FreeBSD/Linux Kernel Cross Reference
sys/net/netmap.h
1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (C) 2011-2014 Matteo Landi, Luigi Rizzo. All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``S IS''AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 /*
30 * $FreeBSD: releng/12.0/sys/net/netmap.h 339906 2018-10-30 08:36:36Z vmaffione $
31 *
32 * Definitions of constants and the structures used by the netmap
33 * framework, for the part visible to both kernel and userspace.
34 * Detailed info on netmap is available with "man netmap" or at
35 *
36 * http://info.iet.unipi.it/~luigi/netmap/
37 *
38 * This API is also used to communicate with the VALE software switch
39 */
40
41 #ifndef _NET_NETMAP_H_
42 #define _NET_NETMAP_H_
43
44 #define NETMAP_API 12 /* current API version */
45
46 #define NETMAP_MIN_API 11 /* min and max versions accepted */
47 #define NETMAP_MAX_API 15
48 /*
49 * Some fields should be cache-aligned to reduce contention.
50 * The alignment is architecture and OS dependent, but rather than
51 * digging into OS headers to find the exact value we use an estimate
52 * that should cover most architectures.
53 */
54 #define NM_CACHE_ALIGN 128
55
56 /*
57 * --- Netmap data structures ---
58 *
59 * The userspace data structures used by netmap are shown below.
60 * They are allocated by the kernel and mmap()ed by userspace threads.
61 * Pointers are implemented as memory offsets or indexes,
62 * so that they can be easily dereferenced in kernel and userspace.
63
64 KERNEL (opaque, obviously)
65
66 ====================================================================
67 |
68 USERSPACE | struct netmap_ring
69 +---->+---------------+
70 / | head,cur,tail |
71 struct netmap_if (nifp, 1 per fd) / | buf_ofs |
72 +---------------+ / | other fields |
73 | ni_tx_rings | / +===============+
74 | ni_rx_rings | / | buf_idx, len | slot[0]
75 | | / | flags, ptr |
76 | | / +---------------+
77 +===============+ / | buf_idx, len | slot[1]
78 | txring_ofs[0] | (rel.to nifp)--' | flags, ptr |
79 | txring_ofs[1] | +---------------+
80 (tx+1 entries) (num_slots entries)
81 | txring_ofs[t] | | buf_idx, len | slot[n-1]
82 +---------------+ | flags, ptr |
83 | rxring_ofs[0] | +---------------+
84 | rxring_ofs[1] |
85 (rx+1 entries)
86 | rxring_ofs[r] |
87 +---------------+
88
89 * For each "interface" (NIC, host stack, PIPE, VALE switch port) bound to
90 * a file descriptor, the mmap()ed region contains a (logically readonly)
91 * struct netmap_if pointing to struct netmap_ring's.
92 *
93 * There is one netmap_ring per physical NIC ring, plus one tx/rx ring
94 * pair attached to the host stack (this pair is unused for non-NIC ports).
95 *
96 * All physical/host stack ports share the same memory region,
97 * so that zero-copy can be implemented between them.
98 * VALE switch ports instead have separate memory regions.
99 *
100 * The netmap_ring is the userspace-visible replica of the NIC ring.
101 * Each slot has the index of a buffer (MTU-sized and residing in the
102 * mmapped region), its length and some flags. An extra 64-bit pointer
103 * is provided for user-supplied buffers in the tx path.
104 *
105 * In user space, the buffer address is computed as
106 * (char *)ring + buf_ofs + index * NETMAP_BUF_SIZE
107 *
108 * Added in NETMAP_API 11:
109 *
110 * + NIOCREGIF can request the allocation of extra spare buffers from
111 * the same memory pool. The desired number of buffers must be in
112 * nr_arg3. The ioctl may return fewer buffers, depending on memory
113 * availability. nr_arg3 will return the actual value, and, once
114 * mapped, nifp->ni_bufs_head will be the index of the first buffer.
115 *
116 * The buffers are linked to each other using the first uint32_t
117 * as the index. On close, ni_bufs_head must point to the list of
118 * buffers to be released.
