sch_dsmark.c 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510
  1. /* net/sched/sch_dsmark.c - Differentiated Services field marker */
  2. /* Written 1998-2000 by Werner Almesberger, EPFL ICA */
  3. #include <linux/module.h>
  4. #include <linux/init.h>
  5. #include <linux/slab.h>
  6. #include <linux/types.h>
  7. #include <linux/string.h>
  8. #include <linux/errno.h>
  9. #include <linux/skbuff.h>
  10. #include <linux/rtnetlink.h>
  11. #include <linux/bitops.h>
  12. #include <net/pkt_sched.h>
  13. #include <net/dsfield.h>
  14. #include <net/inet_ecn.h>
  15. #include <asm/byteorder.h>
  16. /*
  17. * classid class marking
  18. * ------- ----- -------
  19. * n/a 0 n/a
  20. * x:0 1 use entry [0]
  21. * ... ... ...
  22. * x:y y>0 y+1 use entry [y]
  23. * ... ... ...
  24. * x:indices-1 indices use entry [indices-1]
  25. * ... ... ...
  26. * x:y y+1 use entry [y & (indices-1)]
  27. * ... ... ...
  28. * 0xffff 0x10000 use entry [indices-1]
  29. */
  30. #define NO_DEFAULT_INDEX (1 << 16)
  31. struct dsmark_qdisc_data {
  32. struct Qdisc *q;
  33. struct tcf_proto *filter_list;
  34. u8 *mask; /* "owns" the array */
  35. u8 *value;
  36. u16 indices;
  37. u32 default_index; /* index range is 0...0xffff */
  38. int set_tc_index;
  39. };
  40. static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
  41. {
  42. return (index <= p->indices && index > 0);
  43. }
  44. /* ------------------------- Class/flow operations ------------------------- */
  45. static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
  46. struct Qdisc *new, struct Qdisc **old)
  47. {
  48. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  49. pr_debug("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
  50. sch, p, new, old);
  51. if (new == NULL) {
  52. new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
  53. sch->handle);
  54. if (new == NULL)
  55. new = &noop_qdisc;
  56. }
  57. sch_tree_lock(sch);
  58. *old = p->q;
  59. p->q = new;
  60. qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
  61. qdisc_reset(*old);
  62. sch_tree_unlock(sch);
  63. return 0;
  64. }
  65. static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
  66. {
  67. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  68. return p->q;
  69. }
  70. static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
  71. {
  72. pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
  73. sch, qdisc_priv(sch), classid);
  74. return TC_H_MIN(classid) + 1;
  75. }
  76. static unsigned long dsmark_bind_filter(struct Qdisc *sch,
  77. unsigned long parent, u32 classid)
  78. {
  79. return dsmark_get(sch, classid);
  80. }
  81. static void dsmark_put(struct Qdisc *sch, unsigned long cl)
  82. {
  83. }
  84. static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
  85. [TCA_DSMARK_INDICES] = { .type = NLA_U16 },
  86. [TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
  87. [TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
  88. [TCA_DSMARK_MASK] = { .type = NLA_U8 },
  89. [TCA_DSMARK_VALUE] = { .type = NLA_U8 },
  90. };
  91. static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
  92. struct nlattr **tca, unsigned long *arg)
  93. {
  94. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  95. struct nlattr *opt = tca[TCA_OPTIONS];
  96. struct nlattr *tb[TCA_DSMARK_MAX + 1];
  97. int err = -EINVAL;
  98. u8 mask = 0;
  99. pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
  100. "arg 0x%lx\n", sch, p, classid, parent, *arg);
  101. if (!dsmark_valid_index(p, *arg)) {
  102. err = -ENOENT;
  103. goto errout;
  104. }
  105. if (!opt)
  106. goto errout;
  107. err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
  108. if (err < 0)
  109. goto errout;
  110. if (tb[TCA_DSMARK_MASK])
  111. mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
  112. if (tb[TCA_DSMARK_VALUE])
  113. p->value[*arg - 1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
  114. if (tb[TCA_DSMARK_MASK])
  115. p->mask[*arg - 1] = mask;
  116. err = 0;
