2 * Copyright (c) 2009-2019 Apple Inc. All rights reserved.
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
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.
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
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.
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
29 #include <sys/systm.h>
30 #include <sys/kernel.h>
31 #include <sys/types.h>
32 #include <sys/filedesc.h>
33 #include <sys/file_internal.h>
35 #include <sys/socket.h>
36 #include <sys/socketvar.h>
37 #include <sys/errno.h>
38 #include <sys/protosw.h>
39 #include <sys/domain.h>
41 #include <sys/queue.h>
42 #include <sys/sysctl.h>
43 #include <sys/sysproto.h>
46 #include <net/if_var.h>
47 #include <net/route.h>
49 #include <netinet/in.h>
50 #include <netinet/in_var.h>
51 #include <netinet/in_pcb.h>
52 #include <netinet/ip.h>
53 #include <netinet/ip_var.h>
54 #include <netinet/ip6.h>
55 #include <netinet6/ip6_var.h>
56 #include <netinet/udp.h>
57 #include <netinet/udp_var.h>
58 #include <netinet/tcp.h>
59 #include <netinet/tcp_var.h>
60 #include <netinet/tcp_cc.h>
61 #include <netinet/lro_ext.h>
62 #include <netinet/in_tclass.h>
68 static inline int so_throttle_best_effort(struct socket
*, struct ifnet
*);
69 static void set_dscp_to_wifi_ac_map(const struct dcsp_msc_map
*, int);
70 static errno_t
dscp_msc_map_from_netsvctype_dscp_map(struct netsvctype_dscp_map
*, size_t,
71 struct dcsp_msc_map
*);
73 static lck_grp_attr_t
*tclass_lck_grp_attr
= NULL
; /* mutex group attributes */
74 static lck_grp_t
*tclass_lck_grp
= NULL
; /* mutex group definition */
75 static lck_attr_t
*tclass_lck_attr
= NULL
; /* mutex attributes */
76 decl_lck_mtx_data(static, tclass_lock_data
);
77 static lck_mtx_t
*tclass_lock
= &tclass_lock_data
;
79 SYSCTL_NODE(_net
, OID_AUTO
, qos
,
80 CTLFLAG_RW
| CTLFLAG_LOCKED
, 0, "QoS");
82 static int sysctl_default_netsvctype_to_dscp_map SYSCTL_HANDLER_ARGS
;
83 SYSCTL_PROC(_net_qos
, OID_AUTO
, default_netsvctype_to_dscp_map
,
84 CTLTYPE_STRUCT
| CTLFLAG_RW
| CTLFLAG_LOCKED
,
85 0, 0, sysctl_default_netsvctype_to_dscp_map
, "S", "");
87 static int sysctl_dscp_to_wifi_ac_map SYSCTL_HANDLER_ARGS
;
88 SYSCTL_PROC(_net_qos
, OID_AUTO
, dscp_to_wifi_ac_map
,
89 CTLTYPE_STRUCT
| CTLFLAG_RW
| CTLFLAG_LOCKED
,
90 0, 0, sysctl_dscp_to_wifi_ac_map
, "S", "");
92 static int sysctl_reset_dscp_to_wifi_ac_map SYSCTL_HANDLER_ARGS
;
93 SYSCTL_PROC(_net_qos
, OID_AUTO
, reset_dscp_to_wifi_ac_map
,
94 CTLTYPE_INT
| CTLFLAG_WR
| CTLFLAG_LOCKED
,
95 0, 0, sysctl_reset_dscp_to_wifi_ac_map
, "I", "");
97 int net_qos_verbose
= 0;
98 SYSCTL_INT(_net_qos
, OID_AUTO
, verbose
,
99 CTLFLAG_RW
| CTLFLAG_LOCKED
, &net_qos_verbose
, 0, "");
102 * Fastlane QoS policy:
103 * By Default allow all apps to get traffic class to DSCP mapping
105 SYSCTL_NODE(_net_qos
, OID_AUTO
, policy
,
106 CTLFLAG_RW
| CTLFLAG_LOCKED
, 0, "");
108 int net_qos_policy_restricted
= 0;
109 SYSCTL_INT(_net_qos_policy
, OID_AUTO
, restricted
,
110 CTLFLAG_RW
| CTLFLAG_LOCKED
, &net_qos_policy_restricted
, 0, "");
112 int net_qos_policy_restrict_avapps
= 0;
113 SYSCTL_INT(_net_qos_policy
, OID_AUTO
, restrict_avapps
,
114 CTLFLAG_RW
| CTLFLAG_LOCKED
, &net_qos_policy_restrict_avapps
, 0, "");
116 int net_qos_policy_wifi_enabled
= 0;
117 SYSCTL_INT(_net_qos_policy
, OID_AUTO
, wifi_enabled
,
118 CTLFLAG_RW
| CTLFLAG_LOCKED
, &net_qos_policy_wifi_enabled
, 0, "");
120 int net_qos_policy_none_wifi_enabled
= 0;
121 SYSCTL_INT(_net_qos_policy
, OID_AUTO
, none_wifi_enabled
,
122 CTLFLAG_RW
| CTLFLAG_LOCKED
, &net_qos_policy_none_wifi_enabled
, 0, "");
124 int net_qos_policy_capable_enabled
= 0;
125 SYSCTL_INT(_net_qos_policy
, OID_AUTO
, capable_enabled
,
126 CTLFLAG_RW
| CTLFLAG_LOCKED
, &net_qos_policy_capable_enabled
, 0, "");
129 * Socket traffic class from network service type
131 const int sotc_by_netservicetype
[_NET_SERVICE_TYPE_COUNT
] = {
132 SO_TC_BE
, /* NET_SERVICE_TYPE_BE */
133 SO_TC_BK_SYS
, /* NET_SERVICE_TYPE_BK */
134 SO_TC_VI
, /* NET_SERVICE_TYPE_SIG */
135 SO_TC_VI
, /* NET_SERVICE_TYPE_VI */
136 SO_TC_VO
, /* NET_SERVICE_TYPE_VO */
137 SO_TC_RV
, /* NET_SERVICE_TYPE_RV */
138 SO_TC_AV
, /* NET_SERVICE_TYPE_AV */
139 SO_TC_OAM
, /* NET_SERVICE_TYPE_OAM */
140 SO_TC_RD
/* NET_SERVICE_TYPE_RD */
144 * DSCP mappings for QoS Fastlane as based on network service types
147 struct netsvctype_dscp_map fastlane_netsvctype_dscp_map
[_NET_SERVICE_TYPE_COUNT
] = {
148 { NET_SERVICE_TYPE_BE
, _DSCP_DF
},
149 { NET_SERVICE_TYPE_BK
, _DSCP_AF11
},
150 { NET_SERVICE_TYPE_SIG
, _DSCP_CS3
},
151 { NET_SERVICE_TYPE_VI
, _DSCP_AF41
},
152 { NET_SERVICE_TYPE_VO
, _DSCP_EF
},
153 { NET_SERVICE_TYPE_RV
, _DSCP_CS4
},
154 { NET_SERVICE_TYPE_AV
, _DSCP_AF31
},
155 { NET_SERVICE_TYPE_OAM
, _DSCP_CS2
},
156 { NET_SERVICE_TYPE_RD
, _DSCP_AF21
},
159 static struct net_qos_dscp_map default_net_qos_dscp_map
;
162 * The size is one more than the max because DSCP start at zero
164 #define DSCP_ARRAY_SIZE (_MAX_DSCP + 1)
167 * The DSCP to UP mapping (via mbuf service class) for WiFi follows is the mapping
168 * that implemented at the 802.11 driver level when the mbuf service class is
171 * This clashes with the recommended mapping documented by the IETF document
172 * draft-szigeti-tsvwg-ieee-802-11e-01.txt but we keep the mapping to maintain
173 * binary compatibility. Applications should use the network service type socket
174 * option instead to select L2 QoS marking instead of IP_TOS or IPV6_TCLASS.
