1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net/core/gen_stats.c
4 *
5 * Authors: Thomas Graf <tgraf@suug.ch>
6 * Jamal Hadi Salim
7 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
8 *
9 * See Documentation/networking/gen_stats.rst
10 */
11
12#include <linux/types.h>
13#include <linux/kernel.h>
14#include <linux/module.h>
15#include <linux/interrupt.h>
16#include <linux/socket.h>
17#include <linux/rtnetlink.h>
18#include <linux/gen_stats.h>
19#include <net/netlink.h>
20#include <net/gen_stats.h>
21#include <net/sch_generic.h>
22
23static inline int
24gnet_stats_copy(struct gnet_dump *d, int type, void *buf, int size, int padattr)
25{
26 if (nla_put_64bit(skb: d->skb, attrtype: type, attrlen: size, data: buf, padattr))
27 goto nla_put_failure;
28 return 0;
29
30nla_put_failure:
31 if (d->lock)
32 spin_unlock_bh(lock: d->lock);
33 kfree(objp: d->xstats);
34 d->xstats = NULL;
35 d->xstats_len = 0;
36 return -1;
37}
38
39/**
40 * gnet_stats_start_copy_compat - start dumping procedure in compatibility mode
41 * @skb: socket buffer to put statistics TLVs into
42 * @type: TLV type for top level statistic TLV
43 * @tc_stats_type: TLV type for backward compatibility struct tc_stats TLV
44 * @xstats_type: TLV type for backward compatibility xstats TLV
45 * @lock: statistics lock
46 * @d: dumping handle
47 * @padattr: padding attribute
48 *
49 * Initializes the dumping handle, grabs the statistic lock and appends
50 * an empty TLV header to the socket buffer for use a container for all
51 * other statistic TLVS.
52 *
53 * The dumping handle is marked to be in backward compatibility mode telling
54 * all gnet_stats_copy_XXX() functions to fill a local copy of struct tc_stats.
55 *
56 * Returns 0 on success or -1 if the room in the socket buffer was not sufficient.
57 */
58int
59gnet_stats_start_copy_compat(struct sk_buff *skb, int type, int tc_stats_type,
60 int xstats_type, spinlock_t *lock,
61 struct gnet_dump *d, int padattr)
62 __acquires(lock)
63{
64 memset(s: d, c: 0, n: sizeof(*d));
65
66 if (type)
67 d->tail = (struct nlattr *)skb_tail_pointer(skb);
68 d->skb = skb;
69 d->compat_tc_stats = tc_stats_type;
70 d->compat_xstats = xstats_type;
71 d->padattr = padattr;
72 if (lock) {
73 d->lock = lock;
74 spin_lock_bh(lock);
75 }
76 if (d->tail) {
77 int ret = gnet_stats_copy(d, type, NULL, size: 0, padattr);
78
79 /* The initial attribute added in gnet_stats_copy() may be
80 * preceded by a padding attribute, in which case d->tail will
81 * end up pointing at the padding instead of the real attribute.
82 * Fix this so gnet_stats_finish_copy() adjusts the length of
83 * the right attribute.
84 */
85 if (ret == 0 && d->tail->nla_type == padattr)
86 d->tail = (struct nlattr *)((char *)d->tail +
87 NLA_ALIGN(d->tail->nla_len));
88 return ret;
89 }
90
91 return 0;
92}
93EXPORT_SYMBOL(gnet_stats_start_copy_compat);
94
95/**
96 * gnet_stats_start_copy - start dumping procedure in compatibility mode
97 * @skb: socket buffer to put statistics TLVs into
98 * @type: TLV type for top level statistic TLV
99 * @lock: statistics lock
100 * @d: dumping handle
101 * @padattr: padding attribute
102 *
103 * Initializes the dumping handle, grabs the statistic lock and appends
104 * an empty TLV header to the socket buffer for use a container for all
105 * other statistic TLVS.
106 *
107 * Returns 0 on success or -1 if the room in the socket buffer was not sufficient.
