1 | /* |
2 | * Copyright (c) 2003-2008 Chelsio, Inc. All rights reserved. |
3 | * |
4 | * This software is available to you under a choice of one of two |
5 | * licenses. You may choose to be licensed under the terms of the GNU |
6 | * General Public License (GPL) Version 2, available from the file |
7 | * COPYING in the main directory of this source tree, or the |
8 | * OpenIB.org BSD license below: |
9 | * |
10 | * Redistribution and use in source and binary forms, with or |
11 | * without modification, are permitted provided that the following |
12 | * conditions are met: |
13 | * |
14 | * - Redistributions of source code must retain the above |
15 | * copyright notice, this list of conditions and the following |
16 | * disclaimer. |
17 | * |
18 | * - Redistributions in binary form must reproduce the above |
19 | * copyright notice, this list of conditions and the following |
20 | * disclaimer in the documentation and/or other materials |
21 | * provided with the distribution. |
22 | * |
23 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, |
24 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF |
25 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND |
26 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS |
27 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN |
28 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN |
29 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
30 | * SOFTWARE. |
31 | */ |
32 | #include <linux/skbuff.h> |
33 | #include <linux/netdevice.h> |
34 | #include <linux/if.h> |
35 | #include <linux/if_vlan.h> |
36 | #include <linux/jhash.h> |
37 | #include <linux/slab.h> |
38 | #include <linux/export.h> |
39 | #include <net/neighbour.h> |
40 | #include "common.h" |
41 | #include "t3cdev.h" |
42 | #include "cxgb3_defs.h" |
43 | #include "l2t.h" |
44 | #include "t3_cpl.h" |
45 | #include "firmware_exports.h" |
46 | |
47 | #define VLAN_NONE 0xfff |
48 | |
49 | /* |
50 | * Module locking notes: There is a RW lock protecting the L2 table as a |
51 | * whole plus a spinlock per L2T entry. Entry lookups and allocations happen |
52 | * under the protection of the table lock, individual entry changes happen |
53 | * while holding that entry's spinlock. The table lock nests outside the |
54 | * entry locks. Allocations of new entries take the table lock as writers so |
55 | * no other lookups can happen while allocating new entries. Entry updates |
56 | * take the table lock as readers so multiple entries can be updated in |
57 | * parallel. An L2T entry can be dropped by decrementing its reference count |
58 | * and therefore can happen in parallel with entry allocation but no entry |
59 | * can change state or increment its ref count during allocation as both of |
60 | * these perform lookups. |
61 | */ |
62 | |
63 | static inline unsigned int vlan_prio(const struct l2t_entry *e) |
64 | { |
65 | return e->vlan >> 13; |
66 | } |
67 | |
68 | static inline unsigned int arp_hash(u32 key, int ifindex, |
69 | const struct l2t_data *d) |
70 | { |
71 | return jhash_2words(a: key, b: ifindex, initval: 0) & (d->nentries - 1); |
72 | } |
73 | |
74 | static inline void neigh_replace(struct l2t_entry *e, struct neighbour *n) |
75 | { |
76 | neigh_hold(n); |