176 static const struct dcsp_msc_map default_dscp_to_wifi_ac_map
[] = {
177 { _DSCP_DF
, MBUF_SC_BE
}, /* RFC 2474 Standard */
178 { 1, MBUF_SC_BE
}, /* */
179 { 2, MBUF_SC_BE
}, /* */
180 { 3, MBUF_SC_BE
}, /* */
181 { 4, MBUF_SC_BE
}, /* */
182 { 5, MBUF_SC_BE
}, /* */
183 { 6, MBUF_SC_BE
}, /* */
184 { 7, MBUF_SC_BE
}, /* */
186 { _DSCP_CS1
, MBUF_SC_BK
}, /* RFC 3662 Low-Priority Data */
187 { 9, MBUF_SC_BK
}, /* */
188 { _DSCP_AF11
, MBUF_SC_BK
}, /* RFC 2597 High-Throughput Data */
189 { 11, MBUF_SC_BK
}, /* */
190 { _DSCP_AF12
, MBUF_SC_BK
}, /* RFC 2597 High-Throughput Data */
191 { 13, MBUF_SC_BK
}, /* */
192 { _DSCP_AF13
, MBUF_SC_BK
}, /* RFC 2597 High-Throughput Data */
193 { 15, MBUF_SC_BK
}, /* */
195 { _DSCP_CS2
, MBUF_SC_BK
}, /* RFC 4594 OAM */
196 { 17, MBUF_SC_BK
}, /* */
197 { _DSCP_AF21
, MBUF_SC_BK
}, /* RFC 2597 Low-Latency Data */
198 { 19, MBUF_SC_BK
}, /* */
199 { _DSCP_AF22
, MBUF_SC_BK
}, /* RFC 2597 Low-Latency Data */
200 { 21, MBUF_SC_BK
}, /* */
201 { _DSCP_AF23
, MBUF_SC_BK
}, /* RFC 2597 Low-Latency Data */
202 { 23, MBUF_SC_BK
}, /* */
204 { _DSCP_CS3
, MBUF_SC_BE
}, /* RFC 2474 Broadcast Video */
205 { 25, MBUF_SC_BE
}, /* */
206 { _DSCP_AF31
, MBUF_SC_BE
}, /* RFC 2597 Multimedia Streaming */
207 { 27, MBUF_SC_BE
}, /* */
208 { _DSCP_AF32
, MBUF_SC_BE
}, /* RFC 2597 Multimedia Streaming */
209 { 29, MBUF_SC_BE
}, /* */
210 { _DSCP_AF33
, MBUF_SC_BE
}, /* RFC 2597 Multimedia Streaming */
211 { 31, MBUF_SC_BE
}, /* */
213 { _DSCP_CS4
, MBUF_SC_VI
}, /* RFC 2474 Real-Time Interactive */
214 { 33, MBUF_SC_VI
}, /* */
215 { _DSCP_AF41
, MBUF_SC_VI
}, /* RFC 2597 Multimedia Conferencing */
216 { 35, MBUF_SC_VI
}, /* */
217 { _DSCP_AF42
, MBUF_SC_VI
}, /* RFC 2597 Multimedia Conferencing */
218 { 37, MBUF_SC_VI
}, /* */
219 { _DSCP_AF43
, MBUF_SC_VI
}, /* RFC 2597 Multimedia Conferencing */
220 { 39, MBUF_SC_VI
}, /* */
222 { _DSCP_CS5
, MBUF_SC_VI
}, /* RFC 2474 Signaling */
223 { 41, MBUF_SC_VI
}, /* */
224 { 42, MBUF_SC_VI
}, /* */
225 { 43, MBUF_SC_VI
}, /* */
226 { _DSCP_VA
, MBUF_SC_VI
}, /* RFC 5865 VOICE-ADMIT */
227 { 45, MBUF_SC_VI
}, /* */
228 { _DSCP_EF
, MBUF_SC_VI
}, /* RFC 3246 Telephony */
229 { 47, MBUF_SC_VI
}, /* */
231 { _DSCP_CS6
, MBUF_SC_VO
}, /* Wi-Fi WMM Certification: Chariot */
232 { 49, MBUF_SC_VO
}, /* */
233 { 50, MBUF_SC_VO
}, /* */
234 { 51, MBUF_SC_VO
}, /* */
235 { 52, MBUF_SC_VO
}, /* Wi-Fi WMM Certification: Sigma */
236 { 53, MBUF_SC_VO
}, /* */
237 { 54, MBUF_SC_VO
}, /* */
238 { 55, MBUF_SC_VO
}, /* */
240 { _DSCP_CS7
, MBUF_SC_VO
}, /* Wi-Fi WMM Certification: Chariot */
241 { 57, MBUF_SC_VO
}, /* */
242 { 58, MBUF_SC_VO
}, /* */
243 { 59, MBUF_SC_VO
}, /* */
244 { 60, MBUF_SC_VO
}, /* */
245 { 61, MBUF_SC_VO
}, /* */
246 { 62, MBUF_SC_VO
}, /* */
247 { 63, MBUF_SC_VO
}, /* */
249 { 255, MBUF_SC_UNSPEC
} /* invalid DSCP to mark last entry */
252 mbuf_svc_class_t wifi_dscp_to_msc_array
[DSCP_ARRAY_SIZE
];
255 * If there is no foreground activity on the interface for bg_switch_time
256 * seconds, the background connections can switch to foreground TCP
257 * congestion control.
