xref: /freebsd/sys/net/netmap_legacy.h (revision 4db78cacdee1b6f3b7880eb8c5560e8edaf91698)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (C) 2011-2014 Matteo Landi, Luigi Rizzo. All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  *
10  *   1. Redistributions of source code must retain the above copyright
11  *      notice, this list of conditions and the following disclaimer.
12  *   2. Redistributions in binary form must reproduce the above copyright
13  *      notice, this list of conditions and the following disclaimer in the
14  *      documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``S IS''AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #ifndef _NET_NETMAP_LEGACY_H_
30 #define _NET_NETMAP_LEGACY_H_
31 
32 /*
33  * $FreeBSD$
34  *
35  * ioctl names and related fields
36  *
37  * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues,
38  *	whose identity is set in NIOCREGIF through nr_ringid.
39  *	These are non blocking and take no argument.
40  *
41  * NIOCGINFO takes a struct ifreq, the interface name is the input,
42  *	the outputs are number of queues and number of descriptor
43  *	for each queue (useful to set number of threads etc.).
44  *	The info returned is only advisory and may change before
45  *	the interface is bound to a file descriptor.
46  *
47  * NIOCREGIF takes an interface name within a struct nmre,
48  *	and activates netmap mode on the interface (if possible).
49  *
50  * The argument to NIOCGINFO/NIOCREGIF overlays struct ifreq so we
51  * can pass it down to other NIC-related ioctls.
52  *
53  * The actual argument (struct nmreq) has a number of options to request
54  * different functions.
55  * The following are used in NIOCREGIF when nr_cmd == 0:
56  *
57  * nr_name	(in)
58  *	The name of the port (em0, valeXXX:YYY, etc.)
59  *	limited to IFNAMSIZ for backward compatibility.
60  *
61  * nr_version	(in/out)
62  *	Must match NETMAP_API as used in the kernel, error otherwise.
63  *	Always returns the desired value on output.
64  *
65  * nr_tx_slots, nr_tx_slots, nr_tx_rings, nr_rx_rings (in/out)
66  *	On input, non-zero values may be used to reconfigure the port
67  *	according to the requested values, but this is not guaranteed.
68  *	On output the actual values in use are reported.
69  *
70  * nr_ringid (in)
71  *	Indicates how rings should be bound to the file descriptors.
72  *	If nr_flags != 0, then the low bits (in NETMAP_RING_MASK)
73  *	are used to indicate the ring number, and nr_flags specifies
74  *	the actual rings to bind. NETMAP_NO_TX_POLL is unaffected.
75  *
76  *	NOTE: THE FOLLOWING (nr_flags == 0) IS DEPRECATED:
77  *	If nr_flags == 0, NETMAP_HW_RING and NETMAP_SW_RING control
78  *	the binding as follows:
79  *	0 (default)			binds all physical rings
80  *	NETMAP_HW_RING | ring number	binds a single ring pair
81  *	NETMAP_SW_RING			binds only the host tx/rx rings
82  *
83  *	NETMAP_NO_TX_POLL can be OR-ed to make select()/poll() push
84  *		packets on tx rings only if POLLOUT is set.
85  *		The default is to push any pending packet.
86  *
87  *	NETMAP_DO_RX_POLL can be OR-ed to make select()/poll() release
88  *		packets on rx rings also when POLLIN is NOT set.
89  *		The default is to touch the rx ring only with POLLIN.
90  *		Note that this is the opposite of TX because it
91  *		reflects the common usage.
92  *
93  *	NOTE: NETMAP_PRIV_MEM IS DEPRECATED, use nr_arg2 instead.
94  *	NETMAP_PRIV_MEM is set on return for ports that do not use
95  *		the global memory allocator.
96  *		This information is not significant and applications
97  *		should look at the region id in nr_arg2
98  *
99  * nr_flags	is the recommended mode to indicate which rings should
100  *		be bound to a file descriptor. Values are NR_REG_*
101  *
102  * nr_arg1 (in)	The number of extra rings to be reserved.
103  *		Especially when allocating a VALE port the system only
104  *		allocates the amount of memory needed for the port.
