1 // SPDX-License-Identifier: GPL-2.0 2 3 #include <linux/mutex.h> 4 #include <linux/netdevice.h> 5 #include <linux/xarray.h> 6 #include <net/net_debug.h> 7 #include <net/page_pool/types.h> 8 #include <net/page_pool/helpers.h> 9 #include <net/sock.h> 10 11 #include "page_pool_priv.h" 12 #include "netdev-genl-gen.h" 13 14 static DEFINE_XARRAY_FLAGS(page_pools, XA_FLAGS_ALLOC1); 15 /* Protects: page_pools, netdevice->page_pools, pool->slow.netdev, pool->user. 16 * Ordering: inside rtnl_lock 17 */ 18 static DEFINE_MUTEX(page_pools_lock); 19 20 /* Page pools are only reachable from user space (via netlink) if they are 21 * linked to a netdev at creation time. Following page pool "visibility" 22 * states are possible: 23 * - normal 24 * - user.list: linked to real netdev, netdev: real netdev 25 * - orphaned - real netdev has disappeared 26 * - user.list: linked to lo, netdev: lo 27 * - invisible - either (a) created without netdev linking, (b) unlisted due 28 * to error, or (c) the entire namespace which owned this pool disappeared 29 * - user.list: unhashed, netdev: unknown 30 */ 31 32 typedef int (*pp_nl_fill_cb)(struct sk_buff *rsp, const struct page_pool *pool, 33 const struct genl_info *info); 34 35 static int 36 netdev_nl_page_pool_get_do(struct genl_info *info, u32 id, pp_nl_fill_cb fill) 37 { 38 struct page_pool *pool; 39 struct sk_buff *rsp; 40 int err; 41 42 mutex_lock(&page_pools_lock); 43 pool = xa_load(&page_pools, id); 44 if (!pool || hlist_unhashed(&pool->user.list) || 45 !net_eq(dev_net(pool->slow.netdev), genl_info_net(info))) { 46 err = -ENOENT; 47 goto err_unlock; 48 } 49 50 rsp = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL); 51 if (!rsp) { 52 err = -ENOMEM; 53 goto err_unlock; 54 } 55 56 err = fill(rsp, pool, info); 57 if (err) 58 goto err_free_msg; 59 60 mutex_unlock(&page_pools_lock); 61 62 return genlmsg_reply(rsp, info); 63 64 err_free_msg: 65 nlmsg_free(rsp); 66 err_unlock: 67 mutex_unlock(&page_pools_lock); 68 return err; 69 } 70 71 struct page_pool_dump_cb { 72 unsigned long ifindex; 73 u32 pp_id; 74 }; 75 76 static int 77 netdev_nl_page_pool_get_dump(struct sk_buff *skb, struct netlink_callback *cb, 78 pp_nl_fill_cb fill) 79 { 80 struct page_pool_dump_cb *state = (void *)cb->ctx; 81 const struct genl_info *info = genl_info_dump(cb); 82 struct net *net = sock_net(skb->sk); 83 struct net_device *netdev; 84 struct page_pool *pool; 85 int err = 0; 86 87 rtnl_lock(); 88 mutex_lock(&page_pools_lock); 89 for_each_netdev_dump(net, netdev, state->ifindex) { 90 hlist_for_each_entry(pool, &netdev->page_pools, user.list) { 91 if (state->pp_id && state->pp_id < pool->user.id) 92 continue; 93 94 state->pp_id = pool->user.id; 95 err = fill(skb, pool, info); 96 if (err) 97 break; 98 } 99 100 state->pp_id = 0; 101 } 102 mutex_unlock(&page_pools_lock); 103 rtnl_unlock(); 104 105 return err; 106 } 107 108 static int 109 page_pool_nl_stats_fill(struct sk_buff *rsp, const struct page_pool *pool, 110 const struct genl_info *info) 111 { 112 #ifdef CONFIG_PAGE_POOL_STATS 113 struct page_pool_stats stats = {}; 114 struct nlattr *nest; 115 void *hdr; 116 117 if (!page_pool_get_stats(pool, &stats)) 118 return 0; 119 120 hdr = genlmsg_iput(rsp, info); 121 if (!hdr) 122 return -EMSGSIZE; 123 124 nest = nla_nest_start(rsp, NETDEV_A_PAGE_POOL_STATS_INFO); 125 126 if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_ID, pool->user.id) || 127 (pool->slow.netdev->ifindex != LOOPBACK_IFINDEX && 128 nla_put_u32(rsp, NETDEV_A_PAGE_POOL_IFINDEX, 129 pool->slow.netdev->ifindex))) 130 goto err_cancel_nest; 131 132 nla_nest_end(rsp, nest); 133 134 if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_FAST, 135 stats.alloc_stats.fast) || 136 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_SLOW, 137 stats.alloc_stats.slow) || 138 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_SLOW_HIGH_ORDER, 139 stats.alloc_stats.slow_high_order) || 