77 | if (e->neigh) |
78 | neigh_release(neigh: e->neigh); |
79 | e->neigh = n; |
80 | } |
81 | |
82 | /* |
83 | * Set up an L2T entry and send any packets waiting in the arp queue. The |
84 | * supplied skb is used for the CPL_L2T_WRITE_REQ. Must be called with the |
85 | * entry locked. |
86 | */ |
87 | static int setup_l2e_send_pending(struct t3cdev *dev, struct sk_buff *skb, |
88 | struct l2t_entry *e) |
89 | { |
90 | struct cpl_l2t_write_req *req; |
91 | struct sk_buff *tmp; |
92 | |
93 | if (!skb) { |
94 | skb = alloc_skb(size: sizeof(*req), GFP_ATOMIC); |
95 | if (!skb) |
96 | return -ENOMEM; |
97 | } |
98 | |
99 | req = __skb_put(skb, len: sizeof(*req)); |
100 | req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD)); |
101 | OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, e->idx)); |
102 | req->params = htonl(V_L2T_W_IDX(e->idx) | V_L2T_W_IFF(e->smt_idx) | |
103 | V_L2T_W_VLAN(e->vlan & VLAN_VID_MASK) | |
104 | V_L2T_W_PRIO(vlan_prio(e))); |
105 | memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac)); |
106 | memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac)); |
107 | skb->priority = CPL_PRIORITY_CONTROL; |
108 | cxgb3_ofld_send(dev, skb); |
109 | |
110 | skb_queue_walk_safe(&e->arpq, skb, tmp) { |
111 | __skb_unlink(skb, list: &e->arpq); |
112 | cxgb3_ofld_send(dev, skb); |
113 | } |
114 | e->state = L2T_STATE_VALID; |
115 | |
116 | return 0; |
117 | } |
118 | |
119 | /* |
120 | * Add a packet to the an L2T entry's queue of packets awaiting resolution. |
121 | * Must be called with the entry's lock held. |
122 | */ |
123 | static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb) |
124 | { |
125 | __skb_queue_tail(list: &e->arpq, newsk: skb); |
126 | } |
127 | |
128 | int t3_l2t_send_slow(struct t3cdev *dev, struct sk_buff *skb, |
129 | struct l2t_entry *e) |
130 | { |
131 | again: |
132 | switch (e->state) { |
133 | case L2T_STATE_STALE: /* entry is stale, kick off revalidation */ |
134 | neigh_event_send(neigh: e->neigh, NULL); |
135 | spin_lock_bh(lock: &e->lock); |
136 | if (e->state == L2T_STATE_STALE) |
137 | e->state = L2T_STATE_VALID; |
138 | spin_unlock_bh(lock: &e->lock); |
139 | fallthrough; |
140 | case L2T_STATE_VALID: /* fast-path, send the packet on */ |
141 | return cxgb3_ofld_send(dev, skb); |
142 | case L2T_STATE_RESOLVING: |
143 | spin_lock_bh(lock: &e->lock); |
144 | if (e->state != L2T_STATE_RESOLVING) { |
145 | /* ARP already completed */ |
146 | spin_unlock_bh(lock: &e->lock); |
147 | goto again; |
148 | } |
149 | arpq_enqueue(e, skb); |
150 | spin_unlock_bh(lock: &e->lock); |
151 | |
152 | /* |
153 | * Only the first packet added to the arpq should kick off |
154 | * resolution. However, because the alloc_skb below can fail, |
155 | * we allow each packet added to the arpq to retry resolution |
156 | * as a way of recovering from transient memory exhaustion. |
157 | * A better way would be to use a work request to retry L2T |
158 | * entries when there's no memory. |
159 | */ |
160 | if (!neigh_event_send(neigh: e->neigh, NULL)) { |
161 | skb = alloc_skb(size: sizeof(struct cpl_l2t_write_req), |
162 | GFP_ATOMIC); |
163 | if (!skb) |
164 | break; |
165 | |
166 | spin_lock_bh(lock: &e->lock); |
