Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * sch_plug.c Queue traffic until an explicit release command
4 *
5 * There are two ways to use this qdisc:
6 * 1. A simple "instantaneous" plug/unplug operation, by issuing an alternating
7 * sequence of TCQ_PLUG_BUFFER & TCQ_PLUG_RELEASE_INDEFINITE commands.
8 *
9 * 2. For network output buffering (a.k.a output commit) functionality.
10 * Output commit property is commonly used by applications using checkpoint
11 * based fault-tolerance to ensure that the checkpoint from which a system
12 * is being restored is consistent w.r.t outside world.
13 *
14 * Consider for e.g. Remus - a Virtual Machine checkpointing system,
15 * wherein a VM is checkpointed, say every 50ms. The checkpoint is replicated
16 * asynchronously to the backup host, while the VM continues executing the
17 * next epoch speculatively.
18 *
19 * The following is a typical sequence of output buffer operations:
20 * 1.At epoch i, start_buffer(i)
21 * 2. At end of epoch i (i.e. after 50ms):
22 * 2.1 Stop VM and take checkpoint(i).
23 * 2.2 start_buffer(i+1) and Resume VM
24 * 3. While speculatively executing epoch(i+1), asynchronously replicate
25 * checkpoint(i) to backup host.
26 * 4. When checkpoint_ack(i) is received from backup, release_buffer(i)
27 * Thus, this Qdisc would receive the following sequence of commands:
28 * TCQ_PLUG_BUFFER (epoch i)
29 * .. TCQ_PLUG_BUFFER (epoch i+1)
30 * ....TCQ_PLUG_RELEASE_ONE (epoch i)
31 * ......TCQ_PLUG_BUFFER (epoch i+2)
32 * ........
33 */
34
35#include <linux/module.h>
36#include <linux/types.h>
37#include <linux/kernel.h>
38#include <linux/errno.h>
39#include <linux/netdevice.h>
40#include <linux/skbuff.h>
41#include <net/pkt_sched.h>
42
43/*
44 * State of the queue, when used for network output buffering:
45 *
46 * plug(i+1) plug(i) head
47 * ------------------+--------------------+---------------->
48 * | |
49 * | |
50 * pkts_current_epoch| pkts_last_epoch |pkts_to_release
51 * ----------------->|<--------+--------->|+--------------->
52 * v v
53 *
54 */
55
56struct plug_sched_data {
57 /* If true, the dequeue function releases all packets
58 * from head to end of the queue. The queue turns into
59 * a pass-through queue for newly arriving packets.
60 */
61 bool unplug_indefinite;
62
63 bool throttled;
64
65 /* Queue Limit in bytes */
66 u32 limit;
67
68 /* Number of packets (output) from the current speculatively
69 * executing epoch.
70 */
71 u32 pkts_current_epoch;
72
73 /* Number of packets corresponding to the recently finished
74 * epoch. These will be released when we receive a
75 * TCQ_PLUG_RELEASE_ONE command. This command is typically
76 * issued after committing a checkpoint at the target.
77 */
78 u32 pkts_last_epoch;
79
80 /*
81 * Number of packets from the head of the queue, that can
82 * be released (committed checkpoint).
83 */
84 u32 pkts_to_release;
85};
86
87static int plug_enqueue(struct sk_buff *skb, struct Qdisc *sch,
88 struct sk_buff **to_free)
89{
90 struct plug_sched_data *q = qdisc_priv(sch);
91
92 if (likely(sch->qstats.backlog + skb->len <= q->limit)) {
93 if (!q->unplug_indefinite)
94 q->pkts_current_epoch++;
95 return qdisc_enqueue_tail(skb, sch);
96 }
97
98 return qdisc_drop(skb, sch, to_free);
99}
100
101static struct sk_buff *plug_dequeue(struct Qdisc *sch)
102{
103 struct plug_sched_data *q = qdisc_priv(sch);
104
105 if (q->throttled)
106 return NULL;
107
108 if (!q->unplug_indefinite) {
109 if (!q->pkts_to_release) {
110 /* No more packets to dequeue. Block the queue
111 * and wait for the next release command.