259 #define TCP_BG_SWITCH_TIME 2 /* seconds */
261 #if (DEVELOPMENT || DEBUG)
263 static int tfp_count
= 0;
265 static TAILQ_HEAD(, tclass_for_proc
) tfp_head
=
266 TAILQ_HEAD_INITIALIZER(tfp_head
);
268 struct tclass_for_proc
{
269 TAILQ_ENTRY(tclass_for_proc
) tfp_link
;
272 char tfp_pname
[(2 * MAXCOMLEN
) + 1];
273 u_int32_t tfp_qos_mode
;
276 static int get_pid_tclass(struct so_tcdbg
*);
277 static int get_pname_tclass(struct so_tcdbg
*);
278 static int set_pid_tclass(struct so_tcdbg
*);
279 static int set_pname_tclass(struct so_tcdbg
*);
280 static int flush_pid_tclass(struct so_tcdbg
*);
281 static int purge_tclass_for_proc(void);
282 static int flush_tclass_for_proc(void);
283 static void set_tclass_for_curr_proc(struct socket
*);
286 * Must be called with tclass_lock held
288 static struct tclass_for_proc
*
289 find_tfp_by_pid(pid_t pid
)
291 struct tclass_for_proc
*tfp
;
293 TAILQ_FOREACH(tfp
, &tfp_head
, tfp_link
) {
294 if (tfp
->tfp_pid
== pid
) {
302 * Must be called with tclass_lock held
304 static struct tclass_for_proc
*
305 find_tfp_by_pname(const char *pname
)
307 struct tclass_for_proc
*tfp
;
309 TAILQ_FOREACH(tfp
, &tfp_head
, tfp_link
) {
310 if (strncmp(pname
, tfp
->tfp_pname
,
311 sizeof(tfp
->tfp_pname
)) == 0) {
318 __private_extern__
void
319 set_tclass_for_curr_proc(struct socket
*so
)
321 struct tclass_for_proc
*tfp
= NULL
;
322 proc_t p
= current_proc(); /* Not ref counted */
323 pid_t pid
= proc_pid(p
);
324 char *pname
= proc_best_name(p
);
326 lck_mtx_lock(tclass_lock
);
328 TAILQ_FOREACH(tfp
, &tfp_head
, tfp_link
) {
329 if ((tfp
->tfp_pid
== pid
) || (tfp
->tfp_pid
== -1 &&
330 strncmp(pname
, tfp
->tfp_pname
,
331 sizeof(tfp
->tfp_pname
)) == 0)) {
332 if (tfp
->tfp_class
!= SO_TC_UNSPEC
) {
333 so
->so_traffic_class
= tfp
->tfp_class
;
336 if (tfp
->tfp_qos_mode
== QOS_MODE_MARKING_POLICY_ENABLE
) {
337 so
->so_flags1
|= SOF1_QOSMARKING_ALLOWED
;
338 } else if (tfp
->tfp_qos_mode
== QOS_MODE_MARKING_POLICY_DISABLE
) {
339 so
->so_flags1
&= ~SOF1_QOSMARKING_ALLOWED
;
345 lck_mtx_unlock(tclass_lock
);
349 * Purge entries with PIDs of exited processes
352 purge_tclass_for_proc(void)
355 struct tclass_for_proc
*tfp
, *tvar
;
357 lck_mtx_lock(tclass_lock
);
359 TAILQ_FOREACH_SAFE(tfp
, &tfp_head
, tfp_link
, tvar
) {
362 if (tfp
->tfp_pid
== -1) {
365 if ((p
= proc_find(tfp
->tfp_pid
)) == NULL
) {
367 TAILQ_REMOVE(&tfp_head
, tfp
, tfp_link
);
375 lck_mtx_unlock(tclass_lock
);
382 * Must be called with tclass_lock held
385 free_tclass_for_proc(struct tclass_for_proc
*tfp
)
391 TAILQ_REMOVE(&tfp_head
, tfp
, tfp_link
);
399 flush_tclass_for_proc(void)
402 struct tclass_for_proc
*tfp
, *tvar
;
404 lck_mtx_lock(tclass_lock
);
406 TAILQ_FOREACH_SAFE(tfp
, &tfp_head
, tfp_link
, tvar
) {
407 free_tclass_for_proc(tfp
);
410 lck_mtx_unlock(tclass_lock
);
416 * Must be called with tclass_lock held
418 static struct tclass_for_proc
*
419 alloc_tclass_for_proc(pid_t pid
, const char *pname
)
421 struct tclass_for_proc
*tfp
;
423 if (pid
== -1 && pname
== NULL
) {
427 tfp
= _MALLOC(sizeof(struct tclass_for_proc
), M_TEMP
, M_NOWAIT
| M_ZERO
);
434 * Add per pid entries before per proc name so we can find
435 * a specific instance of a process before the general name base entry.
438 TAILQ_INSERT_HEAD(&tfp_head
, tfp
, tfp_link
);
440 strlcpy(tfp
->tfp_pname
, pname
, sizeof(tfp
->tfp_pname
));
441 TAILQ_INSERT_TAIL(&tfp_head
, tfp
, tfp_link
);
450 * SO_TC_UNSPEC for tclass means to remove the entry
453 set_pid_tclass(struct so_tcdbg
*so_tcdbg
)
457 struct filedesc
*fdp
;
459 struct tclass_for_proc
*tfp
;
461 pid_t pid
= so_tcdbg
->so_tcdbg_pid
;
462 int tclass
= so_tcdbg
->so_tcdbg_tclass
;
463 int netsvctype
= so_tcdbg
->so_tcdbg_netsvctype
;
467 printf("%s proc_find(%d) failed\n", __func__
, pid
);
472 lck_mtx_lock(tclass_lock
);
474 tfp
= find_tfp_by_pid(pid
);
476 tfp
= alloc_tclass_for_proc(pid
, NULL
);
478 lck_mtx_unlock(tclass_lock
);
483 tfp
->tfp_class
= tclass
;
484 tfp
->tfp_qos_mode
= so_tcdbg
->so_tcbbg_qos_mode
;
486 lck_mtx_unlock(tclass_lock
);
492 for (i
= 0; i
< fdp
->fd_nfiles
; i
++) {
495 fp
= fdp
->fd_ofiles
[i
];
497 (fdp
->fd_ofileflags
[i
] & UF_RESERVED
) != 0 ||
498 FILEGLOB_DTYPE(fp
->f_fglob
) != DTYPE_SOCKET
) {
502 so
= (struct socket
*)fp
->f_fglob
->fg_data
;
503 if (SOCK_DOM(so
) != PF_INET
&& SOCK_DOM(so
) != PF_INET6
) {
508 if (tfp
->tfp_qos_mode
== QOS_MODE_MARKING_POLICY_ENABLE
) {
509 so
->so_flags1
|= SOF1_QOSMARKING_ALLOWED
;
510 } else if (tfp
->tfp_qos_mode
== QOS_MODE_MARKING_POLICY_DISABLE
) {
511 so
->so_flags1
&= ~SOF1_QOSMARKING_ALLOWED
;
513 socket_unlock(so
, 1);
515 if (netsvctype
!= _NET_SERVICE_TYPE_UNSPEC
) {
516 error
= sock_setsockopt(so
, SOL_SOCKET
,
517 SO_NET_SERVICE_TYPE
, &netsvctype
, sizeof(int));
519 if (tclass
!= SO_TC_UNSPEC
) {
520 error
= sock_setsockopt(so
, SOL_SOCKET
,
521 SO_TRAFFIC_CLASS
, &tclass
, sizeof(int));
538 set_pname_tclass(struct so_tcdbg
*so_tcdbg
)
541 struct tclass_for_proc
*tfp
;
543 lck_mtx_lock(tclass_lock
);
545 tfp
= find_tfp_by_pname(so_tcdbg
->so_tcdbg_pname
);
547 tfp
= alloc_tclass_for_proc(-1, so_tcdbg
->so_tcdbg_pname
);
549 lck_mtx_unlock(tclass_lock
);
554 tfp
->tfp_class
= so_tcdbg
->so_tcdbg_tclass
;
555 tfp
->tfp_qos_mode
= so_tcdbg
->so_tcbbg_qos_mode
;
557 lck_mtx_unlock(tclass_lock
);
566 flush_pid_tclass(struct so_tcdbg
*so_tcdbg
)