119 *
120 * + NIOCREGIF can request space for extra rings (and buffers)
121 * allocated in the same memory space. The number of extra rings
122 * is in nr_arg1, and is advisory. This is a no-op on NICs where
123 * the size of the memory space is fixed.
124 *
125 * + NIOCREGIF can attach to PIPE rings sharing the same memory
126 * space with a parent device. The ifname indicates the parent device,
127 * which must already exist. Flags in nr_flags indicate if we want to
128 * bind the master or slave side, the index (from nr_ringid)
129 * is just a cookie and does not need to be sequential.
130 *
131 * + NIOCREGIF can also attach to 'monitor' rings that replicate
132 * the content of specific rings, also from the same memory space.
133 *
134 * Extra flags in nr_flags support the above functions.
135 * Application libraries may use the following naming scheme:
136 * netmap:foo all NIC ring pairs
137 * netmap:foo^ only host ring pair
138 * netmap:foo+ all NIC ring + host ring pairs
139 * netmap:foo-k the k-th NIC ring pair
140 * netmap:foo{k PIPE ring pair k, master side
141 * netmap:foo}k PIPE ring pair k, slave side
142 *
143 * Some notes about host rings:
144 *
145 * + The RX host ring is used to store those packets that the host network
146 * stack is trying to transmit through a NIC queue, but only if that queue
147 * is currently in netmap mode. Netmap will not intercept host stack mbufs
148 * designated to NIC queues that are not in netmap mode. As a consequence,
149 * registering a netmap port with netmap:foo^ is not enough to intercept
150 * mbufs in the RX host ring; the netmap port should be registered with
151 * netmap:foo*, or another registration should be done to open at least a
152 * NIC TX queue in netmap mode.
153 *
154 * + Netmap is not currently able to deal with intercepted trasmit mbufs which
155 * require offloadings like TSO, UFO, checksumming offloadings, etc. It is
156 * responsibility of the user to disable those offloadings (e.g. using
157 * ifconfig on FreeBSD or ethtool -K on Linux) for an interface that is being
158 * used in netmap mode. If the offloadings are not disabled, GSO and/or
159 * unchecksummed packets may be dropped immediately or end up in the host RX
160 * ring, and will be dropped as soon as the packet reaches another netmap
161 * adapter.
162 */
163
164 /*
165 * struct netmap_slot is a buffer descriptor
166 */
167 struct netmap_slot {
168 uint32_t buf_idx; /* buffer index */
169 uint16_t len; /* length for this slot */
170 uint16_t flags; /* buf changed, etc. */
171 uint64_t ptr; /* pointer for indirect buffers */
172 };
173
174 /*
175 * The following flags control how the slot is used
176 */
177
178 #define NS_BUF_CHANGED 0x0001 /* buf_idx changed */
179 /*
180 * must be set whenever buf_idx is changed (as it might be
181 * necessary to recompute the physical address and mapping)
182 *
183 * It is also set by the kernel whenever the buf_idx is
184 * changed internally (e.g., by pipes). Applications may
185 * use this information to know when they can reuse the
186 * contents of previously prepared buffers.
187 */
188
189 #define NS_REPORT 0x0002 /* ask the hardware to report results */
190 /*
191 * Request notification when slot is used by the hardware.
192 * Normally transmit completions are handled lazily and
193 * may be unreported. This flag lets us know when a slot
194 * has been sent (e.g. to terminate the sender).
195 */
196
197 #define NS_FORWARD 0x0004 /* pass packet 'forward' */
198 /*
199 * (Only for physical ports, rx rings with NR_FORWARD set).
200 * Slot released to the kernel (i.e. before ring->head) with
201 * this flag set are passed to the peer ring (host/NIC),
202 * thus restoring the host-NIC connection for these slots.
203 * This supports efficient traffic monitoring or firewalling.
204 */
205
206 #define NS_NO_LEARN 0x0008 /* disable bridge learning */
207 /*
208 * On a VALE switch, do not 'learn' the source port for
209 * this buffer.
210 */
211
212 #define NS_INDIRECT 0x0010 /* userspace buffer */
213 /*
214 * (VALE tx rings only) data is in a userspace buffer,
215 * whose address is in the 'ptr' field in the slot.