  117. errout:
  118. return err;
  119. }
  120. static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
  121. {
  122. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  123. if (!dsmark_valid_index(p, arg))
  124. return -EINVAL;
  125. p->mask[arg - 1] = 0xff;
  126. p->value[arg - 1] = 0;
  127. return 0;
  128. }
  129. static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
  130. {
  131. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  132. int i;
  133. pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
  134. if (walker->stop)
  135. return;
  136. for (i = 0; i < p->indices; i++) {
  137. if (p->mask[i] == 0xff && !p->value[i])
  138. goto ignore;
  139. if (walker->count >= walker->skip) {
  140. if (walker->fn(sch, i + 1, walker) < 0) {
  141. walker->stop = 1;
  142. break;
  143. }
  144. }
  145. ignore:
  146. walker->count++;
  147. }
  148. }
  149. static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
  150. unsigned long cl)
  151. {
  152. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  153. return &p->filter_list;
  154. }
  155. /* --------------------------- Qdisc operations ---------------------------- */
  156. static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
  157. {
  158. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  159. int err;
  160. pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
  161. if (p->set_tc_index) {
  162. switch (skb->protocol) {
  163. case htons(ETH_P_IP):
  164. if (skb_cow_head(skb, sizeof(struct iphdr)))
  165. goto drop;
  166. skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
  167. & ~INET_ECN_MASK;
  168. break;
  169. case htons(ETH_P_IPV6):
  170. if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
  171. goto drop;
  172. skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
  173. & ~INET_ECN_MASK;
  174. break;
  175. default:
  176. skb->tc_index = 0;
  177. break;
  178. }
  179. }
  180. if (TC_H_MAJ(skb->priority) == sch->handle)
  181. skb->tc_index = TC_H_MIN(skb->priority);
  182. else {
  183. struct tcf_result res;
  184. int result = tc_classify(skb, p->filter_list, &res);
  185. pr_debug("result %d class 0x%04x\n", result, res.classid);
  186. switch (result) {
  187. #ifdef CONFIG_NET_CLS_ACT
  188. case TC_ACT_QUEUED:
  189. case TC_ACT_STOLEN:
  190. kfree_skb(skb);
  191. return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
  192. case TC_ACT_SHOT:
  193. goto drop;
  194. #endif
  195. case TC_ACT_OK:
  196. skb->tc_index = TC_H_MIN(res.classid);
  197. break;
  198. default:
  199. if (p->default_index != NO_DEFAULT_INDEX)
  200. skb->tc_index = p->default_index;
  201. break;
  202. }
  203. }
  204. err = qdisc_enqueue(skb, p->q);
  205. if (err != NET_XMIT_SUCCESS) {
  206. if (net_xmit_drop_count(err))
  207. sch->qstats.drops++;
  208. return err;
  209. }
  210. sch->q.qlen++;
  211. return NET_XMIT_SUCCESS;
  212. drop:
  213. kfree_skb(skb);
  214. sch->qstats.drops++;
  215. return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
  216. }
  217. static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
  218. {
  219. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  220. struct sk_buff *skb;
  221. u32 index;
  222. pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
  223. skb = p->q->ops->dequeue(p->q);
  224. if (skb == NULL)
  225. return NULL;
  226. qdisc_bstats_update(sch, skb);
  227. sch->q.qlen--;
  228. index = skb->tc_index & (p->indices - 1);
  229. pr_debug("index %d->%d\n", skb->tc_index, index);
  230. switch (skb->protocol) {
  231. case htons(ETH_P_IP):
  232. ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
  233. p->value[index]);
  234. break;
  235. case htons(ETH_P_IPV6):
  236. ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
  237. p->value[index]);
  238. break;
  239. default:
  240. /*
  241. * Only complain if a change was actually attempted.
  242. * This way, we can send non-IP traffic through dsmark
  243. * and don't need yet another qdisc as a bypass.