568 pid_t pid
= so_tcdbg
->so_tcdbg_pid
;
569 int tclass
= so_tcdbg
->so_tcdbg_tclass
;
570 struct filedesc
*fdp
;
576 if (p
== PROC_NULL
) {
577 printf("%s proc_find(%d) failed\n", __func__
, pid
);
583 for (i
= 0; i
< fdp
->fd_nfiles
; i
++) {
587 fp
= fdp
->fd_ofiles
[i
];
589 (fdp
->fd_ofileflags
[i
] & UF_RESERVED
) != 0 ||
590 FILEGLOB_DTYPE(fp
->f_fglob
) != DTYPE_SOCKET
) {
594 so
= (struct socket
*)fp
->f_fglob
->fg_data
;
595 error
= sock_setsockopt(so
, SOL_SOCKET
, SO_FLUSH
, &tclass
,
598 printf("%s: setsockopt(SO_FLUSH) (so=0x%llx, fd=%d, "
599 "tclass=%d) failed %d\n", __func__
,
600 (uint64_t)VM_KERNEL_ADDRPERM(so
), i
, tclass
,
609 if (p
!= PROC_NULL
) {
617 get_pid_tclass(struct so_tcdbg
*so_tcdbg
)
621 struct tclass_for_proc
*tfp
;
622 pid_t pid
= so_tcdbg
->so_tcdbg_pid
;
624 so_tcdbg
->so_tcdbg_tclass
= SO_TC_UNSPEC
; /* Means not set */
628 printf("%s proc_find(%d) failed\n", __func__
, pid
);
633 lck_mtx_lock(tclass_lock
);
635 tfp
= find_tfp_by_pid(pid
);
637 so_tcdbg
->so_tcdbg_tclass
= tfp
->tfp_class
;
638 so_tcdbg
->so_tcbbg_qos_mode
= tfp
->tfp_qos_mode
;
641 lck_mtx_unlock(tclass_lock
);
651 get_pname_tclass(struct so_tcdbg
*so_tcdbg
)
654 struct tclass_for_proc
*tfp
;
656 so_tcdbg
->so_tcdbg_tclass
= SO_TC_UNSPEC
; /* Means not set */
659 lck_mtx_lock(tclass_lock
);
661 tfp
= find_tfp_by_pname(so_tcdbg
->so_tcdbg_pname
);
663 so_tcdbg
->so_tcdbg_tclass
= tfp
->tfp_class
;
664 so_tcdbg
->so_tcbbg_qos_mode
= tfp
->tfp_qos_mode
;
667 lck_mtx_unlock(tclass_lock
);
673 delete_tclass_for_pid_pname(struct so_tcdbg
*so_tcdbg
)
676 pid_t pid
= so_tcdbg
->so_tcdbg_pid
;
677 struct tclass_for_proc
*tfp
= NULL
;
679 lck_mtx_lock(tclass_lock
);
682 tfp
= find_tfp_by_pid(pid
);
684 tfp
= find_tfp_by_pname(so_tcdbg
->so_tcdbg_pname
);
688 free_tclass_for_proc(tfp
);
692 lck_mtx_unlock(tclass_lock
);
698 * Setting options requires privileges
700 __private_extern__
int
701 so_set_tcdbg(struct socket
*so
, struct so_tcdbg
*so_tcdbg
)
705 if ((so
->so_state
& SS_PRIV
) == 0) {
709 socket_unlock(so
, 0);
711 switch (so_tcdbg
->so_tcdbg_cmd
) {
713 error
= set_pid_tclass(so_tcdbg
);
717 error
= set_pname_tclass(so_tcdbg
);
721 error
= purge_tclass_for_proc();
725 error
= flush_tclass_for_proc();
728 case SO_TCDBG_DELETE
:
729 error
= delete_tclass_for_pid_pname(so_tcdbg
);
732 case SO_TCDBG_TCFLUSH_PID
:
733 error
= flush_pid_tclass(so_tcdbg
);
747 * Not required to be privileged to get
749 __private_extern__
int
750 sogetopt_tcdbg(struct socket
*so
, struct sockopt
*sopt
)
753 struct so_tcdbg so_tcdbg
;
755 size_t len
= sopt
->sopt_valsize
;
757 error
= sooptcopyin(sopt
, &so_tcdbg
, sizeof(struct so_tcdbg
),
758 sizeof(struct so_tcdbg
));
763 sopt
->sopt_valsize
= len
;
765 socket_unlock(so
, 0);
767 switch (so_tcdbg
.so_tcdbg_cmd
) {
769 error
= get_pid_tclass(&so_tcdbg
);
773 error
= get_pname_tclass(&so_tcdbg
);
777 lck_mtx_lock(tclass_lock
);
778 so_tcdbg
.so_tcdbg_count
= tfp_count
;
779 lck_mtx_unlock(tclass_lock
);
782 case SO_TCDBG_LIST
: {
783 struct tclass_for_proc
*tfp
;
785 struct so_tcdbg
*ptr
;
787 lck_mtx_lock(tclass_lock
);
788 if ((alloc_count
= tfp_count
) == 0) {
789 lck_mtx_unlock(tclass_lock
);
793 len
= alloc_count
* sizeof(struct so_tcdbg
);
794 lck_mtx_unlock(tclass_lock
);
796 buf
= _MALLOC(len
, M_TEMP
, M_WAITOK
| M_ZERO
);
802 lck_mtx_lock(tclass_lock
);
804 ptr
= (struct so_tcdbg
*)buf
;
805 TAILQ_FOREACH(tfp
, &tfp_head
, tfp_link
) {
806 if (++n
> alloc_count
) {
809 if (tfp
->tfp_pid
!= -1) {
810 ptr
->so_tcdbg_cmd
= SO_TCDBG_PID
;
811 ptr
->so_tcdbg_pid
= tfp
->tfp_pid
;
813 ptr
->so_tcdbg_cmd
= SO_TCDBG_PNAME
;
814 ptr
->so_tcdbg_pid
= -1;
815 strlcpy(ptr
->so_tcdbg_pname
,
817 sizeof(ptr
->so_tcdbg_pname
));
819 ptr
->so_tcdbg_tclass
= tfp
->tfp_class
;
820 ptr
->so_tcbbg_qos_mode
= tfp
->tfp_qos_mode
;
824 lck_mtx_unlock(tclass_lock
);
837 error
= sooptcopyout(sopt
, &so_tcdbg
,
838 sizeof(struct so_tcdbg
));
840 error
= sooptcopyout(sopt
, buf
, len
);
847 #endif /* (DEVELOPMENT || DEBUG) */
850 so_get_netsvc_marking_level(struct socket
*so
)
852 int marking_level
= NETSVC_MRKNG_UNKNOWN
;
853 struct ifnet
*ifp
= NULL
;
855 switch (SOCK_DOM(so
)) {
857 struct inpcb
*inp
= sotoinpcb(so
);
860 ifp
= inp
->inp_last_outifp
;
865 struct in6pcb
*in6p
= sotoin6pcb(so
);
868 ifp
= in6p
->in6p_last_outifp
;
876 if ((ifp
->if_eflags
&
877 (IFEF_QOSMARKING_ENABLED
| IFEF_QOSMARKING_CAPABLE
)) ==
878 (IFEF_QOSMARKING_ENABLED
| IFEF_QOSMARKING_CAPABLE
)) {
879 if ((so
->so_flags1
& SOF1_QOSMARKING_ALLOWED
)) {
880 marking_level
= NETSVC_MRKNG_LVL_L3L2_ALL
;
882 marking_level
= NETSVC_MRKNG_LVL_L3L2_BK
;
885 marking_level
= NETSVC_MRKNG_LVL_L2
;
888 return marking_level
;
891 __private_extern__
int
892 so_set_traffic_class(struct socket
*so
, int optval
)
896 if (optval
< SO_TC_BE
|| optval
> SO_TC_CTL
) {
910 if (!SO_VALID_TC(optval
)) {
917 int oldval
= so
->so_traffic_class
;
919 VERIFY(SO_VALID_TC(optval
));
920 so
->so_traffic_class
= optval
;
922 if ((SOCK_DOM(so
) == PF_INET
||
923 SOCK_DOM(so
) == PF_INET6
) &&
924 SOCK_TYPE(so
) == SOCK_STREAM
) {
925 set_tcp_stream_priority(so
);
928 if ((SOCK_DOM(so
) == PF_INET
||
929 SOCK_DOM(so
) == PF_INET6
) &&
930 optval
!= oldval
&& (optval
== SO_TC_BK_SYS
||
931 oldval
== SO_TC_BK_SYS
)) {
933 * If the app switches from BK_SYS to something
934 * else, resume the socket if it was suspended.
