| 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
| 2 | /* |
| 3 | * net/core/netprio_cgroup.c Priority Control Group |
| 4 | * |
| 5 | * Authors: Neil Horman <nhorman@tuxdriver.com> |
| 6 | */ |
| 7 | |
| 8 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 9 | |
| 10 | #include <linux/module.h> |
| 11 | #include <linux/slab.h> |
| 12 | #include <linux/types.h> |
| 13 | #include <linux/string.h> |
| 14 | #include <linux/errno.h> |
| 15 | #include <linux/skbuff.h> |
| 16 | #include <linux/cgroup.h> |
| 17 | #include <linux/rcupdate.h> |
| 18 | #include <linux/atomic.h> |
| 19 | #include <linux/sched/task.h> |
| 20 | |
| 21 | #include <net/rtnetlink.h> |
| 22 | #include <net/pkt_cls.h> |
| 23 | #include <net/sock.h> |
| 24 | #include <net/netprio_cgroup.h> |
| 25 | |
| 26 | #include <linux/fdtable.h> |
| 27 | |
| 28 | /* |
| 29 | * netprio allocates per-net_device priomap array which is indexed by |
| 30 | * css->id. Limiting css ID to 16bits doesn't lose anything. |
| 31 | */ |
| 32 | #define NETPRIO_ID_MAX USHRT_MAX |
| 33 | |
| 34 | #define PRIOMAP_MIN_SZ 128 |
| 35 | |
| 36 | /* |
| 37 | * Extend @dev->priomap so that it's large enough to accommodate |
| 38 | * @target_idx. @dev->priomap.priomap_len > @target_idx after successful |
| 39 | * return. Must be called under rtnl lock. |
| 40 | */ |
| 41 | static int extend_netdev_table(struct net_device *dev, u32 target_idx) |
| 42 | { |
| 43 | struct netprio_map *old, *new; |
| 44 | size_t new_sz, new_len; |
| 45 | |
| 46 | /* is the existing priomap large enough? */ |
| 47 | old = rtnl_dereference(dev->priomap); |
| 48 | if (old && old->priomap_len > target_idx) |
| 49 | return 0; |
| 50 | |
| 51 | /* |
| 52 | * Determine the new size. Let's keep it power-of-two. We start |
| 53 | * from PRIOMAP_MIN_SZ and double it until it's large enough to |
| 54 | * accommodate @target_idx. |
| 55 | */ |
| 56 | new_sz = PRIOMAP_MIN_SZ; |
| 57 | while (true) { |
| 58 | new_len = (new_sz - offsetof(struct netprio_map, priomap)) / |
| 59 | sizeof(new->priomap[0]); |
| 60 | if (new_len > target_idx) |
| 61 | break; |
| 62 | new_sz *= 2; |
| 63 | /* overflowed? */ |
| 64 | if (WARN_ON(new_sz < PRIOMAP_MIN_SZ)) |
| 65 | return -ENOSPC; |
| 66 | } |
| 67 | |
| 68 | /* allocate & copy */ |
| 69 | new = kzalloc(new_sz, GFP_KERNEL); |
| 70 | if (!new) |
| 71 | return -ENOMEM; |
| 72 | |
| 73 | if (old) |
| 74 | memcpy(new->priomap, old->priomap, |
| 75 | old->priomap_len * sizeof(old->priomap[0])); |
| 76 | |
| 77 | new->priomap_len = new_len; |
| 78 | |
| 79 | /* install the new priomap */ |
| 80 | rcu_assign_pointer(dev->priomap, new); |
| 81 | if (old) |
| 82 | kfree_rcu(old, rcu); |
| 83 | return 0; |
| 84 | } |
| 85 | |
| 86 | /** |
| 87 | * netprio_prio - return the effective netprio of a cgroup-net_device pair |
| 88 | * @css: css part of the target pair |
| 89 | * @dev: net_device part of the target pair |
| 90 | * |
| 91 | * Should be called under RCU read or rtnl lock. |
| 92 | */ |
| 93 | static u32 netprio_prio(struct cgroup_subsys_state *css, struct net_device *dev) |
| 94 | { |
