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/busy_poll.h> 7 #include <net/net_debug.h> 8 #include <net/netdev_rx_queue.h> 9 #include <net/page_pool/helpers.h> 10 #include <net/page_pool/types.h> 11 #include <net/page_pool/memory_provider.h> 12 #include <net/sock.h> 13 14 #include "page_pool_priv.h" 15 #include "netdev-genl-gen.h" 16 17 static DEFINE_XARRAY_FLAGS(page_pools, XA_FLAGS_ALLOC1); 18 /* Protects: page_pools, netdevice->page_pools, pool->p.napi, pool->slow.netdev, 19 * pool->user. 20 * Ordering: inside rtnl_lock 21 */ 22 DEFINE_MUTEX(page_pools_lock); 23 24 /* Page pools are only reachable from user space (via netlink) if they are 25 * linked to a netdev at creation time. Following page pool "visibility" 26 * states are possible: 27 * - normal 28 * - user.list: linked to real netdev, netdev: real netdev 29 * - orphaned - real netdev has disappeared 30 * - user.list: linked to lo, netdev: lo 31 * - invisible - either (a) created without netdev linking, (b) unlisted due 32 * to error, or (c) the entire namespace which owned this pool disappeared 33 * - user.list: unhashed, netdev: unknown 34 */ 35 36 typedef int (*pp_nl_fill_cb)(struct sk_buff *rsp, const struct page_pool *pool, 37 const struct genl_info *info); 38 39 static int 40 netdev_nl_page_pool_get_do(struct genl_info *info, u32 id, pp_nl_fill_cb fill) 41 { 42 struct page_pool *pool; 43 struct sk_buff *rsp; 44 int err; 45 46 mutex_lock(&page_pools_lock); 47 pool = xa_load(&page_pools, id); 48 if (!pool || hlist_unhashed(&pool->user.list) || 49 !net_eq(dev_net(pool->slow.netdev), genl_info_net(info))) { 50 err = -ENOENT; 51 goto err_unlock; 52 } 53 54 rsp = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL); 55 if (!rsp) { 56 err = -ENOMEM; 57 goto err_unlock; 58 } 59 60 err = fill(rsp, pool, info); 61 if (err) 62 goto err_free_msg; 63 64 mutex_unlock(&page_pools_lock); 65 66 return genlmsg_reply(rsp, info); 67 68 err_free_msg: 69 nlmsg_free(rsp); 70 err_unlock: 71 mutex_unlock(&page_pools_lock); 72 return err; 73 } 74 75 struct page_pool_dump_cb { 76 unsigned long ifindex; 77 u32 pp_id; 78 }; 79 80 static int 81 netdev_nl_page_pool_get_dump(struct sk_buff *skb, struct netlink_callback *cb, 82 pp_nl_fill_cb fill, struct nlattr *ifindex_attr) 83 { 84 struct page_pool_dump_cb *state = (void *)cb->ctx; 85 const struct genl_info *info = genl_info_dump(cb); 86 struct net *net = sock_net(skb->sk); 87 struct net_device *netdev; 88 struct page_pool *pool; 89 int err = 0; 90 91 if (ifindex_attr) 92 state->ifindex = nla_get_u32(ifindex_attr); 93 94 rtnl_lock(); 95 mutex_lock(&page_pools_lock); 96 for_each_netdev_dump(net, netdev, state->ifindex) { 97 /* Either the provided ifindex doesn't exist or done dumping */ 98 if (ifindex_attr && 99 netdev->ifindex != nla_get_u32(ifindex_attr)) 100 break; 101 102 hlist_for_each_entry(pool, &netdev->page_pools, user.list) { 103 if (state->pp_id && state->pp_id < pool->user.id) 104 continue; 105 106 state->pp_id = pool->user.id; 107 err = fill(skb, pool, info); 108 if (err) 109 goto out; 110 } 111 112 state->pp_id = 0; 113 } 114 out: 115 mutex_unlock(&page_pools_lock); 116 rtnl_unlock(); 117 118 return err; 119 } 120 121 static int 122 page_pool_nl_stats_fill(struct sk_buff *rsp, const struct page_pool *pool, 123 const struct genl_info *info) 124 { 125 #ifdef CONFIG_PAGE_POOL_STATS 126 struct page_pool_stats stats = {}; 127 struct nlattr *nest; 128 void *hdr; 129 130 