1 // SPDX-License-Identifier: GPL-2.0
2 /* Multipath TCP
3 *
4 * Copyright (c) 2022, SUSE.
5 */
6
7 #define pr_fmt(fmt) "MPTCP: " fmt
8
9 #include <linux/kernel.h>
10 #include <linux/module.h>
11 #include <linux/list.h>
12 #include <linux/rculist.h>
13 #include <linux/spinlock.h>
14 #include "protocol.h"
15
16 static DEFINE_SPINLOCK(mptcp_sched_list_lock);
17 static LIST_HEAD(mptcp_sched_list);
18
mptcp_sched_default_get_subflow(struct mptcp_sock * msk,struct mptcp_sched_data * data)19 static int mptcp_sched_default_get_subflow(struct mptcp_sock *msk,
20 struct mptcp_sched_data *data)
21 {
22 struct sock *ssk;
23
24 ssk = data->reinject ? mptcp_subflow_get_retrans(msk) :
25 mptcp_subflow_get_send(msk);
26 if (!ssk)
27 return -EINVAL;
28
29 mptcp_subflow_set_scheduled(mptcp_subflow_ctx(ssk), true);
30 return 0;
31 }
32
33 static struct mptcp_sched_ops mptcp_sched_default = {
34 .get_subflow = mptcp_sched_default_get_subflow,
35 .name = "default",
36 .owner = THIS_MODULE,
37 };
38
39 /* Must be called with rcu read lock held */
mptcp_sched_find(const char * name)40 struct mptcp_sched_ops *mptcp_sched_find(const char *name)
41 {
42 struct mptcp_sched_ops *sched, *ret = NULL;
43
44 list_for_each_entry_rcu(sched, &mptcp_sched_list, list) {
45 if (!strcmp(sched->name, name)) {
46 ret = sched;
47 break;
48 }
49 }
50
51 return ret;
52 }
53
54 /* Build string with list of available scheduler values.
55 * Similar to tcp_get_available_congestion_control()
56 */
mptcp_get_available_schedulers(char * buf,size_t maxlen)57 void mptcp_get_available_schedulers(char *buf, size_t maxlen)
58 {
59 struct mptcp_sched_ops *sched;
60 size_t offs = 0;
61
62 rcu_read_lock();
63 spin_lock(&mptcp_sched_list_lock);
64 list_for_each_entry_rcu(sched, &mptcp_sched_list, list) {
65 offs += snprintf(buf + offs, maxlen - offs,
66 "%s%s",
67 offs == 0 ? "" : " ", sched->name);
68
69 if (WARN_ON_ONCE(offs >= maxlen))
70 break;
71 }
72 spin_unlock(&mptcp_sched_list_lock);
73 rcu_read_unlock();
74 }
75
mptcp_register_scheduler(struct mptcp_sched_ops * sched)76 int mptcp_register_scheduler(struct mptcp_sched_ops *sched)
77 {
78 if (!sched->get_subflow)
79 return -EINVAL;
80
81 spin_lock(&mptcp_sched_list_lock);
82 if (mptcp_sched_find(sched->name)) {
83 spin_unlock(&mptcp_sched_list_lock);
84 return -EEXIST;
85 }
86 list_add_tail_rcu(&sched->list, &mptcp_sched_list);
87 spin_unlock(&mptcp_sched_list_lock);
88
89 pr_debug("%s registered\n", sched->name);
90 return 0;
91 }
92
mptcp_unregister_scheduler(struct mptcp_sched_ops * sched)93 void mptcp_unregister_scheduler(struct mptcp_sched_ops *sched)
94 {
95 if (sched == &mptcp_sched_default)
96 return;
97
98 spin_lock(&mptcp_sched_list_lock);
99 list_del_rcu(&sched->list);
100 spin_unlock(&mptcp_sched_list_lock);
101 }
102
mptcp_sched_init(void)103 void mptcp_sched_init(void)
104 {
105 mptcp_register_scheduler(&mptcp_sched_default);
106 }
107
mptcp_init_sched(struct mptcp_sock * msk,struct mptcp_sched_ops * sched)108 int mptcp_init_sched(struct mptcp_sock *msk,
109 struct mptcp_sched_ops *sched)
110 {
111 if (!sched)
112 sched = &mptcp_sched_default;
113
114 if (!bpf_try_module_get(sched, sched->owner))
115 return -EBUSY;
116
117 msk->sched = sched;
118 if (msk->sched->init)
119 msk->sched->init(msk);
120
121 pr_debug("sched=%s\n", msk->sched->name);
122
123 return 0;
124 }
125
mptcp_release_sched(struct mptcp_sock * msk)126 void mptcp_release_sched(struct mptcp_sock *msk)
127 {
128 struct mptcp_sched_ops *sched = msk->sched;
129
130 if (!sched)
131 return;
132
133 msk->sched = NULL;
134 if (sched->release)
135 sched->release(msk);
136
137 bpf_module_put(sched, sched->owner);
138 }
139
mptcp_subflow_set_scheduled(struct mptcp_subflow_context * subflow,bool scheduled)140 void mptcp_subflow_set_scheduled(struct mptcp_subflow_context *subflow,
141 bool scheduled)
142 {
143 WRITE_ONCE(subflow->scheduled, scheduled);
144 }
145
mptcp_sched_get_send(struct mptcp_sock * msk)146 int mptcp_sched_get_send(struct mptcp_sock *msk)
147 {
148 struct mptcp_subflow_context *subflow;
149 struct mptcp_sched_data data;
150
151 msk_owned_by_me(msk);
152
153 /* the following check is moved out of mptcp_subflow_get_send */
154 if (__mptcp_check_fallback(msk)) {
155 if (msk->first &&
156 __tcp_can_send(msk->first) &&
157 sk_stream_memory_free(msk->first)) {
158 mptcp_subflow_set_scheduled(mptcp_subflow_ctx(msk->first), true);
159 return 0;
160 }
161 return -EINVAL;
162 }
163
164 mptcp_for_each_subflow(msk, subflow) {
165 if (READ_ONCE(subflow->scheduled))
166 return 0;
167 }
168
169 data.reinject = false;
170 if (msk->sched == &mptcp_sched_default || !msk->sched)
171 return mptcp_sched_default_get_subflow(msk, &data);
172 return msk->sched->get_subflow(msk, &data);
173 }
174
mptcp_sched_get_retrans(struct mptcp_sock * msk)175 int mptcp_sched_get_retrans(struct mptcp_sock *msk)
176 {
177 struct mptcp_subflow_context *subflow;
178 struct mptcp_sched_data data;
179
180 msk_owned_by_me(msk);
181
182 /* the following check is moved out of mptcp_subflow_get_retrans */
183 if (__mptcp_check_fallback(msk))
184 return -EINVAL;
185
186 mptcp_for_each_subflow(msk, subflow) {
187 if (READ_ONCE(subflow->scheduled))
188 return 0;
189 }
190
191 data.reinject = true;
192 if (msk->sched == &mptcp_sched_default || !msk->sched)
193 return mptcp_sched_default_get_subflow(msk, &data);
194 return msk->sched->get_subflow(msk, &data);
195 }
196