xref: /linux/net/core/page_pool_user.c (revision e7d759f31ca295d589f7420719c311870bb3166f)
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 	if (skb->len && err == -EMSGSIZE)
106 		return skb->len;
107 	return err;
108 }
109 
110 static int
111 page_pool_nl_stats_fill(struct sk_buff *rsp, const struct page_pool *pool,
112 			const struct genl_info *info)
113 {
114 #ifdef CONFIG_PAGE_POOL_STATS
115 	struct page_pool_stats stats = {};
116 	struct nlattr *nest;
117 	void *hdr;
118 
119 	if (!page_pool_get_stats(pool, &stats))
120 		return 0;
121 
122 	hdr = genlmsg_iput(rsp, info);
123 	if (!hdr)
124 		return -EMSGSIZE;
125 
126 	nest = nla_nest_start(rsp, NETDEV_A_PAGE_POOL_STATS_INFO);
127 
128 	if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_ID, pool->user.id) ||
129 	    (pool->slow.netdev->ifindex != LOOPBACK_IFINDEX &&
130 	     nla_put_u32(rsp, NETDEV_A_PAGE_POOL_IFINDEX,
131 			 pool->slow.netdev->ifindex)))
132 		goto err_cancel_nest;
133 
134 	nla_nest_end(rsp, nest);
135 
136 	if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_FAST,
137 			 stats.alloc_stats.fast) ||
138 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_SLOW,
139 			 stats.alloc_stats.slow) ||
140 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_SLOW_HIGH_ORDER,
141 			 stats.alloc_stats.slow_high_order) ||
142 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_EMPTY,
143 			 stats.alloc_stats.empty) ||
144 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_REFILL,
145 			 stats.alloc_stats.refill) ||
146 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_ALLOC_WAIVE,
147 			 stats.alloc_stats.waive) ||
148 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_CACHED,
149 			 stats.recycle_stats.cached) ||
150 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_CACHE_FULL,
151 			 stats.recycle_stats.cache_full) ||
152 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_RING,
153 			 stats.recycle_stats.ring) ||
154 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_RING_FULL,
155 			 stats.recycle_stats.ring_full) ||
156 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_STATS_RECYCLE_RELEASED_REFCNT,
157 			 stats.recycle_stats.released_refcnt))
158 		goto err_cancel_msg;
159 
160 	genlmsg_end(rsp, hdr);
161 
162 	return 0;
163 err_cancel_nest:
164 	nla_nest_cancel(rsp, nest);
165 err_cancel_msg:
166 	genlmsg_cancel(rsp, hdr);
167 	return -EMSGSIZE;
168 #else
169 	GENL_SET_ERR_MSG(info, "kernel built without CONFIG_PAGE_POOL_STATS");
170 	return -EOPNOTSUPP;
171 #endif
172 }
173 
174 int netdev_nl_page_pool_stats_get_doit(struct sk_buff *skb,
175 				       struct genl_info *info)
176 {
177 	struct nlattr *tb[ARRAY_SIZE(netdev_page_pool_info_nl_policy)];
178 	struct nlattr *nest;
179 	int err;
180 	u32 id;
181 
182 	if (GENL_REQ_ATTR_CHECK(info, NETDEV_A_PAGE_POOL_STATS_INFO))
183 		return -EINVAL;
184 
185 	nest = info->attrs[NETDEV_A_PAGE_POOL_STATS_INFO];
186 	err = nla_parse_nested(tb, ARRAY_SIZE(tb) - 1, nest,
187 			       netdev_page_pool_info_nl_policy,
188 			       info->extack);
189 	if (err)
190 		return err;
191 
192 	if (NL_REQ_ATTR_CHECK(info->extack, nest, tb, NETDEV_A_PAGE_POOL_ID))