page_pool_get_stats(pool, &stats); 131 132 hdr = genlmsg_iput(rsp, info); 133 if (!hdr) 134 return -EMSGSIZE; 135 136 nest = nla_nest_start(rsp, NETDEV_A_PAGE_POOL_STATS_INFO); 137 if (!nest) 138 goto err_cancel_msg; 139 140 if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_ID, pool->user.id) || 141 (pool->slow.netdev->ifindex != LOOPBACK_IFINDEX && 142 nla_put_u32(rsp, NETDEV_A_PAGE_POOL_IFINDEX, 143 pool->slow.netdev->ifindex))) 144 goto err_cancel_nest; 145 146 nla_nest_end(rsp, nest); 147 148 if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_FAST, 149 stats.alloc_stats.fast) || 150 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_SLOW, 151 stats.alloc_stats.slow) || 152 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_SLOW_HIGH_ORDER, 153 stats.alloc_stats.slow_high_order) || 154 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_EMPTY, 155 stats.alloc_stats.empty) || 156 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_REFILL, 157 stats.alloc_stats.refill) || 158 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_WAIVE, 159 stats.alloc_stats.waive) || 160 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_CACHED, 161 stats.recycle_stats.cached) || 162 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_CACHE_FULL, 163 stats.recycle_stats.cache_full) || 164 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_RING, 165 stats.recycle_stats.ring) || 166 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_RING_FULL, 167 stats.recycle_stats.ring_full) || 168 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_RELEASED_REFCNT, 169 stats.recycle_stats.released_refcnt)) 170 goto err_cancel_msg; 171 172 genlmsg_end(rsp, hdr); 173 174 return 0; 175 err_cancel_nest: 176 nla_nest_cancel(rsp, nest); 177 err_cancel_msg: 178 genlmsg_cancel(rsp, hdr); 179 return -EMSGSIZE; 180 #else 181 GENL_SET_ERR_MSG(info, "kernel built without CONFIG_PAGE_POOL_STATS"); 182 return -EOPNOTSUPP; 183 #endif 184 } 185 186 int netdev_nl_page_pool_stats_get_doit(struct sk_buff *skb, 187 struct genl_info *info) 188 { 189 struct nlattr *tb[ARRAY_SIZE(netdev_page_pool_info_nl_policy)]; 190 struct nlattr *nest; 191 int err; 192 u32 id; 193 194 if (GENL_REQ_ATTR_CHECK(info, NETDEV_A_PAGE_POOL_STATS_INFO)) 195 return -EINVAL; 196 197 nest = info->attrs[NETDEV_A_PAGE_POOL_STATS_INFO]; 198 err = nla_parse_nested(tb, ARRAY_SIZE(tb) - 1, nest, 199 netdev_page_pool_info_nl_policy, 200 info->extack); 201 if (err) 202 return err; 203 204 if (NL_REQ_ATTR_CHECK(info->extack, nest, tb, NETDEV_A_PAGE_POOL_ID)) 205 return -EINVAL; 206 if (tb[NETDEV_A_PAGE_POOL_IFINDEX]) { 207 NL_SET_ERR_MSG_ATTR(info->extack, 208 tb[NETDEV_A_PAGE_POOL_IFINDEX], 209 "selecting by ifindex not supported"); 210 return -EINVAL; 211 } 212 213 id = nla_get_uint(tb[NETDEV_A_PAGE_POOL_ID]); 214 215 return netdev_nl_page_pool_get_do(info, id, page_pool_nl_stats_fill); 216 } 217 218 static const struct netlink_range_validation page_pool_ifindex_range = { 219 .min = 1ULL, 220 .max = S32_MAX, 221 }; 222 223 static const struct nla_policy 224 page_pool_stat_info_policy[NETDEV_A_PAGE_POOL_IFINDEX + 1] = { 225 [NETDEV_A_PAGE_POOL_IFINDEX] = 226 NLA_POLICY_FULL_RANGE(NLA_U32, &page_pool_ifindex_range), 227 }; 228 229 int netdev_nl_page_pool_stats_get_dumpit(struct sk_buff *skb, 230 struct netlink_callback *cb) 231 { 232 struct nlattr *tb[ARRAY_SIZE(page_pool_stat_info_policy)]; 233 const struct genl_info *info = genl_info_dump(cb); 234 struct nlattr *ifindex_attr = NULL; 235 236 if (info->attrs[NETDEV_A_PAGE_POOL_STATS_INFO]) { 237 struct nlattr *nest; 238 int err; 239 240 nest = info->attrs[NETDEV_A_PAGE_POOL_STATS_INFO]; 241 err = nla_parse_nested(tb, ARRAY_SIZE(tb) - 1, nest, 242 page_pool_stat_info_policy, 243 info->extack); 244 if (err) 245 return err; 246 247 ifindex_attr = tb[NETDEV_A_PAGE_POOL_IFINDEX]; 248 } 249 250 return netdev_nl_page_pool_get_dump(skb, cb, page_pool_nl_stats_fill, 251 ifindex_attr); 252 } 253 254 static int 255 page_pool_nl_fill(struct sk_buff *rsp, const struct page_pool *pool, 256 const struct genl_info *info) 257 { 258 size_t inflight, refsz; 259 unsigned int napi_id; 260 void *hdr; 261 262 hdr = genlmsg_iput(rsp, info); 263 if (!hdr) 264 return -EMSGSIZE; 265 266 if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_ID, pool->user.id)) 267 goto err_cancel; 268 269 if (pool->slow.netdev->ifindex != LOOPBACK_IFINDEX && 270 nla_put_u32(rsp, NETDEV_A_PAGE_POOL_IFINDEX, 271 pool->slow.netdev->ifindex)) 272 goto err_cancel; 273 274 napi_id = pool->p.napi ? READ_ONCE(pool->p.napi->napi_id) : 0; 275 if (napi_id_valid(napi_id) && 276 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_NAPI_ID, napi_id)) 277 goto err_cancel; 278 279 inflight = page_pool_inflight(pool, false); 280 refsz = PAGE_SIZE << pool->p.order; 281 if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_INFLIGHT, inflight) || 282 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_INFLIGHT_MEM, 283 inflight * refsz)) 284 goto err_cancel; 285 if (pool->user.detach_time && 286 nla_put_uint(rsp, NETDEV_A_PAGE_POOL_DETACH_TIME, 287 ktime_divns(pool->user.detach_time, NSEC_PER_SEC))) 288 goto err_cancel; 289 290 if (pool->mp_ops && pool->mp_ops->nl_fill(pool->mp_priv, rsp, NULL)) 291 goto err_cancel; 292 293 genlmsg_end(rsp, hdr); 294 295 return 0; 296 err_cancel: 297 genlmsg_cancel(rsp, hdr); 298 return -EMSGSIZE; 299 } 300 301 static void netdev_nl_page_pool_event(const struct page_pool *pool, u32 cmd) 302 { 303 struct genl_info info; 304 struct sk_buff *ntf; 305 struct net *net; 306 307 lockdep_assert_held(&page_pools_lock); 308 309 /* 'invisible' page pools don't matter */ 310 if (hlist_unhashed(&pool->user.list)) 311 return; 312 net = dev_net(pool->slow.netdev); 313 314 if (!genl_has_listeners(&netdev_nl_family, net, NETDEV_NLGRP_PAGE_POOL)) 315 return; 316 317 genl_info_init_ntf(&info, &netdev_nl_family, cmd); 318 319 ntf = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL); 320 if (!ntf) 321 return; 322 323 if (page_pool_nl_fill(ntf, pool, &info)) { 324 nlmsg_free(ntf); 325 return; 326 } 327 328 genlmsg_multicast_netns(&netdev_nl_family, net, ntf, 329 0, NETDEV_NLGRP_PAGE_POOL, GFP_KERNEL); 330 } 331 332 int netdev_nl_page_pool_get_doit(struct sk_buff *skb, struct genl_info *info) 333 { 334 u32 id; 335 336 if (GENL_REQ_ATTR_CHECK(info, NETDEV_A_PAGE_POOL_ID)) 337 return -EINVAL; 338 339 id = nla_get_uint(info->attrs[NETDEV_A_PAGE_POOL_ID]); 340 341 return netdev_nl_page_pool_get_do(info, id, page_pool_nl_fill); 342 } 343 344 int netdev_nl_page_pool_get_dumpit(struct sk_buff *skb, 345 struct netlink_callback *cb) 346 { 347 const struct genl_info *info = genl_info_dump(cb); 348 349 return netdev_nl_page_pool_get_dump(skb, cb, page_pool_nl_fill, 350 info->attrs[NETDEV_A_PAGE_POOL_IFINDEX]); 351 } 352 353 int page_pool_list(struct page_pool *pool) 354 { 355 static u32 id_alloc_next; 