216 */
217
218 #define NS_MOREFRAG 0x0020 /* packet has more fragments */
219 /*
220 * (VALE ports, ptnetmap ports and some NIC ports, e.g.
221 * ixgbe and i40e on Linux)
222 * Set on all but the last slot of a multi-segment packet.
223 * The 'len' field refers to the individual fragment.
224 */
225
226 #define NS_PORT_SHIFT 8
227 #define NS_PORT_MASK (0xff << NS_PORT_SHIFT)
228 /*
229 * The high 8 bits of the flag, if not zero, indicate the
230 * destination port for the VALE switch, overriding
231 * the lookup table.
232 */
233
234 #define NS_RFRAGS(_slot) ( ((_slot)->flags >> 8) & 0xff)
235 /*
236 * (VALE rx rings only) the high 8 bits
237 * are the number of fragments.
238 */
239
240 #define NETMAP_MAX_FRAGS 64 /* max number of fragments */
241
242
243 /*
244 * struct netmap_ring
245 *
246 * Netmap representation of a TX or RX ring (also known as "queue").
247 * This is a queue implemented as a fixed-size circular array.
248 * At the software level the important fields are: head, cur, tail.
249 *
250 * In TX rings:
251 *
252 * head first slot available for transmission.
253 * cur wakeup point. select() and poll() will unblock
254 * when 'tail' moves past 'cur'
255 * tail (readonly) first slot reserved to the kernel
256 *
257 * [head .. tail-1] can be used for new packets to send;
258 * 'head' and 'cur' must be incremented as slots are filled
259 * with new packets to be sent;
260 * 'cur' can be moved further ahead if we need more space
261 * for new transmissions. XXX todo (2014-03-12)
262 *
263 * In RX rings:
264 *
265 * head first valid received packet
266 * cur wakeup point. select() and poll() will unblock
267 * when 'tail' moves past 'cur'
268 * tail (readonly) first slot reserved to the kernel
269 *
270 * [head .. tail-1] contain received packets;
271 * 'head' and 'cur' must be incremented as slots are consumed
272 * and can be returned to the kernel;
273 * 'cur' can be moved further ahead if we want to wait for
274 * new packets without returning the previous ones.
275 *
276 * DATA OWNERSHIP/LOCKING:
277 * The netmap_ring, and all slots and buffers in the range
278 * [head .. tail-1] are owned by the user program;
279 * the kernel only accesses them during a netmap system call
280 * and in the user thread context.
281 *
282 * Other slots and buffers are reserved for use by the kernel
283 */
284 struct netmap_ring {
285 /*
286 * buf_ofs is meant to be used through macros.
287 * It contains the offset of the buffer region from this
288 * descriptor.
289 */
290 const int64_t buf_ofs;
291 const uint32_t num_slots; /* number of slots in the ring. */
292 const uint32_t nr_buf_size;
293 const uint16_t ringid;
294 const uint16_t dir; /* 0: tx, 1: rx */
295
296 uint32_t head; /* (u) first user slot */
297 uint32_t cur; /* (u) wakeup point */
298 uint32_t tail; /* (k) first kernel slot */
299
300 uint32_t flags;
301
302 struct timeval ts; /* (k) time of last *sync() */
303
304 /* opaque room for a mutex or similar object */
305 #if !defined(_WIN32) || defined(__CYGWIN__)
306 uint8_t __attribute__((__aligned__(NM_CACHE_ALIGN))) sem[128];
307 #else
308 uint8_t __declspec(align(NM_CACHE_ALIGN)) sem[128];
309 #endif
310
311 /* the slots follow. This struct has variable size */
312 struct netmap_slot slot[0]; /* array of slots. */
313 };
314
315
316 /*
317 * RING FLAGS
318 */
319 #define NR_TIMESTAMP 0x0002 /* set timestamp on *sync() */
320 /*
321 * updates the 'ts' field on each netmap syscall. This saves
322 * saves a separate gettimeofday(), and is not much worse than
323 * software timestamps generated in the interrupt handler.
324 */
325
326 #define NR_FORWARD 0x0004 /* enable NS_FORWARD for ring */
327 /*
328 * Enables the NS_FORWARD slot flag for the ring.