167 | if (!skb_queue_empty(list: &e->arpq)) |
168 | setup_l2e_send_pending(dev, skb, e); |
169 | else /* we lost the race */ |
170 | __kfree_skb(skb); |
171 | spin_unlock_bh(lock: &e->lock); |
172 | } |
173 | } |
174 | return 0; |
175 | } |
176 | |
177 | EXPORT_SYMBOL(t3_l2t_send_slow); |
178 | |
179 | void t3_l2t_send_event(struct t3cdev *dev, struct l2t_entry *e) |
180 | { |
181 | again: |
182 | switch (e->state) { |
183 | case L2T_STATE_STALE: /* entry is stale, kick off revalidation */ |
184 | neigh_event_send(neigh: e->neigh, NULL); |
185 | spin_lock_bh(lock: &e->lock); |
186 | if (e->state == L2T_STATE_STALE) { |
187 | e->state = L2T_STATE_VALID; |
188 | } |
189 | spin_unlock_bh(lock: &e->lock); |
190 | return; |
191 | case L2T_STATE_VALID: /* fast-path, send the packet on */ |
192 | return; |
193 | case L2T_STATE_RESOLVING: |
194 | spin_lock_bh(lock: &e->lock); |
195 | if (e->state != L2T_STATE_RESOLVING) { |
196 | /* ARP already completed */ |
197 | spin_unlock_bh(lock: &e->lock); |
198 | goto again; |
199 | } |
200 | spin_unlock_bh(lock: &e->lock); |
201 | |
202 | /* |
203 | * Only the first packet added to the arpq should kick off |
204 | * resolution. However, because the alloc_skb below can fail, |
205 | * we allow each packet added to the arpq to retry resolution |
206 | * as a way of recovering from transient memory exhaustion. |
207 | * A better way would be to use a work request to retry L2T |
208 | * entries when there's no memory. |
209 | */ |
210 | neigh_event_send(neigh: e->neigh, NULL); |
211 | } |
212 | } |
213 | |
214 | EXPORT_SYMBOL(t3_l2t_send_event); |
215 | |
216 | /* |
217 | * Allocate a free L2T entry. Must be called with l2t_data.lock held. |
218 | */ |
219 | static struct l2t_entry *alloc_l2e(struct l2t_data *d) |
220 | { |
221 | struct l2t_entry *end, *e, **p; |
222 | |
223 | if (!atomic_read(v: &d->nfree)) |
224 | return NULL; |
225 | |
226 | /* there's definitely a free entry */ |
227 | for (e = d->rover, end = &d->l2tab[d->nentries]; e != end; ++e) |
228 | if (atomic_read(v: &e->refcnt) == 0) |
229 | goto found; |
230 | |
231 | for (e = &d->l2tab[1]; atomic_read(v: &e->refcnt); ++e) ; |
232 | found: |
233 | d->rover = e + 1; |
234 | atomic_dec(v: &d->nfree); |
235 | |
236 | /* |
237 | * The entry we found may be an inactive entry that is |
238 | * presently in the hash table. We need to remove it. |
239 | */ |
240 | if (e->state != L2T_STATE_UNUSED) { |
241 | int hash = arp_hash(key: e->addr, ifindex: e->ifindex, d); |
242 | |
243 | for (p = &d->l2tab[hash].first; *p; p = &(*p)->next) |
244 | if (*p == e) { |
245 | *p = e->next; |
246 | break; |
247 | } |
248 | e->state = L2T_STATE_UNUSED; |
249 | } |
250 | return e; |
251 | } |
252 | |
253 | /* |
254 | * Called when an L2T entry has no more users. The entry is left in the hash |
255 | * table since it is likely to be reused but we also bump nfree to indicate |
256 | * that the entry can be reallocated for a different neighbor. We also drop |
257 | * the existing neighbor reference in case the neighbor is going away and is |
258 | * waiting on our reference. |
259 | * |
260 | * Because entries can be reallocated to other neighbors once their ref count |
261 | * drops to 0 we need to take the entry's lock to avoid races with a new |
