1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (C) 2011-2014 Matteo Landi, Luigi Rizzo 5 * Copyright (C) 2013-2016 Universita` di Pisa 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30 /* 31 * $FreeBSD$ 32 * 33 * The header contains the definitions of constants and function 34 * prototypes used only in kernelspace. 35 */ 36 37 #ifndef _NET_NETMAP_KERN_H_ 38 #define _NET_NETMAP_KERN_H_ 39 40 #if defined(linux) 41 42 #if defined(CONFIG_NETMAP_EXTMEM) 43 #define WITH_EXTMEM 44 #endif 45 #if defined(CONFIG_NETMAP_VALE) 46 #define WITH_VALE 47 #endif 48 #if defined(CONFIG_NETMAP_PIPE) 49 #define WITH_PIPES 50 #endif 51 #if defined(CONFIG_NETMAP_MONITOR) 52 #define WITH_MONITOR 53 #endif 54 #if defined(CONFIG_NETMAP_GENERIC) 55 #define WITH_GENERIC 56 #endif 57 #if defined(CONFIG_NETMAP_PTNETMAP) 58 #define WITH_PTNETMAP 59 #endif 60 #if defined(CONFIG_NETMAP_SINK) 61 #define WITH_SINK 62 #endif 63 #if defined(CONFIG_NETMAP_NULL) 64 #define WITH_NMNULL 65 #endif 66 67 #elif defined (_WIN32) 68 #define WITH_VALE // comment out to disable VALE support 69 #define WITH_PIPES 70 #define WITH_MONITOR 71 #define WITH_GENERIC 72 #define WITH_NMNULL 73 74 #else /* neither linux nor windows */ 75 #define WITH_VALE // comment out to disable VALE support 76 #define WITH_PIPES 77 #define WITH_MONITOR 78 #define WITH_GENERIC 79 #define WITH_EXTMEM 80 #define WITH_NMNULL 81 #endif 82 83 #if defined(__FreeBSD__) 84 #include <sys/selinfo.h> 85 86 #define likely(x) __builtin_expect((long)!!(x), 1L) 87 #define unlikely(x) __builtin_expect((long)!!(x), 0L) 88 #define __user 89 90 #define NM_LOCK_T struct mtx /* low level spinlock, used to protect queues */ 91 92 #define NM_MTX_T struct sx /* OS-specific mutex (sleepable) */ 93 #define NM_MTX_INIT(m) sx_init(&(m), #m) 94 #define NM_MTX_DESTROY(m) sx_destroy(&(m)) 95 #define NM_MTX_LOCK(m) sx_xlock(&(m)) 96 #define NM_MTX_SPINLOCK(m) while (!sx_try_xlock(&(m))) ; 97 #define NM_MTX_UNLOCK(m) sx_xunlock(&(m)) 98 #define NM_MTX_ASSERT(m) sx_assert(&(m), SA_XLOCKED) 99 100 #define NM_SELINFO_T struct nm_selinfo 101 #define NM_SELRECORD_T struct thread 102 #define MBUF_LEN(m) ((m)->m_pkthdr.len) 103 #define MBUF_TXQ(m) ((m)->m_pkthdr.flowid) 104 #define MBUF_TRANSMIT(na, ifp, m) ((na)->if_transmit(ifp, m)) 105 #define GEN_TX_MBUF_IFP(m) ((m)->m_pkthdr.rcvif) 106 107 #define NM_ATOMIC_T volatile int /* required by atomic/bitops.h */ 108 /* atomic operations */ 109 #include <machine/atomic.h> 110 #define NM_ATOMIC_TEST_AND_SET(p) (!atomic_cmpset_acq_int((p), 0, 1)) 111 #define NM_ATOMIC_CLEAR(p) atomic_store_rel_int((p), 0) 112 113 #if __FreeBSD_version >= 1100030 114 #define WNA(_ifp) (_ifp)->if_netmap 115 #else /* older FreeBSD */ 116 #define WNA(_ifp) (_ifp)->if_pspare[0] 117 #endif /* older FreeBSD */ 118 119 #if __FreeBSD_version >= 1100005 120 struct netmap_adapter *netmap_getna(if_t ifp); 121 #endif 122 123 #if __FreeBSD_version >= 1100027 124 #define MBUF_REFCNT(m) ((m)->m_ext.ext_count) 125 #define SET_MBUF_REFCNT(m, x) (m)->m_ext.ext_count = x 126 #else 127 #define MBUF_REFCNT(m) ((m)->m_ext.ref_cnt ? *((m)->m_ext.ref_cnt) : -1) 128 #define SET_MBUF_REFCNT(m, x) *((m)->m_ext.ref_cnt) = x 129 #endif 130 131 #define MBUF_QUEUED(m) 1 132 133 struct nm_selinfo { 134 /* Support for select(2) and poll(2). */ 135 struct selinfo si; 136 /* Support for kqueue(9). See comments in netmap_freebsd.c */ 137 struct taskqueue *ntfytq; 138 struct task ntfytask; 139 struct mtx m; 140 char mtxname[32]; 141 int kqueue_users; 142 }; 143 144 145 struct hrtimer { 146 /* Not used in FreeBSD. */ 147 }; 148 149 #define NM_BNS_GET(b) 150 #define NM_BNS_PUT(b) 151 152 #elif defined (linux) 153 154 #define NM_LOCK_T safe_spinlock_t // see bsd_glue.h 155 #define NM_SELINFO_T wait_queue_head_t 156 #define MBUF_LEN(m) ((m)->len) 157 #define MBUF_TRANSMIT(na, ifp, m) \ 158 ({ \ 159 /* Avoid infinite recursion with generic. */ \ 160 m->priority = NM_MAGIC_PRIORITY_TX; \ 161 (((struct net_device_ops *)(na)->if_transmit)->ndo_start_xmit(m, ifp)); \ 162 0; \ 163 }) 164 165 /* See explanation in nm_os_generic_xmit_frame. */ 166 #define GEN_TX_MBUF_IFP(m) ((struct ifnet *)skb_shinfo(m)->destructor_arg) 167 168 #define NM_ATOMIC_T volatile long unsigned int 169 170 #define NM_MTX_T struct mutex /* OS-specific sleepable lock */ 171 #define NM_MTX_INIT(m) mutex_init(&(m)) 172 #define NM_MTX_DESTROY(m) do { (void)(m); } while (0) 173 #define NM_MTX_LOCK(m) mutex_lock(&(m)) 174 #define NM_MTX_UNLOCK(m) mutex_unlock(&(m)) 175 #define NM_MTX_ASSERT(m) mutex_is_locked(&(m)) 176 177 #ifndef DEV_NETMAP 178 #define DEV_NETMAP 179 #endif /* DEV_NETMAP */ 180 181 #elif defined (__APPLE__) 182 183 #warning apple support is incomplete. 184 #define likely(x) __builtin_expect(!!(x), 1) 185 #define unlikely(x) __builtin_expect(!!(x), 0) 186 #define NM_LOCK_T IOLock * 187 #define NM_SELINFO_T struct selinfo 188 #define MBUF_LEN(m) ((m)->m_pkthdr.len) 189 190 #elif defined (_WIN32) 191 #include "../../../WINDOWS/win_glue.h" 192 193 #define NM_SELRECORD_T IO_STACK_LOCATION 194 #define NM_SELINFO_T win_SELINFO // see win_glue.h 195 #define NM_LOCK_T win_spinlock_t // see win_glue.h 196 #define NM_MTX_T KGUARDED_MUTEX /* OS-specific mutex (sleepable) */ 197 198 #define NM_MTX_INIT(m) KeInitializeGuardedMutex(&m); 199 #define NM_MTX_DESTROY(m) do { (void)(m); } while (0) 200 #define NM_MTX_LOCK(m) KeAcquireGuardedMutex(&(m)) 201 #define NM_MTX_UNLOCK(m) KeReleaseGuardedMutex(&(m)) 202 #define NM_MTX_ASSERT(m) assert(&m.Count>0) 203 204 //These linknames are for the NDIS driver 205 #define NETMAP_NDIS_LINKNAME_STRING L"\\DosDevices\\NMAPNDIS" 206 #define NETMAP_NDIS_NTDEVICE_STRING L"\\Device\\NMAPNDIS" 207 208 //Definition of internal driver-to-driver ioctl codes 209 #define NETMAP_KERNEL_XCHANGE_POINTERS _IO('i', 180) 210 #define NETMAP_KERNEL_SEND_SHUTDOWN_SIGNAL _IO_direct('i', 195) 211 212 typedef struct hrtimer{ 213 KTIMER timer; 214 BOOLEAN active; 215 KDPC deferred_proc; 216 }; 217 218 /* MSVC does not have likely/unlikely support */ 219 #ifdef _MSC_VER 220 #define likely(x) (x) 221 #define unlikely(x) (x) 222 #else 223 #define likely(x) __builtin_expect((long)!!(x), 1L) 224 #define unlikely(x) __builtin_expect((long)!!