105  *		If more shared memory rings are desired (e.g. for pipes),
106  *		the first invocation for the same basename/allocator
107  *		should specify a suitable number. Memory cannot be
108  *		extended after the first allocation without closing
109  *		all ports on the same region.
110  *
111  * nr_arg2 (in/out) The identity of the memory region used.
112  *		On input, 0 means the system decides autonomously,
113  *		other values may try to select a specific region.
114  *		On return the actual value is reported.
115  *		Region '1' is the global allocator, normally shared
116  *		by all interfaces. Other values are private regions.
117  *		If two ports the same region zero-copy is possible.
118  *
119  * nr_arg3 (in/out)	number of extra buffers to be allocated.
120  *
121  *
122  *
123  * nr_cmd (in)	if non-zero indicates a special command:
124  *	NETMAP_BDG_ATTACH	 and nr_name = vale*:ifname
125  *		attaches the NIC to the switch; nr_ringid specifies
126  *		which rings to use. Used by vale-ctl -a ...
127  *	    nr_arg1 = NETMAP_BDG_HOST also attaches the host port
128  *		as in vale-ctl -h ...
129  *
130  *	NETMAP_BDG_DETACH	and nr_name = vale*:ifname
131  *		disconnects a previously attached NIC.
132  *		Used by vale-ctl -d ...
133  *
134  *	NETMAP_BDG_LIST
135  *		list the configuration of VALE switches.
136  *
137  *	NETMAP_BDG_VNET_HDR
138  *		Set the virtio-net header length used by the client
139  *		of a VALE switch port.
140  *
141  *	NETMAP_BDG_NEWIF
142  *		create a persistent VALE port with name nr_name.
143  *		Used by vale-ctl -n ...
144  *
145  *	NETMAP_BDG_DELIF
146  *		delete a persistent VALE port. Used by vale-ctl -d ...
147  *
148  * nr_arg1, nr_arg2, nr_arg3  (in/out)		command specific
149  *
150  *
151  *
152  */
153 
154 
155 /*
156  * struct nmreq overlays a struct ifreq (just the name)
157  */
158 struct nmreq {
159 	char		nr_name[IFNAMSIZ];
160 	uint32_t	nr_version;	/* API version */
161 	uint32_t	nr_offset;	/* nifp offset in the shared region */
162 	uint32_t	nr_memsize;	/* size of the shared region */
163 	uint32_t	nr_tx_slots;	/* slots in tx rings */
164 	uint32_t	nr_rx_slots;	/* slots in rx rings */
165 	uint16_t	nr_tx_rings;	/* number of tx rings */
166 	uint16_t	nr_rx_rings;	/* number of rx rings */
167 
168 	uint16_t	nr_ringid;	/* ring(s) we care about */
169 #define NETMAP_HW_RING		0x4000	/* single NIC ring pair */
170 #define NETMAP_SW_RING		0x2000	/* only host ring pair */
171 
172 #define NETMAP_RING_MASK	0x0fff	/* the ring number */
173 
174 #define NETMAP_NO_TX_POLL	0x1000	/* no automatic txsync on poll */
175 
176 #define NETMAP_DO_RX_POLL	0x8000	/* DO automatic rxsync on poll */
177 
178 	uint16_t	nr_cmd;
179 #define NETMAP_BDG_ATTACH	1	/* attach the NIC */
180 #define NETMAP_BDG_DETACH	2	/* detach the NIC */
181 #define NETMAP_BDG_REGOPS	3	/* register bridge callbacks */
182 #define NETMAP_BDG_LIST		4	/* get bridge's info */
183 #define NETMAP_BDG_VNET_HDR     5       /* set the port virtio-net-hdr length */
184 #define NETMAP_BDG_NEWIF	6	/* create a virtual port */
185 #define NETMAP_BDG_DELIF	7	/* destroy a virtual port */
186 #define NETMAP_PT_HOST_CREATE	8	/* create ptnetmap kthreads */
187 #define NETMAP_PT_HOST_DELETE	9	/* delete ptnetmap kthreads */
188 #define NETMAP_BDG_POLLING_ON	10	/* delete polling kthread */
189 #define NETMAP_BDG_POLLING_OFF	11	/* delete polling kthread */
190 #define NETMAP_VNET_HDR_GET	12      /* get the port virtio-net-hdr length */
191 	uint16_t	nr_arg1;	/* reserve extra rings in NIOCREGIF */
192 #define NETMAP_BDG_HOST		1	/* nr_arg1 value for NETMAP_BDG_ATTACH */
193 
194 	uint16_t	nr_arg2;	/* id of the memory allocator */
195 	uint32_t	nr_arg3;	/* req. extra buffers in NIOCREGIF */
196 	uint32_t	nr_flags;	/* specify NR_REG_* mode and other flags */
197 #define NR_REG_MASK		0xf /* to extract NR_REG_* mode from nr_flags */
198 	/* various modes, extends nr_ringid */
199 	uint32_t	spare2[1];
200 };
201 
202 #ifdef _WIN32
203 /*
204  * Windows does not have _IOWR(). _IO(), _IOW() and _IOR() are defined
205  * in ws2def.h but not sure if they are in the form we need.
