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 6 * are met: 7 * 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``S IS''AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 /* 28 * $FreeBSD: head/sys/net/netmap.h 251139 2013-05-30 14:07:14Z luigi $ 29 * 30 * Definitions of constants and the structures used by the netmap 31 * framework, for the part visible to both kernel and userspace. 32 * Detailed info on netmap is available with "man netmap" or at 33 * 34 * http://info.iet.unipi.it/~luigi/netmap/ 35 * 36 * This API is also used to communicate with the VALE software switch 37 */ 38 39 #ifndef _NET_NETMAP_H_ 40 #define _NET_NETMAP_H_ 41 42 /* 43 * --- Netmap data structures --- 44 * 45 * The userspace data structures used by netmap are shown below. 46 * They are allocated by the kernel and mmap()ed by userspace threads. 47 * Pointers are implemented as memory offsets or indexes, 48 * so that they can be easily dereferenced in kernel and userspace. 49 50 KERNEL (opaque, obviously) 51 52 ==================================================================== 53 | 54 USERSPACE | struct netmap_ring 55 +---->+--------------+ 56 / | cur | 57 struct netmap_if (nifp, 1 per fd) / | avail | 58 +---------------+ / | buf_ofs | 59 | ni_tx_rings | / +==============+ 60 | ni_rx_rings | / | buf_idx, len | slot[0] 61 | | / | flags, ptr | 62 | | / +--------------+ 63 +===============+ / | buf_idx, len | slot[1] 64 | txring_ofs[0] | (rel.to nifp)--' | flags, ptr | 65 | txring_ofs[1] | +--------------+ 66 (ni_tx_rings+1 entries) (num_slots entries) 67 | txring_ofs[t] | | buf_idx, len | slot[n-1] 68 +---------------+ | flags, ptr | 69 | rxring_ofs[0] | +--------------+ 70 | rxring_ofs[1] | 71 (ni_rx_rings+1 entries) 72 | rxring_ofs[r] | 73 +---------------+ 74 75 * For each "interface" (NIC, host stack, VALE switch port) attached to a 76 * file descriptor, the mmap()ed region contains a (logically readonly) 77 * struct netmap_if pointing to struct netmap_ring's. 78 * There is one netmap_ring per physical NIC ring, plus one tx/rx ring 79 * pair attached to the host stack (this pair is unused for VALE ports). 80 * 81 * All physical/host stack ports share the same memory region, 82 * so that zero-copy can be implemented between them. 83 * VALE switch ports instead have separate memory regions. 84 * 85 * The netmap_ring is the userspace-visible replica of the NIC ring. 86 * Each slot has the index of a buffer (MTU-sized and residing in the 87 * mmapped region), its length and some flags. An extra 64-bit pointer 88 * is provided for user-supplied buffers in the tx path. 89 * 90 * In user space, the buffer address is computed as 91 * (char *)ring + buf_ofs + index*NETMAP_BUF_SIZE 92 */ 93 94 /* 95 * struct netmap_slot is a buffer descriptor 96 * 97 * buf_idx the index of the buffer associated to the slot. 98 * len the length of the payload 99 * flags control operation on the slot, as defined below 100 * 101 * NS_BUF_CHANGED must be set whenever userspace wants 102 * to change buf_idx (it might be necessary to 103 * reprogram the NIC) 104 * 105 * NS_REPORT must be set if we want the NIC to generate an interrupt 106 * when this slot is used. Leaving it to 0 improves 107 * performance. 108 * 109 * NS_FORWARD if set on a receive ring, and the device is in 110 * transparent mode, buffers released with the flag set 111 * will be forwarded to the 'other' side (host stack 112 * or NIC, respectively) on the next select() or ioctl() 113 * 114 * NS_NO_LEARN on a VALE switch, do not 'learn' the source port for 115 * this packet. 116 * 117 * NS_INDIRECT (tx rings only) data is in a userspace buffer pointed 118 * by the ptr field in the slot. 119 * 120 * NS_MOREFRAG Part of a multi-segment frame. The last (or only) 121 * segment must not have this flag. 122 * Only supported on VALE ports. 123 * 124 * NS_PORT_MASK the high 8 bits of the flag, if not zero, indicate the 125 * destination port for the VALE switch, overriding 126 * the lookup table. 127 */ 128 129 struct netmap_slot { 130 uint32_t buf_idx; /* buffer index */ 131 uint16_t len; /* packet length */ 132 uint16_t flags; /* buf changed, etc. */ 133 #define NS_BUF_CHANGED 0x0001 /* buf_idx changed */ 134 #define NS_REPORT 0x0002 /* ask the hardware to report results 135 * e.g. by generating an interrupt 136 */ 137 #define NS_FORWARD 0x0004 /* pass packet to the other endpoint 138 * (host stack or device) 139 */ 140 #define NS_NO_LEARN 0x0008 141 #define NS_INDIRECT 0x0010 142 #define NS_MOREFRAG 0x0020 143 #define NS_PORT_SHIFT 8 144 #define NS_PORT_MASK (0xff << NS_PORT_SHIFT) 145 /* 146 * in rx rings, the high 8 bits 147 * are the number of fragments. 