2 * Copyright (C) 2011-2013 Matteo Landi, Luigi Rizzo. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions are
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the
16 * 3. Neither the name of the authors nor the names of their contributors
17 * may be used to endorse or promote products derived from this
18 * software without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY MATTEO LANDI AND CONTRIBUTORS "AS IS" AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL MATTEO LANDI OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30 * THE POSSIBILITY OF SUCH DAMAGE.
36 * Definitions of constants and the structures used by the netmap
37 * framework, for the part visible to both kernel and userspace.
38 * Detailed info on netmap is available with "man netmap" or at
40 * http://info.iet.unipi.it/~luigi/netmap/
43 #ifndef _NET_NETMAP_H_
44 #define _NET_NETMAP_H_
47 * --- Netmap data structures ---
49 * The data structures used by netmap are shown below. Those in
50 * capital letters are in an mmapp()ed area shared with userspace,
51 * while others are private to the kernel.
52 * Shared structures do not contain pointers but only memory
53 * offsets, so that addressing is portable between kernel and userspace.
59 | if_pspare[0] ----------+
62 +----------------+<------+
65 | tx_rings *--------------------------------->+---------------+
66 | | netmap_kring | ring *---------.
67 | rx_rings *--------->+---------------+ | nr_hwcur | |
68 +----------------+ | ring *--------. | nr_hwavail | V
69 | nr_hwcur | | | selinfo | |
70 | nr_hwavail | | +---------------+ .
71 | selinfo | | | ... | .
72 +---------------+ | |(ntx+1 entries)|
74 |(nrx+1 entries)| | +---------------+
76 KERNEL +---------------+ |
78 ====================================================================
80 USERSPACE | NETMAP_RING
83 NETMAP_IF (nifp, one per file desc.) / | avail |
84 +---------------+ / | buf_ofs |
85 | ni_tx_rings | / +=============+
86 | ni_rx_rings | / | buf_idx | slot[0]
89 +===============+ / | buf_idx | slot[1]
90 | txring_ofs[0] | (rel.to nifp)--' | len, flags |
91 | txring_ofs[1] | +-------------+
92 (num_rings+1 entries) (nr_num_slots entries)
93 | txring_ofs[n] | | buf_idx | slot[n-1]
94 +---------------+ | len, flags |
95 | rxring_ofs[0] | +-------------+
101 * The private descriptor ('softc' or 'adapter') of each interface
102 * is extended with a "struct netmap_adapter" containing netmap-related
103 * info (see description in dev/netmap/netmap_kernel.h.
104 * Among other things, tx_rings and rx_rings point to the arrays of
105 * "struct netmap_kring" which in turn reache the various
106 * "struct netmap_ring", shared with userspace.
108 * The NETMAP_RING is the userspace-visible replica of the NIC ring.
109 * Each slot has the index of a buffer, its length and some flags.
110 * In user space, the buffer address is computed as
111 * (char *)ring + buf_ofs + index*NETMAP_BUF_SIZE
112 * In the kernel, buffers do not necessarily need to be contiguous,
113 * and the virtual and physical addresses are derived through
116 * struct netmap_slot:
118 * buf_idx is the index of the buffer associated to the slot.
119 * len is the length of the payload
120 * NS_BUF_CHANGED must be set whenever userspace wants
121 * to change buf_idx (it might be necessary to
122 * reprogram the NIC slot)
123 * NS_REPORT must be set if we want the NIC to generate an interrupt
124 * when this slot is used. Leaving it to 0 improves
126 * NS_FORWARD if set on a receive ring, and the device is in
127 * transparent mode, buffers released with the flag set
128 * will be forwarded to the 'other' side (host stack
129 * or NIC, respectively) on the next select() or ioctl()
131 * The following will be supported from NETMAP_API = 5
132 * NS_NO_LEARN on a VALE switch, do not 'learn' the source port for
134 * NS_INDIRECT the netmap buffer contains a 64-bit pointer to
135 * the actual userspace buffer. This may be useful
136 * to reduce copies in a VM environment.
137 * NS_MOREFRAG Part of a multi-segment frame. The last (or only)
138 * segment must not have this flag.
