1 /* 2 * Copyright (C) 2011-2013 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 * 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 39 * 40 * http://info.iet.unipi.it/~luigi/netmap/ 41 * 42 * This API is also used to communicate with the VALE software switch 43 */ 44 45 #ifndef _NET_NETMAP_H_ 46 #define _NET_NETMAP_H_ 47 48 /* 49 * --- Netmap data structures --- 50 * 51 * The userspace data structures used by netmap are shown below. 52 * They are allocated by the kernel and mmap()ed by userspace threads. 53 * Pointers are implemented as memory offsets or indexes, 54 * so that they can be easily dereferenced in kernel and userspace. 55 56 KERNEL (opaque, obviously) 57 58 ==================================================================== 59 | 60 USERSPACE | struct netmap_ring 61 +---->+--------------+ 62 / | cur | 63 struct netmap_if (nifp, 1 per fd) / | avail | 64 +---------------+ / | buf_ofs | 65 | ni_tx_rings | / +==============+ 66 | ni_rx_rings | / | buf_idx, len | slot[0] 67 | | / | flags, ptr | 68 | | / +--------------+ 69 +===============+ / | buf_idx, len | slot[1] 70 | txring_ofs[0] | (rel.to nifp)--' | flags, ptr | 71 | txring_ofs[1] | +--------------+ 72 (ni_tx_rings+1 entries) (num_slots entries) 73 | txring_ofs[t] | | buf_idx, len | slot[n-1] 74 +---------------+ | flags, ptr | 75 | rxring_ofs[0] | +--------------+ 76 | rxring_ofs[1] | 77 (ni_rx_rings+1 entries) 78 | rxring_ofs[r] | 79 +---------------+ 80 81 * For each "interface" (NIC, host stack, VALE switch port) attached to a 82 * file descriptor, the mmap()ed region contains a (logically readonly) 83 * struct netmap_if pointing to struct netmap_ring's. 84 * There is one netmap_ring per physical NIC ring, plus one tx/rx ring 85 * pair attached to the host stack (this pair is unused for VALE ports). 86 * 87 * All physical/host stack ports share the same memory region, 88 * so that zero-copy can be implemented between them. 89 * VALE switch ports instead have separate memory regions. 90 * 91 * The netmap_ring is the userspace-visible replica of the NIC ring. 92 * Each slot has the index of a buffer (MTU-sized and residing in the 93 * mmapped region), its length and some flags. An extra 64-bit pointer 94 * is provided for user-supplied buffers in the tx path. 95 * 96 * In user space, the buffer address is computed as 97 * (char *)ring + buf_ofs + index*NETMAP_BUF_SIZE 98 */ 99 100 /* 101 * struct netmap_slot is a buffer descriptor 102 * 103 * buf_idx the index of the buffer associated to the slot. 104 * len the length of the payload 105 * flags control operation on the slot, as defined below 106 * 107 * NS_BUF_CHANGED must be set whenever userspace wants 108 * to change buf_idx (it might be necessary to 109 * reprogram the NIC) 110 * 111 * NS_REPORT must be set if we want the NIC to generate an interrupt 112 * when this slot is used. Leaving it to 0 improves 113 * performance. 114 * 115 * NS_FORWARD if set on a receive ring, and the device is in 116 * transparent mode, buffers released with the flag set 117 * will be forwarded to the 'other' side (host stack 118 * or NIC, respectively) on the next select() or ioctl() 119 * 120 * NS_NO_LEARN on a VALE switch, do not 'learn' the source port for 121 * this packet. 122 * 123 * NS_INDIRECT (tx rings only) data is in a userspace buffer pointed 124 * by the ptr field in the slot. 125 * 126 * NS_MOREFRAG Part of a multi-segment frame. The last (or only) 127 * segment must not have this flag. 128 * Only supported on VALE ports. 129 * 130 * NS_PORT_MASK the high 8 bits of the flag, if not zero, indicate the 131 * destination port for the VALE switch, overriding 132 * the lookup table. 133 */ 134 135 struct netmap_slot { 136 uint32_t buf_idx; /* buffer index */ 137 uint16_t len; /* packet length */ 138 uint16_t flags; /* buf changed, etc. */ 139 #define NS_BUF_CHANGED 0x0001 /* buf_idx changed */ 140 #define NS_REPORT 0x0002 /* ask the hardware to report results 141 * e.g. by generating an interrupt 142 */ 143 #define NS_FORWARD 0x0004 /* pass packet to the other endpoint 144 * (host stack or device) 145 */ 146 #define NS_NO_LEARN 0x0008 147 #define NS_INDIRECT 0x0010 148 #define NS_MOREFRAG 0x0020 149 #define NS_PORT_SHIFT 8 150 #define NS_PORT_MASK (0xff << NS_PORT_SHIFT) 151 /* 152 * in rx rings, the high 8 bits 153 * are the number of fragments. 