148 */ 149 #define NS_RFRAGS(_slot) ( ((_slot)->flags >> 8) & 0xff) 150 uint64_t ptr; /* pointer for indirect buffers */ 151 }; 152 153 /* 154 * struct netmap_ring 155 * 156 * Netmap representation of a TX or RX ring (also known as "queue"). 157 * This is a queue implemented as a fixed-size circular array. 158 * At the software level, two fields are important: avail and cur. 159 * 160 * In TX rings: 161 * 162 * avail tells how many slots are available for transmission. 163 * It is updated by the kernel in each netmap system call. 164 * It MUST BE decremented by the user when it 165 * adds a new packet to send. 166 * 167 * cur indicates the slot to use for the next packet 168 * to send (i.e. the "tail" of the queue). 169 * It MUST BE incremented by the user before 170 * netmap system calls to reflect the number of newly 171 * sent packets. 172 * It is checked by the kernel on netmap system calls 173 * (normally unmodified by the kernel unless invalid). 174 * 175 * In RX rings: 176 * 177 * avail is the number of packets available (possibly 0). 178 * It is updated by the kernel in each netmap system call. 179 * It MUST BE decremented by the user when it 180 * consumes a packet. 181 * 182 * cur indicates the first slot that contains a packet not 183 * yet processed (the "head" of the queue). 184 * It MUST BE incremented by the user when it consumes 185 * a packet. 186 * 187 * reserved indicates the number of buffers before 'cur' 188 * that the user has not released yet. Normally 0, 189 * it MUST BE incremented by the user when it 190 * does not return the buffer immediately, and decremented 191 * when the buffer is finally freed. 192 * 193 * 194 * DATA OWNERSHIP/LOCKING: 195 * The netmap_ring, all slots, and buffers in the range 196 * [reserved-cur , cur+avail[ are owned by the user program, 197 * and the kernel only touches them in the same thread context 198 * during a system call. 199 * Other buffers are reserved for use by the NIC's DMA engines. 200 * 201 * FLAGS 202 * NR_TIMESTAMP updates the 'ts' field on each syscall. This is 203 * a global timestamp for all packets. 204 * NR_RX_TSTMP if set, the last 64 byte in each buffer will 205 * contain a timestamp for the frame supplied by 206 * the hardware (if supported) 207 * NR_FORWARD if set, the NS_FORWARD flag in each slot of the 208 * RX ring is checked, and if set the packet is 209 * passed to the other side (host stack or device, 210 * respectively). This permits bpf-like behaviour 211 * or transparency for selected packets. 212 */ 213 struct netmap_ring { 214 /* 215 * buf_ofs is meant to be used through macros. 216 * It contains the offset of the buffer region from this 217 * descriptor. 218 */ 219 const ssize_t buf_ofs; 220 const uint32_t num_slots; /* number of slots in the ring. */ 221 uint32_t avail; /* number of usable slots */ 222 uint32_t cur; /* 'current' r/w position */ 223 uint32_t reserved; /* not refilled before current */ 224 225 const uint16_t nr_buf_size; 226 uint16_t flags; 227 #define NR_TIMESTAMP 0x0002 /* set timestamp on *sync() */ 228 #define NR_FORWARD 0x0004 /* enable NS_FORWARD for ring */ 229 #define NR_RX_TSTMP 0x0008 /* set rx timestamp in slots */ 230 231 struct timeval ts; /* time of last *sync() */ 232 233 /* the slots follow. This struct has variable size */ 234 struct netmap_slot slot[0]; /* array of slots. */ 235 }; 236 237 238 /* 239 * Netmap representation of an interface and its queue(s). 240 * This is initialized by the kernel when binding a file 241 * descriptor to a port, and should be considered as readonly 242 * by user programs. The kernel never uses it. 243 * 244 * There is one netmap_if for each file descriptor on which we want 245 * to select/poll. 246 * select/poll operates on one or all pairs depending on the value of 247 * nmr_queueid passed on the ioctl. 248 */ 249 struct netmap_if { 250 char ni_name[IFNAMSIZ]; /* name of the interface. */ 251 const uint32_t ni_version; /* API version, currently unused */ 252 const uint32_t ni_flags; /* properties */ 253 #define NI_PRIV_MEM 0x1 /* private memory region */ 254 255 const uint32_t ni_rx_rings; /* number of rx rings */ 256 const uint32_t ni_tx_rings; /* number of tx rings */ 257 /* 258 * The following array contains the offset of each netmap ring 259 * from this structure. The first ni_tx_rings+1 entries refer 260 * to the tx rings, the next ni_rx_rings+1 refer to the rx rings 261 * (the last entry in each block refers to the host stack rings). 