| 95 | struct netprio_map *map = rcu_dereference_rtnl(dev->priomap); |
| 96 | int id = css->cgroup->id; |
| 97 | |
| 98 | if (map && id < map->priomap_len) |
| 99 | return map->priomap[id]; |
| 100 | return 0; |
| 101 | } |
| 102 | |
| 103 | /** |
| 104 | * netprio_set_prio - set netprio on a cgroup-net_device pair |
| 105 | * @css: css part of the target pair |
| 106 | * @dev: net_device part of the target pair |
| 107 | * @prio: prio to set |
| 108 | * |
| 109 | * Set netprio to @prio on @css-@dev pair. Should be called under rtnl |
| 110 | * lock and may fail under memory pressure for non-zero @prio. |
| 111 | */ |
| 112 | static int netprio_set_prio(struct cgroup_subsys_state *css, |
| 113 | struct net_device *dev, u32 prio) |
| 114 | { |
| 115 | struct netprio_map *map; |
| 116 | int id = css->cgroup->id; |
| 117 | int ret; |
| 118 | |
| 119 | /* avoid extending priomap for zero writes */ |
| 120 | map = rtnl_dereference(dev->priomap); |
| 121 | if (!prio && (!map || map->priomap_len <= id)) |
| 122 | return 0; |
| 123 | |
| 124 | ret = extend_netdev_table(dev, id); |
| 125 | if (ret) |
| 126 | return ret; |
| 127 | |
| 128 | map = rtnl_dereference(dev->priomap); |
| 129 | map->priomap[id] = prio; |
| 130 | return 0; |
| 131 | } |
| 132 | |
| 133 | static struct cgroup_subsys_state * |
| 134 | cgrp_css_alloc(struct cgroup_subsys_state *parent_css) |
| 135 | { |
| 136 | struct cgroup_subsys_state *css; |
| 137 | |
| 138 | css = kzalloc(sizeof(*css), GFP_KERNEL); |
| 139 | if (!css) |
| 140 | return ERR_PTR(-ENOMEM); |
| 141 | |
| 142 | return css; |
| 143 | } |
| 144 | |
| 145 | static int cgrp_css_online(struct cgroup_subsys_state *css) |
| 146 | { |
| 147 | struct cgroup_subsys_state *parent_css = css->parent; |
| 148 | struct net_device *dev; |
| 149 | int ret = 0; |
| 150 | |
| 151 | if (css->id > NETPRIO_ID_MAX) |
| 152 | return -ENOSPC; |
| 153 | |
| 154 | if (!parent_css) |
| 155 | return 0; |
| 156 | |
| 157 | rtnl_lock(); |
| 158 | /* |
| 159 | * Inherit prios from the parent. As all prios are set during |
| 160 | * onlining, there is no need to clear them on offline. |
| 161 | */ |
| 162 | for_each_netdev(&init_net, dev) { |
| 163 | u32 prio = netprio_prio(parent_css, dev); |
| 164 | |
| 165 | ret = netprio_set_prio(css, dev, prio); |
| 166 | if (ret) |
| 167 | break; |
| 168 | } |
| 169 | rtnl_unlock(); |
| 170 | return ret; |
| 171 | } |
| 172 | |
| 173 | static void cgrp_css_free(struct cgroup_subsys_state *css) |
| 174 | { |
| 175 | kfree(css); |
| 176 | } |
| 177 | |
| 178 | static u64 read_prioidx(struct cgroup_subsys_state *css, struct cftype *cft) |
| 179 | { |
| 180 | return css->cgroup->id; |
| 181 | } |
| 182 | |
| 183 | static int read_priomap(struct seq_file *sf, void *v) |
| 184 | { |
| 185 | struct net_device *dev; |
| 186 | |
| 187 | rcu_read_lock(); |
| 188 | for_each_netdev_rcu(&init_net, dev) |
| 189 | seq_printf(sf, "%s %u\n", dev->name, |
| 190 | netprio_prio(seq_css(sf), dev)); |
| 191 | rcu_read_unlock(); |
| 192 | return 0; |
| 193 | } |
| 194 | |
| 195 | static ssize_t write_priomap(struct kernfs_open_file *of, |
