netmap.h (f9790aeb8869bfcedf111517bace712b390e6cc5) | netmap.h (17885a7bfde9d164e45a9833bb172215c55739f9) |
---|---|
1/* | 1/* |
2 * Copyright (C) 2011-2013 Matteo Landi, Luigi Rizzo. All rights reserved. | 2 * Copyright (C) 2011-2014 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 --- 23 unchanged lines hidden (view full) --- 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 | 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 --- 23 unchanged lines hidden (view full) --- 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#define NETMAP_API 10 /* current API version */ 43 |
|
42/* | 44/* |
45 * Some fields should be cache-aligned to reduce contention. 46 * The alignment is architecture and OS dependent, but rather than 47 * digging into OS headers to find the exact value we use an estimate 48 * that should cover most architectures. 49 */ 50#define NM_CACHE_ALIGN 128 51 52/* |
|
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 | 53 * --- Netmap data structures --- 54 * 55 * The userspace data structures used by netmap are shown below. 56 * They are allocated by the kernel and mmap()ed by userspace threads. 57 * Pointers are implemented as memory offsets or indexes, 58 * so that they can be easily dereferenced in kernel and userspace. 59 60 KERNEL (opaque, obviously) 61 62 ==================================================================== 63 | 64 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] | +--------------+ | 65 +---->+---------------+ 66 / | head,cur,tail | 67 struct netmap_if (nifp, 1 per fd) / | buf_ofs | 68 +---------------+ / | other fields | 69 | ni_tx_rings | / +===============+ 70 | ni_rx_rings | / | buf_idx, len | slot[0] 71 | | / | flags, ptr | 72 | | / +---------------+ 73 +===============+ / | buf_idx, len | slot[1] 74 | txring_ofs[0] | (rel.to nifp)--' | flags, ptr | 75 | txring_ofs[1] | +---------------+ 76 (tx+1+extra_tx entries) (num_slots entries) 77 | txring_ofs[t] | | buf_idx, len | slot[n-1] 78 +---------------+ | flags, ptr | 79 | rxring_ofs[0] | +---------------+ |
70 | rxring_ofs[1] | | 80 | rxring_ofs[1] | |
71 (ni_rx_rings+1 entries) | 81 (rx+1+extra_rx 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). --- 8 unchanged lines hidden (view full) --- 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 | 82 | rxring_ofs[r] | 83 +---------------+ 84 85 * For each "interface" (NIC, host stack, VALE switch port) attached to a 86 * file descriptor, the mmap()ed region contains a (logically readonly) 87 * struct netmap_if pointing to struct netmap_ring's. 88 * There is one netmap_ring per physical NIC ring, plus one tx/rx ring 89 * pair attached to the host stack (this pair is unused for VALE ports). --- 8 unchanged lines hidden (view full) --- 98 * is provided for user-supplied buffers in the tx path. 99 * 100 * In user space, the buffer address is computed as 101 * (char *)ring + buf_ofs + index*NETMAP_BUF_SIZE 102 */ 103 104/* 105 * 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 */ | 106 */ |
128 | |
129struct netmap_slot { 130 uint32_t buf_idx; /* buffer index */ | 107struct netmap_slot { 108 uint32_t buf_idx; /* buffer index */ |
131 uint16_t len; /* packet length */ | 109 uint16_t len; /* length for this slot */ |
132 uint16_t flags; /* buf changed, etc. */ | 110 uint16_t flags; /* buf changed, etc. */ |
111 uint64_t ptr; /* pointer for indirect buffers */ 112}; 113 114/* 115 * The following flags control how the slot is used 116 */ 117 |
|