154 */ 155 #define NS_RFRAGS(_slot) ( ((_slot)->flags >> 8) & 0xff) 156 uint64_t ptr; /* pointer for indirect buffers */ 157 }; 158 159 /* 160 * struct netmap_ring 161 * 162 * Netmap representation of a TX or RX ring (also known as "queue"). 163 * This is a queue implemented as a fixed-size circular array. 164 * At the software level, two fields are important: avail and cur. 165 * 166 * In TX rings: 167 * 168 * avail tells how many slots are available for transmission. 169 * It is updated by the kernel in each netmap system call. 170 * It MUST BE decremented by the user when it 171 * adds a new packet to send. 172 * 173 * cur indicates the slot to use for the next packet 174 * to send (i.e. the "tail" of the queue). 175 * It MUST BE incremented by the user before 176 * netmap system calls to reflect the number of newly 177 * sent packets. 178 * It is checked by the kernel on netmap system calls 179 * (normally unmodified by the kernel unless invalid). 180 * 181 * In RX rings: 182 * 183 * avail is the number of packets available (possibly 0). 184 * It is updated by the kernel in each netmap system call. 185 * It MUST BE decremented by the user when it 186 * consumes a packet. 187 * 188 * cur indicates the first slot that contains a packet not 189 * yet processed (the "head" of the queue). 190 * It MUST BE incremented by the user when it consumes 191 * a packet. 192 * 193 * reserved indicates the number of buffers before 'cur' 194 * that the user has not released yet. Normally 0, 195 * it MUST BE incremented by the user when it 196 * does not return the buffer immediately, and decremented 197 * when the buffer is finally freed. 198 * 199 * 200 * DATA OWNERSHIP/LOCKING: 201 * The netmap_ring, all slots, and buffers in the range 202 * [reserved-cur , cur+avail[ are owned by the user program, 203 * and the kernel only touches them in the same thread context 204 * during a system call. 205 * Other buffers are reserved for use by the NIC's DMA engines. 206 * 207 * FLAGS 208 * NR_TIMESTAMP updates the 'ts' field on each syscall. This is 209 * a global timestamp for all packets. 210 * NR_RX_TSTMP if set, the last 64 byte in each buffer will 211 * contain a timestamp for the frame supplied by 212 * the hardware (if supported) 213 * NR_FORWARD if set, the NS_FORWARD flag in each slot of the 214 * RX ring is checked, and if set the packet is 215 * passed to the other side (host stack or device, 216 * respectively). This permits bpf-like behaviour 217 * or transparency for selected packets. 218 */ 219 struct netmap_ring { 220 /* 221 * buf_ofs is meant to be used through macros. 222 * It contains the offset of the buffer region from this 223 * descriptor. 224 */ 225 const ssize_t buf_ofs; 226 const uint32_t num_slots; /* number of slots in the ring. */ 227 uint32_t avail; /* number of usable slots */ 228 uint32_t cur; /* 'current' r/w position */ 229 uint32_t reserved; /* not refilled before current */ 230 231 const uint16_t nr_buf_size; 232 uint16_t flags; 233 #define NR_TIMESTAMP 0x0002 /* set timestamp on *sync() */ 234 #define NR_FORWARD 0x0004 /* enable NS_FORWARD for ring */ 235 #define NR_RX_TSTMP 0x0008 /* set rx timestamp in slots */ 236 237 struct timeval ts; /* time of last *sync() */ 238 239 /* the slots follow. This struct has variable size */ 240 struct netmap_slot slot[0]; /* array of slots. */ 241 }; 242 243 244 /* 245 * Netmap representation of an interface and its queue(s). 246 * This is initialized by the kernel when binding a file 247 * descriptor to a port, and should be considered as readonly 248 * by user programs. The kernel never uses it. 249 * 250 * There is one netmap_if for each file descriptor on which we want 251 * to select/poll. 252 * select/poll operates on one or all pairs depending on the value of 253 * nmr_queueid passed on the ioctl. 254 */ 255 struct netmap_if { 256 char ni_name[IFNAMSIZ]; /* name of the interface. */ 257 const uint32_t ni_version; /* API version, currently unused */ 258 const uint32_t ni_flags; /* properties */ 259 #define NI_PRIV_MEM 0x1 /* private memory region */ 260 261 const uint32_t ni_rx_rings; /* number of rx rings */ 262 const uint32_t ni_tx_rings; /* number of tx rings */ 263 /* 264 * The following array contains the offset of each netmap ring 265 * from this structure. The first ni_tx_rings+1 entries refer 266 * to the tx rings, the next ni_rx_rings+1 refer to the rx rings 267 * (the last entry in each block refers to the host stack rings). 