112 */
113 q->throttled = true;
114 return NULL;
115 }
116 q->pkts_to_release--;
117 }
118
119 return qdisc_dequeue_head(sch);
120}
121
122static int plug_init(struct Qdisc *sch, struct nlattr *opt,
123 struct netlink_ext_ack *extack)
124{
125 struct plug_sched_data *q = qdisc_priv(sch);
126
127 q->pkts_current_epoch = 0;
128 q->pkts_last_epoch = 0;
129 q->pkts_to_release = 0;
130 q->unplug_indefinite = false;
131
132 if (opt == NULL) {
133 q->limit = qdisc_dev(sch)->tx_queue_len
134 * psched_mtu(qdisc_dev(sch));
135 } else {
136 struct tc_plug_qopt *ctl = nla_data(opt);
137
138 if (nla_len(opt) < sizeof(*ctl))
139 return -EINVAL;
140
141 q->limit = ctl->limit;
142 }
143
144 q->throttled = true;
145 return 0;
146}
147
148/* Receives 4 types of messages:
149 * TCQ_PLUG_BUFFER: Inset a plug into the queue and
150 * buffer any incoming packets
151 * TCQ_PLUG_RELEASE_ONE: Dequeue packets from queue head
152 * to beginning of the next plug.
153 * TCQ_PLUG_RELEASE_INDEFINITE: Dequeue all packets from queue.
154 * Stop buffering packets until the next TCQ_PLUG_BUFFER
155 * command is received (just act as a pass-thru queue).
156 * TCQ_PLUG_LIMIT: Increase/decrease queue size
157 */
158static int plug_change(struct Qdisc *sch, struct nlattr *opt,
159 struct netlink_ext_ack *extack)
160{
161 struct plug_sched_data *q = qdisc_priv(sch);
162 struct tc_plug_qopt *msg;
163
164 if (opt == NULL)
165 return -EINVAL;
166
167 msg = nla_data(opt);
168 if (nla_len(opt) < sizeof(*msg))
169 return -EINVAL;
170
171 switch (msg->action) {
172 case TCQ_PLUG_BUFFER:
173 /* Save size of the current buffer */
174 q->pkts_last_epoch = q->pkts_current_epoch;
175 q->pkts_current_epoch = 0;
176 if (q->unplug_indefinite)
177 q->throttled = true;
178 q->unplug_indefinite = false;
179 break;
180 case TCQ_PLUG_RELEASE_ONE:
181 /* Add packets from the last complete buffer to the
182 * packets to be released set.
183 */
184 q->pkts_to_release += q->pkts_last_epoch;
185 q->pkts_last_epoch = 0;
186 q->throttled = false;
187 netif_schedule_queue(sch->dev_queue);
188 break;
189 case TCQ_PLUG_RELEASE_INDEFINITE:
190 q->unplug_indefinite = true;
191 q->pkts_to_release = 0;
192 q->pkts_last_epoch = 0;
193 q->pkts_current_epoch = 0;
194 q->throttled = false;
195 netif_schedule_queue(sch->dev_queue);
196 break;
197 case TCQ_PLUG_LIMIT:
198 /* Limit is supplied in bytes */
199 q->limit = msg->limit;
200 break;
201 default:
202 return -EINVAL;
203 }
204
205 return 0;
206}
207
208static struct Qdisc_ops plug_qdisc_ops __read_mostly = {
209 .id = "plug",
210 .priv_size = sizeof(struct plug_sched_data),
211 .enqueue = plug_enqueue,
212 .dequeue = plug_dequeue,
213 .peek = qdisc_peek_head,
214 .init = plug_init,
215 .change = plug_change,
216 .reset = qdisc_reset_queue,
217 .owner = THIS_MODULE,
218};
219
220static int __init plug_module_init(void)
221{
222 return register_qdisc(&plug_qdisc_ops);
223}
224
225static void __exit plug_module_exit(void)
226{
227 unregister_qdisc(&plug_qdisc_ops);
228}
229module_init(plug_module_init)
230module_exit(plug_module_exit)
231MODULE_LICENSE("GPL");
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * sch_plug.c Queue traffic until an explicit release command
4 *
5 * There are two ways to use this qdisc:
6 * 1. A simple "instantaneous" plug/unplug operation, by issuing an alternating
7 * sequence of TCQ_PLUG_BUFFER & TCQ_PLUG_RELEASE_INDEFINITE commands.
