1 /* 2 * Copyright (C) 2011 Matteo Landi, Luigi Rizzo. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions are 6 * met: 7 * 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 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 14 * distribution. 15 * 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. 19 * 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. 31 */ 32 33 /* 34 * $FreeBSD$ 35 * $Id: netmap.h 11997 2013-01-17 21:59:12Z luigi $ 36 * 37 * Definitions of constants and the structures used by the netmap 38 * framework, for the part visible to both kernel and userspace. 39 * Detailed info on netmap is available with "man netmap" or at 40 * 41 * http://info.iet.unipi.it/~luigi/netmap/ 42 */ 43 44 #ifndef _NET_NETMAP_H_ 45 #define _NET_NETMAP_H_ 46 47 /* 48 * --- Netmap data structures --- 49 * 50 * The data structures used by netmap are shown below. Those in 51 * capital letters are in an mmapp()ed area shared with userspace, 52 * while others are private to the kernel. 53 * Shared structures do not contain pointers but only memory 54 * offsets, so that addressing is portable between kernel and userspace. 55 56 57 softc 58 +----------------+ 59 | standard fields| 60 | if_pspare[0] ----------+ 61 +----------------+ | 62 | 63 +----------------+<------+ 64 |(netmap_adapter)| 65 | | netmap_kring 66 | tx_rings *--------------------------------->+---------------+ 67 | | netmap_kring | ring *---------. 68 | rx_rings *--------->+---------------+ | nr_hwcur | | 69 +----------------+ | ring *--------. | nr_hwavail | V 70 | nr_hwcur | | | selinfo | | 71 | nr_hwavail | | +---------------+ . 72 | selinfo | | | ... | . 73 +---------------+ | |(ntx+1 entries)| 74 | .... | | | | 75 |(nrx+1 entries)| | +---------------+ 76 | | | 77 KERNEL +---------------+ | 78 | 79 ==================================================================== 80 | 81 USERSPACE | NETMAP_RING 82 +---->+-------------+ 83 / | cur | 84 NETMAP_IF (nifp, one per file desc.) / | avail | 85 +---------------+ / | buf_ofs | 86 | ni_tx_rings | / +=============+ 87 | ni_rx_rings | / | buf_idx | slot[0] 88 | | / | len, flags | 89 | | / +-------------+ 90 +===============+ / | buf_idx | slot[1] 91 | txring_ofs[0] | (rel.to nifp)--' | len, flags | 92 | txring_ofs[1] | +-------------+ 93 (num_rings+1 entries) (nr_num_slots entries) 94 | txring_ofs[n] | | buf_idx | slot[n-1] 95 +---------------+ | len, flags | 96 | rxring_ofs[0] | +-------------+ 97 | rxring_ofs[1] | 98 (num_rings+1 entries) 99 | txring_ofs[n] | 100 +---------------+ 101 102 * The private descriptor ('softc' or 'adapter') of each interface 103 * is extended with a "struct netmap_adapter" containing netmap-related 104 * info (see description in dev/netmap/netmap_kernel.h. 105 * Among other things, tx_rings and rx_rings point to the arrays of 106 * "struct netmap_kring" which in turn reache the various 107 * "struct netmap_ring", shared with userspace. 108 109 * The NETMAP_RING is the userspace-visible replica of the NIC ring. 110 * Each slot has the index of a buffer, its length and some flags. 111 * In user space, the buffer address is computed as 112 * (char *)ring + buf_ofs + index*NETMAP_BUF_SIZE 113 * In the kernel, buffers do not necessarily need to be contiguous, 114 * and the virtual and physical addresses are derived through 115 * a lookup table. 116 * 117 * struct netmap_slot: 118 * 119 * buf_idx is the index of the buffer associated to the slot. 120 * len is the length of the payload 121 * NS_BUF_CHANGED must be set whenever userspace wants 122 * to change buf_idx (it might be necessary to 123 * reprogram the NIC slot) 124 * NS_REPORT must be set if we want the NIC to generate an interrupt 125 * when this slot is used. Leaving it to 0 improves 126 * performance. 127 * NS_FORWARD if set on a receive ring, and the device is in 128 * transparent mode, buffers released with the flag set 129 * will be forwarded to the 'other' side (host stack 130 * or NIC, respectively) on the next select() or ioctl() 131 * NS_NO_LEARN on a VALE switch, do not 'learn' the source port for 132 * this packet. 