329 */
330
331 /*
332 * Helper functions for kernel and userspace
333 */
334
335 /*
336 * check if space is available in the ring.
337 */
338 static inline int
339 nm_ring_empty(struct netmap_ring *ring)
340 {
341 return (ring->cur == ring->tail);
342 }
343
344 /*
345 * Netmap representation of an interface and its queue(s).
346 * This is initialized by the kernel when binding a file
347 * descriptor to a port, and should be considered as readonly
348 * by user programs. The kernel never uses it.
349 *
350 * There is one netmap_if for each file descriptor on which we want
351 * to select/poll.
352 * select/poll operates on one or all pairs depending on the value of
353 * nmr_queueid passed on the ioctl.
354 */
355 struct netmap_if {
356 char ni_name[IFNAMSIZ]; /* name of the interface. */
357 const uint32_t ni_version; /* API version, currently unused */
358 const uint32_t ni_flags; /* properties */
359 #define NI_PRIV_MEM 0x1 /* private memory region */
360
361 /*
362 * The number of packet rings available in netmap mode.
363 * Physical NICs can have different numbers of tx and rx rings.
364 * Physical NICs also have a 'host' ring pair.
365 * Additionally, clients can request additional ring pairs to
366 * be used for internal communication.
367 */
368 const uint32_t ni_tx_rings; /* number of HW tx rings */
369 const uint32_t ni_rx_rings; /* number of HW rx rings */
370
371 uint32_t ni_bufs_head; /* head index for extra bufs */
372 uint32_t ni_spare1[5];
373 /*
374 * The following array contains the offset of each netmap ring
375 * from this structure, in the following order:
376 * NIC tx rings (ni_tx_rings); host tx ring (1); extra tx rings;
377 * NIC rx rings (ni_rx_rings); host tx ring (1); extra rx rings.
378 *
379 * The area is filled up by the kernel on NIOCREGIF,
380 * and then only read by userspace code.
381 */
382 const ssize_t ring_ofs[0];
383 };
384
385 /* Legacy interface to interact with a netmap control device.
386 * Included for backward compatibility. The user should not include this
387 * file directly. */
388 #include "netmap_legacy.h"
389
390 /*
391 * New API to control netmap control devices. New applications should only use
392 * nmreq_xyz structs with the NIOCCTRL ioctl() command.
393 *
394 * NIOCCTRL takes a nmreq_header struct, which contains the required
395 * API version, the name of a netmap port, a command type, and pointers
396 * to request body and options.
397 *
398 * nr_name (in)
399 * The name of the port (em0, valeXXX:YYY, eth0{pn1 etc.)
400 *
401 * nr_version (in/out)
402 * Must match NETMAP_API as used in the kernel, error otherwise.
403 * Always returns the desired value on output.
404 *
405 * nr_reqtype (in)
406 * One of the NETMAP_REQ_* command types below
407 *
408 * nr_body (in)
409 * Pointer to a command-specific struct, described by one
410 * of the struct nmreq_xyz below.
411 *
412 * nr_options (in)
413 * Command specific options, if any.
414 *
415 * A NETMAP_REQ_REGISTER command activates netmap mode on the netmap
416 * port (e.g. physical interface) specified by nmreq_header.nr_name.
417 * The request body (struct nmreq_register) has several arguments to
418 * specify how the port is to be registered.
419 *
420 * nr_tx_slots, nr_tx_slots, nr_tx_rings, nr_rx_rings (in/out)
421 * On input, non-zero values may be used to reconfigure the port
422 * according to the requested values, but this is not guaranteed.
423 * On output the actual values in use are reported.
424 *
425 * nr_mode (in)
426 * Indicate what set of rings must be bound to the netmap
427 * device (e.g. all NIC rings, host rings only, NIC and
428 * host rings, ...). Values are in NR_REG_*.
429 *
430 * nr_ringid (in)
431 * If nr_mode == NR_REG_ONE_NIC (only a single couple of TX/RX
432 * rings), indicate which NIC TX and/or RX ring is to be bound
433 * (0..nr_*x_rings-1).
434 *
435 * nr_flags (in)
436 * Indicate special options for how to open the port.