108 */
109int
110gnet_stats_start_copy(struct sk_buff *skb, int type, spinlock_t *lock,
111 struct gnet_dump *d, int padattr)
112{
113 return gnet_stats_start_copy_compat(skb, type, 0, 0, lock, d, padattr);
114}
115EXPORT_SYMBOL(gnet_stats_start_copy);
116
117/* Must not be inlined, due to u64_stats seqcount_t lockdep key */
118void gnet_stats_basic_sync_init(struct gnet_stats_basic_sync *b)
119{
120 u64_stats_set(p: &b->bytes, val: 0);
121 u64_stats_set(p: &b->packets, val: 0);
122 u64_stats_init(syncp: &b->syncp);
123}
124EXPORT_SYMBOL(gnet_stats_basic_sync_init);
125
126static void gnet_stats_add_basic_cpu(struct gnet_stats_basic_sync *bstats,
127 struct gnet_stats_basic_sync __percpu *cpu)
128{
129 u64 t_bytes = 0, t_packets = 0;
130 int i;
131
132 for_each_possible_cpu(i) {
133 struct gnet_stats_basic_sync *bcpu = per_cpu_ptr(cpu, i);
134 unsigned int start;
135 u64 bytes, packets;
136
137 do {
138 start = u64_stats_fetch_begin(syncp: &bcpu->syncp);
139 bytes = u64_stats_read(p: &bcpu->bytes);
140 packets = u64_stats_read(p: &bcpu->packets);
141 } while (u64_stats_fetch_retry(syncp: &bcpu->syncp, start));
142
143 t_bytes += bytes;
144 t_packets += packets;
145 }
146 _bstats_update(bstats, bytes: t_bytes, packets: t_packets);
147}
148
149void gnet_stats_add_basic(struct gnet_stats_basic_sync *bstats,
150 struct gnet_stats_basic_sync __percpu *cpu,
151 struct gnet_stats_basic_sync *b, bool running)
152{
153 unsigned int start;
154 u64 bytes = 0;
155 u64 packets = 0;
156
157 WARN_ON_ONCE((cpu || running) && in_hardirq());
158
159 if (cpu) {
160 gnet_stats_add_basic_cpu(bstats, cpu);
161 return;
162 }
163 do {
164 if (running)
165 start = u64_stats_fetch_begin(syncp: &b->syncp);
166 bytes = u64_stats_read(p: &b->bytes);
167 packets = u64_stats_read(p: &b->packets);
168 } while (running && u64_stats_fetch_retry(syncp: &b->syncp, start));
169
170 _bstats_update(bstats, bytes, packets);
171}
172EXPORT_SYMBOL(gnet_stats_add_basic);
173
174static void gnet_stats_read_basic(u64 *ret_bytes, u64 *ret_packets,
175 struct gnet_stats_basic_sync __percpu *cpu,
176 struct gnet_stats_basic_sync *b, bool running)
177{
178 unsigned int start;
179
180 if (cpu) {
181 u64 t_bytes = 0, t_packets = 0;
182 int i;
183
184 for_each_possible_cpu(i) {
185 struct gnet_stats_basic_sync *bcpu = per_cpu_ptr(cpu, i);
186 unsigned int start;
187 u64 bytes, packets;
188
189 do {
190 start = u64_stats_fetch_begin(syncp: &bcpu->syncp);
191 bytes = u64_stats_read(p: &bcpu->bytes);
192 packets = u64_stats_read(p: &bcpu->packets);
193 } while (u64_stats_fetch_retry(syncp: &bcpu->syncp, start));
194
195 t_bytes += bytes;
196 t_packets += packets;
197 }
198 *ret_bytes = t_bytes;
199 *ret_packets = t_packets;
200 return;
201 }
202 do {
203 if (running)
204 start = u64_stats_fetch_begin(syncp: &b->syncp);
205 *ret_bytes = u64_stats_read(p: &b->bytes);
206 *ret_packets = u64_stats_read(p: &b->packets);
207 } while (running && u64_stats_fetch_retry(syncp: &b->syncp, start));
208}
209
210static int
211___gnet_stats_copy_basic(struct gnet_dump *d,
212 struct gnet_stats_basic_sync __percpu *cpu,
213 struct gnet_stats_basic_sync *b,
214 int type, bool running)
215{
216 u64 bstats_bytes, bstats_packets;
217
218 gnet_stats_read_basic(ret_bytes: &bstats_bytes, ret_packets: &bstats_packets, cpu, b, running);
219
220 if (d->compat_tc_stats && type == TCA_STATS_BASIC) {
221 d->tc_stats.bytes = bstats_bytes;
222 d->tc_stats.packets = bstats_packets;
223 }
224
225 if (d->tail) {
226 struct gnet_stats_basic sb;
227 int res;
228
229 memset(s: &sb, c: 0, n: sizeof(sb));
230 sb.bytes = bstats_bytes;
231 sb.packets = bstats_packets;
232 res = gnet_stats_copy(d, type, buf: &sb, size: sizeof(sb), padattr: TCA_STATS_PAD);
233 if (res < 0 || sb.packets == bstats_packets)
234 return res;
235 /* emit 64bit stats only if needed */
236 return gnet_stats_copy(d, type: TCA_STATS_PKT64, buf: &bstats_packets,
237 size: sizeof(bstats_packets), padattr: TCA_STATS_PAD);
238 }
239 return 0;
240}
241
242/**
243 * gnet_stats_copy_basic - copy basic statistics into statistic TLV
244 * @d: dumping handle
245 * @cpu: copy statistic per cpu
246 * @b: basic statistics
247 * @running: true if @b represents a running qdisc, thus @b's
248 * internal values might change during basic reads.