8 *
9 * 2. For network output buffering (a.k.a output commit) functionality.
10 * Output commit property is commonly used by applications using checkpoint
11 * based fault-tolerance to ensure that the checkpoint from which a system
12 * is being restored is consistent w.r.t outside world.
13 *
14 * Consider for e.g. Remus - a Virtual Machine checkpointing system,
15 * wherein a VM is checkpointed, say every 50ms. The checkpoint is replicated
16 * asynchronously to the backup host, while the VM continues executing the
17 * next epoch speculatively.
18 *
19 * The following is a typical sequence of output buffer operations:
20 * 1.At epoch i, start_buffer(i)
21 * 2. At end of epoch i (i.e. after 50ms):
22 * 2.1 Stop VM and take checkpoint(i).
23 * 2.2 start_buffer(i+1) and Resume VM
24 * 3. While speculatively executing epoch(i+1), asynchronously replicate
25 * checkpoint(i) to backup host.
26 * 4. When checkpoint_ack(i) is received from backup, release_buffer(i)
27 * Thus, this Qdisc would receive the following sequence of commands:
28 * TCQ_PLUG_BUFFER (epoch i)
29 * .. TCQ_PLUG_BUFFER (epoch i+1)
30 * ....TCQ_PLUG_RELEASE_ONE (epoch i)
31 * ......TCQ_PLUG_BUFFER (epoch i+2)
32 * ........
33 */
34
35#include <linux/module.h>
36#include <linux/types.h>
37#include <linux/kernel.h>
38#include <linux/errno.h>
39#include <linux/netdevice.h>
40#include <linux/skbuff.h>
41#include <net/pkt_sched.h>
42
43/*
44 * State of the queue, when used for network output buffering:
45 *
46 * plug(i+1) plug(i) head
47 * ------------------+--------------------+---------------->
48 * | |
49 * | |
50 * pkts_current_epoch| pkts_last_epoch |pkts_to_release
51 * ----------------->|<--------+--------->|+--------------->
52 * v v
53 *
54 */
55
56struct plug_sched_data {
57 /* If true, the dequeue function releases all packets
58 * from head to end of the queue. The queue turns into
59 * a pass-through queue for newly arriving packets.
60 */
61 bool unplug_indefinite;
62
63 bool throttled;
64
65 /* Queue Limit in bytes */
66 u32 limit;
67
68 /* Number of packets (output) from the current speculatively
69 * executing epoch.
70 */
71 u32 pkts_current_epoch;
72
73 /* Number of packets corresponding to the recently finished
74 * epoch. These will be released when we receive a
75 * TCQ_PLUG_RELEASE_ONE command. This command is typically
76 * issued after committing a checkpoint at the target.
77 */
78 u32 pkts_last_epoch;
79
80 /*
81 * Number of packets from the head of the queue, that can
82 * be released (committed checkpoint).
83 */
84 u32 pkts_to_release;
85};
86
87static int plug_enqueue(struct sk_buff *skb, struct Qdisc *sch,
88 struct sk_buff **to_free)
89{
90 struct plug_sched_data *q = qdisc_priv(sch);
91
92 if (likely(sch->qstats.backlog + skb->len <= q->limit)) {
93 if (!q->unplug_indefinite)
94 q->pkts_current_epoch++;
95 return qdisc_enqueue_tail(skb, sch);
96 }
97
98 return qdisc_drop(skb, sch, to_free);
99}
100
101static struct sk_buff *plug_dequeue(struct Qdisc *sch)
102{
103 struct plug_sched_data *q = qdisc_priv(sch);
104
105 if (q->throttled)
106 return NULL;
107
108 if (!q->unplug_indefinite) {
109 if (!q->pkts_to_release) {
110 /* No more packets to dequeue. Block the queue
111 * and wait for the next release command.