193 		return -EINVAL;
194 	if (tb[NETDEV_A_PAGE_POOL_IFINDEX]) {
195 		NL_SET_ERR_MSG_ATTR(info->extack,
196 				    tb[NETDEV_A_PAGE_POOL_IFINDEX],
197 				    "selecting by ifindex not supported");
198 		return -EINVAL;
199 	}
200 
201 	id = nla_get_uint(tb[NETDEV_A_PAGE_POOL_ID]);
202 
203 	return netdev_nl_page_pool_get_do(info, id, page_pool_nl_stats_fill);
204 }
205 
206 int netdev_nl_page_pool_stats_get_dumpit(struct sk_buff *skb,
207 					 struct netlink_callback *cb)
208 {
209 	return netdev_nl_page_pool_get_dump(skb, cb, page_pool_nl_stats_fill);
210 }
211 
212 static int
213 page_pool_nl_fill(struct sk_buff *rsp, const struct page_pool *pool,
214 		  const struct genl_info *info)
215 {
216 	size_t inflight, refsz;
217 	void *hdr;
218 
219 	hdr = genlmsg_iput(rsp, info);
220 	if (!hdr)
221 		return -EMSGSIZE;
222 
223 	if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_ID, pool->user.id))
224 		goto err_cancel;
225 
226 	if (pool->slow.netdev->ifindex != LOOPBACK_IFINDEX &&
227 	    nla_put_u32(rsp, NETDEV_A_PAGE_POOL_IFINDEX,
228 			pool->slow.netdev->ifindex))
229 		goto err_cancel;
230 	if (pool->user.napi_id &&
231 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_NAPI_ID, pool->user.napi_id))
232 		goto err_cancel;
233 
234 	inflight = page_pool_inflight(pool, false);
235 	refsz =	PAGE_SIZE << pool->p.order;
236 	if (nla_put_uint(rsp, NETDEV_A_PAGE_POOL_INFLIGHT, inflight) ||
237 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_INFLIGHT_MEM,
238 			 inflight * refsz))
239 		goto err_cancel;
240 	if (pool->user.detach_time &&
241 	    nla_put_uint(rsp, NETDEV_A_PAGE_POOL_DETACH_TIME,
242 			 pool->user.detach_time))
243 		goto err_cancel;
244 
245 	genlmsg_end(rsp, hdr);
246 
247 	return 0;
248 err_cancel:
249 	genlmsg_cancel(rsp, hdr);
250 	return -EMSGSIZE;
251 }
252 
253 static void netdev_nl_page_pool_event(const struct page_pool *pool, u32 cmd)
254 {
255 	struct genl_info info;
256 	struct sk_buff *ntf;
257 	struct net *net;
258 
259 	lockdep_assert_held(&page_pools_lock);
260 
261 	/* 'invisible' page pools don't matter */
262 	if (hlist_unhashed(&pool->user.list))
263 		return;
264 	net = dev_net(pool->slow.netdev);
265 
266 	if (!genl_has_listeners(&netdev_nl_family, net, NETDEV_NLGRP_PAGE_POOL))
267 		return;
268 
269 	genl_info_init_ntf(&info, &netdev_nl_family, cmd);
270 
271 	ntf = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
272 	if (!ntf)
273 		return;
274 
275 	if (page_pool_nl_fill(ntf, pool, &info)) {
276 		nlmsg_free(ntf);
277 		return;
278 	}
279 
280 	genlmsg_multicast_netns(&netdev_nl_family, net, ntf,
281 				0, NETDEV_NLGRP_PAGE_POOL, GFP_KERNEL);
282 }
283 
284 int netdev_nl_page_pool_get_doit(struct sk_buff *skb, struct genl_info *info)
285 {
286 	u32 id;
287 
288 	if (GENL_REQ_ATTR_CHECK(info, NETDEV_A_PAGE_POOL_ID))
289 		return -EINVAL;
290 
291 	id = nla_get_uint(info->attrs[NETDEV_A_PAGE_POOL_ID]);
292 
293 	return netdev_nl_page_pool_get_do(info, id, page_pool_nl_fill);
294 }
295 
296 int netdev_nl_page_pool_get_dumpit(struct sk_buff *skb,
297 				   struct netlink_callback *cb)
298 {
299 	return netdev_nl_page_pool_get_dump(skb, cb, page_pool_nl_fill);