133 * NS_PORT_MASK the high 8 bits of the flag, if not zero, indicate the 134 * destination port for the VALE switch, overriding 135 * the lookup table. 136 */ 137 138 struct netmap_slot { 139 uint32_t buf_idx; /* buffer index */ 140 uint16_t len; /* packet length, to be copied to/from the hw ring */ 141 uint16_t flags; /* buf changed, etc. */ 142 #define NS_BUF_CHANGED 0x0001 /* must resync the map, buffer changed */ 143 #define NS_REPORT 0x0002 /* ask the hardware to report results 144 * e.g. by generating an interrupt 145 */ 146 #define NS_FORWARD 0x0004 /* pass packet to the other endpoint 147 * (host stack or device) 148 */ 149 #define NS_NO_LEARN 0x0008 150 #define NS_PORT_SHIFT 8 151 #define NS_PORT_MASK (0xff << NS_PORT_SHIFT) 152 }; 153 154 /* 155 * Netmap representation of a TX or RX ring (also known as "queue"). 156 * This is a queue implemented as a fixed-size circular array. 157 * At the software level, two fields are important: avail and cur. 158 * 159 * In TX rings: 160 * avail indicates the number of slots available for transmission. 161 * It is updated by the kernel after every netmap system call. 162 * It MUST BE decremented by the application when it appends a 163 * packet. 164 * cur indicates the slot to use for the next packet 165 * to send (i.e. the "tail" of the queue). 166 * It MUST BE incremented by the application before 167 * netmap system calls to reflect the number of newly 168 * sent packets. 169 * It is checked by the kernel on netmap system calls 170 * (normally unmodified by the kernel unless invalid). 171 * 172 * The kernel side of netmap uses two additional fields in its own 173 * private ring structure, netmap_kring: 174 * nr_hwcur is a copy of nr_cur on an NIOCTXSYNC. 175 * nr_hwavail is the number of slots known as available by the 176 * hardware. It is updated on an INTR (inc by the 177 * number of packets sent) and on a NIOCTXSYNC 178 * (decrease by nr_cur - nr_hwcur) 179 * A special case, nr_hwavail is -1 if the transmit 180 * side is idle (no pending transmits). 181 * 182 * In RX rings: 183 * avail is the number of packets available (possibly 0). 184 * It MUST BE decremented by the application when it consumes 185 * a packet, and it is updated to nr_hwavail on a NIOCRXSYNC 186 * cur indicates the first slot that contains a packet not 187 * processed yet (the "head" of the queue). 188 * It MUST BE incremented by the software when it consumes 189 * a packet. 190 * reserved indicates the number of buffers before 'cur' 191 * that the application has still in use. Normally 0, 192 * it MUST BE incremented by the application when it 193 * does not return the buffer immediately, and decremented 194 * when the buffer is finally freed. 195 * 196 * The kernel side of netmap uses two additional fields in the kring: 197 * nr_hwcur is a copy of nr_cur on an NIOCRXSYNC 198 * nr_hwavail is the number of packets available. It is updated 199 * on INTR (inc by the number of new packets arrived) 200 * and on NIOCRXSYNC (decreased by nr_cur - nr_hwcur). 201 * 202 * DATA OWNERSHIP/LOCKING: 203 * The netmap_ring is owned by the user program and it is only 204 * accessed or modified in the upper half of the kernel during 205 * a system call. 206 * 207 * The netmap_kring is only modified by the upper half of the kernel. 208 * 209 * FLAGS 210 * NR_TIMESTAMP updates the 'ts' field on each syscall. This is 211 * a global timestamp for all packets. 212 * NR_RX_TSTMP if set, the last 64 byte in each buffer will 213 * contain a timestamp for the frame supplied by 214 * the hardware (if supported) 215 * NR_FORWARD if set, the NS_FORWARD flag in each slot of the 216 * RX ring is checked, and if set the packet is 217 * passed to the other side (host stack or device, 218 * respectively). This permits bpf-like behaviour 219 * or transparency for selected packets. 220 */ 221 struct netmap_ring { 222 /* 223 * nr_buf_base_ofs is meant to be used through macros. 