437 *
438 * NR_NO_TX_POLL can be OR-ed to make select()/poll() push
439 * packets on tx rings only if POLLOUT is set.
440 * The default is to push any pending packet.
441 *
442 * NR_DO_RX_POLL can be OR-ed to make select()/poll() release
443 * packets on rx rings also when POLLIN is NOT set.
444 * The default is to touch the rx ring only with POLLIN.
445 * Note that this is the opposite of TX because it
446 * reflects the common usage.
447 *
448 * Other options are NR_MONITOR_TX, NR_MONITOR_RX, NR_ZCOPY_MON,
449 * NR_EXCLUSIVE, NR_RX_RINGS_ONLY, NR_TX_RINGS_ONLY and
450 * NR_ACCEPT_VNET_HDR.
451 *
452 * nr_mem_id (in/out)
453 * The identity of the memory region used.
454 * On input, 0 means the system decides autonomously,
455 * other values may try to select a specific region.
456 * On return the actual value is reported.
457 * Region '1' is the global allocator, normally shared
458 * by all interfaces. Other values are private regions.
459 * If two ports the same region zero-copy is possible.
460 *
461 * nr_extra_bufs (in/out)
462 * Number of extra buffers to be allocated.
463 *
464 * The other NETMAP_REQ_* commands are described below.
465 *
466 */
467
468 /* maximum size of a request, including all options */
469 #define NETMAP_REQ_MAXSIZE 4096
470
471 /* Header common to all request options. */
472 struct nmreq_option {
473 /* Pointer ot the next option. */
474 uint64_t nro_next;
475 /* Option type. */
476 uint32_t nro_reqtype;
477 /* (out) status of the option:
478 * 0: recognized and processed
479 * !=0: errno value
480 */
481 uint32_t nro_status;
482 };
483
484 /* Header common to all requests. Do not reorder these fields, as we need
485 * the second one (nr_reqtype) to know how much to copy from/to userspace. */
486 struct nmreq_header {
487 uint16_t nr_version; /* API version */
488 uint16_t nr_reqtype; /* nmreq type (NETMAP_REQ_*) */
489 uint32_t nr_reserved; /* must be zero */
490 #define NETMAP_REQ_IFNAMSIZ 64
491 char nr_name[NETMAP_REQ_IFNAMSIZ]; /* port name */
492 uint64_t nr_options; /* command-specific options */
493 uint64_t nr_body; /* ptr to nmreq_xyz struct */
494 };
495
496 enum {
497 /* Register a netmap port with the device. */
498 NETMAP_REQ_REGISTER = 1,
499 /* Get information from a netmap port. */
500 NETMAP_REQ_PORT_INFO_GET,
501 /* Attach a netmap port to a VALE switch. */
502 NETMAP_REQ_VALE_ATTACH,
503 /* Detach a netmap port from a VALE switch. */
504 NETMAP_REQ_VALE_DETACH,
505 /* List the ports attached to a VALE switch. */
506 NETMAP_REQ_VALE_LIST,
507 /* Set the port header length (was virtio-net header length). */
508 NETMAP_REQ_PORT_HDR_SET,
509 /* Get the port header length (was virtio-net header length). */
510 NETMAP_REQ_PORT_HDR_GET,
511 /* Create a new persistent VALE port. */
512 NETMAP_REQ_VALE_NEWIF,
513 /* Delete a persistent VALE port. */
514 NETMAP_REQ_VALE_DELIF,
515 /* Enable polling kernel thread(s) on an attached VALE port. */
516 NETMAP_REQ_VALE_POLLING_ENABLE,
517 /* Disable polling kernel thread(s) on an attached VALE port. */
518 NETMAP_REQ_VALE_POLLING_DISABLE,
519 /* Get info about the pools of a memory allocator. */
520 NETMAP_REQ_POOLS_INFO_GET,
521 };
522
523 enum {
524 /* On NETMAP_REQ_REGISTER, ask netmap to use memory allocated
525 * from user-space allocated memory pools (e.g. hugepages). */
526 NETMAP_REQ_OPT_EXTMEM = 1,
527 };
528
529 /*
530 * nr_reqtype: NETMAP_REQ_REGISTER
531 * Bind (register) a netmap port to this control device.
