1 /* 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 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 * $FreeBSD$ 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 #define NETMAP_API 11 /* current API version */ 43 44 #define NETMAP_MIN_API 11 /* min and max versions accepted */ 45 #define NETMAP_MAX_API 15 46 /* 47 * Some fields should be cache-aligned to reduce contention. 48 * The alignment is architecture and OS dependent, but rather than 49 * digging into OS headers to find the exact value we use an estimate 50 * that should cover most architectures. 51 */ 52 #define NM_CACHE_ALIGN 128 53 54 /* 55 * --- Netmap data structures --- 56 * 57 * The userspace data structures used by netmap are shown below. 58 * They are allocated by the kernel and mmap()ed by userspace threads. 59 * Pointers are implemented as memory offsets or indexes, 60 * so that they can be easily dereferenced in kernel and userspace. 61 62 KERNEL (opaque, obviously) 63 64 ==================================================================== 65 | 66 USERSPACE | struct netmap_ring 67 +---->+---------------+ 68 / | head,cur,tail | 69 struct netmap_if (nifp, 1 per fd) / | buf_ofs | 70 +---------------+ / | other fields | 71 | ni_tx_rings | / +===============+ 72 | ni_rx_rings | / | buf_idx, len | slot[0] 73 | | / | flags, ptr | 74 | | / +---------------+ 75 +===============+ / | buf_idx, len | slot[1] 76 | txring_ofs[0] | (rel.to nifp)--' | flags, ptr | 77 | txring_ofs[1] | +---------------+ 78 (tx+1 entries) (num_slots entries) 79 | txring_ofs[t] | | buf_idx, len | slot[n-1] 80 +---------------+ | flags, ptr | 81 | rxring_ofs[0] | +---------------+ 82 | rxring_ofs[1] | 83 (rx+1 entries) 84 | rxring_ofs[r] | 85 +---------------+ 86 87 * For each "interface" (NIC, host stack, PIPE, VALE switch port) bound to 88 * a file descriptor, the mmap()ed region contains a (logically readonly) 89 * struct netmap_if pointing to struct netmap_ring's. 90 * 91 * There is one netmap_ring per physical NIC ring, plus one tx/rx ring 92 * pair attached to the host stack (this pair is unused for non-NIC ports). 93 * 94 * All physical/host stack ports share the same memory region, 95 * so that zero-copy can be implemented between them. 96 * VALE switch ports instead have separate memory regions. 97 * 98 * The netmap_ring is the userspace-visible replica of the NIC ring. 99 * Each slot has the index of a buffer (MTU-sized and residing in the 100 * mmapped region), its length and some flags. An extra 64-bit pointer 101 * is provided for user-supplied buffers in the tx path. 102 * 103 * In user space, the buffer address is computed as 104 * (char *)ring + buf_ofs + index * NETMAP_BUF_SIZE 105 * 106 * Added in NETMAP_API 11: 107 * 108 * + NIOCREGIF can request the allocation of extra spare buffers from 109 * the same memory pool. The desired number of buffers must be in 110 * nr_arg3. The ioctl may return fewer buffers, depending on memory 111 * availability. nr_arg3 will return the actual value, and, once 112 * mapped, nifp->ni_bufs_head will be the index of the first buffer. 113 * 114 * The buffers are linked to each other using the first uint32_t 115 * as the index. On close, ni_bufs_head must point to the list of 116 * buffers to be released. 117 * 118 * + NIOCREGIF can request space for extra rings (and buffers) 119 * allocated in the same memory space. The number of extra rings 120 * is in nr_arg1, and is advisory. This is a no-op on NICs where 121 * the size of the memory space is fixed. 122 * 123 * + NIOCREGIF can attach to PIPE rings sharing the same memory 124 * space with a parent device. The ifname indicates the parent device, 125 * which must already exist. Flags in nr_flags indicate if we want to 126 * bind the master or slave side, the index (from nr_ringid) 127 * is just a cookie and does not need to be sequential. 128 * 129 * + NIOCREGIF can also attach to 'monitor' rings that replicate 130 * the content of specific rings, also from the same memory space. 