224 * It contains the offset of the buffer region from this 225 * descriptor. 226 */ 227 const ssize_t buf_ofs; 228 const uint32_t num_slots; /* number of slots in the ring. */ 229 uint32_t avail; /* number of usable slots */ 230 uint32_t cur; /* 'current' r/w position */ 231 uint32_t reserved; /* not refilled before current */ 232 233 const uint16_t nr_buf_size; 234 uint16_t flags; 235 #define NR_TIMESTAMP 0x0002 /* set timestamp on *sync() */ 236 #define NR_FORWARD 0x0004 /* enable NS_FORWARD for ring */ 237 #define NR_RX_TSTMP 0x0008 /* set rx timestamp in slots */ 238 239 struct timeval ts; /* time of last *sync() */ 240 241 /* the slots follow. This struct has variable size */ 242 struct netmap_slot slot[0]; /* array of slots. */ 243 }; 244 245 246 /* 247 * Netmap representation of an interface and its queue(s). 248 * There is one netmap_if for each file descriptor on which we want 249 * to select/poll. We assume that on each interface has the same number 250 * of receive and transmit queues. 251 * select/poll operates on one or all pairs depending on the value of 252 * nmr_queueid passed on the ioctl. 253 */ 254 struct netmap_if { 255 char ni_name[IFNAMSIZ]; /* name of the interface. */ 256 const u_int ni_version; /* API version, currently unused */ 257 const u_int ni_rx_rings; /* number of rx rings */ 258 const u_int ni_tx_rings; /* if zero, same as ni_rx_rings */ 259 /* 260 * The following array contains the offset of each netmap ring 261 * from this structure. The first ni_tx_queues+1 entries refer 262 * to the tx rings, the next ni_rx_queues+1 refer to the rx rings 263 * (the last entry in each block refers to the host stack rings). 264 * The area is filled up by the kernel on NIOCREG, 265 * and then only read by userspace code. 266 */ 267 const ssize_t ring_ofs[0]; 268 }; 269 270 #ifndef NIOCREGIF 271 /* 272 * ioctl names and related fields 273 * 274 * NIOCGINFO takes a struct ifreq, the interface name is the input, 275 * the outputs are number of queues and number of descriptor 276 * for each queue (useful to set number of threads etc.). 277 * 278 * NIOCREGIF takes an interface name within a struct ifreq, 279 * and activates netmap mode on the interface (if possible). 280 * 281 * NIOCUNREGIF unregisters the interface associated to the fd. 282 * 283 * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues, 284 * whose identity is set in NIOCREGIF through nr_ringid 285 */ 286 287 /* 288 * struct nmreq overlays a struct ifreq 289 */ 290 struct nmreq { 291 char nr_name[IFNAMSIZ]; 292 uint32_t nr_version; /* API version */ 293 #define NETMAP_API 3 /* current version */ 294 uint32_t nr_offset; /* nifp offset in the shared region */ 295 uint32_t nr_memsize; /* size of the shared region */ 296 uint32_t nr_tx_slots; /* slots in tx rings */ 297 uint32_t nr_rx_slots; /* slots in rx rings */ 298 uint16_t nr_tx_rings; /* number of tx rings */ 299 uint16_t nr_rx_rings; /* number of rx rings */ 300 uint16_t nr_ringid; /* ring(s) we care about */ 301 #define NETMAP_HW_RING 0x4000 /* low bits indicate one hw ring */ 302 #define NETMAP_SW_RING 0x2000 /* process the sw ring */ 303 #define NETMAP_NO_TX_POLL 0x1000 /* no automatic txsync on poll */ 304 #define NETMAP_RING_MASK 0xfff /* the ring number */ 305 uint16_t spare1; 306 uint32_t spare2[4]; 307 }; 308 309 /* 310 * FreeBSD uses the size value embedded in the _IOWR to determine 311 * how much to copy in/out. So we need it to match the actual 312 * data structure we pass. We put some spares in the structure 313 * to ease compatibility with other versions 314 */ 315 #define NIOCGINFO _IOWR('i', 145, struct nmreq) /* return IF info */ 316 #define NIOCREGIF _IOWR('i', 146, struct nmreq) /* interface register */ 317 #define NIOCUNREGIF _IO('i', 147) /* interface unregister */ 318 #define NIOCTXSYNC _IO('i', 148) /* sync tx queues */ 319 #define NIOCRXSYNC _IO('i', 149) /* sync rx queues */ 320 #endif /* !NIOCREGIF */ 321 322 #endif /* _NET_NETMAP_H_ */ 323