140 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_EMPTY, 141 stats.alloc_stats.empty) || 142 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_REFILL, 143 stats.alloc_stats.refill) || 144 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_WAIVE, 145 stats.alloc_stats.waive) || 146 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_CACHED, 147 stats.recycle_stats.cached) || 148 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_CACHE_FULL, 149 stats.recycle_stats.cache_full) || 150 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_RING, 151 stats.recycle_stats.ring) || 152 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_RING_FULL, 153 stats.recycle_stats.ring_full) || 154 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_RELEASED_REFCNT, 155 stats.recycle_stats.released_refcnt)) 156 goto err_cancel_msg; 157 158 genlmsg_end(rsp, hdr); 159 160 return 0; 161 err_cancel_nest: 162 nla_nest_cancel(rsp, nest); 163 err_cancel_msg: 164 genlmsg_cancel(rsp, hdr); 165 return -EMSGSIZE; 166 #else 167 GENL_SET_ERR_MSG(info, "kernel built without CONFIG_PAGE_POOL_STATS"); 168 return -EOPNOTSUPP; 169 #endif 170 } 171 172 int netdev_nl_page_pool_stats_get_doit(struct sk_buff *skb, 173 struct genl_info *info) 174 { 175 struct nlattr *tb[ARRAY_SIZE(netdev_page_pool_info_nl_policy)]; 176 struct nlattr *nest; 177 int err; 178 u32 id; 179 180 if (GENL_REQ_ATTR_CHECK(info, NETDEV_A_PAGE_POOL_STATS_INFO)) 181 return -EINVAL; 182 183 nest = info->attrs[NETDEV_A_PAGE_POOL_STATS_INFO]; 184 err = nla_parse_nested(tb, ARRAY_SIZE(tb) - 1, nest, 185 netdev_page_pool_info_nl_policy, 186 info->extack); 187 if (err) 188 return err; 189 190 if (NL_REQ_ATTR_CHECK(info->extack, nest, tb, NETDEV_A_PAGE_POOL_ID)) 191 return -EINVAL; 192 if (tb[NETDEV_A_PAGE_POOL_IFINDEX]) { 193 NL_SET_ERR_MSG_ATTR(info->extack, 194 tb[NETDEV_A_PAGE_POOL_IFINDEX], 195 "selecting by ifindex not supported"); 196 return -EINVAL; 197 } 198 199 id = nla_get_uint(tb[NETDEV_A_PAGE_POOL_ID]); 200 201 return netdev_nl_page_pool_get_do(info, id, page_pool_nl_stats_fill); 202 } 203 204 int netdev_nl_page_pool_stats_get_dumpit(struct sk_buff *skb, 205 struct netlink_callback *cb) 206 { 207 return netdev_nl_page_pool_get_dump(skb, cb, page_pool_nl_stats_fill); 208 } 209 210 static int 211 page_pool_nl_fill(struct sk_buff *rsp, const struct page_pool *pool, 212 const struct genl_info *info) 213 { 214 size_t inflight, refsz; 215 void *hdr; 216 217 hdr = genlmsg_iput(rsp, info); 218 if (!hdr) 219 return -EMSGSIZE; 220 221 if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_ID, pool->user.id)) 222 goto err_cancel; 223 224 if (pool->slow.netdev->ifindex != LOOPBACK_IFINDEX && 225 nla_put_u32(rsp, NETDEV_A_PAGE_POOL_IFINDEX, 226 pool->slow.netdev->ifindex)) 227 goto err_cancel; 228 if (pool->user.napi_id && 229 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_NAPI_ID, pool->user.napi_id)) 230 goto err_cancel; 231 232 inflight = page_pool_inflight(pool, false); 233 refsz = PAGE_SIZE << pool->p.order; 234 if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_INFLIGHT, inflight) || 235 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_INFLIGHT_MEM, 236 inflight * refsz)) 237 goto err_cancel; 238 if (pool->user.detach_time && 239 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_DETACH_TIME, 240 pool->user.detach_time)) 241 goto err_cancel; 242 243 genlmsg_end(rsp, hdr); 244 245 return 0; 246 err_cancel: 247 genlmsg_cancel(rsp, hdr); 248 return -EMSGSIZE; 249 } 250 251 static void netdev_nl_page_pool_event(const struct page_pool *pool, u32 cmd) 252 { 253 struct genl_info info; 254 struct sk_buff *ntf; 255 struct net *net; 256 257 lockdep_assert_held(&page_pools_lock); 258 259 /* 'invisible' page pools don't matter */ 260 if (hlist_unhashed(&pool->user.list)) 261 return; 262 net = dev_net(pool->slow.netdev); 263 264 if (!genl_has_listeners(&netdev_nl_family, net, NETDEV_NLGRP_PAGE_POOL)) 