1/*
2 * Copyright (c) 2012-2017 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28
29/*
30 * Flow Control and Feedback Advisory
31 *
32 * Each mbuf that is being sent out through an interface is tagged with a
33 * unique 32-bit ID which will help to identify all the packets that belong
34 * to a particular flow at the interface layer. Packets carrying such ID
35 * would need to be marked with PKTF_FLOW_ID. Normally, this ID is computed
36 * by the module that generates the flow. There are 3 kinds of flow sources
37 * that are currently recognized:
38 *
39 * a. INPCB (INET/INET6 Protocol Control Block). When a socket is
40 * connected, the flow hash for the socket is computed and stored in
41 * the PCB. Further transmissions on the socket will cause the hash
42 * value to be carried within the mbuf as the flow ID.
43 *
44 * b. Interface. When an interface is attached, the flow hash for the
45 * interface is computed and stored in the ifnet. This value is
46 * normally ignored for most network drivers, except for those that
47 * reside atop another driver, e.g. a virtual interface performing
48 * encapsulation/encryption on the original packet and sending the
49 * newly-generated packet to another interface. Such interface needs
50 * to associate all generated packets with the interface flow hash
51 * value as the flow ID.
52 *
53 * c. PF (Packet Filter). When a packet goes through PF and it is not
54 * already associated with a flow ID, PF will compute a flow hash and
55 * store it in the packet as flow ID. When the packet is associated
56 * with a PF state, the state record will have the flow ID stored
57 * within, in order to avoid recalculating the flow hash. Although PF
58 * is capable of generating flow IDs, it does not participate in flow
59 * advisory, and therefore packets whose IDs are computed by PF will
60 * not have their PKTF_FLOW_ADV packet flag set.
61 *
62 * Activation of flow advisory mechanism is done by setting the PKTF_FLOW_ADV
63 * packet flag; because a flow ID is required, the mechanism will not take
64 * place unless PKTF_FLOW_ID is set as well. The packet must also carry one
65 * of the flow source types FLOWSRC_{INPCB,IFNET} in order to identify where
66 * the flow advisory notification should be delivered to. As noted above,
67 * FLOWSRC_PF does not participate in this mechanism.
68 *
69 * The classq module configured on the interface is responsible for exerting
70 * flow control to the upper layers. This occurs when the number of packets
71 * queued for a flow reaches a limit. The module generating the flow will
72 * cease transmission until further flow advisory notice, and the flow will
73 * be inserted into the classq's flow control list.
74 *
75 * When packets are dequeued from the classq and the number of packets for
76 * a flow goes below a limit, the classq will transfer its flow control list
77 * to the global fadv_list. This will then trigger the flow advisory thread
78 * to run, which will cause the flow source modules to be notified that data
79 * can now be generated for those previously flow-controlled flows.
80 */
81
82#include <sys/param.h>
83#include <sys/systm.h>
84#include <sys/kernel.h>
85#include <sys/mcache.h>
86#include <sys/mbuf.h>
87#include <sys/proc_internal.h>
88#include <sys/socketvar.h>
89
90#include <kern/assert.h>
91#include <kern/thread.h>
92#include <kern/locks.h>
93#include <kern/zalloc.h>
94
95#include <netinet/in_pcb.h>
96#include <net/flowadv.h>
97
98/* Lock group and attribute for fadv_lock */
99static lck_grp_t *fadv_lock_grp;
100static lck_grp_attr_t *fadv_lock_grp_attr;
101decl_lck_mtx_data(static, fadv_lock);
102
103/* protected by fadv_lock */
104static STAILQ_HEAD(fadv_head, flowadv_fcentry) fadv_list;
105static thread_t fadv_thread = THREAD_NULL;
106static uint32_t fadv_active;