133#define NS_BUF_CHANGED 0x0001 /* buf_idx changed */ | 118#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 | 119 /* 120 * must be set whenever buf_idx is changed (as it might be 121 * necessary to recompute the physical address and mapping) 122 */ 123 124#define NS_REPORT 0x0002 /* ask the hardware to report results */ 125 /* 126 * Request notification when slot is used by the hardware. 127 * Normally transmit completions are handled lazily and 128 * may be unreported. This flag lets us know when a slot 129 * has been sent (e.g. to terminate the sender). 130 */ 131 132#define NS_FORWARD 0x0004 /* pass packet 'forward' */ 133 /* 134 * (Only for physical ports, rx rings with NR_FORWARD set). 135 * Slot released to the kernel (i.e. before ring->head) with 136 * this flag set are passed to the peer ring (host/NIC), 137 * thus restoring the host-NIC connection for these slots. 138 * This supports efficient traffic monitoring or firewalling. 139 */ 140 141#define NS_NO_LEARN 0x0008 /* disable bridge learning */ 142 /* 143 * On a VALE switch, do not 'learn' the source port for 144 * this buffer. 145 */ 146 147#define NS_INDIRECT 0x0010 /* userspace buffer */ 148 /* 149 * (VALE tx rings only) data is in a userspace buffer, 150 * whose address is in the 'ptr' field in the slot. 151 */ 152 153#define NS_MOREFRAG 0x0020 /* packet has more fragments */ 154 /* 155 * (VALE ports only) 156 * Set on all but the last slot of a multi-segment packet. 157 * The 'len' field refers to the individual fragment. 158 */ 159 |
143#define NS_PORT_SHIFT 8 144#define NS_PORT_MASK (0xff << NS_PORT_SHIFT) | 160#define NS_PORT_SHIFT 8 161#define NS_PORT_MASK (0xff << NS_PORT_SHIFT) |
145 /* 146 * in rx rings, the high 8 bits 147 * are the number of fragments. 148 */ | 162 /* 163 * The high 8 bits of the flag, if not zero, indicate the 164 * destination port for the VALE switch, overriding 165 * the lookup table. 166 */ 167 |
149#define NS_RFRAGS(_slot) ( ((_slot)->flags >> 8) & 0xff) | 168#define NS_RFRAGS(_slot) ( ((_slot)->flags >> 8) & 0xff) |
150 uint64_t ptr; /* pointer for indirect buffers */ 151}; | 169 /* 170 * (VALE rx rings only) the high 8 bits 171 * are the number of fragments. 172 */ |
152 | 173 |
174 |
|
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. | 175/* 176 * struct netmap_ring 177 * 178 * Netmap representation of a TX or RX ring (also known as "queue"). 179 * This is a queue implemented as a fixed-size circular array. |
158 * At the software level, two fields are important: avail and cur. | 180 * At the software level the important fields are: head, cur, tail. |
159 * 160 * In TX rings: 161 * | 181 * 182 * In TX rings: 183 * |
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. | 184 * head first slot available for transmission. 185 * cur wakeup point. select() and poll() will unblock 186 * when 'tail' moves past 'cur' 187 * tail (readonly) first slot reserved to the kernel |
166 * | 188 * |
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). | 189 * [head .. tail-1] can be used for new packets to send; 190 * 'head' and 'cur' must be incremented as slots are filled 191 * with new packets to be sent; 192 * 'cur' can be moved further ahead if we need more space 193 * for new transmissions. |
174 * 175 * In RX rings: 176 * | 194 * 195 * In RX rings: 196 * |
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. | 197 * head first valid received packet 198 * cur wakeup point. select() and poll() will unblock 199 * when 'tail' moves past 'cur' 200 * tail (readonly) first slot reserved to the kernel |
181 * | 201 * |
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. | 202 * [head .. tail-1] contain received packets; 203 * 'head' and 'cur' must be incremented as slots are consumed 204 * and can be returned to the kernel; 205 * 'cur' can be moved further ahead if we want to wait for 206 * new packets without returning the previous ones. |
186 * | 207 * |
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: | 208 * 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. | 209 * The netmap_ring, and all slots and buffers in the range 210 * [head .. tail-1] are owned by the user program; 211 * the kernel only accesses them during a netmap system call 212 * and in the user thread context. |