139 * NS_PORT_MASK the high 8 bits of the flag, if not zero, indicate the
140 * destination port for the VALE switch, overriding
145 uint32_t buf_idx; /* buffer index */
146 uint16_t len; /* packet length, to be copied to/from the hw ring */
147 uint16_t flags; /* buf changed, etc. */
148 #define NS_BUF_CHANGED 0x0001 /* must resync the map, buffer changed */
149 #define NS_REPORT 0x0002 /* ask the hardware to report results
150 * e.g. by generating an interrupt
152 #define NS_FORWARD 0x0004 /* pass packet to the other endpoint
153 * (host stack or device)
155 #define NS_NO_LEARN 0x0008
156 #define NS_INDIRECT 0x0010
157 #define NS_MOREFRAG 0x0020
158 #define NS_PORT_SHIFT 8
159 #define NS_PORT_MASK (0xff << NS_PORT_SHIFT)
163 * Netmap representation of a TX or RX ring (also known as "queue").
164 * This is a queue implemented as a fixed-size circular array.
165 * At the software level, two fields are important: avail and cur.
168 * avail indicates the number of slots available for transmission.
169 * It is updated by the kernel after every netmap system call.
170 * It MUST BE decremented by the application when it appends a
172 * cur indicates the slot to use for the next packet
173 * to send (i.e. the "tail" of the queue).
174 * It MUST BE incremented by the application before
175 * netmap system calls to reflect the number of newly
177 * It is checked by the kernel on netmap system calls
178 * (normally unmodified by the kernel unless invalid).
180 * The kernel side of netmap uses two additional fields in its own
181 * private ring structure, netmap_kring:
182 * nr_hwcur is a copy of nr_cur on an NIOCTXSYNC.
183 * nr_hwavail is the number of slots known as available by the
184 * hardware. It is updated on an INTR (inc by the
185 * number of packets sent) and on a NIOCTXSYNC
186 * (decrease by nr_cur - nr_hwcur)
187 * A special case, nr_hwavail is -1 if the transmit
188 * side is idle (no pending transmits).
191 * avail is the number of packets available (possibly 0).
192 * It MUST BE decremented by the application when it consumes
193 * a packet, and it is updated to nr_hwavail on a NIOCRXSYNC
194 * cur indicates the first slot that contains a packet not
195 * processed yet (the "head" of the queue).
196 * It MUST BE incremented by the software when it consumes
198 * reserved indicates the number of buffers before 'cur'
199 * that the application has still in use. Normally 0,
200 * it MUST BE incremented by the application when it
201 * does not return the buffer immediately, and decremented
202 * when the buffer is finally freed.
204 * The kernel side of netmap uses two additional fields in the kring:
205 * nr_hwcur is a copy of nr_cur on an NIOCRXSYNC
206 * nr_hwavail is the number of packets available. It is updated
207 * on INTR (inc by the number of new packets arrived)
208 * and on NIOCRXSYNC (decreased by nr_cur - nr_hwcur).
210 * DATA OWNERSHIP/LOCKING:
211 * The netmap_ring is owned by the user program and it is only
212 * accessed or modified in the upper half of the kernel during
215 * The netmap_kring is only modified by the upper half of the kernel.
218 * NR_TIMESTAMP updates the 'ts' field on each syscall. This is
219 * a global timestamp for all packets.
220 * NR_RX_TSTMP if set, the last 64 byte in each buffer will
221 * contain a timestamp for the frame supplied by
222 * the hardware (if supported)
223 * NR_FORWARD if set, the NS_FORWARD flag in each slot of the
224 * RX ring is checked, and if set the packet is
225 * passed to the other side (host stack or device,
226 * respectively). This permits bpf-like behaviour
227 * or transparency for selected packets.
231 * nr_buf_base_ofs is meant to be used through macros.
232 * It contains the offset of the buffer region from this
235 const ssize_t buf_ofs;
236 const uint32_t num_slots; /* number of slots in the ring. */
237 uint32_t avail; /* number of usable slots */
238 uint32_t cur; /* 'current' r/w position */
239 uint32_t reserved; /* not refilled before current */
241 const uint16_t nr_buf_size;
243 #define NR_TIMESTAMP 0x0002 /* set timestamp on *sync() */
244 #define NR_FORWARD 0x0004 /* enable NS_FORWARD for ring */
245 #define NR_RX_TSTMP 0x0008 /* set rx timestamp in slots */
247 struct timeval ts; /* time of last *sync() */
249 /* the slots follow. This struct has variable size */
250 struct netmap_slot slot[0]; /* array of slots. */
255 * Netmap representation of an interface and its queue(s).