107
108static unsigned int fadv_zone_size; /* size of flowadv_fcentry */
109static struct zone *fadv_zone; /* zone for flowadv_fcentry */
110
111#define FADV_ZONE_MAX 32 /* maximum elements in zone */
112#define FADV_ZONE_NAME "fadv_zone" /* zone name */
113
114static int flowadv_thread_cont(int);
115static void flowadv_thread_func(void *, wait_result_t);
116
117void
118flowadv_init(void)
119{
120 STAILQ_INIT(&fadv_list);
121
122 /* Setup lock group and attribute for fadv_lock */
123 fadv_lock_grp_attr = lck_grp_attr_alloc_init();
124 fadv_lock_grp = lck_grp_alloc_init("fadv_lock", fadv_lock_grp_attr);
125 lck_mtx_init(&fadv_lock, fadv_lock_grp, NULL);
126
127 fadv_zone_size = P2ROUNDUP(sizeof (struct flowadv_fcentry),
128 sizeof (u_int64_t));
129 fadv_zone = zinit(fadv_zone_size,
130 FADV_ZONE_MAX * fadv_zone_size, 0, FADV_ZONE_NAME);
131 if (fadv_zone == NULL) {
132 panic("%s: failed allocating %s", __func__, FADV_ZONE_NAME);
133 /* NOTREACHED */
134 }
135 zone_change(fadv_zone, Z_EXPAND, TRUE);
136 zone_change(fadv_zone, Z_CALLERACCT, FALSE);
137
138 if (kernel_thread_start(flowadv_thread_func, NULL, &fadv_thread) !=
139 KERN_SUCCESS) {
140 panic("%s: couldn't create flow event advisory thread",
141 __func__);
142 /* NOTREACHED */
143 }
144 thread_deallocate(fadv_thread);
145}
146
147struct flowadv_fcentry *
148flowadv_alloc_entry(int how)
149{
150 struct flowadv_fcentry *fce;
151
152 fce = (how == M_WAITOK) ? zalloc(fadv_zone) : zalloc_noblock(fadv_zone);
153 if (fce != NULL)
154 bzero(fce, fadv_zone_size);
155
156 return (fce);
157}
158
159void
160flowadv_free_entry(struct flowadv_fcentry *fce)
161{
162 zfree(fadv_zone, fce);
163}
164
165void
166flowadv_add(struct flowadv_fclist *fcl)
167{
168 if (STAILQ_EMPTY(fcl))
169 return;
170
171 lck_mtx_lock_spin(&fadv_lock);
172
173 STAILQ_CONCAT(&fadv_list, fcl);
174 VERIFY(!STAILQ_EMPTY(&fadv_list));
175
176 if (!fadv_active && fadv_thread != THREAD_NULL)
177 wakeup_one((caddr_t)&fadv_list);
178
179 lck_mtx_unlock(&fadv_lock);
180}
181
182void
183flowadv_add_entry(struct flowadv_fcentry *fce) {
184 lck_mtx_lock_spin(&fadv_lock);
185 STAILQ_INSERT_HEAD(&fadv_list, fce, fce_link);
186 VERIFY(!STAILQ_EMPTY(&fadv_list));
187
188 if (!fadv_active && fadv_thread != THREAD_NULL)
189 wakeup_one((caddr_t)&fadv_list);
190
191 lck_mtx_unlock(&fadv_lock);
192}
193
194static int
195flowadv_thread_cont(int err)
196{
197#pragma unused(err)
198 for (;;) {
199 LCK_MTX_ASSERT(&fadv_lock, LCK_MTX_ASSERT_OWNED);
200 while (STAILQ_EMPTY(&fadv_list)) {
201 VERIFY(!fadv_active);
202 (void) msleep0(&fadv_list, &fadv_lock, (PSOCK | PSPIN),
203 "flowadv_cont", 0, flowadv_thread_cont);
204 /* NOTREACHED */
205 }
206
207 fadv_active = 1;
208 for (;;) {
209 struct flowadv_fcentry *fce;
210
211 VERIFY(!STAILQ_EMPTY(&fadv_list));
212 fce = STAILQ_FIRST(&fadv_list);
213 STAILQ_REMOVE(&fadv_list, fce,
214 flowadv_fcentry, fce_link);
215 STAILQ_NEXT(fce, fce_link) = NULL;
216
217 lck_mtx_unlock(&fadv_lock);
218 switch (fce->fce_flowsrc_type) {
219 case FLOWSRC_INPCB:
220 inp_flowadv(fce->fce_flowid);
221 break;
222
223 case FLOWSRC_IFNET:
224 ifnet_flowadv(fce->fce_flowid);
225 break;
226
227
228 case FLOWSRC_PF:
229 default:
230 break;
231 }
232 flowadv_free_entry(fce);
233 lck_mtx_lock_spin(&fadv_lock);
234
235 /* if there's no pending request, we're done */
236 if (STAILQ_EMPTY(&fadv_list))
237 break;
238 }
239 fadv_active = 0;
240 }
241}
242
243static void
244flowadv_thread_func(void *v, wait_result_t w)
245{
246#pragma unused(v, w)
247 lck_mtx_lock(&fadv_lock);
248 (void) msleep0(&fadv_list, &fadv_lock, (PSOCK | PSPIN),
249 "flowadv", 0, flowadv_thread_cont);
250 /*
251 * msleep0() shouldn't have returned as PCATCH was not set;
252 * therefore assert in this case.
253 */
254 lck_mtx_unlock(&fadv_lock);
255 VERIFY(0);
256}
257