131 * 132 * Extra flags in nr_flags support the above functions. 133 * Application libraries may use the following naming scheme: 134 * netmap:foo all NIC ring pairs 135 * netmap:foo^ only host ring pair 136 * netmap:foo+ all NIC ring + host ring pairs 137 * netmap:foo-k the k-th NIC ring pair 138 * netmap:foo{k PIPE ring pair k, master side 139 * netmap:foo}k PIPE ring pair k, slave side 140 * 141 * Some notes about host rings: 142 * 143 * + The RX host ring is used to store those packets that the host network 144 * stack is trying to transmit through a NIC queue, but only if that queue 145 * is currently in netmap mode. Netmap will not intercept host stack mbufs 146 * designated to NIC queues that are not in netmap mode. As a consequence, 147 * registering a netmap port with netmap:foo^ is not enough to intercept 148 * mbufs in the RX host ring; the netmap port should be registered with 149 * netmap:foo*, or another registration should be done to open at least a 150 * NIC TX queue in netmap mode. 151 * 152 * + Netmap is not currently able to deal with intercepted trasmit mbufs which 153 * require offloadings like TSO, UFO, checksumming offloadings, etc. It is 154 * responsibility of the user to disable those offloadings (e.g. using 155 * ifconfig on FreeBSD or ethtool -K on Linux) for an interface that is being 156 * used in netmap mode. If the offloadings are not disabled, GSO and/or 157 * unchecksummed packets may be dropped immediately or end up in the host RX 158 * ring, and will be dropped as soon as the packet reaches another netmap 159 * adapter. 160 */ 161 162 /* 163 * struct netmap_slot is a buffer descriptor 164 */ 165 struct netmap_slot { 166 uint32_t buf_idx; /* buffer index */ 167 uint16_t len; /* length for this slot */ 168 uint16_t flags; /* buf changed, etc. */ 169 uint64_t ptr; /* pointer for indirect buffers */ 170 }; 171 172 /* 173 * The following flags control how the slot is used 174 */ 175 176 #define NS_BUF_CHANGED 0x0001 /* buf_idx changed */ 177 /* 178 * must be set whenever buf_idx is changed (as it might be 179 * necessary to recompute the physical address and mapping) 180 * 181 * It is also set by the kernel whenever the buf_idx is 182 * changed internally (e.g., by pipes). Applications may 183 * use this information to know when they can reuse the 184 * contents of previously prepared buffers. 185 */ 186 187 #define NS_REPORT 0x0002 /* ask the hardware to report results */ 188 /* 189 * Request notification when slot is used by the hardware. 190 * Normally transmit completions are handled lazily and 191 * may be unreported. This flag lets us know when a slot 192 * has been sent (e.g. to terminate the sender). 193 */ 194 195 #define NS_FORWARD 0x0004 /* pass packet 'forward' */ 196 /* 197 * (Only for physical ports, rx rings with NR_FORWARD set). 198 * Slot released to the kernel (i.e. before ring->head) with 199 * this flag set are passed to the peer ring (host/NIC), 200 * thus restoring the host-NIC connection for these slots. 201 * This supports efficient traffic monitoring or firewalling. 202 */ 203 204 #define NS_NO_LEARN 0x0008 /* disable bridge learning */ 205 /* 206 * On a VALE switch, do not 'learn' the source port for 207 * this buffer. 208 */ 209 210 #define NS_INDIRECT 0x0010 /* userspace buffer */ 211 /* 212 * (VALE tx rings only) data is in a userspace buffer, 213 * whose address is in the 'ptr' field in the slot. 214 */ 215 216 #define NS_MOREFRAG 0x0020 /* packet has more fragments */ 217 /* 218 * (VALE ports only) 219 * Set on all but the last slot of a multi-segment packet. 220 * The 'len' field refers to the individual fragment. 221 */ 222 223 #define NS_PORT_SHIFT 8 224 #define NS_PORT_MASK (0xff << NS_PORT_SHIFT) 225 /* 226 * The high 8 bits of the flag, if not zero, indicate the 227 * destination port for the VALE switch, overriding 228 * the lookup table. 