->active_cnt++;
|
|
|
|
if (pm->advertised <= (ADV_ACTIVE * 3 / 4))
|
|
|
|
tipc_group_proto_xmit(grp, pm, GRP_ADV_MSG, xmitq);
|
|
|
|
return;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
default:
|
|
|
|
pr_warn("Received unknown GROUP_PROTO message\n");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-10-13 17:04:26 +08:00
|
|
|
/* tipc_group_member_evt() - receive and handle a member up/down event
|
|
|
|
*/
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
void tipc_group_member_evt(struct tipc_group *grp,
|
2017-10-13 17:04:26 +08:00
|
|
|
bool *usr_wakeup,
|
|
|
|
int *sk_rcvbuf,
|
2018-01-09 04:03:26 +08:00
|
|
|
struct tipc_msg *hdr,
|
2017-10-13 17:04:25 +08:00
|
|
|
struct sk_buff_head *inputq,
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
struct sk_buff_head *xmitq)
|
|
|
|
{
|
|
|
|
struct tipc_event *evt = (void *)msg_data(hdr);
|
2017-10-13 17:04:25 +08:00
|
|
|
u32 instance = evt->found_lower;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
u32 node = evt->port.node;
|
|
|
|
u32 port = evt->port.ref;
|
2017-10-13 17:04:25 +08:00
|
|
|
int event = evt->event;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
struct tipc_member *m;
|
|
|
|
struct net *net;
|
|
|
|
u32 self;
|
|
|
|
|
|
|
|
if (!grp)
|
2018-01-09 04:03:26 +08:00
|
|
|
return;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
|
|
|
|
net = grp->net;
|
|
|
|
self = tipc_own_addr(net);
|
|
|
|
if (!grp->loopback && node == self && port == grp->portid)
|
2018-01-09 04:03:26 +08:00
|
|
|
return;
|
2017-10-13 17:04:25 +08:00
|
|
|
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
m = tipc_group_find_member(grp, node, port);
|
|
|
|
|
2018-01-09 04:03:28 +08:00
|
|
|
switch (event) {
|
|
|
|
case TIPC_PUBLISHED:
|
|
|
|
/* Send and wait for arrival of JOIN message if necessary */
|
|
|
|
if (!m) {
|
|
|
|
m = tipc_group_create_member(grp, node, port, instance,
|
|
|
|
MBR_PUBLISHED);
|
|
|
|
if (!m)
|
|
|
|
break;
|
|
|
|
tipc_group_update_member(m, 0);
|
|
|
|
tipc_group_proto_xmit(grp, m, GRP_JOIN_MSG, xmitq);
|
|
|
|
break;
|
|
|
|
}
|
2018-01-09 04:03:26 +08:00
|
|
|
|
2018-01-09 04:03:28 +08:00
|
|
|
if (m->state != MBR_JOINING)
|
|
|
|
break;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
|
2018-01-09 04:03:28 +08:00
|
|
|
/* Member can be taken into service */
|
|
|
|
m->instance = instance;
|
|
|
|
m->state = MBR_JOINED;
|
2018-01-09 04:03:31 +08:00
|
|
|
tipc_group_open(m, usr_wakeup);
|
2018-01-04 22:20:45 +08:00
|
|
|
tipc_group_update_member(m, 0);
|
2018-01-09 04:03:28 +08:00
|
|
|
tipc_group_proto_xmit(grp, m, GRP_JOIN_MSG, xmitq);
|
|
|
|
tipc_group_create_event(grp, m, TIPC_PUBLISHED,
|
|
|
|
m->bc_syncpt, inputq);
|
|
|
|
break;
|
|
|
|
case TIPC_WITHDRAWN:
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
if (!m)
|
2018-01-09 04:03:28 +08:00
|
|
|
break;
|
2017-10-13 17:04:25 +08:00
|
|
|
|
2018-01-09 04:03:27 +08:00
|
|
|
tipc_group_decr_active(grp, m);
|
|
|
|
m->state = MBR_LEAVING;
|
2017-12-19 01:13:34 +08:00
|
|
|
list_del_init(&m->list);
|
2018-01-09 04:03:31 +08:00
|
|
|
tipc_group_open(m, usr_wakeup);
|
2018-01-09 04:03:27 +08:00
|
|
|
|
|
|
|
/* Only send event if no LEAVE message can be expected */
|
|
|
|
if (!tipc_node_is_up(net, node))
|
|
|
|
tipc_group_create_event(grp, m, TIPC_WITHDRAWN,
|
|
|
|
m->bc_rcv_nxt, inputq);
|
2018-01-09 04:03:28 +08:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
}
|
2017-10-13 17:04:26 +08:00
|
|
|
*sk_rcvbuf = tipc_group_rcvbuf_limit(grp);
|
tipc: introduce communication groups
As a preparation for introducing flow control for multicast and datagram
messaging we need a more strictly defined framework than we have now. A
socket must be able keep track of exactly how many and which other
sockets it is allowed to communicate with at any moment, and keep the
necessary state for those.