256 * There is one netmap_if for each file descriptor on which we want
257 * to select/poll. We assume that on each interface has the same number
258 * of receive and transmit queues.
259 * select/poll operates on one or all pairs depending on the value of
260 * nmr_queueid passed on the ioctl.
263 char ni_name[IFNAMSIZ]; /* name of the interface. */
264 const u_int ni_version; /* API version, currently unused */
265 const u_int ni_rx_rings; /* number of rx rings */
266 const u_int ni_tx_rings; /* if zero, same as ni_rx_rings */
268 * The following array contains the offset of each netmap ring
269 * from this structure. The first ni_tx_queues+1 entries refer
270 * to the tx rings, the next ni_rx_queues+1 refer to the rx rings
271 * (the last entry in each block refers to the host stack rings).
272 * The area is filled up by the kernel on NIOCREG,
273 * and then only read by userspace code.
275 const ssize_t ring_ofs[0];
280 * ioctl names and related fields
282 * NIOCGINFO takes a struct ifreq, the interface name is the input,
283 * the outputs are number of queues and number of descriptor
284 * for each queue (useful to set number of threads etc.).
286 * NIOCREGIF takes an interface name within a struct ifreq,
287 * and activates netmap mode on the interface (if possible).
289 * For vale ports, starting with NETMAP_API = 5,
290 * nr_tx_rings and nr_rx_rings specify how many software rings
291 * are created (0 means 1).
293 * NIOCREGIF is also used to attach a NIC to a VALE switch.
294 * In this case the name is vale*:ifname, and "nr_cmd"
295 * is set to 'NETMAP_BDG_ATTACH' or 'NETMAP_BDG_DETACH'.
296 * nr_ringid specifies which rings should be attached, 0 means all,
297 * NETMAP_HW_RING + n means only the n-th ring.
298 * The process can terminate after the interface has been attached.
300 * NIOCUNREGIF unregisters the interface associated to the fd.
301 * this is deprecated and will go away.
303 * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues,
304 * whose identity is set in NIOCREGIF through nr_ringid
306 * NETMAP_API is the API version.
310 * struct nmreq overlays a struct ifreq
313 char nr_name[IFNAMSIZ];
314 uint32_t nr_version; /* API version */
315 #define NETMAP_API 4 /* current version */
316 uint32_t nr_offset; /* nifp offset in the shared region */
317 uint32_t nr_memsize; /* size of the shared region */
318 uint32_t nr_tx_slots; /* slots in tx rings */
319 uint32_t nr_rx_slots; /* slots in rx rings */
320 uint16_t nr_tx_rings; /* number of tx rings */
321 uint16_t nr_rx_rings; /* number of rx rings */
322 uint16_t nr_ringid; /* ring(s) we care about */
323 #define NETMAP_HW_RING 0x4000 /* low bits indicate one hw ring */
324 #define NETMAP_SW_RING 0x2000 /* process the sw ring */
325 #define NETMAP_NO_TX_POLL 0x1000 /* no automatic txsync on poll */
326 #define NETMAP_RING_MASK 0xfff /* the ring number */
328 #define NETMAP_BDG_ATTACH 1 /* attach the NIC */
329 #define NETMAP_BDG_DETACH 2 /* detach the NIC */
330 #define NETMAP_BDG_LOOKUP_REG 3 /* register lookup function */
331 #define NETMAP_BDG_LIST 4 /* get bridge's info */
333 #define NETMAP_BDG_HOST 1 /* attach the host stack on ATTACH */
339 * FreeBSD uses the size value embedded in the _IOWR to determine
340 * how much to copy in/out. So we need it to match the actual
341 * data structure we pass. We put some spares in the structure
342 * to ease compatibility with other versions
344 #define NIOCGINFO _IOWR('i', 145, struct nmreq) /* return IF info */
345 #define NIOCREGIF _IOWR('i', 146, struct nmreq) /* interface register */
346 #define NIOCUNREGIF _IO('i', 147) /* interface unregister */
347 #define NIOCTXSYNC _IO('i', 148) /* sync tx queues */
348 #define NIOCRXSYNC _IO('i', 149) /* sync rx queues */
349 #endif /* !NIOCREGIF */
351 #endif /* _NET_NETMAP_H_ */