532 */
533 struct nmreq_register {
534 uint64_t nr_offset; /* nifp offset in the shared region */
535 uint64_t nr_memsize; /* size of the shared region */
536 uint32_t nr_tx_slots; /* slots in tx rings */
537 uint32_t nr_rx_slots; /* slots in rx rings */
538 uint16_t nr_tx_rings; /* number of tx rings */
539 uint16_t nr_rx_rings; /* number of rx rings */
540
541 uint16_t nr_mem_id; /* id of the memory allocator */
542 uint16_t nr_ringid; /* ring(s) we care about */
543 uint32_t nr_mode; /* specify NR_REG_* modes */
544
545 uint64_t nr_flags; /* additional flags (see below) */
546 /* monitors use nr_ringid and nr_mode to select the rings to monitor */
547 #define NR_MONITOR_TX 0x100
548 #define NR_MONITOR_RX 0x200
549 #define NR_ZCOPY_MON 0x400
550 /* request exclusive access to the selected rings */
551 #define NR_EXCLUSIVE 0x800
552 /* request ptnetmap host support */
553 #define NR_PASSTHROUGH_HOST NR_PTNETMAP_HOST /* deprecated */
554 #define NR_PTNETMAP_HOST 0x1000
555 #define NR_RX_RINGS_ONLY 0x2000
556 #define NR_TX_RINGS_ONLY 0x4000
557 /* Applications set this flag if they are able to deal with virtio-net headers,
558 * that is send/receive frames that start with a virtio-net header.
559 * If not set, NIOCREGIF will fail with netmap ports that require applications
560 * to use those headers. If the flag is set, the application can use the
561 * NETMAP_VNET_HDR_GET command to figure out the header length. */
562 #define NR_ACCEPT_VNET_HDR 0x8000
563 /* The following two have the same meaning of NETMAP_NO_TX_POLL and
564 * NETMAP_DO_RX_POLL. */
565 #define NR_DO_RX_POLL 0x10000
566 #define NR_NO_TX_POLL 0x20000
567
568 uint32_t nr_extra_bufs; /* number of requested extra buffers */
569 };
570
571 /* Valid values for nmreq_register.nr_mode (see above). */
572 enum { NR_REG_DEFAULT = 0, /* backward compat, should not be used. */
573 NR_REG_ALL_NIC = 1,
574 NR_REG_SW = 2,
575 NR_REG_NIC_SW = 3,
576 NR_REG_ONE_NIC = 4,
577 NR_REG_PIPE_MASTER = 5, /* deprecated, use "x{y" port name syntax */
578 NR_REG_PIPE_SLAVE = 6, /* deprecated, use "x}y" port name syntax */
579 };
580
581 /* A single ioctl number is shared by all the new API command.
582 * Demultiplexing is done using the nr_hdr.nr_reqtype field.
583 * FreeBSD uses the size value embedded in the _IOWR to determine
584 * how much to copy in/out, so we define the ioctl() command
585 * specifying only nmreq_header, and copyin/copyout the rest. */
586 #define NIOCCTRL _IOWR('i', 151, struct nmreq_header)
587
588 /* The ioctl commands to sync TX/RX netmap rings.
589 * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues,
590 * whose identity is set in NIOCREGIF through nr_ringid.
591 * These are non blocking and take no argument. */
592 #define NIOCTXSYNC _IO('i', 148) /* sync tx queues */
593 #define NIOCRXSYNC _IO('i', 149) /* sync rx queues */
594
595 /*
596 * nr_reqtype: NETMAP_REQ_PORT_INFO_GET
597 * Get information about a netmap port, including number of rings.
598 * slots per ring, id of the memory allocator, etc.