200 * | 213 * |
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. | 214 * Other slots and buffers are reserved for use by the kernel |
212 */ 213struct 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 */ | 215 */ 216struct netmap_ring { 217 /* 218 * buf_ofs is meant to be used through macros. 219 * It contains the offset of the buffer region from this 220 * descriptor. 221 */ |
219 const ssize_t buf_ofs; | 222 const int64_t buf_ofs; |
220 const uint32_t num_slots; /* number of slots in the ring. */ | 223 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 const uint32_t nr_buf_size; 225 const uint16_t ringid; 226 const uint16_t dir; /* 0: tx, 1: rx */ |
224 | 227 |
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 */ | 228 uint32_t head; /* (u) first user slot */ 229 uint32_t cur; /* (u) wakeup point */ 230 uint32_t tail; /* (k) first kernel slot */ |
230 | 231 |
231 struct timeval ts; /* time of last *sync() */ | 232 uint32_t flags; |
232 | 233 |
234 struct timeval ts; /* (k) time of last *sync() */ 235 236 /* opaque room for a mutex or similar object */ 237 uint8_t sem[128] __attribute__((__aligned__(NM_CACHE_ALIGN))); 238 |
|
233 /* the slots follow. This struct has variable size */ 234 struct netmap_slot slot[0]; /* array of slots. */ 235}; 236 237 238/* | 239 /* the slots follow. This struct has variable size */ 240 struct netmap_slot slot[0]; /* array of slots. */ 241}; 242 243 244/* |
245 * RING FLAGS 246 */ 247#define NR_TIMESTAMP 0x0002 /* set timestamp on *sync() */ 248 /* 249 * updates the 'ts' field on each netmap syscall. This saves 250 * saves a separate gettimeofday(), and is not much worse than 251 * software timestamps generated in the interrupt handler. 252 */ 253 254#define NR_FORWARD 0x0004 /* enable NS_FORWARD for ring */ 255 /* 256 * Enables the NS_FORWARD slot flag for the ring. 257 */ 258 259 260/* |
|
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 */ 249struct 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 | 261 * Netmap representation of an interface and its queue(s). 262 * This is initialized by the kernel when binding a file 263 * descriptor to a port, and should be considered as readonly 264 * by user programs. The kernel never uses it. 265 * 266 * There is one netmap_if for each file descriptor on which we want 267 * to select/poll. 268 * select/poll operates on one or all pairs depending on the value of 269 * nmr_queueid passed on the ioctl. 270 */ 271struct netmap_if { 272 char ni_name[IFNAMSIZ]; /* name of the interface. */ 273 const uint32_t ni_version; /* API version, currently unused */ 274 const uint32_t ni_flags; /* properties */ 275#define NI_PRIV_MEM 0x1 /* private memory region */ 276 |
255 const uint32_t ni_rx_rings; /* number of rx rings */ 256 const uint32_t ni_tx_rings; /* number of tx rings */ | |
257 /* | 277 /* |
278 * The number of packet rings available in netmap mode. 279 * Physical NICs can have different numbers of tx and rx rings. 280 * Physical NICs also have a 'host' ring pair. 281 * Additionally, clients can request additional ring pairs to 282 * be used for internal communication. 283 */ 284 const uint32_t ni_tx_rings; /* number of HW tx rings */ 285 const uint32_t ni_rx_rings; /* number of HW rx rings */ 286 287 const uint32_t ni_extra_tx_rings; 288 const uint32_t ni_extra_rx_rings; 289 /* |
|
258 * The following array contains the offset of each netmap ring | 290 * 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). | 291 * from this structure, in the following order: 292 * NIC tx rings (ni_tx_rings); host tx ring (1); extra tx rings; 293 * NIC rx rings (ni_rx_rings); host tx ring (1); extra rx rings. 294 * |