936 if (oldval
== SO_TC_BK_SYS
) {
937 inp_reset_fc_state(so
->so_pcb
);
940 SOTHROTTLELOG("throttle[%d]: so 0x%llx "
941 "[%d,%d] opportunistic %s\n", so
->last_pid
,
942 (uint64_t)VM_KERNEL_ADDRPERM(so
),
943 SOCK_DOM(so
), SOCK_TYPE(so
),
944 (optval
== SO_TC_BK_SYS
) ? "ON" : "OFF");
951 __private_extern__
int
952 so_set_net_service_type(struct socket
*so
, int netsvctype
)
957 if (!IS_VALID_NET_SERVICE_TYPE(netsvctype
)) {
961 sotc
= sotc_by_netservicetype
[netsvctype
];
962 error
= so_set_traffic_class(so
, sotc
);
966 so
->so_netsvctype
= netsvctype
;
967 so
->so_flags1
|= SOF1_TC_NET_SERV_TYPE
;
972 __private_extern__
void
973 so_set_default_traffic_class(struct socket
*so
)
975 so
->so_traffic_class
= SO_TC_BE
;
977 if ((SOCK_DOM(so
) == PF_INET
|| SOCK_DOM(so
) == PF_INET6
)) {
978 if (net_qos_policy_restricted
== 0) {
979 so
->so_flags1
|= SOF1_QOSMARKING_ALLOWED
;
981 #if (DEVELOPMENT || DEBUG)
983 set_tclass_for_curr_proc(so
);
985 #endif /* (DEVELOPMENT || DEBUG) */
989 __private_extern__
int
990 so_set_opportunistic(struct socket
*so
, int optval
)
992 return so_set_traffic_class(so
, (optval
== 0) ?
993 SO_TC_BE
: SO_TC_BK_SYS
);
996 __private_extern__
int
997 so_get_opportunistic(struct socket
*so
)
999 return so
->so_traffic_class
== SO_TC_BK_SYS
;
1002 __private_extern__
int
1003 so_tc_from_control(struct mbuf
*control
, int *out_netsvctype
)
1006 int sotc
= SO_TC_UNSPEC
;
1008 *out_netsvctype
= _NET_SERVICE_TYPE_UNSPEC
;
1010 for (cm
= M_FIRST_CMSGHDR(control
);
1011 is_cmsg_valid(control
, cm
);
1012 cm
= M_NXT_CMSGHDR(control
, cm
)) {
1015 if (cm
->cmsg_level
!= SOL_SOCKET
||
1016 cm
->cmsg_len
!= CMSG_LEN(sizeof(int))) {
1019 val
= *(int *)(void *)CMSG_DATA(cm
);
1021 * The first valid option wins
1023 switch (cm
->cmsg_type
) {
1024 case SO_TRAFFIC_CLASS
:
1025 if (SO_VALID_TC(val
)) {
1029 } else if (val
< SO_TC_NET_SERVICE_OFFSET
) {
1033 * Handle the case SO_NET_SERVICE_TYPE values are
1034 * passed using SO_TRAFFIC_CLASS
1036 val
= val
- SO_TC_NET_SERVICE_OFFSET
;
1038 case SO_NET_SERVICE_TYPE
:
1039 if (!IS_VALID_NET_SERVICE_TYPE(val
)) {
1042 *out_netsvctype
= val
;
1043 sotc
= sotc_by_netservicetype
[val
];
1054 __private_extern__
void
1055 so_recv_data_stat(struct socket
*so
, struct mbuf
*m
, size_t off
)
1057 uint32_t mtc
= m_get_traffic_class(m
);
1059 if (mtc
>= SO_TC_STATS_MAX
) {
1063 so
->so_tc_stats
[mtc
].rxpackets
+= 1;
1064 so
->so_tc_stats
[mtc
].rxbytes
+=
1065 ((m
->m_flags
& M_PKTHDR
) ? m
->m_pkthdr
.len
: 0) + off
;
1068 __private_extern__
void
1069 so_inc_recv_data_stat(struct socket
*so
, size_t pkts
, size_t bytes
,
1072 if (mtc
>= SO_TC_STATS_MAX
) {
1076 so
->so_tc_stats
[mtc
].rxpackets
+= pkts
;
1077 so
->so_tc_stats
[mtc
].rxbytes
+= bytes
;
1081 so_throttle_best_effort(struct socket
*so
, struct ifnet
*ifp
)
1083 u_int32_t uptime
= net_uptime();
1084 return soissrcbesteffort(so
) &&
1085 net_io_policy_throttle_best_effort
== 1 &&
1086 ifp
->if_rt_sendts
> 0 &&
1087 (int)(uptime
- ifp
->if_rt_sendts
) <= TCP_BG_SWITCH_TIME
;
1090 __private_extern__
void
1091 set_tcp_stream_priority(struct socket
*so
)
1093 struct inpcb
*inp
= sotoinpcb(so
);
1094 struct tcpcb
*tp
= intotcpcb(inp
);
1095 struct ifnet
*outifp
;
1096 u_char old_cc
= tp
->tcp_cc_index
;
1097 int recvbg
= IS_TCP_RECV_BG(so
);
1098 bool is_local
= false, fg_active
= false;
1101 VERIFY((SOCK_CHECK_DOM(so
, PF_INET
) ||
1102 SOCK_CHECK_DOM(so
, PF_INET6
)) &&
1103 SOCK_CHECK_TYPE(so
, SOCK_STREAM
) &&
1104 SOCK_CHECK_PROTO(so
, IPPROTO_TCP
));
1106 /* Return if the socket is in a terminal state */
1107 if (inp
->inp_state
== INPCB_STATE_DEAD
) {
1111 outifp
= inp
->inp_last_outifp
;
1112 uptime
= net_uptime();
1115 * If the socket was marked as a background socket or if the
1116 * traffic class is set to background with traffic class socket
1117 * option then make both send and recv side of the stream to be
1118 * background. The variable sotcdb which can be set with sysctl
1119 * is used to disable these settings for testing.
1121 if (outifp
== NULL
|| (outifp
->if_flags
& IFF_LOOPBACK
)) {
1125 /* Check if there has been recent foreground activity */
1126 if (outifp
!= NULL
) {
1128 * If the traffic source is background, check if
1129 * if it can be switched to foreground. This can
1130 * happen when there is no indication of foreground
1133 if (soissrcbackground(so
) && outifp
->if_fg_sendts
> 0 &&
1134 (int)(uptime
- outifp
->if_fg_sendts
) <= TCP_BG_SWITCH_TIME
) {
1139 * The traffic source is best-effort -- check if
1140 * the policy to throttle best effort is enabled
1141 * and there was realtime activity on this
1142 * interface recently. If this is true, enable
1143 * algorithms that respond to increased latency
1144 * on best-effort traffic.
1146 if (so_throttle_best_effort(so
, outifp
)) {
1152 * System initiated background traffic like cloud uploads should
1153 * always use background delay sensitive algorithms. This will
1154 * make the stream more responsive to other streams on the user's
1155 * network and it will minimize latency induced.
1157 if (fg_active
|| IS_SO_TC_BACKGROUNDSYSTEM(so
->so_traffic_class
)) {
1159 * If the interface that the connection is using is
1160 * loopback, do not use background congestion
1161 * control algorithm.
1163 * If there has been recent foreground activity or if
1164 * there was an indication that a foreground application
1165 * is going to use networking (net_io_policy_throttled),
1166 * switch the backgroung streams to use background
1167 * congestion control algorithm. Otherwise, even background
1168 * flows can move into foreground.