(x), 0L) 225 #endif //_MSC_VER 226 227 #else 228 229 #error unsupported platform 230 231 #endif /* end - platform-specific code */ 232 233 #ifndef _WIN32 /* support for emulated sysctl */ 234 #define SYSBEGIN(x) 235 #define SYSEND 236 #endif /* _WIN32 */ 237 238 #define NM_ACCESS_ONCE(x) (*(volatile __typeof__(x) *)&(x)) 239 240 #define NMG_LOCK_T NM_MTX_T 241 #define NMG_LOCK_INIT() NM_MTX_INIT(netmap_global_lock) 242 #define NMG_LOCK_DESTROY() NM_MTX_DESTROY(netmap_global_lock) 243 #define NMG_LOCK() NM_MTX_LOCK(netmap_global_lock) 244 #define NMG_UNLOCK() NM_MTX_UNLOCK(netmap_global_lock) 245 #define NMG_LOCK_ASSERT() NM_MTX_ASSERT(netmap_global_lock) 246 247 #if defined(__FreeBSD__) 248 #define nm_prerr_int printf 249 #define nm_prinf_int printf 250 #elif defined (_WIN32) 251 #define nm_prerr_int DbgPrint 252 #define nm_prinf_int DbgPrint 253 #elif defined(linux) 254 #define nm_prerr_int(fmt, arg...) printk(KERN_ERR fmt, ##arg) 255 #define nm_prinf_int(fmt, arg...) printk(KERN_INFO fmt, ##arg) 256 #endif 257 258 #define nm_prinf(format, ...) \ 259 do { \ 260 struct timeval __xxts; \ 261 microtime(&__xxts); \ 262 nm_prinf_int("%03d.%06d [%4d] %-25s " format "\n",\ 263 (int)__xxts.tv_sec % 1000, (int)__xxts.tv_usec, \ 264 __LINE__, __FUNCTION__, ##__VA_ARGS__); \ 265 } while (0) 266 267 #define nm_prerr(format, ...) \ 268 do { \ 269 struct timeval __xxts; \ 270 microtime(&__xxts); \ 271 nm_prerr_int("%03d.%06d [%4d] %-25s " format "\n",\ 272 (int)__xxts.tv_sec % 1000, (int)__xxts.tv_usec, \ 273 __LINE__, __FUNCTION__, ##__VA_ARGS__); \ 274 } while (0) 275 276 /* Disabled printf (used to be nm_prdis). */ 277 #define nm_prdis(format, ...) 278 279 /* Rate limited, lps indicates how many per second. */ 280 #define nm_prlim(lps, format, ...) \ 281 do { \ 282 static int t0, __cnt; \ 283 if (t0 != time_second) { \ 284 t0 = time_second; \ 285 __cnt = 0; \ 286 } \ 287 if (__cnt++ < lps) \ 288 nm_prinf(format, ##__VA_ARGS__); \ 289 } while (0) 290 291 struct netmap_adapter; 292 struct nm_bdg_fwd; 293 struct nm_bridge; 294 struct netmap_priv_d; 295 struct nm_bdg_args; 296 297 /* os-specific NM_SELINFO_T initialzation/destruction functions */ 298 int nm_os_selinfo_init(NM_SELINFO_T *, const char *name); 299 void nm_os_selinfo_uninit(NM_SELINFO_T *); 300 301 const char *nm_dump_buf(char *p, int len, int lim, char *dst); 302 303 void nm_os_selwakeup(NM_SELINFO_T *si); 304 void nm_os_selrecord(NM_SELRECORD_T *sr, NM_SELINFO_T *si); 305 306 int nm_os_ifnet_init(void); 307 void nm_os_ifnet_fini(void); 308 void nm_os_ifnet_lock(void); 309 void nm_os_ifnet_unlock(void); 310 311 unsigned nm_os_ifnet_mtu(struct ifnet *ifp); 312 313 void nm_os_get_module(void); 314 void nm_os_put_module(void); 315 316 void netmap_make_zombie(struct ifnet *); 317 void netmap_undo_zombie(struct ifnet *); 318 319 /* os independent alloc/realloc/free */ 320 void *nm_os_malloc(size_t); 321 void *nm_os_vmalloc(size_t); 322 void *nm_os_realloc(void *, size_t new_size, size_t old_size); 323 void nm_os_free(void *); 324 void nm_os_vfree(void *); 325 326 /* os specific attach/detach enter/exit-netmap-mode routines */ 327 void nm_os_onattach(struct ifnet *); 328 void nm_os_ondetach(struct ifnet *); 329 void nm_os_onenter(struct ifnet *); 330 void nm_os_onexit(struct ifnet *); 331 332 /* passes a packet up to the host stack. 333 * If the packet is sent (or dropped) immediately it returns NULL, 334 * otherwise it links the packet to prev and returns m. 335 * In this case, a final call with m=NULL and prev != NULL will send up 336 * the entire chain to the host stack. 337 */ 338 void *nm_os_send_up(struct ifnet *, struct mbuf *m, struct mbuf *prev); 339 340 int nm_os_mbuf_has_seg_offld(struct mbuf *m); 341 int nm_os_mbuf_has_csum_offld(struct mbuf *m); 342 343 #include "netmap_mbq.h" 344 345 extern NMG_LOCK_T netmap_global_lock; 346 347 enum txrx { NR_RX = 0, NR_TX = 1, NR_TXRX }; 348 349 static __inline const char* 350 nm_txrx2str(enum txrx t) 351 { 352 return (t== NR_RX ? "RX" : "TX"); 353 } 354 355 static __inline enum txrx 356 nm_txrx_swap(enum txrx t) 357 { 358 return (t== NR_RX ? NR_TX : NR_RX); 359 } 360 361 #define for_rx_tx(t) for ((t) = 0; (t) < NR_TXRX; (t)++) 362 363 #ifdef WITH_MONITOR 364 struct netmap_zmon_list { 365 struct netmap_kring *next; 366 struct netmap_kring *prev; 367 }; 368 #endif /* WITH_MONITOR */ 369 370 /* 371 * private, kernel view of a ring. Keeps track of the status of 372 * a ring across system calls. 373 * 374 * nr_hwcur index of the next buffer to refill. 375 * It corresponds to ring->head 376 * at the time the system call returns. 377 * 378 * nr_hwtail index of the first buffer owned by the kernel. 379 * On RX, hwcur->hwtail are receive buffers 380 * not yet released. hwcur is advanced following 381 * ring->head, hwtail is advanced on incoming packets, 382 * and a wakeup is generated when hwtail passes ring->cur 383 * On TX, hwcur->rcur have been filled by the sender 384 * but not sent yet to the NIC; rcur->hwtail are available 385 * for new transmissions, and hwtail->hwcur-1 are pending 386 * transmissions not yet acknowledged. 387 * 388 * The indexes in the NIC and netmap rings are offset by nkr_hwofs slots. 389 * This is so that, on a reset, buffers owned by userspace are not 390 * modified by the kernel. In particular: 391 * RX rings: the next empty buffer (hwtail + hwofs) coincides with 392 * the next empty buffer as known by the hardware (next_to_check or so). 393 * TX rings: hwcur + hwofs coincides with next_to_send 394 * 395 * The following fields are used to implement lock-free copy of packets 396 * from input to output ports in VALE switch: 397 * nkr_hwlease buffer after the last one being copied. 398 * A writer in nm_bdg_flush reserves N buffers 399 * from nr_hwlease, advances it, then does the 400 * copy outside the lock. 401 * In RX rings (used for VALE ports), 402 * nkr_hwtail <= nkr_hwlease < nkr_hwcur+N-1 403 * In TX rings (used for NIC or host stack ports) 404 * nkr_hwcur <= nkr_hwlease < nkr_hwtail 405 * nkr_leases array of nkr_num_slots where writers can report 406 * completion of their block. NR_NOSLOT (~0) indicates 407 * that the writer has not finished yet 408 * nkr_lease_idx index of next free slot in nr_leases, to be assigned 409 * 410 * The kring is manipulated by txsync/rxsync and generic netmap function. 