356 int err; 357 358 mutex_lock(&page_pools_lock); 359 err = xa_alloc_cyclic(&page_pools, &pool->user.id, pool, xa_limit_32b, 360 &id_alloc_next, GFP_KERNEL); 361 if (err < 0) 362 goto err_unlock; 363 364 INIT_HLIST_NODE(&pool->user.list); 365 if (pool->slow.netdev) { 366 hlist_add_head(&pool->user.list, 367 &pool->slow.netdev->page_pools); 368 netdev_nl_page_pool_event(pool, NETDEV_CMD_PAGE_POOL_ADD_NTF); 369 } 370 371 mutex_unlock(&page_pools_lock); 372 return 0; 373 374 err_unlock: 375 mutex_unlock(&page_pools_lock); 376 return err; 377 } 378 379 void page_pool_detached(struct page_pool *pool) 380 { 381 mutex_lock(&page_pools_lock); 382 pool->user.detach_time = ktime_get_boottime(); 383 netdev_nl_page_pool_event(pool, NETDEV_CMD_PAGE_POOL_CHANGE_NTF); 384 mutex_unlock(&page_pools_lock); 385 } 386 387 void page_pool_unlist(struct page_pool *pool) 388 { 389 mutex_lock(&page_pools_lock); 390 netdev_nl_page_pool_event(pool, NETDEV_CMD_PAGE_POOL_DEL_NTF); 391 xa_erase(&page_pools, pool->user.id); 392 if (!hlist_unhashed(&pool->user.list)) 393 hlist_del(&pool->user.list); 394 mutex_unlock(&page_pools_lock); 395 } 396 397 int page_pool_check_memory_provider(struct net_device *dev, 398 struct netdev_rx_queue *rxq) 399 { 400 void *binding = rxq->mp_params.mp_priv; 401 struct page_pool *pool; 402 struct hlist_node *n; 403 404 if (!binding) 405 return 0; 406 407 mutex_lock(&page_pools_lock); 408 hlist_for_each_entry_safe(pool, n, &dev->page_pools, user.list) { 409 if (pool->mp_priv != binding) 410 continue; 411 412 if (pool->slow.queue_idx == get_netdev_rx_queue_index(rxq)) { 413 mutex_unlock(&page_pools_lock); 414 return 0; 415 } 416 } 417 mutex_unlock(&page_pools_lock); 418 return -ENODATA; 419 } 420 421 static void page_pool_unreg_netdev_wipe(struct net_device *netdev) 422 { 423 struct page_pool *pool; 424 struct hlist_node *n; 425 426 mutex_lock(&page_pools_lock); 427 hlist_for_each_entry_safe(pool, n, &netdev->page_pools, user.list) { 428 hlist_del_init(&pool->user.list); 429 pool->slow.netdev = NET_PTR_POISON; 430 } 431 mutex_unlock(&page_pools_lock); 432 } 433 434 static void page_pool_unreg_netdev(struct net_device *netdev) 435 { 436 struct page_pool *pool, *last; 437 struct net_device *lo; 438 439 lo = dev_net(netdev)->loopback_dev; 440 441 mutex_lock(&page_pools_lock); 442 last = NULL; 443 hlist_for_each_entry(pool, &netdev->page_pools, user.list) { 444 pool->slow.netdev = lo; 445 netdev_nl_page_pool_event(pool, 446 NETDEV_CMD_PAGE_POOL_CHANGE_NTF); 447 last = pool; 448 } 449 if (last) 450 hlist_splice_init(&netdev->page_pools, &last->user.list, 451 &lo->page_pools); 452 mutex_unlock(&page_pools_lock); 453 } 454 455 static int 456 page_pool_netdevice_event(struct notifier_block *nb, 457 unsigned long event, void *ptr) 458 { 459 struct net_device *netdev = netdev_notifier_info_to_dev(ptr); 460 461 if (event != NETDEV_UNREGISTER) 462 return NOTIFY_DONE; 463 464 if (hlist_empty(&netdev->page_pools)) 465 return NOTIFY_OK; 466 467 if (netdev->ifindex != LOOPBACK_IFINDEX) 468 page_pool_unreg_netdev(netdev); 469 else 470 page_pool_unreg_netdev_wipe(netdev); 471 return NOTIFY_OK; 472 } 473 474 static struct notifier_block page_pool_netdevice_nb = { 475 .notifier_call = page_pool_netdevice_event, 476 }; 477 478 static int __init page_pool_user_init(void) 479 { 480 return register_netdevice_notifier(&page_pool_netdevice_nb); 481 } 482 483 subsys_initcall(page_pool_user_init); 484