1170 if ((sotcdb
& SOTCDB_NO_SENDTCPBG
) != 0 || is_local
||
1171 !IS_SO_TC_BACKGROUNDSYSTEM(so
->so_traffic_class
)) {
1172 if (old_cc
== TCP_CC_ALGO_BACKGROUND_INDEX
) {
1173 tcp_set_foreground_cc(so
);
1176 if (old_cc
!= TCP_CC_ALGO_BACKGROUND_INDEX
) {
1177 tcp_set_background_cc(so
);
1181 /* Set receive side background flags */
1182 if ((sotcdb
& SOTCDB_NO_RECVTCPBG
) != 0 || is_local
||
1183 !IS_SO_TC_BACKGROUNDSYSTEM(so
->so_traffic_class
)) {
1184 tcp_clear_recv_bg(so
);
1186 tcp_set_recv_bg(so
);
1189 tcp_clear_recv_bg(so
);
1190 if (old_cc
== TCP_CC_ALGO_BACKGROUND_INDEX
) {
1191 tcp_set_foreground_cc(so
);
1195 if (old_cc
!= tp
->tcp_cc_index
|| recvbg
!= IS_TCP_RECV_BG(so
)) {
1196 SOTHROTTLELOG("throttle[%d]: so 0x%llx [%d,%d] TCP %s send; "
1197 "%s recv\n", so
->last_pid
,
1198 (uint64_t)VM_KERNEL_ADDRPERM(so
),
1199 SOCK_DOM(so
), SOCK_TYPE(so
),
1200 (tp
->tcp_cc_index
== TCP_CC_ALGO_BACKGROUND_INDEX
) ?
1201 "background" : "foreground",
1202 IS_TCP_RECV_BG(so
) ? "background" : "foreground");
1207 * Set traffic class to an IPv4 or IPv6 packet
1209 * - set the DSCP code following the WMM mapping
1211 __private_extern__
void
1212 set_packet_service_class(struct mbuf
*m
, struct socket
*so
,
1213 int sotc
, u_int32_t flags
)
1215 mbuf_svc_class_t msc
= MBUF_SC_BE
; /* Best effort by default */
1216 struct inpcb
*inp
= sotoinpcb(so
); /* in6pcb and inpcb are the same */
1218 if (!(m
->m_flags
& M_PKTHDR
)) {
1223 * Here is the precedence:
1224 * 1) TRAFFIC_MGT_SO_BACKGROUND trumps all
1225 * 2) Traffic class passed via ancillary data to sendmsdg(2)
1226 * 3) Traffic class socket option last
1228 if (sotc
!= SO_TC_UNSPEC
) {
1229 VERIFY(SO_VALID_TC(sotc
));
1230 msc
= so_tc2msc(sotc
);
1231 /* Assert because tc must have been valid */
1232 VERIFY(MBUF_VALID_SC(msc
));
1236 * If TRAFFIC_MGT_SO_BACKGROUND is set or policy to throttle
1237 * best effort is set, depress the priority.
1239 if (!IS_MBUF_SC_BACKGROUND(msc
) && soisthrottled(so
)) {
1243 if (IS_MBUF_SC_BESTEFFORT(msc
) && inp
->inp_last_outifp
!= NULL
&&
1244 so_throttle_best_effort(so
, inp
->inp_last_outifp
)) {
1248 if (soissrcbackground(so
)) {
1249 m
->m_pkthdr
.pkt_flags
|= PKTF_SO_BACKGROUND
;
1252 if (soissrcrealtime(so
) || IS_MBUF_SC_REALTIME(msc
)) {
1253 m
->m_pkthdr
.pkt_flags
|= PKTF_SO_REALTIME
;
1256 * Set the traffic class in the mbuf packet header svc field
1258 if (sotcdb
& SOTCDB_NO_MTC
) {
1263 * Elevate service class if the packet is a pure TCP ACK.
1264 * We can do this only when the flow is not a background
1265 * flow and the outgoing interface supports
1266 * transmit-start model.
1268 if (!IS_MBUF_SC_BACKGROUND(msc
) &&
1269 (flags
& (PKT_SCF_TCP_ACK
| PKT_SCF_TCP_SYN
)) != 0) {
1273 (void) m_set_service_class(m
, msc
);
1276 * Set the privileged traffic auxiliary flag if applicable,
1279 if (!(sotcdb
& SOTCDB_NO_PRIVILEGED
) && soisprivilegedtraffic(so
) &&
1280 msc
!= MBUF_SC_UNSPEC
) {
1281 m
->m_pkthdr
.pkt_flags
|= PKTF_PRIO_PRIVILEGED
;
1283 m
->m_pkthdr
.pkt_flags
&= ~PKTF_PRIO_PRIVILEGED
;
1288 * For TCP with background traffic class switch CC algo based on sysctl
1290 if (so
->so_type
== SOCK_STREAM
) {
1291 set_tcp_stream_priority(so
);
1294 so_tc_update_stats(m
, so
, msc
);
1297 __private_extern__
void
1298 so_tc_update_stats(struct mbuf
*m
, struct socket
*so
, mbuf_svc_class_t msc
)
1300 mbuf_traffic_class_t mtc
;
1303 * Assume socket and mbuf traffic class values are the same
1304 * Also assume the socket lock is held. Note that the stats
1305 * at the socket layer are reduced down to the legacy traffic
1306 * classes; we could/should potentially expand so_tc_stats[].
1308 mtc
= MBUF_SC2TC(msc
);
1309 VERIFY(mtc
< SO_TC_STATS_MAX
);
1310 so
->so_tc_stats
[mtc
].txpackets
+= 1;
1311 so
->so_tc_stats
[mtc
].txbytes
+= m
->m_pkthdr
.len
;
1314 __private_extern__
void
1315 socket_tclass_init(void)
1317 _CASSERT(_SO_TC_MAX
== SO_TC_STATS_MAX
);
1319 tclass_lck_grp_attr
= lck_grp_attr_alloc_init();
1320 tclass_lck_grp
= lck_grp_alloc_init("tclass", tclass_lck_grp_attr
);
1321 tclass_lck_attr
= lck_attr_alloc_init();
1322 lck_mtx_init(tclass_lock
, tclass_lck_grp
, tclass_lck_attr
);
1325 __private_extern__ mbuf_svc_class_t
1328 mbuf_svc_class_t msc
;
1332 msc
= MBUF_SC_BK_SYS
;
1357 case SO_TC_NETSVC_SIG
:
1369 msc
= MBUF_SC_UNSPEC
;
1376 __private_extern__
int
1377 so_svc2tc(mbuf_svc_class_t svc
)
1380 case MBUF_SC_BK_SYS
:
1381 return SO_TC_BK_SYS
;
1397 return SO_TC_NETSVC_SIG
;
1402 case MBUF_SC_UNSPEC
:
1409 * LRO is turned on for AV streaming class.