229 */ 230 231 #define NS_RFRAGS(_slot) ( ((_slot)->flags >> 8) & 0xff) 232 /* 233 * (VALE rx rings only) the high 8 bits 234 * are the number of fragments. 235 */ 236 237 238 /* 239 * struct netmap_ring 240 * 241 * Netmap representation of a TX or RX ring (also known as "queue"). 242 * This is a queue implemented as a fixed-size circular array. 243 * At the software level the important fields are: head, cur, tail. 244 * 245 * In TX rings: 246 * 247 * head first slot available for transmission. 248 * cur wakeup point. select() and poll() will unblock 249 * when 'tail' moves past 'cur' 250 * tail (readonly) first slot reserved to the kernel 251 * 252 * [head .. tail-1] can be used for new packets to send; 253 * 'head' and 'cur' must be incremented as slots are filled 254 * with new packets to be sent; 255 * 'cur' can be moved further ahead if we need more space 256 * for new transmissions. XXX todo (2014-03-12) 257 * 258 * In RX rings: 259 * 260 * head first valid received packet 261 * cur wakeup point. select() and poll() will unblock 262 * when 'tail' moves past 'cur' 263 * tail (readonly) first slot reserved to the kernel 264 * 265 * [head .. tail-1] contain received packets; 266 * 'head' and 'cur' must be incremented as slots are consumed 267 * and can be returned to the kernel; 268 * 'cur' can be moved further ahead if we want to wait for 269 * new packets without returning the previous ones. 270 * 271 * DATA OWNERSHIP/LOCKING: 272 * The netmap_ring, and all slots and buffers in the range 273 * [head .. tail-1] are owned by the user program; 274 * the kernel only accesses them during a netmap system call 275 * and in the user thread context. 276 * 277 * Other slots and buffers are reserved for use by the kernel 278 */ 279 struct netmap_ring { 280 /* 281 * buf_ofs is meant to be used through macros. 282 * It contains the offset of the buffer region from this 283 * descriptor. 284 */ 285 const int64_t buf_ofs; 286 const uint32_t num_slots; /* number of slots in the ring. */ 287 const uint32_t nr_buf_size; 288 const uint16_t ringid; 289 const uint16_t dir; /* 0: tx, 1: rx */ 290 291 uint32_t head; /* (u) first user slot */ 292 uint32_t cur; /* (u) wakeup point */ 293 uint32_t tail; /* (k) first kernel slot */ 294 295 uint32_t flags; 296 297 struct timeval ts; /* (k) time of last *sync() */ 298 299 /* opaque room for a mutex or similar object */ 300 #if !defined(_WIN32) || defined(__CYGWIN__) 301 uint8_t __attribute__((__aligned__(NM_CACHE_ALIGN))) sem[128]; 302 #else 303 uint8_t __declspec(align(NM_CACHE_ALIGN)) sem[128]; 304 #endif 305 306 /* the slots follow. This struct has variable size */ 307 struct netmap_slot slot[0]; /* array of slots. */ 308 }; 309 310 311 /* 312 * RING FLAGS 313 */ 314 #define NR_TIMESTAMP 0x0002 /* set timestamp on *sync() */ 315 /* 316 * updates the 'ts' field on each netmap syscall. This saves 317 * saves a separate gettimeofday(), and is not much worse than 318 * software timestamps generated in the interrupt handler. 319 */ 320 321 #define NR_FORWARD 0x0004 /* enable NS_FORWARD for ring */ 322 /* 323 * Enables the NS_FORWARD slot flag for the ring. 324 */ 325 326 327 /* 328 * Netmap representation of an interface and its queue(s). 329 * This is initialized by the kernel when binding a file 330 * descriptor to a port, and should be considered as readonly 331 * by user programs. The kernel never uses it. 332 * 333 * There is one netmap_if for each file descriptor on which we want 334 * to select/poll. 335 * select/poll operates on one or all pairs depending on the value of 336 * nmr_queueid passed on the ioctl. 337 */ 338 struct netmap_if { 339 char ni_name[IFNAMSIZ]; /* name of the interface. */ 340 const uint32_t ni_version; /* API version, currently unused */ 341 const uint32_t ni_flags; /* properties */ 342 #define NI_PRIV_MEM 0x1 /* private memory region */ 343 344 /* 345 * The number of packet rings available in netmap mode. 346 * Physical NICs can have different numbers of tx and rx rings. 