Blame view
net/sched/sch_plug.c
6.4 KB
2874c5fd2 treewide: Replace... |
1 |
// SPDX-License-Identifier: GPL-2.0-or-later |
c3059be16 net/sched: sch_pl... |
2 3 4 |
/* * sch_plug.c Queue traffic until an explicit release command * |
c3059be16 net/sched: sch_pl... |
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 |
* There are two ways to use this qdisc: * 1. A simple "instantaneous" plug/unplug operation, by issuing an alternating * sequence of TCQ_PLUG_BUFFER & TCQ_PLUG_RELEASE_INDEFINITE commands. * * 2. For network output buffering (a.k.a output commit) functionality. * Output commit property is commonly used by applications using checkpoint * based fault-tolerance to ensure that the checkpoint from which a system * is being restored is consistent w.r.t outside world. * * Consider for e.g. Remus - a Virtual Machine checkpointing system, * wherein a VM is checkpointed, say every 50ms. The checkpoint is replicated * asynchronously to the backup host, while the VM continues executing the * next epoch speculatively. * * The following is a typical sequence of output buffer operations: * 1.At epoch i, start_buffer(i) * 2. At end of epoch i (i.e. after 50ms): * 2.1 Stop VM and take checkpoint(i). * 2.2 start_buffer(i+1) and Resume VM * 3. While speculatively executing epoch(i+1), asynchronously replicate * checkpoint(i) to backup host. * 4. When checkpoint_ack(i) is received from backup, release_buffer(i) * Thus, this Qdisc would receive the following sequence of commands: * TCQ_PLUG_BUFFER (epoch i) * .. TCQ_PLUG_BUFFER (epoch i+1) * ....TCQ_PLUG_RELEASE_ONE (epoch i) * ......TCQ_PLUG_BUFFER (epoch i+2) * ........ */ #include <linux/module.h> #include <linux/types.h> #include <linux/kernel.h> #include <linux/errno.h> #include <linux/netdevice.h> #include <linux/skbuff.h> #include <net/pkt_sched.h> /* * State of the queue, when used for network output buffering: * * plug(i+1) plug(i) head * ------------------+--------------------+----------------> * | | * | | * pkts_current_epoch| pkts_last_epoch |pkts_to_release * ----------------->|<--------+--------->|+---------------> * v v * */ struct plug_sched_data { /* If true, the dequeue function releases all packets * from head to end of the queue. The queue turns into * a pass-through queue for newly arriving packets. */ bool unplug_indefinite; |
8fe6a79fb net_sched: sch_pl... |
62 |
bool throttled; |
c3059be16 net/sched: sch_pl... |
63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 |
/* Queue Limit in bytes */ u32 limit; /* Number of packets (output) from the current speculatively * executing epoch. */ u32 pkts_current_epoch; /* Number of packets corresponding to the recently finished * epoch. These will be released when we receive a * TCQ_PLUG_RELEASE_ONE command. This command is typically * issued after committing a checkpoint at the target. */ u32 pkts_last_epoch; /* * Number of packets from the head of the queue, that can * be released (committed checkpoint). */ u32 pkts_to_release; }; |
520ac30f4 net_sched: drop p... |
84 85 |
static int plug_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free) |
c3059be16 net/sched: sch_pl... |
86 87 88 89 90 91 92 93 |
{ struct plug_sched_data *q = qdisc_priv(sch); if (likely(sch->qstats.backlog + skb->len <= q->limit)) { if (!q->unplug_indefinite) q->pkts_current_epoch++; return qdisc_enqueue_tail(skb, sch); } |
520ac30f4 net_sched: drop p... |
94 |
return qdisc_drop(skb, sch, to_free); |
c3059be16 net/sched: sch_pl... |
95 96 97 98 99 |
} static struct sk_buff *plug_dequeue(struct Qdisc *sch) { struct plug_sched_data *q = qdisc_priv(sch); |
8fe6a79fb net_sched: sch_pl... |
100 |
if (q->throttled) |
c3059be16 net/sched: sch_pl... |
101 102 103 104 105 106 107 |
return NULL; if (!q->unplug_indefinite) { if (!q->pkts_to_release) { /* No more packets to dequeue. Block the queue * and wait for the next release command. */ |