262 | * incarnation. |
263 | */ |
264 | void t3_l2e_free(struct l2t_data *d, struct l2t_entry *e) |
265 | { |
266 | spin_lock_bh(lock: &e->lock); |
267 | if (atomic_read(v: &e->refcnt) == 0) { /* hasn't been recycled */ |
268 | if (e->neigh) { |
269 | neigh_release(neigh: e->neigh); |
270 | e->neigh = NULL; |
271 | } |
272 | } |
273 | spin_unlock_bh(lock: &e->lock); |
274 | atomic_inc(v: &d->nfree); |
275 | } |
276 | |
277 | EXPORT_SYMBOL(t3_l2e_free); |
278 | |
279 | /* |
280 | * Update an L2T entry that was previously used for the same next hop as neigh. |
281 | * Must be called with softirqs disabled. |
282 | */ |
283 | static inline void reuse_entry(struct l2t_entry *e, struct neighbour *neigh) |
284 | { |
285 | unsigned int nud_state; |
286 | |
287 | spin_lock(lock: &e->lock); /* avoid race with t3_l2t_free */ |
288 | |
289 | if (neigh != e->neigh) |
290 | neigh_replace(e, n: neigh); |
291 | nud_state = neigh->nud_state; |
292 | if (memcmp(p: e->dmac, q: neigh->ha, size: sizeof(e->dmac)) || |
293 | !(nud_state & NUD_VALID)) |
294 | e->state = L2T_STATE_RESOLVING; |
295 | else if (nud_state & NUD_CONNECTED) |
296 | e->state = L2T_STATE_VALID; |
297 | else |
298 | e->state = L2T_STATE_STALE; |
299 | spin_unlock(lock: &e->lock); |
300 | } |
301 | |
302 | struct l2t_entry *t3_l2t_get(struct t3cdev *cdev, struct dst_entry *dst, |
303 | struct net_device *dev, const void *daddr) |
304 | { |
305 | struct l2t_entry *e = NULL; |
306 | struct neighbour *neigh; |
307 | struct port_info *p; |
308 | struct l2t_data *d; |
309 | int hash; |
310 | u32 addr; |
311 | int ifidx; |
312 | int smt_idx; |
313 | |
314 | rcu_read_lock(); |
315 | neigh = dst_neigh_lookup(dst, daddr); |
316 | if (!neigh) |
317 | goto done_rcu; |
318 | |
319 | addr = *(u32 *) neigh->primary_key; |
320 | ifidx = neigh->dev->ifindex; |
321 | |
322 | if (!dev) |
323 | dev = neigh->dev; |
324 | p = netdev_priv(dev); |
325 | smt_idx = p->port_id; |
326 | |
327 | d = L2DATA(cdev); |
328 | if (!d) |
329 | goto done_rcu; |
330 | |
331 | hash = arp_hash(key: addr, ifindex: ifidx, d); |
332 | |
333 | write_lock_bh(&d->lock); |
334 | for (e = d->l2tab[hash].first; e; e = e->next) |
335 | if (e->addr == addr && e->ifindex == ifidx && |
336 | e->smt_idx == smt_idx) { |
337 | l2t_hold(d, e); |
338 | if (atomic_read(v: &e->refcnt) == 1) |
339 | reuse_entry(e, neigh); |
340 | goto done_unlock; |
341 | } |
342 | |
343 | /* Need to allocate a new entry */ |
344 | e = alloc_l2e(d); |
345 | if (e) { |
346 | spin_lock(lock: &e->lock); /* avoid race with t3_l2t_free */ |
347 | e->next = d->l2tab[hash].first; |
348 | d->l2tab[hash].first = e; |
349 | e->state = L2T_STATE_RESOLVING; |
350 | e->addr = addr; |
351 | e->ifindex = ifidx; |
352 | e->smt_idx = smt_idx; |
353 | atomic_set(v: &e->refcnt, i: 1); |
354 | neigh_replace(e, n: neigh); |
355 | if (is_vlan_dev(dev: neigh->dev)) |
356 | e->vlan = vlan_dev_vlan_id(dev: neigh->dev); |
357 | else |
358 | e->vlan = VLAN_NONE; |
359 | spin_unlock(lock: &e->lock); |
360 | } |
361 | done_unlock: |
362 | write_unlock_bh(&d->lock); |
363 | done_rcu: |
364 | if (neigh) |
365 | neigh_release(neigh); |
366 | rcu_read_unlock(); |
367 | return e; |
368 | } |