347 * Physical NICs also have a 'host' ring pair. 348 * Additionally, clients can request additional ring pairs to 349 * be used for internal communication. 350 */ 351 const uint32_t ni_tx_rings; /* number of HW tx rings */ 352 const uint32_t ni_rx_rings; /* number of HW rx rings */ 353 354 uint32_t ni_bufs_head; /* head index for extra bufs */ 355 uint32_t ni_spare1[5]; 356 /* 357 * The following array contains the offset of each netmap ring 358 * from this structure, in the following order: 359 * NIC tx rings (ni_tx_rings); host tx ring (1); extra tx rings; 360 * NIC rx rings (ni_rx_rings); host tx ring (1); extra rx rings. 361 * 362 * The area is filled up by the kernel on NIOCREGIF, 363 * and then only read by userspace code. 364 */ 365 const ssize_t ring_ofs[0]; 366 }; 367 368 369 #ifndef NIOCREGIF 370 /* 371 * ioctl names and related fields 372 * 373 * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues, 374 * whose identity is set in NIOCREGIF through nr_ringid. 375 * These are non blocking and take no argument. 376 * 377 * NIOCGINFO takes a struct ifreq, the interface name is the input, 378 * the outputs are number of queues and number of descriptor 379 * for each queue (useful to set number of threads etc.). 380 * The info returned is only advisory and may change before 381 * the interface is bound to a file descriptor. 382 * 383 * NIOCREGIF takes an interface name within a struct nmre, 384 * and activates netmap mode on the interface (if possible). 385 * 386 * The argument to NIOCGINFO/NIOCREGIF overlays struct ifreq so we 387 * can pass it down to other NIC-related ioctls. 388 * 389 * The actual argument (struct nmreq) has a number of options to request 390 * different functions. 391 * The following are used in NIOCREGIF when nr_cmd == 0: 392 * 393 * nr_name (in) 394 * The name of the port (em0, valeXXX:YYY, etc.) 395 * limited to IFNAMSIZ for backward compatibility. 396 * 397 * nr_version (in/out) 398 * Must match NETMAP_API as used in the kernel, error otherwise. 399 * Always returns the desired value on output. 400 * 401 * nr_tx_slots, nr_tx_slots, nr_tx_rings, nr_rx_rings (in/out) 402 * On input, non-zero values may be used to reconfigure the port 403 * according to the requested values, but this is not guaranteed. 404 * On output the actual values in use are reported. 405 * 406 * nr_ringid (in) 407 * Indicates how rings should be bound to the file descriptors. 408 * If nr_flags != 0, then the low bits (in NETMAP_RING_MASK) 409 * are used to indicate the ring number, and nr_flags specifies 410 * the actual rings to bind. NETMAP_NO_TX_POLL is unaffected. 411 * 412 * NOTE: THE FOLLOWING (nr_flags == 0) IS DEPRECATED: 413 * If nr_flags == 0, NETMAP_HW_RING and NETMAP_SW_RING control 414 * the binding as follows: 415 * 0 (default) binds all physical rings 416 * NETMAP_HW_RING | ring number binds a single ring pair 417 * NETMAP_SW_RING binds only the host tx/rx rings 418 * 419 * NETMAP_NO_TX_POLL can be OR-ed to make select()/poll() push 420 * packets on tx rings only if POLLOUT is set. 421 * The default is to push any pending packet. 422 * 423 * NETMAP_DO_RX_POLL can be OR-ed to make select()/poll() release 424 * packets on rx rings also when POLLIN is NOT set. 425 * The default is to touch the rx ring only with POLLIN. 426 * Note that this is the opposite of TX because it 427 * reflects the common usage. 428 * 429 * NOTE: NETMAP_PRIV_MEM IS DEPRECATED, use nr_arg2 instead. 430 * NETMAP_PRIV_MEM is set on return for ports that do not use 431 * the global memory allocator. 432 * This information is not significant and applications 433 * should look at the region id in nr_arg2 434 * 435 * nr_flags is the recommended mode to indicate which rings should 436 * be bound to a file descriptor. Values are NR_REG_* 437 * 438 * nr_arg1 (in) The number of extra rings to be reserved. 