262 * The area is filled up by the kernel on NIOCREGIF, 263 * and then only read by userspace code. 264 */ 265 const ssize_t ring_ofs[0]; 266 }; 267 268 #ifndef NIOCREGIF 269 /* 270 * ioctl names and related fields 271 * 272 * NIOCGINFO takes a struct ifreq, the interface name is the input, 273 * the outputs are number of queues and number of descriptor 274 * for each queue (useful to set number of threads etc.). 275 * The info returned is only advisory and may change before 276 * the interface is bound to a file descriptor. 277 * 278 * NIOCREGIF takes an interface name within a struct ifreq, 279 * and activates netmap mode on the interface (if possible). 280 * 281 * nr_name is the name of the interface 282 * 283 * nr_tx_slots, nr_tx_slots, nr_tx_rings, nr_rx_rings 284 * indicate the configuration of the port on return. 285 * 286 * On input, non-zero values for nr_tx_rings, nr_tx_slots and the 287 * rx counterparts may be used to reconfigure the port according 288 * to the requested values, but this is not guaranteed. 289 * The actual values are returned on completion of the ioctl(). 290 * 291 * nr_ringid 292 * indicates how rings should be bound to the file descriptors. 293 * The default (0) means all physical rings of a NIC are bound. 294 * NETMAP_HW_RING plus a ring number lets you bind just 295 * a single ring pair. 296 * NETMAP_SW_RING binds only the host tx/rx rings 297 * NETMAP_NO_TX_POLL prevents select()/poll() from pushing 298 * out packets on the tx ring unless POLLOUT is specified. 299 * 300 * NETMAP_PRIV_MEM is a return value used to indicate that 301 * this ring is in a private memory region hence buffer 302 * swapping cannot be used 303 * 304 * nr_cmd is used to configure NICs attached to a VALE switch, 305 * or to dump the configuration of a VALE switch. 306 * 307 * nr_cmd = NETMAP_BDG_ATTACH and nr_name = vale*:ifname 308 * attaches the NIC to the switch, with nr_ringid specifying 309 * which rings to use 310 * 311 * nr_cmd = NETMAP_BDG_DETACH and nr_name = vale*:ifname 312 * disconnects a previously attached NIC 313 * 314 * nr_cmd = NETMAP_BDG_LIST is used to list the configuration 315 * of VALE switches, with additional arguments. 316 * 317 * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues, 318 * whose identity is set in NIOCREGIF through nr_ringid 319 * 320 * NETMAP_API is the API version. 321 */ 322 323 /* 324 * struct nmreq overlays a struct ifreq 325 */ 326 struct nmreq { 327 char nr_name[IFNAMSIZ]; 328 uint32_t nr_version; /* API version */ 329 #define NETMAP_API 5 /* current version */ 330 uint32_t nr_offset; /* nifp offset in the shared region */ 331 uint32_t nr_memsize; /* size of the shared region */ 332 uint32_t nr_tx_slots; /* slots in tx rings */ 333 uint32_t nr_rx_slots; /* slots in rx rings */ 334 uint16_t nr_tx_rings; /* number of tx rings */ 335 uint16_t nr_rx_rings; /* number of rx rings */ 336 uint16_t nr_ringid; /* ring(s) we care about */ 337 #define NETMAP_PRIV_MEM 0x8000 /* rings use private memory */ 338 #define NETMAP_HW_RING 0x4000 /* low bits indicate one hw ring */ 339 #define NETMAP_SW_RING 0x2000 /* process the sw ring */ 340 #define NETMAP_NO_TX_POLL 0x1000 /* no automatic txsync on poll */ 341 #define NETMAP_RING_MASK 0xfff /* the ring number */ 342 uint16_t nr_cmd; 343 #define NETMAP_BDG_ATTACH 1 /* attach the NIC */ 344 #define NETMAP_BDG_DETACH 2 /* detach the NIC */ 345 #define NETMAP_BDG_LOOKUP_REG 3 /* register lookup function */ 346 #define NETMAP_BDG_LIST 4 /* get bridge's info */ 347 uint16_t nr_arg1; 348 #define NETMAP_BDG_HOST 1 /* attach the host stack on ATTACH */ 349 uint16_t nr_arg2; 350 uint32_t spare2[3]; 351 }; 352 353 /* 354 * FreeBSD uses the size value embedded in the _IOWR to determine 355 * how much to copy in/out. So we need it to match the actual 356 * data structure we pass. We put some spares in the structure 357 * to ease compatibility with other versions 358 */ 359 #define NIOCGINFO _IOWR('i', 145, struct nmreq) /* return IF info */ 360 #define NIOCREGIF _IOWR('i', 146, struct nmreq) /* interface register */ 361 #define NIOCUNREGIF _IO('i', 147) /* deprecated. Was interface unregister */ 362 #define NIOCTXSYNC _IO('i', 148) /* sync tx queues */ 363 #define NIOCRXSYNC _IO('i', 149) /* sync rx queues */ 364 #endif /* !NIOCREGIF */ 365 366 #endif /* _NET_NETMAP_H_ */ 367