249 * Only used if @cpu is NULL
250 *
251 * Context: task; must not be run from IRQ or BH contexts
252 *
253 * Appends the basic statistics to the top level TLV created by
254 * gnet_stats_start_copy().
255 *
256 * Returns 0 on success or -1 with the statistic lock released
257 * if the room in the socket buffer was not sufficient.
258 */
259int
260gnet_stats_copy_basic(struct gnet_dump *d,
261 struct gnet_stats_basic_sync __percpu *cpu,
262 struct gnet_stats_basic_sync *b,
263 bool running)
264{
265 return ___gnet_stats_copy_basic(d, cpu, b, type: TCA_STATS_BASIC, running);
266}
267EXPORT_SYMBOL(gnet_stats_copy_basic);
268
269/**
270 * gnet_stats_copy_basic_hw - copy basic hw statistics into statistic TLV
271 * @d: dumping handle
272 * @cpu: copy statistic per cpu
273 * @b: basic statistics
274 * @running: true if @b represents a running qdisc, thus @b's
275 * internal values might change during basic reads.
276 * Only used if @cpu is NULL
277 *
278 * Context: task; must not be run from IRQ or BH contexts
279 *
280 * Appends the basic statistics to the top level TLV created by
281 * gnet_stats_start_copy().
282 *
283 * Returns 0 on success or -1 with the statistic lock released
284 * if the room in the socket buffer was not sufficient.
285 */
286int
287gnet_stats_copy_basic_hw(struct gnet_dump *d,
288 struct gnet_stats_basic_sync __percpu *cpu,
289 struct gnet_stats_basic_sync *b,
290 bool running)
291{
292 return ___gnet_stats_copy_basic(d, cpu, b, type: TCA_STATS_BASIC_HW, running);
293}
294EXPORT_SYMBOL(gnet_stats_copy_basic_hw);
295
296/**
297 * gnet_stats_copy_rate_est - copy rate estimator statistics into statistics TLV
298 * @d: dumping handle
299 * @rate_est: rate estimator
300 *
301 * Appends the rate estimator statistics to the top level TLV created by
302 * gnet_stats_start_copy().
303 *
304 * Returns 0 on success or -1 with the statistic lock released
305 * if the room in the socket buffer was not sufficient.
306 */
307int
308gnet_stats_copy_rate_est(struct gnet_dump *d,
309 struct net_rate_estimator __rcu **rate_est)
310{
311 struct gnet_stats_rate_est64 sample;
312 struct gnet_stats_rate_est est;
313 int res;
314
315 if (!gen_estimator_read(ptr: rate_est, sample: &sample))
316 return 0;
317 est.bps = min_t(u64, UINT_MAX, sample.bps);
318 /* we have some time before reaching 2^32 packets per second */
319 est.pps = sample.pps;
320
321 if (d->compat_tc_stats) {
322 d->tc_stats.bps = est.bps;
323 d->tc_stats.pps = est.pps;
324 }
325
326 if (d->tail) {
327 res = gnet_stats_copy(d, type: TCA_STATS_RATE_EST, buf: &est, size: sizeof(est),
328 padattr: TCA_STATS_PAD);
329 if (res < 0 || est.bps == sample.bps)
330 return res;
331 /* emit 64bit stats only if needed */
332 return gnet_stats_copy(d, type: TCA_STATS_RATE_EST64, buf: &sample,
333 size: sizeof(sample), padattr: TCA_STATS_PAD);
334 }
335
336 return 0;
337}
338EXPORT_SYMBOL(gnet_stats_copy_rate_est);
339
340static void gnet_stats_add_queue_cpu(struct gnet_stats_queue *qstats,
341 const struct gnet_stats_queue __percpu *q)
342{
343 int i;
344
345 for_each_possible_cpu(i) {
346 const struct gnet_stats_queue *qcpu = per_cpu_ptr(q, i);
347
348 qstats->qlen += qcpu->qlen;
349 qstats->backlog += qcpu->backlog;
350 qstats->drops += qcpu->drops;
351 qstats->requeues += qcpu->requeues;
352 qstats->overlimits += qcpu->overlimits;
353 }
354}
355
356void gnet_stats_add_queue(struct gnet_stats_queue *qstats,
357 const struct gnet_stats_queue __percpu *cpu,
358 const struct gnet_stats_queue *q)
359{
360 if (cpu) {
361 gnet_stats_add_queue_cpu(qstats, q: cpu);
362 } else {
363 qstats->qlen += q->qlen;
364 qstats->backlog += q->backlog;
365 qstats->drops += q->drops;
366 qstats->requeues += q->requeues;
367 qstats->overlimits += q->overlimits;
368 }
369}
370EXPORT_SYMBOL(gnet_stats_add_queue);
371
372/**
373 * gnet_stats_copy_queue - copy queue statistics into statistics TLV
374 * @d: dumping handle
375 * @cpu_q: per cpu queue statistics
376 * @q: queue statistics
377 * @qlen: queue length statistics
378 *
379 * Appends the queue statistics to the top level TLV created by
380 * gnet_stats_start_copy(). Using per cpu queue statistics if
381 * they are available.