112 */
113 q->throttled = true;
114 return NULL;
115 }
116 q->pkts_to_release--;
117 }
118
119 return qdisc_dequeue_head(sch);
120}
121
122static int plug_init(struct Qdisc *sch, struct nlattr *opt,
123 struct netlink_ext_ack *extack)
124{
125 struct plug_sched_data *q = qdisc_priv(sch);
126
127 q->pkts_current_epoch = 0;
128 q->pkts_last_epoch = 0;
129 q->pkts_to_release = 0;
130 q->unplug_indefinite = false;
131
132 if (opt == NULL) {
133 q->limit = qdisc_dev(sch)->tx_queue_len
134 * psched_mtu(qdisc_dev(sch));
135 } else {
136 struct tc_plug_qopt *ctl = nla_data(opt);
137
138 if (nla_len(opt) < sizeof(*ctl))
139 return -EINVAL;
140
141 q->limit = ctl->limit;
142 }
143
144 q->throttled = true;
145 return 0;
146}
147
148/* Receives 4 types of messages:
149 * TCQ_PLUG_BUFFER: Inset a plug into the queue and
150 * buffer any incoming packets
151 * TCQ_PLUG_RELEASE_ONE: Dequeue packets from queue head
152 * to beginning of the next plug.
153 * TCQ_PLUG_RELEASE_INDEFINITE: Dequeue all packets from queue.
154 * Stop buffering packets until the next TCQ_PLUG_BUFFER
155 * command is received (just act as a pass-thru queue).
156 * TCQ_PLUG_LIMIT: Increase/decrease queue size
157 */
158static int plug_change(struct Qdisc *sch, struct nlattr *opt,
159 struct netlink_ext_ack *extack)
160{
161 struct plug_sched_data *q = qdisc_priv(sch);
162 struct tc_plug_qopt *msg;
163
164 msg = nla_data(opt);
165 if (nla_len(opt) < sizeof(*msg))
166 return -EINVAL;
167
168 switch (msg->action) {
169 case TCQ_PLUG_BUFFER:
170 /* Save size of the current buffer */
171 q->pkts_last_epoch = q->pkts_current_epoch;
172 q->pkts_current_epoch = 0;
173 if (q->unplug_indefinite)
174 q->throttled = true;
175 q->unplug_indefinite = false;
176 break;
177 case TCQ_PLUG_RELEASE_ONE:
178 /* Add packets from the last complete buffer to the
179 * packets to be released set.
180 */
181 q->pkts_to_release += q->pkts_last_epoch;
182 q->pkts_last_epoch = 0;
183 q->throttled = false;
184 netif_schedule_queue(sch->dev_queue);
185 break;
186 case TCQ_PLUG_RELEASE_INDEFINITE:
187 q->unplug_indefinite = true;
188 q->pkts_to_release = 0;
189 q->pkts_last_epoch = 0;
190 q->pkts_current_epoch = 0;
191 q->throttled = false;
192 netif_schedule_queue(sch->dev_queue);
193 break;
194 case TCQ_PLUG_LIMIT:
195 /* Limit is supplied in bytes */
196 q->limit = msg->limit;
197 break;
198 default:
199 return -EINVAL;
200 }
201
202 return 0;
203}
204
205static struct Qdisc_ops plug_qdisc_ops __read_mostly = {
206 .id = "plug",
207 .priv_size = sizeof(struct plug_sched_data),
208 .enqueue = plug_enqueue,
209 .dequeue = plug_dequeue,
210 .peek = qdisc_peek_dequeued,
211 .init = plug_init,
212 .change = plug_change,
213 .reset = qdisc_reset_queue,
214 .owner = THIS_MODULE,
215};
216
217static int __init plug_module_init(void)
218{
219 return register_qdisc(&plug_qdisc_ops);
220}
221
222static void __exit plug_module_exit(void)
223{
224 unregister_qdisc(&plug_qdisc_ops);
225}
226module_init(plug_module_init)
227module_exit(plug_module_exit)
228MODULE_LICENSE("GPL");
229MODULE_DESCRIPTION("Qdisc to plug and unplug traffic via netlink control");