We therefore introduce a new concept we have named Communication Group.
Sockets can join a group via a new setsockopt() call TIPC_GROUP_JOIN.
The call takes four parameters: 'type' serves as group identifier,
'instance' serves as an logical member identifier, and 'scope' indicates
the visibility of the group (node/cluster/zone). Finally, 'flags' makes
it possible to set certain properties for the member. For now, there is
only one flag, indicating if the creator of the socket wants to receive
a copy of broadcast or multicast messages it is sending via the socket,
and if wants to be eligible as destination for its own anycasts.
A group is closed, i.e., sockets which have not joined a group will
not be able to send messages to or receive messages from members of
the group, and vice versa.
Any member of a group can send multicast ('group broadcast') messages
to all group members, optionally including itself, using the primitive
send(). The messages are received via the recvmsg() primitive. A socket
can only be member of one group at a time.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-13 17:04:23 +08:00
|
|
|
}
|
2018-06-29 19:26:18 +08:00
|
|
|
|
|
|
|
int tipc_group_fill_sock_diag(struct tipc_group *grp, struct sk_buff *skb)
|
|
|
|
{
|
2019-04-26 17:13:06 +08:00
|
|
|
struct nlattr *group = nla_nest_start_noflag(skb, TIPC_NLA_SOCK_GROUP);
|
2018-06-29 19:26:18 +08:00
|
|
|
|
2019-03-16 01:11:59 +08:00
|
|
|
if (!group)
|
|
|
|
return -EMSGSIZE;
|
|
|
|
|
2018-06-29 19:26:18 +08:00
|
|
|
if (nla_put_u32(skb, TIPC_NLA_SOCK_GROUP_ID,
|
|
|
|
grp->type) ||
|
|
|
|
nla_put_u32(skb, TIPC_NLA_SOCK_GROUP_INSTANCE,
|
|
|
|
grp->instance) ||
|
|
|
|
nla_put_u32(skb, TIPC_NLA_SOCK_GROUP_BC_SEND_NEXT,
|
|
|
|
grp->bc_snd_nxt))
|
|
|
|
goto group_msg_cancel;
|
|
|
|
|
|
|
|
if (grp->scope == TIPC_NODE_SCOPE)
|
|
|
|
if (nla_put_flag(skb, TIPC_NLA_SOCK_GROUP_NODE_SCOPE))
|
|
|
|
goto group_msg_cancel;
|
|
|
|
|
|
|
|
if (grp->scope == TIPC_CLUSTER_SCOPE)
|
|
|
|
if (nla_put_flag(skb, TIPC_NLA_SOCK_GROUP_CLUSTER_SCOPE))
|
|
|
|
goto group_msg_cancel;
|
|
|
|
|
|
|
|
if (*grp->open)
|
|
|
|
if (nla_put_flag(skb, TIPC_NLA_SOCK_GROUP_OPEN))
|
|
|
|
goto group_msg_cancel;
|
|
|
|
|
|
|
|
nla_nest_end(skb, group);
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
group_msg_cancel:
|
|
|
|
nla_nest_cancel(skb, group);
|
|
|
|
return -1;
|
|
|
|
}
|