268 * The area is filled up by the kernel on NIOCREGIF, 269 * and then only read by userspace code. 270 */ 271 const ssize_t ring_ofs[0]; 272 }; 273 274 #ifndef NIOCREGIF 275 /* 276 * ioctl names and related fields 277 * 278 * NIOCGINFO takes a struct ifreq, the interface name is the input, 279 * the outputs are number of queues and number of descriptor 280 * for each queue (useful to set number of threads etc.). 281 * The info returned is only advisory and may change before 282 * the interface is bound to a file descriptor. 283 * 284 * NIOCREGIF takes an interface name within a struct ifreq, 285 * and activates netmap mode on the interface (if possible). 286 * 287 * nr_name is the name of the interface 288 * 289 * nr_tx_slots, nr_tx_slots, nr_tx_rings, nr_rx_rings 290 * indicate the configuration of the port on return. 291 * 292 * On input, non-zero values for nr_tx_rings, nr_tx_slots and the 293 * rx counterparts may be used to reconfigure the port according 294 * to the requested values, but this is not guaranteed. 295 * The actual values are returned on completion of the ioctl(). 296 * 297 * nr_ringid 298 * indicates how rings should be bound to the file descriptors. 299 * The default (0) means all physical rings of a NIC are bound. 300 * NETMAP_HW_RING plus a ring number lets you bind just 301 * a single ring pair. 302 * NETMAP_SW_RING binds only the host tx/rx rings 303 * NETMAP_NO_TX_POLL prevents select()/poll() from pushing 304 * out packets on the tx ring unless POLLOUT is specified. 305 * 306 * NETMAP_PRIV_MEM is a return value used to indicate that 307 * this ring is in a private memory region hence buffer 308 * swapping cannot be used 309 * 310 * nr_cmd is used to configure NICs attached to a VALE switch, 311 * or to dump the configuration of a VALE switch. 312 * 313 * nr_cmd = NETMAP_BDG_ATTACH and nr_name = vale*:ifname 314 * attaches the NIC to the switch, with nr_ringid specifying 315 * which rings to use 316 * 317 * nr_cmd = NETMAP_BDG_DETACH and nr_name = vale*:ifname 318 * disconnects a previously attached NIC 319 * 320 * nr_cmd = NETMAP_BDG_LIST is used to list the configuration 321 * of VALE switches, with additional arguments. 322 * 323 * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues, 324 * whose identity is set in NIOCREGIF through nr_ringid 325 * 326 * NETMAP_API is the API version. 327 */ 328 329 /* 330 * struct nmreq overlays a struct ifreq 331 */ 332 struct nmreq { 333 char nr_name[IFNAMSIZ]; 334 uint32_t nr_version; /* API version */ 335 #define NETMAP_API 5 /* current version */ 336 uint32_t nr_offset; /* nifp offset in the shared region */ 337 uint32_t nr_memsize; /* size of the shared region */ 338 uint32_t nr_tx_slots; /* slots in tx rings */ 339 uint32_t nr_rx_slots; /* slots in rx rings */ 340 uint16_t nr_tx_rings; /* number of tx rings */ 341 uint16_t nr_rx_rings; /* number of rx rings */ 342 uint16_t nr_ringid; /* ring(s) we care about */ 343 #define NETMAP_PRIV_MEM 0x8000 /* rings use private memory */ 344 #define NETMAP_HW_RING 0x4000 /* low bits indicate one hw ring */ 345 #define NETMAP_SW_RING 0x2000 /* process the sw ring */ 346 #define NETMAP_NO_TX_POLL 0x1000 /* no automatic txsync on poll */ 347 #define NETMAP_RING_MASK 0xfff /* the ring number */ 348 uint16_t nr_cmd; 349 #define NETMAP_BDG_ATTACH 1 /* attach the NIC */ 350 #define NETMAP_BDG_DETACH 2 /* detach the NIC */ 351 #define NETMAP_BDG_LOOKUP_REG 3 /* register lookup function */ 352 #define NETMAP_BDG_LIST 4 /* get bridge's info */ 353 uint16_t nr_arg1; 354 #define NETMAP_BDG_HOST 1 /* attach the host stack on ATTACH */ 355 uint16_t nr_arg2; 356 uint32_t spare2[3]; 357 }; 358 359 /* 360 * FreeBSD uses the size value embedded in the _IOWR to determine 361 * how much to copy in/out. So we need it to match the actual 362 * data structure we pass. We put some spares in the structure 363 * to ease compatibility with other versions 364 */ 365 #define NIOCGINFO _IOWR('i', 145, struct nmreq) /* return IF info */ 366 #define NIOCREGIF _IOWR('i', 146, struct nmreq) /* interface register */ 367 #define NIOCUNREGIF _IO('i', 147) /* deprecated. Was interface unregister */ 368 #define NIOCTXSYNC _IO('i', 148) /* sync tx queues */ 369 #define NIOCRXSYNC _IO('i', 149) /* sync rx queues */ 370 #endif /* !NIOCREGIF */ 371 372 #endif /* _NET_NETMAP_H_ */ 373