411 * 412 * Concurrent rxsync or txsync on the same ring are prevented through 413 * by nm_kr_(try)lock() which in turn uses nr_busy. This is all we need 414 * for NIC rings, and for TX rings attached to the host stack. 415 * 416 * RX rings attached to the host stack use an mbq (rx_queue) on both 417 * rxsync_from_host() and netmap_transmit(). The mbq is protected 418 * by its internal lock. 419 * 420 * RX rings attached to the VALE switch are accessed by both senders 421 * and receiver. They are protected through the q_lock on the RX ring. 422 */ 423 struct netmap_kring { 424 struct netmap_ring *ring; 425 426 uint32_t nr_hwcur; /* should be nr_hwhead */ 427 uint32_t nr_hwtail; 428 429 /* 430 * Copies of values in user rings, so we do not need to look 431 * at the ring (which could be modified). These are set in the 432 * *sync_prologue()/finalize() routines. 433 */ 434 uint32_t rhead; 435 uint32_t rcur; 436 uint32_t rtail; 437 438 uint32_t nr_kflags; /* private driver flags */ 439 #define NKR_PENDINTR 0x1 // Pending interrupt. 440 #define NKR_EXCLUSIVE 0x2 /* exclusive binding */ 441 #define NKR_FORWARD 0x4 /* (host ring only) there are 442 packets to forward 443 */ 444 #define NKR_NEEDRING 0x8 /* ring needed even if users==0 445 * (used internally by pipes and 446 * by ptnetmap host ports) 447 */ 448 #define NKR_NOINTR 0x10 /* don't use interrupts on this ring */ 449 #define NKR_FAKERING 0x20 /* don't allocate/free buffers */ 450 451 uint32_t nr_mode; 452 uint32_t nr_pending_mode; 453 #define NKR_NETMAP_OFF 0x0 454 #define NKR_NETMAP_ON 0x1 455 456 uint32_t nkr_num_slots; 457 458 /* 459 * On a NIC reset, the NIC ring indexes may be reset but the 460 * indexes in the netmap rings remain the same. nkr_hwofs 461 * keeps track of the offset between the two. 462 */ 463 int32_t nkr_hwofs; 464 465 /* last_reclaim is opaque marker to help reduce the frequency 466 * of operations such as reclaiming tx buffers. A possible use 467 * is set it to ticks and do the reclaim only once per tick. 468 */ 469 uint64_t last_reclaim; 470 471 472 NM_SELINFO_T si; /* poll/select wait queue */ 473 NM_LOCK_T q_lock; /* protects kring and ring. */ 474 NM_ATOMIC_T nr_busy; /* prevent concurrent syscalls */ 475 476 /* the adapter the owns this kring */ 477 struct netmap_adapter *na; 478 479 /* the adapter that wants to be notified when this kring has 480 * new slots avaialable. This is usually the same as the above, 481 * but wrappers may let it point to themselves 482 */ 483 struct netmap_adapter *notify_na; 484 485 /* The following fields are for VALE switch support */ 486 struct nm_bdg_fwd *nkr_ft; 487 uint32_t *nkr_leases; 488 #define NR_NOSLOT ((uint32_t)~0) /* used in nkr_*lease* */ 489 uint32_t nkr_hwlease; 490 uint32_t nkr_lease_idx; 491 492 /* while nkr_stopped is set, no new [tr]xsync operations can 493 * be started on this kring. 494 * This is used by netmap_disable_all_rings() 495 * to find a synchronization point where critical data 496 * structures pointed to by the kring can be added or removed 497 */ 498 volatile int nkr_stopped; 499 500 /* Support for adapters without native netmap support. 501 * On tx rings we preallocate an array of tx buffers 502 * (same size as the netmap ring), on rx rings we 503 * store incoming mbufs in a queue that is drained by 504 * a rxsync. 505 */ 506 struct mbuf **tx_pool; 507 struct mbuf *tx_event; /* TX event used as a notification */ 508 NM_LOCK_T tx_event_lock; /* protects the tx_event mbuf */ 509 struct mbq rx_queue; /* intercepted rx mbufs. */ 510 511 uint32_t users; /* existing bindings for this ring */ 512 513 uint32_t ring_id; /* kring identifier */ 514 enum txrx tx; /* kind of ring (tx or rx) */ 515 char name[64]; /* diagnostic */ 516 517 /* [tx]sync callback for this kring. 518 * The default nm_kring_create callback (netmap_krings_create) 519 * sets the nm_sync callback of each hardware tx(rx) kring to 520 * the corresponding nm_txsync(nm_rxsync) taken from the 521 * netmap_adapter; moreover, it sets the sync callback 522 * of the host tx(rx) ring to netmap_txsync_to_host 523 * (netmap_rxsync_from_host). 524 * 525 * Overrides: the above configuration is not changed by 526 * any of the nm_krings_create callbacks. 527 */ 528 int (*nm_sync)(struct netmap_kring *kring, int flags); 529 int (*nm_notify)(struct netmap_kring *kring, int flags); 530 531 #ifdef WITH_PIPES 532 struct netmap_kring *pipe; /* if this is a pipe ring, 533 * pointer to the other end 534 */ 535 uint32_t pipe_tail; /* hwtail updated by the other end */ 536 #endif /* WITH_PIPES */ 537 538 int (*save_notify)(struct netmap_kring *kring, int flags); 539 540 #ifdef WITH_MONITOR 541 /* array of krings that are monitoring this kring */ 542 struct netmap_kring **monitors; 543 uint32_t max_monitors; /* current size of the monitors array */ 544 uint32_t n_monitors; /* next unused entry in the monitor array */ 545 uint32_t mon_pos[NR_TXRX]; /* index of this ring in the monitored ring array */ 546 uint32_t mon_tail; /* last seen slot on rx */ 547 548 /* circular list of zero-copy monitors */ 549 struct netmap_zmon_list zmon_list[NR_TXRX]; 550 551 /* 552 * Monitors work by intercepting the sync and notify callbacks of the 553 * monitored krings. This is implemented by replacing the pointers 554 * above and saving the previous ones in mon_* pointers below 555 */ 556 int (*mon_sync)(struct netmap_kring *kring, int flags); 557 int (*mon_notify)(struct netmap_kring *kring, int flags); 558 559 #endif 560 } 561 #ifdef _WIN32 562 __declspec(align(64)); 563 #else 564 __attribute__((__aligned__(64))); 565 #endif 566 567 /* return 1 iff the kring needs to be turned on */ 568 static inline int 569 nm_kring_pending_on(struct netmap_kring *kring) 570 { 571 return kring->nr_pending_mode == NKR_NETMAP_ON && 572 kring->nr_mode == NKR_NETMAP_OFF; 573 } 574 575 /* return 1 iff the kring needs to be turned off */ 576 static inline int 577 nm_kring_pending_off(struct netmap_kring *kring) 578 { 579 return kring->nr_pending_mode == NKR_NETMAP_OFF && 580 kring->nr_mode == NKR_NETMAP_ON; 581 } 582 583 /* return the next index, with wraparound */ 584 static inline uint32_t 585 nm_next(uint32_t i, uint32_t lim) 586 { 587 return unlikely (i == lim) ? 