206  * We therefore redefine them in a convenient way to use for DeviceIoControl
207  * signatures.
208  */
209 #undef _IO	// ws2def.h
210 #define _WIN_NM_IOCTL_TYPE 40000
211 #define _IO(_c, _n)	CTL_CODE(_WIN_NM_IOCTL_TYPE, ((_n) + 0x800) , \
212 		METHOD_BUFFERED, FILE_ANY_ACCESS  )
213 #define _IO_direct(_c, _n)	CTL_CODE(_WIN_NM_IOCTL_TYPE, ((_n) + 0x800) , \
214 		METHOD_OUT_DIRECT, FILE_ANY_ACCESS  )
215 
216 #define _IOWR(_c, _n, _s)	_IO(_c, _n)
217 
218 /* We havesome internal sysctl in addition to the externally visible ones */
219 #define NETMAP_MMAP _IO_direct('i', 160)	// note METHOD_OUT_DIRECT
220 #define NETMAP_POLL _IO('i', 162)
221 
222 /* and also two setsockopt for sysctl emulation */
223 #define NETMAP_SETSOCKOPT _IO('i', 140)
224 #define NETMAP_GETSOCKOPT _IO('i', 141)
225 
226 
227 /* These linknames are for the Netmap Core Driver */
228 #define NETMAP_NT_DEVICE_NAME			L"\\Device\\NETMAP"
229 #define NETMAP_DOS_DEVICE_NAME			L"\\DosDevices\\netmap"
230 
231 /* Definition of a structure used to pass a virtual address within an IOCTL */
232 typedef struct _MEMORY_ENTRY {
233 	PVOID       pUsermodeVirtualAddress;
234 } MEMORY_ENTRY, *PMEMORY_ENTRY;
235 
236 typedef struct _POLL_REQUEST_DATA {
237 	int events;
238 	int timeout;
239 	int revents;
240 } POLL_REQUEST_DATA;
241 #endif /* _WIN32 */
242 
243 /*
244  * Opaque structure that is passed to an external kernel
245  * module via ioctl(fd, NIOCCONFIG, req) for a user-owned
246  * bridge port (at this point ephemeral VALE interface).
247  */
248 #define NM_IFRDATA_LEN 256
249 struct nm_ifreq {
250 	char nifr_name[IFNAMSIZ];
251 	char data[NM_IFRDATA_LEN];
252 };
253 
254 /*
255  * FreeBSD uses the size value embedded in the _IOWR to determine
256  * how much to copy in/out. So we need it to match the actual
257  * data structure we pass. We put some spares in the structure
258  * to ease compatibility with other versions
259  */
260 #define NIOCGINFO	_IOWR('i', 145, struct nmreq) /* return IF info */
261 #define NIOCREGIF	_IOWR('i', 146, struct nmreq) /* interface register */
262 #define NIOCCONFIG	_IOWR('i',150, struct nm_ifreq) /* for ext. modules */
263 
264 #endif /* _NET_NETMAP_LEGACY_H_ */
265