8fe6a79fb net_sched: sch_pl... |
108 |
q->throttled = true; |
c3059be16 net/sched: sch_pl... |
109 110 111 112 113 114 115 |
return NULL; } q->pkts_to_release--; } return qdisc_dequeue_head(sch); } |
e63d7dfd2 net: sched: sch: ... |
116 117 |
static int plug_init(struct Qdisc *sch, struct nlattr *opt, struct netlink_ext_ack *extack) |
c3059be16 net/sched: sch_pl... |
118 119 120 121 122 123 124 125 126 |
{ struct plug_sched_data *q = qdisc_priv(sch); q->pkts_current_epoch = 0; q->pkts_last_epoch = 0; q->pkts_to_release = 0; q->unplug_indefinite = false; if (opt == NULL) { |
348e3435c net: sched: drop ... |
127 128 |
q->limit = qdisc_dev(sch)->tx_queue_len * psched_mtu(qdisc_dev(sch)); |
c3059be16 net/sched: sch_pl... |
129 130 131 132 133 134 135 136 |
} else { struct tc_plug_qopt *ctl = nla_data(opt); if (nla_len(opt) < sizeof(*ctl)) return -EINVAL; q->limit = ctl->limit; } |
8fe6a79fb net_sched: sch_pl... |
137 |
q->throttled = true; |
c3059be16 net/sched: sch_pl... |
138 139 140 141 142 143 144 145 146 147 148 149 150 |
return 0; } /* Receives 4 types of messages: * TCQ_PLUG_BUFFER: Inset a plug into the queue and * buffer any incoming packets * TCQ_PLUG_RELEASE_ONE: Dequeue packets from queue head * to beginning of the next plug. * TCQ_PLUG_RELEASE_INDEFINITE: Dequeue all packets from queue. * Stop buffering packets until the next TCQ_PLUG_BUFFER * command is received (just act as a pass-thru queue). * TCQ_PLUG_LIMIT: Increase/decrease queue size */ |
2030721cc net: sched: sch: ... |
151 152 |
static int plug_change(struct Qdisc *sch, struct nlattr *opt, struct netlink_ext_ack *extack) |
c3059be16 net/sched: sch_pl... |
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 |
{ struct plug_sched_data *q = qdisc_priv(sch); struct tc_plug_qopt *msg; if (opt == NULL) return -EINVAL; msg = nla_data(opt); if (nla_len(opt) < sizeof(*msg)) return -EINVAL; switch (msg->action) { case TCQ_PLUG_BUFFER: /* Save size of the current buffer */ q->pkts_last_epoch = q->pkts_current_epoch; q->pkts_current_epoch = 0; if (q->unplug_indefinite) |
8fe6a79fb net_sched: sch_pl... |
170 |
q->throttled = true; |
c3059be16 net/sched: sch_pl... |
171 172 173 174 175 176 177 178 |
q->unplug_indefinite = false; break; case TCQ_PLUG_RELEASE_ONE: /* Add packets from the last complete buffer to the * packets to be released set. */ q->pkts_to_release += q->pkts_last_epoch; q->pkts_last_epoch = 0; |
8fe6a79fb net_sched: sch_pl... |
179 |
q->throttled = false; |
c3059be16 net/sched: sch_pl... |
180 181 182 183 184 185 186 |
netif_schedule_queue(sch->dev_queue); break; case TCQ_PLUG_RELEASE_INDEFINITE: q->unplug_indefinite = true; q->pkts_to_release = 0; q->pkts_last_epoch = 0; q->pkts_current_epoch = 0; |
8fe6a79fb net_sched: sch_pl... |
187 |
q->throttled = false; |
c3059be16 net/sched: sch_pl... |
188 189 190 191 192 193 194 195 196 197 198 199 |
netif_schedule_queue(sch->dev_queue); break; case TCQ_PLUG_LIMIT: /* Limit is supplied in bytes */ q->limit = msg->limit; break; default: return -EINVAL; } return 0; } |
2132cf643 net_sched: sch_pl... |
200 |
static struct Qdisc_ops plug_qdisc_ops __read_mostly = { |
c3059be16 net/sched: sch_pl... |
201 202 203 204 205 206 207 |
.id = "plug", .priv_size = sizeof(struct plug_sched_data), .enqueue = plug_enqueue, .dequeue = plug_dequeue, .peek = qdisc_peek_head, .init = plug_init, .change = plug_change, |
fe6bea7f1 sch_plug: purge b... |
208 |
.reset = qdisc_reset_queue, |
c3059be16 net/sched: sch_pl... |
209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 |
.owner = THIS_MODULE, }; static int __init plug_module_init(void) { return register_qdisc(&plug_qdisc_ops); } static void __exit plug_module_exit(void) { unregister_qdisc(&plug_qdisc_ops); } module_init(plug_module_init) module_exit(plug_module_exit) MODULE_LICENSE("GPL"); |