0 : i + 1; 588 } 589 590 591 /* return the previous index, with wraparound */ 592 static inline uint32_t 593 nm_prev(uint32_t i, uint32_t lim) 594 { 595 return unlikely (i == 0) ? lim : i - 1; 596 } 597 598 599 /* 600 * 601 * Here is the layout for the Rx and Tx rings. 602 603 RxRING TxRING 604 605 +-----------------+ +-----------------+ 606 | | | | 607 | free | | free | 608 +-----------------+ +-----------------+ 609 head->| owned by user |<-hwcur | not sent to nic |<-hwcur 610 | | | yet | 611 +-----------------+ | | 612 cur->| available to | | | 613 | user, not read | +-----------------+ 614 | yet | cur->| (being | 615 | | | prepared) | 616 | | | | 617 +-----------------+ + ------ + 618 tail->| |<-hwtail | |<-hwlease 619 | (being | ... | | ... 620 | prepared) | ... | | ... 621 +-----------------+ ... | | ... 622 | |<-hwlease +-----------------+ 623 | | tail->| |<-hwtail 624 | | | | 625 | | | | 626 | | | | 627 +-----------------+ +-----------------+ 628 629 * The cur/tail (user view) and hwcur/hwtail (kernel view) 630 * are used in the normal operation of the card. 631 * 632 * When a ring is the output of a switch port (Rx ring for 633 * a VALE port, Tx ring for the host stack or NIC), slots 634 * are reserved in blocks through 'hwlease' which points 635 * to the next unused slot. 636 * On an Rx ring, hwlease is always after hwtail, 637 * and completions cause hwtail to advance. 638 * On a Tx ring, hwlease is always between cur and hwtail, 639 * and completions cause cur to advance. 640 * 641 * nm_kr_space() returns the maximum number of slots that 642 * can be assigned. 643 * nm_kr_lease() reserves the required number of buffers, 644 * advances nkr_hwlease and also returns an entry in 645 * a circular array where completions should be reported. 646 */ 647 648 struct lut_entry; 649 #ifdef __FreeBSD__ 650 #define plut_entry lut_entry 651 #endif 652 653 struct netmap_lut { 654 struct lut_entry *lut; 655 struct plut_entry *plut; 656 uint32_t objtotal; /* max buffer index */ 657 uint32_t objsize; /* buffer size */ 658 }; 659 660 struct netmap_vp_adapter; // forward 661 struct nm_bridge; 662 663 /* Struct to be filled by nm_config callbacks. */ 664 struct nm_config_info { 665 unsigned num_tx_rings; 666 unsigned num_rx_rings; 667 unsigned num_tx_descs; 668 unsigned num_rx_descs; 669 unsigned rx_buf_maxsize; 670 }; 671 672 /* 673 * default type for the magic field. 674 * May be overriden in glue code. 675 */ 676 #ifndef NM_OS_MAGIC 677 #define NM_OS_MAGIC uint32_t 678 #endif /* !NM_OS_MAGIC */ 679 680 /* 681 * The "struct netmap_adapter" extends the "struct adapter" 682 * (or equivalent) device descriptor. 683 * It contains all base fields needed to support netmap operation. 684 * There are in fact different types of netmap adapters 685 * (native, generic, VALE switch...) so a netmap_adapter is 686 * just the first field in the derived type. 687 */ 688 struct netmap_adapter { 689 /* 690 * On linux we do not have a good way to tell if an interface 691 * is netmap-capable. So we always use the following trick: 692 * NA(ifp) points here, and the first entry (which hopefully 693 * always exists and is at least 32 bits) contains a magic 694 * value which we can use to detect that the interface is good. 695 */ 696 NM_OS_MAGIC magic; 697 uint32_t na_flags; /* enabled, and other flags */ 698 #define NAF_SKIP_INTR 1 /* use the regular interrupt handler. 699 * useful during initialization 700 */ 701 #define NAF_SW_ONLY 2 /* forward packets only to sw adapter */ 702 #define NAF_BDG_MAYSLEEP 4 /* the bridge is allowed to sleep when 703 * forwarding packets coming from this 704 * interface 705 */ 706 #define NAF_MEM_OWNER 8 /* the adapter uses its own memory area 707 * that cannot be changed 708 */ 709 #define NAF_NATIVE 16 /* the adapter is native. 710 * Virtual ports (non persistent vale ports, 711 * pipes, monitors...) should never use 712 * this flag. 713 */ 714 #define NAF_NETMAP_ON 32 /* netmap is active (either native or 715 * emulated). Where possible (e.g. FreeBSD) 716 * IFCAP_NETMAP also mirrors this flag. 717 */ 718 #define NAF_HOST_RINGS 64 /* the adapter supports the host rings */ 719 #define NAF_FORCE_NATIVE 128 /* the adapter is always NATIVE */ 720 /* free */ 721 #define NAF_MOREFRAG 512 /* the adapter supports NS_MOREFRAG */ 722 #define NAF_ZOMBIE (1U<<30) /* the nic driver has been unloaded */ 723 #define NAF_BUSY (1U<<31) /* the adapter is used internally and 724 * cannot be registered from userspace 725 */ 726 int active_fds; /* number of user-space descriptors using this 727 interface, which is equal to the number of 728 struct netmap_if objs in the mapped region. */ 729 730 u_int num_rx_rings; /* number of adapter receive rings */ 731 u_int num_tx_rings; /* number of adapter transmit rings */ 732 u_int num_host_rx_rings; /* number of host receive rings */ 733 u_int num_host_tx_rings; /* number of host transmit rings */ 734 735 u_int num_tx_desc; /* number of descriptor in each queue */ 736 u_int num_rx_desc; 737 738 /* tx_rings and rx_rings are private but allocated as a 739 * contiguous chunk of memory. Each array has N+K entries, 740 * N for the hardware rings and K for the host rings. 741 */ 742 struct netmap_kring **tx_rings; /* array of TX rings. */ 743 struct netmap_kring **rx_rings; /* array of RX rings. */ 744 745 void *tailroom; /* space below the rings array */ 746 /* (used for leases) */ 747 748 749 NM_SELINFO_T si[NR_TXRX]; /* global wait queues */ 750 751 /* count users of the global wait queues */ 752 int si_users[NR_TXRX]; 753 754 void *pdev; /* used to store pci device */ 755 756 /* copy of if_qflush and if_transmit pointers, to intercept 757 * packets from the network stack when netmap is active. 758 */ 759 int (*if_transmit)(struct ifnet *, struct mbuf *); 760 761 /* copy of if_input for netmap_send_up() */ 762 void (*if_input)(struct ifnet *, struct mbuf *); 763 764 /* Back reference to the parent ifnet struct. Used for 765 * hardware ports (emulated netmap included). */ 766 struct ifnet *ifp; /* adapter is ifp->if_softc */ 767 768 /*---- callbacks for this netmap adapter -----*/ 769 /* 770 * nm_dtor() is the cleanup routine called when destroying 771 * the adapter. 