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 | 295 * The area is filled up by the kernel on NIOCREGIF, 296 * and then only read by userspace code. 297 */ 298 const ssize_t ring_ofs[0]; 299}; 300 |
301 |
|
268#ifndef NIOCREGIF 269/* 270 * ioctl names and related fields 271 * | 302#ifndef NIOCREGIF 303/* 304 * ioctl names and related fields 305 * |
306 * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues, 307 * whose identity is set in NIOCREGIF through nr_ringid. 308 * These are non blocking and take no argument. 309 * |
|
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 * | 310 * NIOCGINFO takes a struct ifreq, the interface name is the input, 311 * the outputs are number of queues and number of descriptor 312 * for each queue (useful to set number of threads etc.). 313 * The info returned is only advisory and may change before 314 * the interface is bound to a file descriptor. 315 * |
278 * NIOCREGIF takes an interface name within a struct ifreq, | 316 * NIOCREGIF takes an interface name within a struct nmre, |
279 * and activates netmap mode on the interface (if possible). 280 * | 317 * and activates netmap mode on the interface (if possible). 318 * |
281 * nr_name is the name of the interface | 319 * The argument to NIOCGINFO/NIOCREGIF overlays struct ifreq so we 320 * can pass it down to other NIC-related ioctls. |
282 * | 321 * |
283 * nr_tx_slots, nr_tx_slots, nr_tx_rings, nr_rx_rings 284 * indicate the configuration of the port on return. | 322 * The actual argument (struct nmreq) has a number of options to request 323 * different functions. |
285 * | 324 * |
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(). | 325 * nr_name (in) 326 * The name of the port (em0, valeXXX:YYY, etc.) 327 * limited to IFNAMSIZ for backward compatibility. |
290 * | 328 * |
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. | 329 * nr_version (in/out) 330 * Must match NETMAP_API as used in the kernel, error otherwise. 331 * Always returns the desired value on output. |
299 * | 332 * |
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 | 333 * nr_tx_slots, nr_tx_slots, nr_tx_rings, nr_rx_rings (in/out) 334 * On input, non-zero values may be used to reconfigure the port 335 * according to the requested values, but this is not guaranteed. 336 * On output the actual values in use are reported. |
303 * | 337 * |
304 * nr_cmd is used to configure NICs attached to a VALE switch, 305 * or to dump the configuration of a VALE switch. | 338 * nr_ringid (in) 339 * Indicates how rings should be bound to the file descriptors. 340 * 0 (default) binds all physical rings 341 * NETMAP_HW_RING | ring number binds a single ring pair 342 * NETMAP_SW_RING binds only the host tx/rx rings |
306 * | 343 * |
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 | 344 * NETMAP_NO_TX_POLL can be OR-ed to make select()/poll() push 345 * packets on tx rings only if POLLOUT is set. 346 * The default is to push any pending packet. |
310 * | 347 * |
311 * nr_cmd = NETMAP_BDG_DETACH and nr_name = vale*:ifname 312 * disconnects a previously attached NIC | 348 * NETMAP_PRIV_MEM is set on return for ports that use private 349 * memory regions and cannot use buffer swapping. |
313 * | 350 * |
314 * nr_cmd = NETMAP_BDG_LIST is used to list the configuration 315 * of VALE switches, with additional arguments. | 351 * nr_cmd (in) if non-zero indicates a special command: 352 * NETMAP_BDG_ATTACH and nr_name = vale*:ifname 353 * attaches the NIC to the switch; nr_ringid specifies 354 * which rings to use. Used by vale-ctl -a ... 355 * nr_arg1 = NETMAP_BDG_HOST also attaches the host port 356 * as in vale-ctl -h ... |
316 * | 357 * |
317 * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues, 318 * whose identity is set in NIOCREGIF through nr_ringid | 358 * NETMAP_BDG_DETACH and nr_name = vale*:ifname 359 * disconnects a previously attached NIC. 360 * Used by vale-ctl -d ... |