439 * Especially when allocating a VALE port the system only 440 * allocates the amount of memory needed for the port. 441 * If more shared memory rings are desired (e.g. for pipes), 442 * the first invocation for the same basename/allocator 443 * should specify a suitable number. Memory cannot be 444 * extended after the first allocation without closing 445 * all ports on the same region. 446 * 447 * nr_arg2 (in/out) The identity of the memory region used. 448 * On input, 0 means the system decides autonomously, 449 * other values may try to select a specific region. 450 * On return the actual value is reported. 451 * Region '1' is the global allocator, normally shared 452 * by all interfaces. Other values are private regions. 453 * If two ports the same region zero-copy is possible. 454 * 455 * nr_arg3 (in/out) number of extra buffers to be allocated. 456 * 457 * 458 * 459 * nr_cmd (in) if non-zero indicates a special command: 460 * NETMAP_BDG_ATTACH and nr_name = vale*:ifname 461 * attaches the NIC to the switch; nr_ringid specifies 462 * which rings to use. Used by vale-ctl -a ... 463 * nr_arg1 = NETMAP_BDG_HOST also attaches the host port 464 * as in vale-ctl -h ... 465 * 466 * NETMAP_BDG_DETACH and nr_name = vale*:ifname 467 * disconnects a previously attached NIC. 468 * Used by vale-ctl -d ... 469 * 470 * NETMAP_BDG_LIST 471 * list the configuration of VALE switches. 472 * 473 * NETMAP_BDG_VNET_HDR 474 * Set the virtio-net header length used by the client 475 * of a VALE switch port. 476 * 477 * NETMAP_BDG_NEWIF 478 * create a persistent VALE port with name nr_name. 479 * Used by vale-ctl -n ... 480 * 481 * NETMAP_BDG_DELIF 482 * delete a persistent VALE port. Used by vale-ctl -d ... 483 * 484 * nr_arg1, nr_arg2, nr_arg3 (in/out) command specific 485 * 486 * 487 * 488 */ 489 490 491 /* 492 * struct nmreq overlays a struct ifreq (just the name) 493 */ 494 struct nmreq { 495 char nr_name[IFNAMSIZ]; 496 uint32_t nr_version; /* API version */ 497 uint32_t nr_offset; /* nifp offset in the shared region */ 498 uint32_t nr_memsize; /* size of the shared region */ 499 uint32_t nr_tx_slots; /* slots in tx rings */ 500 uint32_t nr_rx_slots; /* slots in rx rings */ 501 uint16_t nr_tx_rings; /* number of tx rings */ 502 uint16_t nr_rx_rings; /* number of rx rings */ 503 504 uint16_t nr_ringid; /* ring(s) we care about */ 505 #define NETMAP_HW_RING 0x4000 /* single NIC ring pair */ 506 #define NETMAP_SW_RING 0x2000 /* only host ring pair */ 507 508 #define NETMAP_RING_MASK 0x0fff /* the ring number */ 509 510 #define NETMAP_NO_TX_POLL 0x1000 /* no automatic txsync on poll */ 511 512 #define NETMAP_DO_RX_POLL 0x8000 /* DO automatic rxsync on poll */ 513 514 uint16_t nr_cmd; 515 #define NETMAP_BDG_ATTACH 1 /* attach the NIC */ 516 #define NETMAP_BDG_DETACH 2 /* detach the NIC */ 517 #define NETMAP_BDG_REGOPS 3 /* register bridge callbacks */ 518 #define NETMAP_BDG_LIST 4 /* get bridge's info */ 519 #define NETMAP_BDG_VNET_HDR 5 /* set the port virtio-net-hdr length */ 520 #define NETMAP_BDG_OFFSET NETMAP_BDG_VNET_HDR /* deprecated alias */ 521 #define NETMAP_BDG_NEWIF 6 /* create a virtual port */ 522 #define NETMAP_BDG_DELIF 7 /* destroy a virtual port */ 523 #define NETMAP_PT_HOST_CREATE 8 /* create ptnetmap kthreads */ 524 #define NETMAP_PT_HOST_DELETE 9 /* delete ptnetmap kthreads */ 525 #define NETMAP_BDG_POLLING_ON 10 /* delete polling kthread */ 526 #define NETMAP_BDG_POLLING_OFF 11 /* delete polling kthread */ 527 #define NETMAP_VNET_HDR_GET 12 /* get the port virtio-net-hdr length */ 528 #define NETMAP_POOLS_INFO_GET 13 /* get memory allocator pools info */ 529 uint16_t nr_arg1; /* reserve extra rings in NIOCREGIF */ 530 #define NETMAP_BDG_HOST 1 /* attach the host stack on ATTACH */ 531 532 uint16_t nr_arg2; 533 uint32_t nr_arg3; /* req. extra buffers in NIOCREGIF */ 534 uint32_t nr_flags; 535 /* various modes, extends nr_ringid */ 536 uint32_t spare2[1]; 537 }; 538 539 #define NR_REG_MASK 0xf /* values for nr_flags */ 540 enum { NR_REG_DEFAULT = 0, /* backward compat, should not be used. */ 541 NR_REG_ALL_NIC = 1, 542 NR_REG_SW = 2, 543 NR_REG_NIC_SW = 3, 544 NR_REG_ONE_NIC = 4, 545 NR_REG_PIPE_MASTER = 5, 546 NR_REG_PIPE_SLAVE = 6, 547 }; 548 /* monitor uses the NR_REG to select the rings to monitor */ 549 #define NR_MONITOR_TX 0x100 550 #define NR_MONITOR_RX 0x200 551 #define NR_ZCOPY_MON 0x400 552 /* request exclusive access to the selected rings */ 553 #define NR_EXCLUSIVE 0x800 554 /* request ptnetmap host support */ 555 #define NR_PASSTHROUGH_HOST NR_PTNETMAP_HOST /* deprecated */ 556 #define NR_PTNETMAP_HOST 0x1000 557 #define NR_RX_RINGS_ONLY 0x2000 558 #define NR_TX_RINGS_ONLY 0x4000 559 /* Applications set this flag if they are able to deal with virtio-net headers, 560 * that is send/receive frames that start with a virtio-net header. 561 * If not set, NIOCREGIF will fail with netmap ports that require applications 562 * to use those headers. If the flag is set, the application can use the 563 * NETMAP_VNET_HDR_GET command to figure out the header length. */ 564 #define NR_ACCEPT_VNET_HDR 0x8000 565 566 #define NM_BDG_NAME "vale" /* prefix for bridge port name */ 567 568 /* 569 * Windows does not have _IOWR(). _IO(), _IOW() and _IOR() are defined 570 * in ws2def.h but not sure if they are in the form we need. 571 * XXX so we redefine them 572 * in a convenient way to use for DeviceIoControl signatures 573 */ 574 #ifdef _WIN32 575 #undef _IO // ws2def.h 576 #define _WIN_NM_IOCTL_TYPE 40000 577 #define _IO(_c, _n) CTL_CODE(_WIN_NM_IOCTL_TYPE, ((_n) + 0x800) , \ 578 METHOD_BUFFERED, FILE_ANY_ACCESS ) 579 #define _IO_direct(_c, _n) CTL_CODE(_WIN_NM_IOCTL_TYPE, ((_n) + 0x800) , \ 580 METHOD_OUT_DIRECT, FILE_ANY_ACCESS ) 581 582 #define _IOWR(_c, _n, _s) _IO(_c, _n) 583 584 /* We havesome internal sysctl in addition to the externally visible ones */ 585 #define NETMAP_MMAP _IO_direct('i', 160) // note METHOD_OUT_DIRECT 586 #define NETMAP_POLL _IO('i', 162) 587 588 /* and also two setsockopt for sysctl emulation */ 589 #define NETMAP_SETSOCKOPT _IO('i', 140) 590 #define NETMAP_GETSOCKOPT _IO('i', 141) 591 592 593 //These linknames are for the Netmap Core Driver 594 #define NETMAP_NT_DEVICE_NAME L"\\Device\\NETMAP" 595 #define NETMAP_DOS_DEVICE_NAME L"\\DosDevices\\netmap" 596 597 //Definition of a structure used to pass a virtual address within an IOCTL 598 typedef struct _MEMORY_ENTRY { 599 PVOID pUsermodeVirtualAddress; 600 } MEMORY_ENTRY, *PMEMORY_ENTRY; 601 602 typedef struct _POLL_REQUEST_DATA { 603 int events; 604 int timeout; 605 int revents; 606 } POLL_REQUEST_DATA; 607 608 #endif /* _WIN32 */ 609 610 /* 611 * FreeBSD uses the size value embedded in the _IOWR to determine 612 * how much to copy in/out. So we need it to match the actual 613 * data structure we pass. We put some spares in the structure 614 * to ease compatibility with other versions 615 */ 616 #define NIOCGINFO _IOWR('i', 145, struct nmreq) /* return IF info */ 617 #define NIOCREGIF _IOWR('i', 146, struct nmreq) /* interface register */ 618 #define NIOCTXSYNC _IO('i', 148) /* sync tx queues */ 619 #define NIOCRXSYNC _IO('i', 149) /* sync rx queues */ 620 #define NIOCCONFIG _IOWR('i',150, struct nm_ifreq) /* for ext. modules */ 621 #endif /* !NIOCREGIF */ 622 623 624 /* 625 * Helper functions for kernel and userspace 626 */ 627 628 /* 629 * check if space is available in the ring. 630 */ 631 static inline int 632 nm_ring_empty(struct netmap_ring *ring) 633 { 634 return (ring->cur == ring->tail); 635 } 636 637 /* 638 * Opaque structure that is passed to an external kernel 639 * module via ioctl(fd, NIOCCONFIG, req) for a user-owned 640 * bridge port (at this point ephemeral VALE interface). 641 */ 642 #define NM_IFRDATA_LEN 256 643 struct nm_ifreq { 644 char nifr_name[IFNAMSIZ]; 645 char data[NM_IFRDATA_LEN]; 646 }; 647 648 #endif /* _NET_NETMAP_H_ */ 649