772 * Called with NMG_LOCK held. 773 * 774 * nm_register() is called on NIOCREGIF and close() to enter 775 * or exit netmap mode on the NIC 776 * Called with NNG_LOCK held. 777 * 778 * nm_txsync() pushes packets to the underlying hw/switch 779 * 780 * nm_rxsync() collects packets from the underlying hw/switch 781 * 782 * nm_config() returns configuration information from the OS 783 * Called with NMG_LOCK held. 784 * 785 * nm_krings_create() create and init the tx_rings and 786 * rx_rings arrays of kring structures. In particular, 787 * set the nm_sync callbacks for each ring. 788 * There is no need to also allocate the corresponding 789 * netmap_rings, since netmap_mem_rings_create() will always 790 * be called to provide the missing ones. 791 * Called with NNG_LOCK held. 792 * 793 * nm_krings_delete() cleanup and delete the tx_rings and rx_rings 794 * arrays 795 * Called with NMG_LOCK held. 796 * 797 * nm_notify() is used to act after data have become available 798 * (or the stopped state of the ring has changed) 799 * For hw devices this is typically a selwakeup(), 800 * but for NIC/host ports attached to a switch (or vice-versa) 801 * we also need to invoke the 'txsync' code downstream. 802 * This callback pointer is actually used only to initialize 803 * kring->nm_notify. 804 * Return values are the same as for netmap_rx_irq(). 805 */ 806 void (*nm_dtor)(struct netmap_adapter *); 807 808 int (*nm_register)(struct netmap_adapter *, int onoff); 809 void (*nm_intr)(struct netmap_adapter *, int onoff); 810 811 int (*nm_txsync)(struct netmap_kring *kring, int flags); 812 int (*nm_rxsync)(struct netmap_kring *kring, int flags); 813 int (*nm_notify)(struct netmap_kring *kring, int flags); 814 #define NAF_FORCE_READ 1 815 #define NAF_FORCE_RECLAIM 2 816 #define NAF_CAN_FORWARD_DOWN 4 817 /* return configuration information */ 818 int (*nm_config)(struct netmap_adapter *, struct nm_config_info *info); 819 int (*nm_krings_create)(struct netmap_adapter *); 820 void (*nm_krings_delete)(struct netmap_adapter *); 821 /* 822 * nm_bdg_attach() initializes the na_vp field to point 823 * to an adapter that can be attached to a VALE switch. If the 824 * current adapter is already a VALE port, na_vp is simply a cast; 825 * otherwise, na_vp points to a netmap_bwrap_adapter. 826 * If applicable, this callback also initializes na_hostvp, 827 * that can be used to connect the adapter host rings to the 828 * switch. 829 * Called with NMG_LOCK held. 830 * 831 * nm_bdg_ctl() is called on the actual attach/detach to/from 832 * to/from the switch, to perform adapter-specific 833 * initializations 834 * Called with NMG_LOCK held. 835 */ 836 int (*nm_bdg_attach)(const char *bdg_name, struct netmap_adapter *, 837 struct nm_bridge *); 838 int (*nm_bdg_ctl)(struct nmreq_header *, struct netmap_adapter *); 839 840 /* adapter used to attach this adapter to a VALE switch (if any) */ 841 struct netmap_vp_adapter *na_vp; 842 /* adapter used to attach the host rings of this adapter 843 * to a VALE switch (if any) */ 844 struct netmap_vp_adapter *na_hostvp; 845 846 /* standard refcount to control the lifetime of the adapter 847 * (it should be equal to the lifetime of the corresponding ifp) 848 */ 849 int na_refcount; 850 851 /* memory allocator (opaque) 852 * We also cache a pointer to the lut_entry for translating 853 * buffer addresses, the total number of buffers and the buffer size. 854 */ 855 struct netmap_mem_d *nm_mem; 856 struct netmap_mem_d *nm_mem_prev; 857 struct netmap_lut na_lut; 858 859 /* additional information attached to this adapter 860 * by other netmap subsystems. Currently used by 861 * bwrap, LINUX/v1000 and ptnetmap 862 */ 863 void *na_private; 864 865 /* array of pipes that have this adapter as a parent */ 866 struct netmap_pipe_adapter **na_pipes; 867 int na_next_pipe; /* next free slot in the array */ 868 int na_max_pipes; /* size of the array */ 869 870 /* Offset of ethernet header for each packet. */ 871 u_int virt_hdr_len; 872 873 /* Max number of bytes that the NIC can store in the buffer 874 * referenced by each RX descriptor. This translates to the maximum 875 * bytes that a single netmap slot can reference. Larger packets 876 * require NS_MOREFRAG support. */ 877 unsigned rx_buf_maxsize; 878 879 char name[NETMAP_REQ_IFNAMSIZ]; /* used at least by pipes */ 880 881 #ifdef WITH_MONITOR 882 unsigned long monitor_id; /* debugging */ 883 #endif 884 }; 885 886 static __inline u_int 887 nma_get_ndesc(struct netmap_adapter *na, enum txrx t) 888 { 889 return (t == NR_TX ? na->num_tx_desc : na->num_rx_desc); 890 } 891 892 static __inline void 893 nma_set_ndesc(struct netmap_adapter *na, enum txrx t, u_int v) 894 { 895 if (t == NR_TX) 896 na->num_tx_desc = v; 897 else 898 na->num_rx_desc = v; 899 } 900 901 static __inline u_int 902 nma_get_nrings(struct netmap_adapter *na, enum txrx t) 903 { 904 return (t == NR_TX ? na->num_tx_rings : na->num_rx_rings); 905 } 906 907 static __inline u_int 908 nma_get_host_nrings(struct netmap_adapter *na, enum txrx t) 909 { 910 return (t == NR_TX ? na->num_host_tx_rings : na->num_host_rx_rings); 911 } 912 913 static __inline void 914 nma_set_nrings(struct netmap_adapter *na, enum txrx t, u_int v) 915 { 916 if (t == NR_TX) 917 na->num_tx_rings = v; 918 else 919 na->num_rx_rings = v; 920 } 921 922 static __inline void 923 nma_set_host_nrings(struct netmap_adapter *na, enum txrx t, u_int v) 924 { 925 if (t == NR_TX) 926 na->num_host_tx_rings = v; 927 else 928 na->num_host_rx_rings = v; 929 } 930 931 static __inline struct netmap_kring** 932 NMR(struct netmap_adapter *na, enum txrx t) 933 { 934 return (t == NR_TX ? na->tx_rings : na->rx_rings); 935 } 936 937 int nma_intr_enable(struct netmap_adapter *na, int onoff); 938 939 /* 940 * If the NIC is owned by the kernel 941 * (i.e., bridge), neither another bridge nor user can use it; 942 * if the NIC is owned by a user, only users can share it. 943 * Evaluation must be done under NMG_LOCK(). 