| 196 | char *buf, size_t nbytes, loff_t off) |
| 197 | { |
| 198 | char devname[IFNAMSIZ + 1]; |
| 199 | struct net_device *dev; |
| 200 | u32 prio; |
| 201 | int ret; |
| 202 | |
| 203 | if (sscanf(buf, "%"__stringify(IFNAMSIZ)"s %u", devname, &prio) != 2) |
| 204 | return -EINVAL; |
| 205 | |
| 206 | dev = dev_get_by_name(&init_net, devname); |
| 207 | if (!dev) |
| 208 | return -ENODEV; |
| 209 | |
| 210 | cgroup_sk_alloc_disable(); |
| 211 | |
| 212 | rtnl_lock(); |
| 213 | |
| 214 | ret = netprio_set_prio(of_css(of), dev, prio); |
| 215 | |
| 216 | rtnl_unlock(); |
| 217 | dev_put(dev); |
| 218 | return ret ?: nbytes; |
| 219 | } |
| 220 | |
| 221 | static int update_netprio(const void *v, struct file *file, unsigned n) |
| 222 | { |
| 223 | int err; |
| 224 | struct socket *sock = sock_from_file(file, &err); |
| 225 | if (sock) { |
| 226 | spin_lock(&cgroup_sk_update_lock); |
| 227 | sock_cgroup_set_prioidx(&sock->sk->sk_cgrp_data, |
| 228 | (unsigned long)v); |
| 229 | spin_unlock(&cgroup_sk_update_lock); |
| 230 | } |
| 231 | return 0; |
| 232 | } |
| 233 | |
| 234 | static void net_prio_attach(struct cgroup_taskset *tset) |
| 235 | { |
| 236 | struct task_struct *p; |
| 237 | struct cgroup_subsys_state *css; |
| 238 | |
| 239 | cgroup_taskset_for_each(p, css, tset) { |
| 240 | void *v = (void *)(unsigned long)css->cgroup->id; |
| 241 | |
| 242 | task_lock(p); |
| 243 | iterate_fd(p->files, 0, update_netprio, v); |
| 244 | task_unlock(p); |
| 245 | } |
| 246 | } |
| 247 | |
| 248 | static struct cftype ss_files[] = { |
| 249 | { |
| 250 | .name = "prioidx", |
| 251 | .read_u64 = read_prioidx, |
| 252 | }, |
| 253 | { |
| 254 | .name = "ifpriomap", |
| 255 | .seq_show = read_priomap, |
| 256 | .write = write_priomap, |
| 257 | }, |
| 258 | { } /* terminate */ |
| 259 | }; |
| 260 | |
| 261 | struct cgroup_subsys net_prio_cgrp_subsys = { |
| 262 | .css_alloc = cgrp_css_alloc, |
| 263 | .css_online = cgrp_css_online, |
| 264 | .css_free = cgrp_css_free, |
| 265 | .attach = net_prio_attach, |
| 266 | .legacy_cftypes = ss_files, |
| 267 | }; |
| 268 | |
| 269 | static int netprio_device_event(struct notifier_block *unused, |
| 270 | unsigned long event, void *ptr) |
| 271 | { |
| 272 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
| 273 | struct netprio_map *old; |
| 274 | |
| 275 | /* |
| 276 | * Note this is called with rtnl_lock held so we have update side |
| 277 | * protection on our rcu assignments |
| 278 | */ |
| 279 | |
| 280 | switch (event) { |
| 281 | case NETDEV_UNREGISTER: |
| 282 | old = rtnl_dereference(dev->priomap); |
| 283 | RCU_INIT_POINTER(dev->priomap, NULL); |
| 284 | if (old) |
| 285 | kfree_rcu(old, rcu); |
| 286 | break; |
| 287 | } |
| 288 | return NOTIFY_DONE; |
| 289 | } |
| 290 | |
| 291 | static struct notifier_block netprio_device_notifier = { |
| 292 | .notifier_call = netprio_device_event |
| 293 | }; |
| 294 | |
| 295 | static int __init init_cgroup_netprio(void) |
| 296 | { |
| 297 | register_netdevice_notifier(&netprio_device_notifier); |
| 298 | return 0; |
| 299 | } |
| 300 | subsys_initcall(init_cgroup_netprio); |