369 | |
370 | EXPORT_SYMBOL(t3_l2t_get); |
371 | |
372 | /* |
373 | * Called when address resolution fails for an L2T entry to handle packets |
374 | * on the arpq head. If a packet specifies a failure handler it is invoked, |
375 | * otherwise the packets is sent to the offload device. |
376 | * |
377 | * XXX: maybe we should abandon the latter behavior and just require a failure |
378 | * handler. |
379 | */ |
380 | static void handle_failed_resolution(struct t3cdev *dev, struct sk_buff_head *arpq) |
381 | { |
382 | struct sk_buff *skb, *tmp; |
383 | |
384 | skb_queue_walk_safe(arpq, skb, tmp) { |
385 | struct l2t_skb_cb *cb = L2T_SKB_CB(skb); |
386 | |
387 | __skb_unlink(skb, list: arpq); |
388 | if (cb->arp_failure_handler) |
389 | cb->arp_failure_handler(dev, skb); |
390 | else |
391 | cxgb3_ofld_send(dev, skb); |
392 | } |
393 | } |
394 | |
395 | /* |
396 | * Called when the host's ARP layer makes a change to some entry that is |
397 | * loaded into the HW L2 table. |
398 | */ |
399 | void t3_l2t_update(struct t3cdev *dev, struct neighbour *neigh) |
400 | { |
401 | struct sk_buff_head arpq; |
402 | struct l2t_entry *e; |
403 | struct l2t_data *d = L2DATA(dev); |
404 | u32 addr = *(u32 *) neigh->primary_key; |
405 | int ifidx = neigh->dev->ifindex; |
406 | int hash = arp_hash(key: addr, ifindex: ifidx, d); |
407 | |
408 | read_lock_bh(&d->lock); |
409 | for (e = d->l2tab[hash].first; e; e = e->next) |
410 | if (e->addr == addr && e->ifindex == ifidx) { |
411 | spin_lock(lock: &e->lock); |
412 | goto found; |
413 | } |
414 | read_unlock_bh(&d->lock); |
415 | return; |
416 | |
417 | found: |
418 | __skb_queue_head_init(list: &arpq); |
419 | |
420 | read_unlock(&d->lock); |
421 | if (atomic_read(v: &e->refcnt)) { |
422 | if (neigh != e->neigh) |
423 | neigh_replace(e, n: neigh); |
424 | |
425 | if (e->state == L2T_STATE_RESOLVING) { |
426 | if (neigh->nud_state & NUD_FAILED) { |
427 | skb_queue_splice_init(list: &e->arpq, head: &arpq); |
428 | } else if (neigh->nud_state & (NUD_CONNECTED|NUD_STALE)) |
429 | setup_l2e_send_pending(dev, NULL, e); |
430 | } else { |
431 | e->state = neigh->nud_state & NUD_CONNECTED ? |
432 | L2T_STATE_VALID : L2T_STATE_STALE; |
433 | if (!ether_addr_equal(addr1: e->dmac, addr2: neigh->ha)) |
434 | setup_l2e_send_pending(dev, NULL, e); |
435 | } |
436 | } |
437 | spin_unlock_bh(lock: &e->lock); |
438 | |
439 | if (!skb_queue_empty(list: &arpq)) |
440 | handle_failed_resolution(dev, arpq: &arpq); |
441 | } |
442 | |
443 | struct l2t_data *t3_init_l2t(unsigned int l2t_capacity) |
444 | { |
445 | struct l2t_data *d; |
446 | int i; |
447 | |
448 | d = kvzalloc(struct_size(d, l2tab, l2t_capacity), GFP_KERNEL); |
449 | if (!d) |
450 | return NULL; |
451 | |
452 | d->nentries = l2t_capacity; |
453 | d->rover = &d->l2tab[1]; /* entry 0 is not used */ |
454 | atomic_set(v: &d->nfree, i: l2t_capacity - 1); |
455 | rwlock_init(&d->lock); |
456 | |
457 | for (i = 0; i < l2t_capacity; ++i) { |
458 | d->l2tab[i].idx = i; |
459 | d->l2tab[i].state = L2T_STATE_UNUSED; |
460 | __skb_queue_head_init(list: &d->l2tab[i].arpq); |
461 | spin_lock_init(&d->l2tab[i].lock); |
462 | atomic_set(v: &d->l2tab[i].refcnt, i: 0); |
463 | } |
464 | return d; |
465 | } |
466 | |