944 */ 945 #define NETMAP_OWNED_BY_KERN(na) ((na)->na_flags & NAF_BUSY) 946 #define NETMAP_OWNED_BY_ANY(na) \ 947 (NETMAP_OWNED_BY_KERN(na) || ((na)->active_fds > 0)) 948 949 /* 950 * derived netmap adapters for various types of ports 951 */ 952 struct netmap_vp_adapter { /* VALE software port */ 953 struct netmap_adapter up; 954 955 /* 956 * Bridge support: 957 * 958 * bdg_port is the port number used in the bridge; 959 * na_bdg points to the bridge this NA is attached to. 960 */ 961 int bdg_port; 962 struct nm_bridge *na_bdg; 963 int retry; 964 int autodelete; /* remove the ifp on last reference */ 965 966 /* Maximum Frame Size, used in bdg_mismatch_datapath() */ 967 u_int mfs; 968 /* Last source MAC on this port */ 969 uint64_t last_smac; 970 }; 971 972 973 struct netmap_hw_adapter { /* physical device */ 974 struct netmap_adapter up; 975 976 #ifdef linux 977 struct net_device_ops nm_ndo; 978 struct ethtool_ops nm_eto; 979 #endif 980 const struct ethtool_ops* save_ethtool; 981 982 int (*nm_hw_register)(struct netmap_adapter *, int onoff); 983 }; 984 985 #ifdef WITH_GENERIC 986 /* Mitigation support. */ 987 struct nm_generic_mit { 988 struct hrtimer mit_timer; 989 int mit_pending; 990 int mit_ring_idx; /* index of the ring being mitigated */ 991 struct netmap_adapter *mit_na; /* backpointer */ 992 }; 993 994 struct netmap_generic_adapter { /* emulated device */ 995 struct netmap_hw_adapter up; 996 997 /* Pointer to a previously used netmap adapter. */ 998 struct netmap_adapter *prev; 999 1000 /* Emulated netmap adapters support: 1001 * - save_if_input saves the if_input hook (FreeBSD); 1002 * - mit implements rx interrupt mitigation; 1003 */ 1004 void (*save_if_input)(struct ifnet *, struct mbuf *); 1005 1006 struct nm_generic_mit *mit; 1007 #ifdef linux 1008 netdev_tx_t (*save_start_xmit)(struct mbuf *, struct ifnet *); 1009 #endif 1010 /* Is the adapter able to use multiple RX slots to scatter 1011 * each packet pushed up by the driver? */ 1012 int rxsg; 1013 1014 /* Is the transmission path controlled by a netmap-aware 1015 * device queue (i.e. qdisc on linux)? */ 1016 int txqdisc; 1017 }; 1018 #endif /* WITH_GENERIC */ 1019 1020 static __inline u_int 1021 netmap_real_rings(struct netmap_adapter *na, enum txrx t) 1022 { 1023 return nma_get_nrings(na, t) + 1024 !!(na->na_flags & NAF_HOST_RINGS) * nma_get_host_nrings(na, t); 1025 } 1026 1027 /* account for fake rings */ 1028 static __inline u_int 1029 netmap_all_rings(struct netmap_adapter *na, enum txrx t) 1030 { 1031 return max(nma_get_nrings(na, t) + 1, netmap_real_rings(na, t)); 1032 } 1033 1034 int netmap_default_bdg_attach(const char *name, struct netmap_adapter *na, 1035 struct nm_bridge *); 1036 struct nm_bdg_polling_state; 1037 /* 1038 * Bridge wrapper for non VALE ports attached to a VALE switch. 1039 * 1040 * The real device must already have its own netmap adapter (hwna). 1041 * The bridge wrapper and the hwna adapter share the same set of 1042 * netmap rings and buffers, but they have two separate sets of 1043 * krings descriptors, with tx/rx meanings swapped: 1044 * 1045 * netmap 1046 * bwrap krings rings krings hwna 1047 * +------+ +------+ +-----+ +------+ +------+ 1048 * |tx_rings->| |\ /| |----| |<-tx_rings| 1049 * | | +------+ \ / +-----+ +------+ | | 1050 * | | X | | 1051 * | | / \ | | 1052 * | | +------+/ \+-----+ +------+ | | 1053 * |rx_rings->| | | |----| |<-rx_rings| 1054 * | | +------+ +-----+ +------+ | | 1055 * +------+ +------+ 1056 * 1057 * - packets coming from the bridge go to the brwap rx rings, 1058 * which are also the hwna tx rings. The bwrap notify callback 1059 * will then complete the hwna tx (see netmap_bwrap_notify). 1060 * 1061 * - packets coming from the outside go to the hwna rx rings, 1062 * which are also the bwrap tx rings. The (overwritten) hwna 1063 * notify method will then complete the bridge tx 1064 * (see netmap_bwrap_intr_notify). 1065 * 1066 * The bridge wrapper may optionally connect the hwna 'host' rings 1067 * to the bridge. This is done by using a second port in the 1068 * bridge and connecting it to the 'host' netmap_vp_adapter 1069 * contained in the netmap_bwrap_adapter. The brwap host adapter 1070 * cross-links the hwna host rings in the same way as shown above. 1071 * 1072 * - packets coming from the bridge and directed to the host stack 1073 * are handled by the bwrap host notify callback 1074 * (see netmap_bwrap_host_notify) 1075 * 1076 * - packets coming from the host stack are still handled by the 1077 * overwritten hwna notify callback (netmap_bwrap_intr_notify), 1078 * but are diverted to the host adapter depending on the ring number. 1079 * 1080 */ 1081 struct netmap_bwrap_adapter { 1082 struct netmap_vp_adapter up; 1083 struct netmap_vp_adapter host; /* for host rings */ 1084 struct netmap_adapter *hwna; /* the underlying device */ 1085 1086 /* 1087 * When we attach a physical interface to the bridge, we 1088 * allow the controlling process to terminate, so we need 1089 * a place to store the n_detmap_priv_d data structure. 1090 * This is only done when physical interfaces 1091 * are attached to a bridge. 1092 */ 1093 struct netmap_priv_d *na_kpriv; 1094 struct nm_bdg_polling_state *na_polling_state; 1095 /* we overwrite the hwna->na_vp pointer, so we save 1096 * here its original value, to be restored at detach 1097 */ 1098 struct netmap_vp_adapter *saved_na_vp; 1099 }; 1100 int nm_bdg_polling(struct nmreq_header *hdr); 1101 1102 #ifdef WITH_VALE 1103 int netmap_vale_attach(struct nmreq_header *hdr, void *auth_token); 1104 int netmap_vale_detach(struct nmreq_header *hdr, void *auth_token); 1105 int netmap_vale_list(struct nmreq_header *hdr); 1106 int netmap_vi_create(struct nmreq_header *hdr, int); 1107 int nm_vi_create(struct nmreq_header *); 1108 int nm_vi_destroy(const char *name); 1109 #else /* !WITH_VALE */ 1110 #define netmap_vi_create(hdr, a) (EOPNOTSUPP) 1111 #endif /* WITH_VALE */ 1112 1113 #ifdef WITH_PIPES 1114 1115 #define NM_MAXPIPES 64 /* max number of pipes per adapter */ 1116 1117 struct netmap_pipe_adapter { 1118 /* pipe identifier is up.name */ 1119 struct netmap_adapter up; 1120 1121 #define NM_PIPE_ROLE_MASTER 0x1 1122 #define NM_PIPE_ROLE_SLAVE 0x2 1123 int role; /* either NM_PIPE_ROLE_MASTER or NM_PIPE_ROLE_SLAVE */ 1124 1125 struct netmap_adapter *parent; /* adapter that owns the memory */ 1126 struct netmap_pipe_adapter *peer; /* the other end of the pipe */ 1127 int peer_ref; /* 1 iff we are holding a ref to the peer */ 1128 struct ifnet *parent_ifp; /* maybe null */ 1129 1130 u_int parent_slot; /* index in the parent