1412 so_set_lro(struct socket
*so
, int optval
)
1414 if (optval
== SO_TC_AV
) {
1415 so
->so_flags
|= SOF_USELRO
;
1417 if (so
->so_flags
& SOF_USELRO
) {
1418 /* transition to non LRO class */
1419 so
->so_flags
&= ~SOF_USELRO
;
1420 struct inpcb
*inp
= sotoinpcb(so
);
1421 struct tcpcb
*tp
= NULL
;
1423 tp
= intotcpcb(inp
);
1424 if (tp
&& (tp
->t_flagsext
& TF_LRO_OFFLOADED
)) {
1425 tcp_lro_remove_state(inp
->inp_laddr
,
1429 tp
->t_flagsext
&= ~TF_LRO_OFFLOADED
;
1437 sotc_index(int sotc
)
1441 return SOTCIX_BK_SYS
;
1471 * Unknown traffic class value
1477 * Pass NULL ifp for default map
1480 set_netsvctype_dscp_map(size_t in_count
,
1481 const struct netsvctype_dscp_map
*netsvctype_dscp_map
)
1484 struct net_qos_dscp_map
*net_qos_dscp_map
= NULL
;
1488 * Do not accept more that max number of distinct DSCPs
1490 if (in_count
> _MAX_DSCP
|| netsvctype_dscp_map
== NULL
) {
1495 * Validate input parameters
1497 for (i
= 0; i
< in_count
; i
++) {
1498 if (!IS_VALID_NET_SERVICE_TYPE(netsvctype_dscp_map
[i
].netsvctype
)) {
1501 if (netsvctype_dscp_map
[i
].dscp
> _MAX_DSCP
) {
1506 net_qos_dscp_map
= &default_net_qos_dscp_map
;
1508 for (i
= 0; i
< in_count
; i
++) {
1509 netsvctype
= netsvctype_dscp_map
[i
].netsvctype
;
1511 net_qos_dscp_map
->netsvctype_to_dscp
[netsvctype
] =
1512 netsvctype_dscp_map
[i
].dscp
;
1514 for (netsvctype
= 0; netsvctype
< _NET_SERVICE_TYPE_COUNT
; netsvctype
++) {
1515 switch (netsvctype
) {
1516 case NET_SERVICE_TYPE_BE
:
1517 case NET_SERVICE_TYPE_BK
:
1518 case NET_SERVICE_TYPE_VI
:
1519 case NET_SERVICE_TYPE_VO
:
1520 case NET_SERVICE_TYPE_RV
:
1521 case NET_SERVICE_TYPE_AV
:
1522 case NET_SERVICE_TYPE_OAM
:
1523 case NET_SERVICE_TYPE_RD
: {
1526 sotcix
= sotc_index(sotc_by_netservicetype
[netsvctype
]);
1527 if (sotcix
!= SIZE_T_MAX
) {
1528 net_qos_dscp_map
->sotc_to_dscp
[sotcix
] =
1529 netsvctype_dscp_map
[netsvctype
].dscp
;
1533 case NET_SERVICE_TYPE_SIG
:
1534 /* Signaling does not have its own traffic class */
1537 /* We should not be here */
1541 /* Network control socket traffic class is always best effort */
1542 net_qos_dscp_map
->sotc_to_dscp
[SOTCIX_CTL
] = _DSCP_DF
;
1544 /* Backround socket traffic class DSCP same as backround system */
1545 net_qos_dscp_map
->sotc_to_dscp
[SOTCIX_BK
] =
1546 net_qos_dscp_map
->sotc_to_dscp
[SOTCIX_BK_SYS
];
1552 * out_count is an input/ouput parameter
1555 get_netsvctype_dscp_map(size_t *out_count
,
1556 struct netsvctype_dscp_map
*netsvctype_dscp_map
)
1559 struct net_qos_dscp_map
*net_qos_dscp_map
= NULL
;
1562 * Do not accept more that max number of distinct DSCPs
1564 if (out_count
== NULL
|| netsvctype_dscp_map
== NULL
) {
1567 if (*out_count
> _MAX_DSCP
) {
1571 net_qos_dscp_map
= &default_net_qos_dscp_map
;
1573 for (i
= 0; i
< MIN(_NET_SERVICE_TYPE_COUNT
, *out_count
); i
++) {
1574 netsvctype_dscp_map
[i
].netsvctype
= i
;
1575 netsvctype_dscp_map
[i
].dscp
= net_qos_dscp_map
->netsvctype_to_dscp
[i
];
1588 * By default use the Fastlane DSCP mappngs
1590 error
= set_netsvctype_dscp_map(_NET_SERVICE_TYPE_COUNT
,
1591 fastlane_netsvctype_dscp_map
);
1595 * No DSCP mapping for network control
1597 default_net_qos_dscp_map
.sotc_to_dscp
[SOTCIX_CTL
] = _DSCP_DF
;
1599 set_dscp_to_wifi_ac_map(default_dscp_to_wifi_ac_map
, 1);
1603 sysctl_default_netsvctype_to_dscp_map SYSCTL_HANDLER_ARGS
1605 #pragma unused(oidp, arg1, arg2)
1607 const size_t max_netsvctype_to_dscp_map_len
=
1608 _NET_SERVICE_TYPE_COUNT
* sizeof(struct netsvctype_dscp_map
);
1610 struct netsvctype_dscp_map netsvctype_dscp_map
[_NET_SERVICE_TYPE_COUNT
] = {};
1613 if (req
->oldptr
== USER_ADDR_NULL
) {
1615 _NET_SERVICE_TYPE_COUNT
* sizeof(struct netsvctype_dscp_map
);
1616 } else if (req
->oldlen
> 0) {
1617 count
= _NET_SERVICE_TYPE_COUNT
;
1618 error
= get_netsvctype_dscp_map(&count
, netsvctype_dscp_map
);
1622 len
= count
* sizeof(struct netsvctype_dscp_map
);
1623 error
= SYSCTL_OUT(req
, netsvctype_dscp_map
,
1624 MIN(len
, req
->oldlen
));
1630 if (req
->newptr
== USER_ADDR_NULL
) {
1634 error
= proc_suser(current_proc());
1640 * Check input length
1642 if (req
->newlen
> max_netsvctype_to_dscp_map_len
) {
1647 * Cap the number of entries to copy from input buffer
1649 error
= SYSCTL_IN(req
, netsvctype_dscp_map
, req
->newlen
);
1654 count
= req
->newlen
/ sizeof(struct netsvctype_dscp_map
);
1655 error
= set_netsvctype_dscp_map(count
, netsvctype_dscp_map
);
1660 __private_extern__ errno_t
1661 set_packet_qos(struct mbuf
*m
, struct ifnet
*ifp
, boolean_t qos_allowed
,
1662 int sotc
, int netsvctype
, u_int8_t
*dscp_inout
)
1664 if (ifp
== NULL
|| dscp_inout
== NULL
) {
1668 if ((ifp
->if_eflags
&
1669 (IFEF_QOSMARKING_ENABLED
| IFEF_QOSMARKING_CAPABLE
)) ==
1670 (IFEF_QOSMARKING_ENABLED
| IFEF_QOSMARKING_CAPABLE
)) {
1674 * When on a Fastlane network, IP_TOS/IPV6_TCLASS are no-ops
1679 * For DSCP use the network service type is specified, otherwise
1680 * use the socket traffic class
1682 * When not whitelisted by the policy, set DSCP only for best
1683 * effort and background, and set the mbuf service class to
1684 * best effort as well so the packet will be queued and
1685 * scheduled at a lower priority.