382 *
383 * Returns 0 on success or -1 with the statistic lock released
384 * if the room in the socket buffer was not sufficient.
385 */
386int
387gnet_stats_copy_queue(struct gnet_dump *d,
388 struct gnet_stats_queue __percpu *cpu_q,
389 struct gnet_stats_queue *q, __u32 qlen)
390{
391 struct gnet_stats_queue qstats = {0};
392
393 gnet_stats_add_queue(&qstats, cpu_q, q);
394 qstats.qlen = qlen;
395
396 if (d->compat_tc_stats) {
397 d->tc_stats.drops = qstats.drops;
398 d->tc_stats.qlen = qstats.qlen;
399 d->tc_stats.backlog = qstats.backlog;
400 d->tc_stats.overlimits = qstats.overlimits;
401 }
402
403 if (d->tail)
404 return gnet_stats_copy(d, type: TCA_STATS_QUEUE,
405 buf: &qstats, size: sizeof(qstats),
406 padattr: TCA_STATS_PAD);
407
408 return 0;
409}
410EXPORT_SYMBOL(gnet_stats_copy_queue);
411
412/**
413 * gnet_stats_copy_app - copy application specific statistics into statistics TLV
414 * @d: dumping handle
415 * @st: application specific statistics data
416 * @len: length of data
417 *
418 * Appends the application specific statistics to the top level TLV created by
419 * gnet_stats_start_copy() and remembers the data for XSTATS if the dumping
420 * handle is in backward compatibility mode.
421 *
422 * Returns 0 on success or -1 with the statistic lock released
423 * if the room in the socket buffer was not sufficient.
424 */
425int
426gnet_stats_copy_app(struct gnet_dump *d, void *st, int len)
427{
428 if (d->compat_xstats) {
429 d->xstats = kmemdup(st, len, GFP_ATOMIC);
430 if (!d->xstats)
431 goto err_out;
432 d->xstats_len = len;
433 }
434
435 if (d->tail)
436 return gnet_stats_copy(d, type: TCA_STATS_APP, buf: st, size: len,
437 padattr: TCA_STATS_PAD);
438
439 return 0;
440
441err_out:
442 if (d->lock)
443 spin_unlock_bh(lock: d->lock);
444 d->xstats_len = 0;
445 return -1;
446}
447EXPORT_SYMBOL(gnet_stats_copy_app);
448
449/**
450 * gnet_stats_finish_copy - finish dumping procedure
451 * @d: dumping handle
452 *
453 * Corrects the length of the top level TLV to include all TLVs added
454 * by gnet_stats_copy_XXX() calls. Adds the backward compatibility TLVs
455 * if gnet_stats_start_copy_compat() was used and releases the statistics
456 * lock.
457 *
458 * Returns 0 on success or -1 with the statistic lock released
459 * if the room in the socket buffer was not sufficient.
460 */
461int
462gnet_stats_finish_copy(struct gnet_dump *d)
463{
464 if (d->tail)
465 d->tail->nla_len = skb_tail_pointer(skb: d->skb) - (u8 *)d->tail;
466
467 if (d->compat_tc_stats)
468 if (gnet_stats_copy(d, type: d->compat_tc_stats, buf: &d->tc_stats,
469 size: sizeof(d->tc_stats), padattr: d->padattr) < 0)
470 return -1;
471
472 if (d->compat_xstats && d->xstats) {
473 if (gnet_stats_copy(d, type: d->compat_xstats, buf: d->xstats,
474 size: d->xstats_len, padattr: d->padattr) < 0)
475 return -1;
476 }
477
478 if (d->lock)
479 spin_unlock_bh(lock: d->lock);
480 kfree(objp: d->xstats);
481 d->xstats = NULL;
482 d->xstats_len = 0;
483 return 0;
484}
485EXPORT_SYMBOL(gnet_stats_finish_copy);
486