599 */
600 struct nmreq_port_info_get {
601 uint64_t nr_offset; /* nifp offset in the shared region */
602 uint64_t nr_memsize; /* size of the shared region */
603 uint32_t nr_tx_slots; /* slots in tx rings */
604 uint32_t nr_rx_slots; /* slots in rx rings */
605 uint16_t nr_tx_rings; /* number of tx rings */
606 uint16_t nr_rx_rings; /* number of rx rings */
607 uint16_t nr_mem_id; /* id of the memory allocator */
608 };
609
610 #define NM_BDG_NAME "vale" /* prefix for bridge port name */
611
612 /*
613 * nr_reqtype: NETMAP_REQ_VALE_ATTACH
614 * Attach a netmap port to a VALE switch. Both the name of the netmap
615 * port and the VALE switch are specified through the nr_name argument.
616 * The attach operation could need to register a port, so at least
617 * the same arguments are available.
618 * port_index will contain the index where the port has been attached.
619 */
620 struct nmreq_vale_attach {
621 struct nmreq_register reg;
622 uint32_t port_index;
623 };
624
625 /*
626 * nr_reqtype: NETMAP_REQ_VALE_DETACH
627 * Detach a netmap port from a VALE switch. Both the name of the netmap
628 * port and the VALE switch are specified through the nr_name argument.
629 * port_index will contain the index where the port was attached.
630 */
631 struct nmreq_vale_detach {
632 uint32_t port_index;
633 };
634
635 /*
636 * nr_reqtype: NETMAP_REQ_VALE_LIST
637 * List the ports of a VALE switch.
638 */
639 struct nmreq_vale_list {
640 /* Name of the VALE port (valeXXX:YYY) or empty. */
641 uint16_t nr_bridge_idx;
642 uint32_t nr_port_idx;
643 };
644
645 /*
646 * nr_reqtype: NETMAP_REQ_PORT_HDR_SET or NETMAP_REQ_PORT_HDR_GET
647 * Set the port header length.
648 */
649 struct nmreq_port_hdr {
650 uint32_t nr_hdr_len;
651 };
652
653 /*
654 * nr_reqtype: NETMAP_REQ_VALE_NEWIF
655 * Create a new persistent VALE port.
656 */
657 struct nmreq_vale_newif {
658 uint32_t nr_tx_slots; /* slots in tx rings */
659 uint32_t nr_rx_slots; /* slots in rx rings */
660 uint16_t nr_tx_rings; /* number of tx rings */
661 uint16_t nr_rx_rings; /* number of rx rings */
662 uint16_t nr_mem_id; /* id of the memory allocator */
663 };
664
665 /*
666 * nr_reqtype: NETMAP_REQ_VALE_POLLING_ENABLE or NETMAP_REQ_VALE_POLLING_DISABLE
667 * Enable or disable polling kthreads on a VALE port.
668 */
669 struct nmreq_vale_polling {
670 uint32_t nr_mode;
671 #define NETMAP_POLLING_MODE_SINGLE_CPU 1
672 #define NETMAP_POLLING_MODE_MULTI_CPU 2
673 uint32_t nr_first_cpu_id;
674 uint32_t nr_num_polling_cpus;
675 };
676
677 /*
678 * nr_reqtype: NETMAP_REQ_POOLS_INFO_GET
679 * Get info about the pools of the memory allocator of the port bound
680 * to a given netmap control device (used i.e. by a ptnetmap-enabled
681 * hypervisor). The nr_hdr.nr_name field is ignored.
682 */
683 struct nmreq_pools_info {
684 uint64_t nr_memsize;
685 uint16_t nr_mem_id;
686 uint64_t nr_if_pool_offset;
687 uint32_t nr_if_pool_objtotal;
688 uint32_t nr_if_pool_objsize;
689 uint64_t nr_ring_pool_offset;
690 uint32_t nr_ring_pool_objtotal;
691 uint32_t nr_ring_pool_objsize;
692 uint64_t nr_buf_pool_offset;
693 uint32_t nr_buf_pool_objtotal;
694 uint32_t nr_buf_pool_objsize;
695 };
696
697 /*
698 * data for NETMAP_REQ_OPT_* options
699 */
700
701 struct nmreq_opt_extmem {
702 struct nmreq_option nro_opt; /* common header */
703 uint64_t nro_usrptr; /* (in) ptr to usr memory */
704 struct nmreq_pools_info nro_info; /* (in/out) */
705 };
706
707 #endif /* _NET_NETMAP_H_ */
Cache object: 93fc91e4642b79684973b36e577e8fef
|