1686 * We still want to prioritize control traffic on the interface
1687 * so we do not change the mbuf service class for SO_TC_CTL
1689 if (IS_VALID_NET_SERVICE_TYPE(netsvctype
) &&
1690 netsvctype
!= NET_SERVICE_TYPE_BE
) {
1691 dscp
= default_net_qos_dscp_map
.netsvctype_to_dscp
[netsvctype
];
1693 if (qos_allowed
== FALSE
&&
1694 netsvctype
!= NET_SERVICE_TYPE_BE
&&
1695 netsvctype
!= NET_SERVICE_TYPE_BK
) {
1697 if (sotc
!= SO_TC_CTL
) {
1698 m_set_service_class(m
, MBUF_SC_BE
);
1701 } else if (sotc
!= SO_TC_UNSPEC
) {
1702 size_t sotcix
= sotc_index(sotc
);
1703 if (sotcix
!= SIZE_T_MAX
) {
1704 dscp
= default_net_qos_dscp_map
.sotc_to_dscp
[sotcix
];
1706 if (qos_allowed
== FALSE
&& sotc
!= SO_TC_BE
&&
1707 sotc
!= SO_TC_BK
&& sotc
!= SO_TC_BK_SYS
&&
1708 sotc
!= SO_TC_CTL
) {
1710 if (sotc
!= SO_TC_CTL
) {
1711 m_set_service_class(m
, MBUF_SC_BE
);
1716 if (net_qos_verbose
!= 0) {
1717 printf("%s qos_allowed %d sotc %u netsvctype %u dscp %u\n",
1718 __func__
, qos_allowed
, sotc
, netsvctype
, dscp
);
1721 if (*dscp_inout
!= dscp
) {
1724 } else if (*dscp_inout
!= _DSCP_DF
&& IFNET_IS_WIFI_INFRA(ifp
)) {
1725 mbuf_svc_class_t msc
= m_get_service_class(m
);
1728 * For WiFi infra, when the mbuf service class is best effort
1729 * and the DSCP is not default, set the service class based
1732 if (msc
== MBUF_SC_BE
) {
1733 msc
= wifi_dscp_to_msc_array
[*dscp_inout
];
1735 if (msc
!= MBUF_SC_BE
) {
1736 m_set_service_class(m
, msc
);
1738 if (net_qos_verbose
!= 0) {
1739 printf("%s set msc %u for dscp %u\n",
1740 __func__
, msc
, *dscp_inout
);
1750 set_dscp_to_wifi_ac_map(const struct dcsp_msc_map
*map
, int clear
)
1755 bzero(wifi_dscp_to_msc_array
, sizeof(wifi_dscp_to_msc_array
));
1758 for (i
= 0; i
< DSCP_ARRAY_SIZE
; i
++) {
1759 const struct dcsp_msc_map
*elem
= map
+ i
;
1761 if (elem
->dscp
> _MAX_DSCP
|| elem
->msc
== MBUF_SC_UNSPEC
) {
1764 switch (elem
->msc
) {
1765 case MBUF_SC_BK_SYS
:
1767 wifi_dscp_to_msc_array
[elem
->dscp
] = MBUF_SC_BK
;
1773 wifi_dscp_to_msc_array
[elem
->dscp
] = MBUF_SC_BE
;
1778 wifi_dscp_to_msc_array
[elem
->dscp
] = MBUF_SC_VI
;
1782 wifi_dscp_to_msc_array
[elem
->dscp
] = MBUF_SC_VO
;
1789 dscp_msc_map_from_netsvctype_dscp_map(struct netsvctype_dscp_map
*netsvctype_dscp_map
,
1790 size_t count
, struct dcsp_msc_map
*dcsp_msc_map
)
1796 * Validate input parameters
1798 for (i
= 0; i
< count
; i
++) {
1799 if (!SO_VALID_TC(netsvctype_dscp_map
[i
].netsvctype
)) {
1803 if (netsvctype_dscp_map
[i
].dscp
> _MAX_DSCP
) {
1809 bzero(dcsp_msc_map
, DSCP_ARRAY_SIZE
* sizeof(struct dcsp_msc_map
));
1811 for (i
= 0; i
< count
; i
++) {
1812 dcsp_msc_map
[i
].dscp
= netsvctype_dscp_map
[i
].dscp
;
1813 dcsp_msc_map
[i
].msc
= so_tc2msc(netsvctype_dscp_map
[i
].netsvctype
);
1820 sysctl_dscp_to_wifi_ac_map SYSCTL_HANDLER_ARGS
1822 #pragma unused(oidp, arg1, arg2)
1824 size_t len
= DSCP_ARRAY_SIZE
* sizeof(struct netsvctype_dscp_map
);
1825 struct netsvctype_dscp_map netsvctype_dscp_map
[DSCP_ARRAY_SIZE
] = {};
1826 struct dcsp_msc_map dcsp_msc_map
[DSCP_ARRAY_SIZE
];
1830 if (req
->oldptr
== USER_ADDR_NULL
) {
1832 } else if (req
->oldlen
> 0) {
1833 for (i
= 0; i
< DSCP_ARRAY_SIZE
; i
++) {
1834 netsvctype_dscp_map
[i
].dscp
= i
;
1835 netsvctype_dscp_map
[i
].netsvctype
=
1836 so_svc2tc(wifi_dscp_to_msc_array
[i
]);
1838 error
= SYSCTL_OUT(req
, netsvctype_dscp_map
,
1839 MIN(len
, req
->oldlen
));
1845 if (req
->newptr
== USER_ADDR_NULL
) {
1849 error
= proc_suser(current_proc());
1855 * Check input length
1857 if (req
->newlen
> len
) {
1862 * Cap the number of entries to copy from input buffer
1864 if (len
> req
->newlen
) {
1867 error
= SYSCTL_IN(req
, netsvctype_dscp_map
, len
);
1871 count
= len
/ sizeof(struct netsvctype_dscp_map
);
1872 bzero(dcsp_msc_map
, sizeof(dcsp_msc_map
));
1873 error
= dscp_msc_map_from_netsvctype_dscp_map(netsvctype_dscp_map
, count
,
1878 set_dscp_to_wifi_ac_map(dcsp_msc_map
, 0);
1884 sysctl_reset_dscp_to_wifi_ac_map SYSCTL_HANDLER_ARGS
1886 #pragma unused(oidp, arg1, arg2)
1890 error
= sysctl_handle_int(oidp
, &val
, 0, req
);
1891 if (error
|| !req
->newptr
) {
1895 set_dscp_to_wifi_ac_map(default_dscp_to_wifi_ac_map
, 1);
1901 * Returns whether a large upload or download transfer should be marked as
1902 * BK service type for network activity. This is a system level
1903 * hint/suggestion to classify application traffic based on statistics
1904 * collected from the current network attachment
1906 * Returns 1 for BK and 0 for default
1910 net_qos_guideline(struct proc
*p
, struct net_qos_guideline_args
*arg
,
1914 #define RETURN_USE_BK 1
1915 #define RETURN_USE_DEFAULT 0
1916 struct net_qos_param qos_arg
;
1917 struct ifnet
*ipv4_primary
, *ipv6_primary
;
1920 if (arg
->param
== USER_ADDR_NULL
|| retval
== NULL
||
1921 arg
->param_len
!= sizeof(qos_arg
)) {
1924 err
= copyin(arg
->param
, (caddr_t
) &qos_arg
, sizeof(qos_arg
));
1929 *retval
= RETURN_USE_DEFAULT
;
1930 ipv4_primary
= ifindex2ifnet
[get_primary_ifscope(AF_INET
)];
1931 ipv6_primary
= ifindex2ifnet
[get_primary_ifscope(AF_INET6
)];
1934 * If either of the interfaces is in Low Internet mode, enable
1935 * background delay based algorithms on this transfer
1937 if (qos_arg
.nq_uplink
) {
1938 if ((ipv4_primary
!= NULL
&&
1939 (ipv4_primary
->if_xflags
& IFXF_LOW_INTERNET_UL
)) ||
1940 (ipv6_primary
!= NULL
&&
1941 (ipv6_primary
->if_xflags
& IFXF_LOW_INTERNET_UL
))) {
1942 *retval
= RETURN_USE_BK
;
1946 if ((ipv4_primary
!= NULL
&&
1947 (ipv4_primary
->if_xflags
& IFXF_LOW_INTERNET_DL
)) ||
1948 (ipv6_primary
!= NULL
&&
1949 (ipv6_primary
->if_xflags
& IFXF_LOW_INTERNET_DL
))) {
1950 *retval
= RETURN_USE_BK
;
1956 * Some times IPv4 and IPv6 primary interfaces can be different.
1957 * In this case, if either of them is non-cellular, we should mark
1958 * the transfer as BK as it can potentially get used based on
1959 * the host name resolution
1961 if (ipv4_primary
!= NULL
&& IFNET_IS_EXPENSIVE(ipv4_primary
) &&
1962 ipv6_primary
!= NULL
&& IFNET_IS_EXPENSIVE(ipv6_primary
)) {
1963 if (qos_arg
.nq_use_expensive
) {
1966 *retval
= RETURN_USE_BK
;
1970 if (qos_arg
.nq_transfer_size
>= 5 * 1024 * 1024) {
1971 *retval
= RETURN_USE_BK
;
1976 #undef RETURN_USE_BK
1977 #undef RETURN_USE_DEFAULT