  244. */
  245. if (p->mask[index] != 0xff || p->value[index])
  246. pr_warning("dsmark_dequeue: unsupported protocol %d\n",
  247. ntohs(skb->protocol));
  248. break;
  249. }
  250. return skb;
  251. }
  252. static struct sk_buff *dsmark_peek(struct Qdisc *sch)
  253. {
  254. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  255. pr_debug("dsmark_peek(sch %p,[qdisc %p])\n", sch, p);
  256. return p->q->ops->peek(p->q);
  257. }
  258. static unsigned int dsmark_drop(struct Qdisc *sch)
  259. {
  260. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  261. unsigned int len;
  262. pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
  263. if (p->q->ops->drop == NULL)
  264. return 0;
  265. len = p->q->ops->drop(p->q);
  266. if (len)
  267. sch->q.qlen--;
  268. return len;
  269. }
  270. static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
  271. {
  272. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  273. struct nlattr *tb[TCA_DSMARK_MAX + 1];
  274. int err = -EINVAL;
  275. u32 default_index = NO_DEFAULT_INDEX;
  276. u16 indices;
  277. u8 *mask;
  278. pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
  279. if (!opt)
  280. goto errout;
  281. err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
  282. if (err < 0)
  283. goto errout;
  284. err = -EINVAL;
  285. indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
  286. if (hweight32(indices) != 1)
  287. goto errout;
  288. if (tb[TCA_DSMARK_DEFAULT_INDEX])
  289. default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
  290. mask = kmalloc(indices * 2, GFP_KERNEL);
  291. if (mask == NULL) {
  292. err = -ENOMEM;
  293. goto errout;
  294. }
  295. p->mask = mask;
  296. memset(p->mask, 0xff, indices);
  297. p->value = p->mask + indices;
  298. memset(p->value, 0, indices);
  299. p->indices = indices;
  300. p->default_index = default_index;
  301. p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
  302. p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
  303. if (p->q == NULL)
  304. p->q = &noop_qdisc;
  305. pr_debug("dsmark_init: qdisc %p\n", p->q);
  306. err = 0;
  307. errout:
  308. return err;
  309. }
  310. static void dsmark_reset(struct Qdisc *sch)
  311. {
  312. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  313. pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
  314. qdisc_reset(p->q);
  315. sch->q.qlen = 0;
  316. }
  317. static void dsmark_destroy(struct Qdisc *sch)
  318. {
  319. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  320. pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
  321. tcf_destroy_chain(&p->filter_list);
  322. qdisc_destroy(p->q);
  323. kfree(p->mask);
  324. }
  325. static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
  326. struct sk_buff *skb, struct tcmsg *tcm)
  327. {
  328. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  329. struct nlattr *opts = NULL;
  330. pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
  331. if (!dsmark_valid_index(p, cl))
  332. return -EINVAL;
  333. tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
  334. tcm->tcm_info = p->q->handle;
  335. opts = nla_nest_start(skb, TCA_OPTIONS);
  336. if (opts == NULL)
  337. goto nla_put_failure;
  338. NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl - 1]);
  339. NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl - 1]);
  340. return nla_nest_end(skb, opts);
  341. nla_put_failure:
  342. nla_nest_cancel(skb, opts);
  343. return -EMSGSIZE;
  344. }
  345. static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
  346. {
  347. struct dsmark_qdisc_data *p = qdisc_priv(sch);
  348. struct nlattr *opts = NULL;
  349. opts = nla_nest_start(skb, TCA_OPTIONS);
  350. if (opts == NULL)
  351. goto nla_put_failure;
  352. NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
  353. if (p->default_index != NO_DEFAULT_INDEX)
  354. NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
  355. if (p->set_tc_index)
  356. NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
  357. return nla_nest_end(skb, opts);
  358. nla_put_failure:
  359. nla_nest_cancel(skb, opts);
  360. return -EMSGSIZE;
  361. }
  362. static const struct Qdisc_class_ops dsmark_class_ops = {
  363. .graft = dsmark_graft,
  364. .leaf = dsmark_leaf,
  365. .get = dsmark_get,
  366. .put = dsmark_put,
  367. .change = dsmark_change,
  368. .delete = dsmark_delete,
  369. .walk = dsmark_walk,
  370. .tcf_chain = dsmark_find_tcf,
  371. .bind_tcf = dsmark_bind_filter,
  372. .unbind_tcf = dsmark_put,
  373. .dump = dsmark_dump_class,
  374. };
  375. static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
  376. .next = NULL,
  377. .cl_ops = &dsmark_class_ops,
  378. .id = "dsmark",
  379. .priv_size = sizeof(struct dsmark_qdisc_data),
  380. .enqueue = dsmark_enqueue,
  381. .dequeue = dsmark_dequeue,
  382. .peek = dsmark_peek,
  383. .drop = dsmark_drop,
  384. .init = dsmark_init,
  385. .reset = dsmark_reset,
  386. .destroy = dsmark_destroy,
  387. .change = NULL,
  388. .dump = dsmark_dump,
  389. .owner = THIS_MODULE,
  390. };
  391. static int __init dsmark_module_init(void)
  392. {
  393. return register_qdisc(&dsmark_qdisc_ops);
  394. }
  395. static void __exit dsmark_module_exit(void)
  396. {
  397. unregister_qdisc(&dsmark_qdisc_ops);
  398. }
  399. module_init(dsmark_module_init)
  400. module_exit(dsmark_module_exit)
  401. MODULE_LICENSE("GPL");