300 }
301 
302 int page_pool_list(struct page_pool *pool)
303 {
304 	static u32 id_alloc_next;
305 	int err;
306 
307 	mutex_lock(&page_pools_lock);
308 	err = xa_alloc_cyclic(&page_pools, &pool->user.id, pool, xa_limit_32b,
309 			      &id_alloc_next, GFP_KERNEL);
310 	if (err < 0)
311 		goto err_unlock;
312 
313 	INIT_HLIST_NODE(&pool->user.list);
314 	if (pool->slow.netdev) {
315 		hlist_add_head(&pool->user.list,
316 			       &pool->slow.netdev->page_pools);
317 		pool->user.napi_id = pool->p.napi ? pool->p.napi->napi_id : 0;
318 
319 		netdev_nl_page_pool_event(pool, NETDEV_CMD_PAGE_POOL_ADD_NTF);
320 	}
321 
322 	mutex_unlock(&page_pools_lock);
323 	return 0;
324 
325 err_unlock:
326 	mutex_unlock(&page_pools_lock);
327 	return err;
328 }
329 
330 void page_pool_detached(struct page_pool *pool)
331 {
332 	mutex_lock(&page_pools_lock);
333 	pool->user.detach_time = ktime_get_boottime_seconds();
334 	netdev_nl_page_pool_event(pool, NETDEV_CMD_PAGE_POOL_CHANGE_NTF);
335 	mutex_unlock(&page_pools_lock);
336 }
337 
338 void page_pool_unlist(struct page_pool *pool)
339 {
340 	mutex_lock(&page_pools_lock);
341 	netdev_nl_page_pool_event(pool, NETDEV_CMD_PAGE_POOL_DEL_NTF);
342 	xa_erase(&page_pools, pool->user.id);
343 	if (!hlist_unhashed(&pool->user.list))
344 		hlist_del(&pool->user.list);
345 	mutex_unlock(&page_pools_lock);
346 }
347 
348 static void page_pool_unreg_netdev_wipe(struct net_device *netdev)
349 {
350 	struct page_pool *pool;
351 	struct hlist_node *n;
352 
353 	mutex_lock(&page_pools_lock);
354 	hlist_for_each_entry_safe(pool, n, &netdev->page_pools, user.list) {
355 		hlist_del_init(&pool->user.list);
356 		pool->slow.netdev = NET_PTR_POISON;
357 	}
358 	mutex_unlock(&page_pools_lock);
359 }
360 
361 static void page_pool_unreg_netdev(struct net_device *netdev)
362 {
363 	struct page_pool *pool, *last;
364 	struct net_device *lo;
365 
366 	lo = dev_net(netdev)->loopback_dev;
367 
368 	mutex_lock(&page_pools_lock);
369 	last = NULL;
370 	hlist_for_each_entry(pool, &netdev->page_pools, user.list) {
371 		pool->slow.netdev = lo;
372 		netdev_nl_page_pool_event(pool,
373 					  NETDEV_CMD_PAGE_POOL_CHANGE_NTF);
374 		last = pool;
375 	}
376 	if (last)
377 		hlist_splice_init(&netdev->page_pools, &last->user.list,
378 				  &lo->page_pools);
379 	mutex_unlock(&page_pools_lock);
380 }
381 
382 static int
383 page_pool_netdevice_event(struct notifier_block *nb,
384 			  unsigned long event, void *ptr)
385 {
386 	struct net_device *netdev = netdev_notifier_info_to_dev(ptr);
387 
388 	if (event != NETDEV_UNREGISTER)
389 		return NOTIFY_DONE;
390 
391 	if (hlist_empty(&netdev->page_pools))
392 		return NOTIFY_OK;
393 
394 	if (netdev->ifindex != LOOPBACK_IFINDEX)
395 		page_pool_unreg_netdev(netdev);
396 	else
397 		page_pool_unreg_netdev_wipe(netdev);
398 	return NOTIFY_OK;
399 }
400 
401 static struct notifier_block page_pool_netdevice_nb = {
402 	.notifier_call = page_pool_netdevice_event,
403 };
404 
405 static int __init page_pool_user_init(void)
406 {
407 	return register_netdevice_notifier(&page_pool_netdevice_nb);
408 }
409 
410 subsys_initcall(page_pool_user_init);
411