319 * | 361 * |
320 * NETMAP_API is the API version. | 362 * NETMAP_BDG_LIST 363 * list the configuration of VALE switches. 364 * 365 * NETMAP_BDG_OFFSET XXX ? 366 * Set the offset of data in packets. Used with VALE 367 * switches where the clients use the vhost header. 368 * 369 * nr_arg1, nr_arg2 (in/out) command specific 370 * |
321 */ 322 | 371 */ 372 |
373 |
|
323/* 324 * struct nmreq overlays a struct ifreq 325 */ 326struct nmreq { 327 char nr_name[IFNAMSIZ]; 328 uint32_t nr_version; /* API version */ | 374/* 375 * struct nmreq overlays a struct ifreq 376 */ 377struct nmreq { 378 char nr_name[IFNAMSIZ]; 379 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 */ | 380 uint32_t nr_offset; /* nifp offset in the shared region */ 381 uint32_t nr_memsize; /* size of the shared region */ 382 uint32_t nr_tx_slots; /* slots in tx rings */ 383 uint32_t nr_rx_slots; /* slots in rx rings */ 384 uint16_t nr_tx_rings; /* number of tx rings */ 385 uint16_t nr_rx_rings; /* number of rx rings */ 386 uint16_t nr_ringid; /* ring(s) we care about */ 387#define NETMAP_PRIV_MEM 0x8000 /* rings use private memory */ 388#define NETMAP_HW_RING 0x4000 /* low bits indicate one hw ring */ 389#define NETMAP_SW_RING 0x2000 /* process the sw ring */ 390#define NETMAP_NO_TX_POLL 0x1000 /* no automatic txsync on poll */ 391#define NETMAP_RING_MASK 0xfff /* the ring number */ |
392 |
|
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#define NETMAP_BDG_OFFSET 5 /* set the port offset */ | 393 uint16_t nr_cmd; 394#define NETMAP_BDG_ATTACH 1 /* attach the NIC */ 395#define NETMAP_BDG_DETACH 2 /* detach the NIC */ 396#define NETMAP_BDG_LOOKUP_REG 3 /* register lookup function */ 397#define NETMAP_BDG_LIST 4 /* get bridge's info */ 398#define NETMAP_BDG_OFFSET 5 /* set the port offset */ |
399 |
|
348 uint16_t nr_arg1; 349#define NETMAP_BDG_HOST 1 /* attach the host stack on ATTACH */ 350#define NETMAP_BDG_MAX_OFFSET 12 | 400 uint16_t nr_arg1; 401#define NETMAP_BDG_HOST 1 /* attach the host stack on ATTACH */ 402#define NETMAP_BDG_MAX_OFFSET 12 |
403 |
|
351 uint16_t nr_arg2; 352 uint32_t spare2[3]; 353}; 354 | 404 uint16_t nr_arg2; 405 uint32_t spare2[3]; 406}; 407 |
408 |
|
355/* 356 * FreeBSD uses the size value embedded in the _IOWR to determine 357 * how much to copy in/out. So we need it to match the actual 358 * data structure we pass. We put some spares in the structure 359 * to ease compatibility with other versions 360 */ 361#define NIOCGINFO _IOWR('i', 145, struct nmreq) /* return IF info */ 362#define NIOCREGIF _IOWR('i', 146, struct nmreq) /* interface register */ | 409/* 410 * FreeBSD uses the size value embedded in the _IOWR to determine 411 * how much to copy in/out. So we need it to match the actual 412 * data structure we pass. We put some spares in the structure 413 * to ease compatibility with other versions 414 */ 415#define NIOCGINFO _IOWR('i', 145, struct nmreq) /* return IF info */ 416#define NIOCREGIF _IOWR('i', 146, struct nmreq) /* interface register */ |
363#define NIOCUNREGIF _IO('i', 147) /* deprecated. Was interface unregister */ | |
364#define NIOCTXSYNC _IO('i', 148) /* sync tx queues */ 365#define NIOCRXSYNC _IO('i', 149) /* sync rx queues */ 366#endif /* !NIOCREGIF */ 367 | 417#define NIOCTXSYNC _IO('i', 148) /* sync tx queues */ 418#define NIOCRXSYNC _IO('i', 149) /* sync rx queues */ 419#endif /* !NIOCREGIF */ 420 |
421 422/* 423 * Helper functions for kernel and userspace 424 */ 425 426/* 427 * check if space is available in the ring. 428 */ 429static inline int 430nm_ring_empty(struct netmap_ring *ring) 431{ 432 return (ring->cur == ring->tail); 433} 434 |
|
368#endif /* _NET_NETMAP_H_ */ | 435#endif /* _NET_NETMAP_H_ */ |