265 return; 266 267 genl_info_init_ntf(&info, &netdev_nl_family, cmd); 268 269 ntf = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL); 270 if (!ntf) 271 return; 272 273 if (page_pool_nl_fill(ntf, pool, &info)) { 274 nlmsg_free(ntf); 275 return; 276 } 277 278 genlmsg_multicast_netns(&netdev_nl_family, net, ntf, 279 0, NETDEV_NLGRP_PAGE_POOL, GFP_KERNEL); 280 } 281 282 int netdev_nl_page_pool_get_doit(struct sk_buff *skb, struct genl_info *info) 283 { 284 u32 id; 285 286 if (GENL_REQ_ATTR_CHECK(info, NETDEV_A_PAGE_POOL_ID)) 287 return -EINVAL; 288 289 id = nla_get_uint(info->attrs[NETDEV_A_PAGE_POOL_ID]); 290 291 return netdev_nl_page_pool_get_do(info, id, page_pool_nl_fill); 292 } 293 294 int netdev_nl_page_pool_get_dumpit(struct sk_buff *skb, 295 struct netlink_callback *cb) 296 { 297 return netdev_nl_page_pool_get_dump(skb, cb, page_pool_nl_fill); 298 } 299 300 int page_pool_list(struct page_pool *pool) 301 { 302 static u32 id_alloc_next; 303 int err; 304 305 mutex_lock(&page_pools_lock); 306 err = xa_alloc_cyclic(&page_pools, &pool->user.id, pool, xa_limit_32b, 307 &id_alloc_next, GFP_KERNEL); 308 if (err < 0) 309 goto err_unlock; 310 311 INIT_HLIST_NODE(&pool->user.list); 312 if (pool->slow.netdev) { 313 hlist_add_head(&pool->user.list, 314 &pool->slow.netdev->page_pools); 315 pool->user.napi_id = pool->p.napi ? pool->p.napi->napi_id : 0; 316 317 netdev_nl_page_pool_event(pool, NETDEV_CMD_PAGE_POOL_ADD_NTF); 318 } 319 320 mutex_unlock(&page_pools_lock); 321 return 0; 322 323 err_unlock: 324 mutex_unlock(&page_pools_lock); 325 return err; 326 } 327 328 void page_pool_detached(struct page_pool *pool) 329 { 330 mutex_lock(&page_pools_lock); 331 pool->user.detach_time = ktime_get_boottime_seconds(); 332 netdev_nl_page_pool_event(pool, NETDEV_CMD_PAGE_POOL_CHANGE_NTF); 333 mutex_unlock(&page_pools_lock); 334 } 335 336 void page_pool_unlist(struct page_pool *pool) 337 { 338 mutex_lock(&page_pools_lock); 339 netdev_nl_page_pool_event(pool, NETDEV_CMD_PAGE_POOL_DEL_NTF); 340 xa_erase(&page_pools, pool->user.id); 341 if (!hlist_unhashed(&pool->user.list)) 342 hlist_del(&pool->user.list); 343 mutex_unlock(&page_pools_lock); 344 } 345 346 static void page_pool_unreg_netdev_wipe(struct net_device *netdev) 347 { 348 struct page_pool *pool; 349 struct hlist_node *n; 350 351 mutex_lock(&page_pools_lock); 352 hlist_for_each_entry_safe(pool, n, &netdev->page_pools, user.list) { 353 hlist_del_init(&pool->user.list); 354 pool->slow.netdev = NET_PTR_POISON; 355 } 356 mutex_unlock(&page_pools_lock); 357 } 358 359 static void page_pool_unreg_netdev(struct net_device *netdev) 360 { 361 struct page_pool *pool, *last; 362 struct net_device *lo; 363 364 lo = dev_net(netdev)->loopback_dev; 365 366 mutex_lock(&page_pools_lock); 367 last = NULL; 368 hlist_for_each_entry(pool, &netdev->page_pools, user.list) { 369 pool->slow.netdev = lo; 370 netdev_nl_page_pool_event(pool, 371 NETDEV_CMD_PAGE_POOL_CHANGE_NTF); 372 last = pool; 373 } 374 if (last) 375 hlist_splice_init(&netdev->page_pools, &last->user.list, 376 &lo->page_pools); 377 mutex_unlock(&page_pools_lock); 378 } 379 380 static int 381 page_pool_netdevice_event(struct notifier_block *nb, 382 unsigned long event, void *ptr) 383 { 384 struct net_device *netdev = netdev_notifier_info_to_dev(ptr); 385 386 if (event != NETDEV_UNREGISTER) 387 return NOTIFY_DONE; 388 389 if (hlist_empty(&netdev->page_pools)) 390 return NOTIFY_OK; 391 392 if (netdev->ifindex != LOOPBACK_IFINDEX) 393 page_pool_unreg_netdev(netdev); 394 else 395 page_pool_unreg_netdev_wipe(netdev); 396 return NOTIFY_OK; 397 } 398 399 static struct notifier_block page_pool_netdevice_nb = { 400 .notifier_call = page_pool_netdevice_event, 401 }; 402 403 static int __init page_pool_user_init(void) 404 { 405 return register_netdevice_notifier(&page_pool_netdevice_nb); 406 } 407 408 subsys_initcall(page_pool_user_init); 409