pipe array */ 1131 }; 1132 1133 #endif /* WITH_PIPES */ 1134 1135 #ifdef WITH_NMNULL 1136 struct netmap_null_adapter { 1137 struct netmap_adapter up; 1138 }; 1139 #endif /* WITH_NMNULL */ 1140 1141 1142 /* return slots reserved to rx clients; used in drivers */ 1143 static inline uint32_t 1144 nm_kr_rxspace(struct netmap_kring *k) 1145 { 1146 int space = k->nr_hwtail - k->nr_hwcur; 1147 if (space < 0) 1148 space += k->nkr_num_slots; 1149 nm_prdis("preserving %d rx slots %d -> %d", space, k->nr_hwcur, k->nr_hwtail); 1150 1151 return space; 1152 } 1153 1154 /* return slots reserved to tx clients */ 1155 #define nm_kr_txspace(_k) nm_kr_rxspace(_k) 1156 1157 1158 /* True if no space in the tx ring, only valid after txsync_prologue */ 1159 static inline int 1160 nm_kr_txempty(struct netmap_kring *kring) 1161 { 1162 return kring->rhead == kring->nr_hwtail; 1163 } 1164 1165 /* True if no more completed slots in the rx ring, only valid after 1166 * rxsync_prologue */ 1167 #define nm_kr_rxempty(_k) nm_kr_txempty(_k) 1168 1169 /* True if the application needs to wait for more space on the ring 1170 * (more received packets or more free tx slots). 1171 * Only valid after *xsync_prologue. */ 1172 static inline int 1173 nm_kr_wouldblock(struct netmap_kring *kring) 1174 { 1175 return kring->rcur == kring->nr_hwtail; 1176 } 1177 1178 /* 1179 * protect against multiple threads using the same ring. 1180 * also check that the ring has not been stopped or locked 1181 */ 1182 #define NM_KR_BUSY 1 /* some other thread is syncing the ring */ 1183 #define NM_KR_STOPPED 2 /* unbounded stop (ifconfig down or driver unload) */ 1184 #define NM_KR_LOCKED 3 /* bounded, brief stop for mutual exclusion */ 1185 1186 1187 /* release the previously acquired right to use the *sync() methods of the ring */ 1188 static __inline void nm_kr_put(struct netmap_kring *kr) 1189 { 1190 NM_ATOMIC_CLEAR(&kr->nr_busy); 1191 } 1192 1193 1194 /* true if the ifp that backed the adapter has disappeared (e.g., the 1195 * driver has been unloaded) 1196 */ 1197 static inline int nm_iszombie(struct netmap_adapter *na); 1198 1199 /* try to obtain exclusive right to issue the *sync() operations on the ring. 1200 * The right is obtained and must be later relinquished via nm_kr_put() if and 1201 * only if nm_kr_tryget() returns 0. 1202 * If can_sleep is 1 there are only two other possible outcomes: 1203 * - the function returns NM_KR_BUSY 1204 * - the function returns NM_KR_STOPPED and sets the POLLERR bit in *perr 1205 * (if non-null) 1206 * In both cases the caller will typically skip the ring, possibly collecting 1207 * errors along the way. 1208 * If the calling context does not allow sleeping, the caller must pass 0 in can_sleep. 1209 * In the latter case, the function may also return NM_KR_LOCKED and leave *perr 1210 * untouched: ideally, the caller should try again at a later time. 1211 */ 1212 static __inline int nm_kr_tryget(struct netmap_kring *kr, int can_sleep, int *perr) 1213 { 1214 int busy = 1, stopped; 1215 /* check a first time without taking the lock 1216 * to avoid starvation for nm_kr_get() 1217 */ 1218 retry: 1219 stopped = kr->nkr_stopped; 1220 if (unlikely(stopped)) { 1221 goto stop; 1222 } 1223 busy = NM_ATOMIC_TEST_AND_SET(&kr->nr_busy); 1224 /* we should not return NM_KR_BUSY if the ring was 1225 * actually stopped, so check another time after 1226 * the barrier provided by the atomic operation 1227 */ 1228 stopped = kr->nkr_stopped; 1229 if (unlikely(stopped)) { 1230 goto stop; 1231 } 1232 1233 if (unlikely(nm_iszombie(kr->na))) { 1234 stopped = NM_KR_STOPPED; 1235 goto stop; 1236 } 1237 1238 return unlikely(busy) ? NM_KR_BUSY : 0; 1239 1240 stop: 1241 if (!busy) 1242 nm_kr_put(kr); 1243 if (stopped == NM_KR_STOPPED) { 1244 /* if POLLERR is defined we want to use it to simplify netmap_poll(). 1245 * Otherwise, any non-zero value will do. 1246 */ 1247 #ifdef POLLERR 1248 #define NM_POLLERR POLLERR 1249 #else 1250 #define NM_POLLERR 1 1251 #endif /* POLLERR */ 1252 if (perr) 1253 *perr |= NM_POLLERR; 1254 #undef NM_POLLERR 1255 } else if (can_sleep) { 1256 tsleep(kr, 0, "NM_KR_TRYGET", 4); 1257 goto retry; 1258 } 1259 return stopped; 1260 } 1261 1262 /* put the ring in the 'stopped' state and wait for the current user (if any) to 1263 * notice. stopped must be either NM_KR_STOPPED or NM_KR_LOCKED 1264 */ 1265 static __inline void nm_kr_stop(struct netmap_kring *kr, int stopped) 1266 { 1267 kr->nkr_stopped = stopped; 1268 while (NM_ATOMIC_TEST_AND_SET(&kr->nr_busy)) 1269 tsleep(kr, 0, "NM_KR_GET", 4); 1270 } 1271 1272 /* restart a ring after a stop */ 1273 static __inline void nm_kr_start(struct netmap_kring *kr) 1274 { 1275 kr->nkr_stopped = 0; 1276 nm_kr_put(kr); 1277 } 1278 1279 1280 /* 1281 * The following functions are used by individual drivers to 1282 * support netmap operation. 1283 * 1284 * netmap_attach() initializes a struct netmap_adapter, allocating the 1285 * struct netmap_ring's and the struct selinfo. 1286 * 1287 * netmap_detach() frees the memory allocated by netmap_attach(). 1288 * 1289 * netmap_transmit() replaces the if_transmit routine of the interface, 1290 * and is used to intercept packets coming from the stack. 1291 * 1292 * netmap_load_map/netmap_reload_map are helper routines to set/reset 1293 * the dmamap for a packet buffer 1294 * 1295 * netmap_reset() is a helper routine to be called in the hw driver 1296 * when reinitializing a ring. It should not be called by 1297 * virtual ports (vale, pipes, monitor) 1298 */ 1299 int netmap_attach(struct netmap_adapter *); 1300 int netmap_attach_ext(struct netmap_adapter *, size_t size, int override_reg); 1301 void netmap_detach(struct ifnet *); 1302 int netmap_transmit(struct ifnet *, struct mbuf *); 1303 struct netmap_slot *netmap_reset(struct netmap_adapter *na, 1304 enum txrx tx, u_int n, u_int new_cur); 1305 int netmap_ring_reinit(struct netmap_kring *); 1306 int netmap_rings_config_get(struct netmap_adapter *, struct nm_config_info *); 1307 1308 /* Return codes for netmap_*x_irq. */ 1309 enum { 1310 /* Driver should do normal interrupt processing, e.g. because 1311 * the interface is not in netmap mode. */ 1312 NM_IRQ_PASS = 0, 1313 /* Port is in netmap mode, and the interrupt work has been 1314 * completed. The driver does not have to notify netmap 1315 * again before the next interrupt. */ 1316 NM_IRQ_COMPLETED = -1, 1317 /* Port is in netmap mode, but the interrupt work has not been 1318 * completed. The driver has to make sure netmap will be 1319 * notified again soon, even if no more interrupts come (e.g. 1320 * on Linux the driver should not call napi_complete()). */ 1321 NM_IRQ_RESCHED = -2, 1322 }; 1323 1324 /* default functions to handle rx/tx interrupts */ 1325 int netmap_rx_irq(struct ifnet *, u_int, u_int *); 1326 #define netmap_tx_irq(_n, _q) netmap_rx_irq(_n, _q, NULL) 1327 int netmap_common_irq(struct netmap_adapter *, u_int, u_int *work_done); 1328 1329 1330 #ifdef WITH_VALE 1331 /* functions used by external modules to interface with VALE */ 1332 #define netmap_vp_to_ifp(_vp) ((_vp)->up.ifp) 1333 #define netmap_ifp_to_vp(_ifp) (NA(_ifp)->na_vp) 1334 #define netmap_ifp_to_host_vp(_ifp) (NA(_ifp)->na_hostvp) 1335 #define netmap_bdg_idx(_vp) ((_vp)->bdg_port) 1336 const char *netmap_bdg_name(struct netmap_vp_adapter *); 1337 #else /* !WITH_VALE */ 1338 #define netmap_vp_to_ifp(_vp) NULL 1339 #define netmap_ifp_to_vp(_ifp) NULL 1340 #define netmap_ifp_to_host_vp(_ifp) NULL 1341 #define netmap_bdg_idx(_vp) -1 1342 #endif /* WITH_VALE */ 1343 1344 static inline int 1345 nm_netmap_on(struct netmap_adapter *na) 1346 { 1347 return na && na->na_flags & NAF_NETMAP_ON; 1348 } 1349 1350 static inline int 1351 nm_native_on(struct netmap_adapter *na) 1352 { 1353 return nm_netmap_on(na) && (na->na_flags & NAF_NATIVE); 1354 } 1355 1356 static inline struct netmap_kring * 1357 netmap_kring_on(struct netmap_adapter *na, u_int q, enum txrx t) 1358 { 1359 struct netmap_kring *kring = NULL; 1360 1361 if (!nm_native_on(na)) 1362 return NULL; 1363 1364 if (t == NR_RX && q < na->num_rx_rings) 1365 kring = na->rx_rings[q]; 1366 else if (t == NR_TX && q < na->num_tx_rings) 1367 kring = na->tx_rings[q]; 1368 else 1369 return NULL; 1370 1371 return (kring->nr_mode == NKR_NETMAP_ON) ? kring : NULL; 1372 } 1373 1374 static inline int 1375 nm_iszombie(struct netmap_adapter *na) 1376 { 1377 return na == NULL || (na->na_flags & NAF_ZOMBIE); 1378 } 1379 1380 static inline void 1381 nm_update_hostrings_mode(struct netmap_adapter *na) 1382 { 1383 /* Process nr_mode and nr_pending_mode for host rings. */ 1384 na->tx_rings[na->num_tx_rings]->nr_mode = 1385 na->tx_rings[na->num_tx_rings]->nr_pending_mode; 1386 na->rx_rings[na->num_rx_rings]->nr_mode = 1387 na->rx_rings[na->num_rx_rings]->nr_pending_mode; 1388 } 1389 1390 void nm_set_native_flags(struct netmap_adapter *); 1391 void nm_clear_native_flags(struct netmap_adapter *); 1392 1393 void netmap_krings_mode_commit(struct netmap_adapter *na, int onoff); 1394 1395 /* 1396 * nm_*sync_prologue() functions are used in ioctl/poll and ptnetmap 1397 * kthreads. 1398 * We need netmap_ring* parameter, because in ptnetmap it is decoupled 1399 * from host kring. 1400 * The user-space ring pointers (head/cur/tail) are shared through 1401 * CSB between host and guest. 1402 */ 1403 1404 /* 1405 * validates parameters in the ring/kring, returns a value for head 1406 * If any error, returns ring_size to force a reinit. 1407 */ 1408 uint32_t nm_txsync_prologue(struct netmap_kring *, struct netmap_ring *); 1409 1410 1411 /* 1412 * validates parameters in the ring/kring, returns a value for head 1413 * If any error, returns ring_size lim to force a reinit. 1414 */ 1415 uint32_t nm_rxsync_prologue(struct netmap_kring *, struct netmap_ring *); 1416 1417 1418 /* check/fix address and len in tx rings */ 1419 #if 1 /* debug version */ 1420 #define NM_CHECK_ADDR_LEN(_na, _a, _l) do { \ 1421 if (_a == NETMAP_BUF_BASE(_na) || _l > NETMAP_BUF_SIZE(_na)) { \ 1422 nm_prlim(5, "bad addr/len ring %d slot %d idx %d len %d", \ 1423 kring->ring_id, nm_i, slot->buf_idx, len); \ 1424 if (_l > NETMAP_BUF_SIZE(_na)) \ 1425 _l = NETMAP_BUF_SIZE(_na); \ 1426 } } while (0) 1427 #else /* no debug version */ 1428 #define NM_CHECK_ADDR_LEN(_na, _a, _l) do { \ 1429 if (_l > NETMAP_BUF_SIZE(_na)) \ 1430 _l = NETMAP_BUF_SIZE(_na); \ 1431 } while (0) 1432 #endif 1433 1434 1435 /*---------------------------------------------------------------*/ 1436 /* 1437 * Support routines used by netmap subsystems 1438 * (native drivers, VALE, generic, pipes, monitors, ...) 1439 */ 1440 1441 1442 /* common routine for all functions that create a netmap adapter. It performs 1443 * two main tasks: 1444 * - if the na points to an ifp, mark the ifp as netmap capable 1445 * using na as its native adapter; 1446 * - provide defaults for the setup callbacks and the memory allocator 1447 */ 1448 int netmap_attach_common(struct netmap_adapter *); 1449 /* fill priv->np_[tr]xq{first,last} using the ringid and flags information 1450 * coming from a struct nmreq_register 1451 */ 1452 int netmap_interp_ringid(struct netmap_priv_d *priv, struct nmreq_header *hdr); 1453 /* update the ring parameters (number and size of tx and rx rings). 1454 * It calls the nm_config callback, if available. 1455 */ 1456 int netmap_update_config(struct netmap_adapter *na); 1457 /* create and initialize the common fields of the krings array. 1458 * using the information that must be already available in the na. 1459 * tailroom can be used to request the allocation of additional 1460 * tailroom bytes after the krings array. This is used by 1461 * netmap_vp_adapter's (i.e., VALE ports) to make room for 1462 * leasing-related data structures 1463 */ 1464 int netmap_krings_create(struct netmap_adapter *na, u_int tailroom); 1465 /* deletes the kring array of the adapter. The array must have 1466 * been created using netmap_krings_create 1467 */ 1468 void netmap_krings_delete(struct netmap_adapter *na); 1469 1470 int netmap_hw_krings_create(struct netmap_adapter *na); 1471 void netmap_hw_krings_delete(struct netmap_adapter *na); 1472 1473 /* set the stopped/enabled status of ring 1474 * When stopping, they also wait for all current activity on the ring to 1475 * terminate. The status change is then notified using the na nm_notify 1476 * callback. 1477 */ 1478 void netmap_set_ring(struct netmap_adapter *, u_int ring_id, enum txrx, int stopped); 1479 /* set the stopped/enabled status of all rings of the adapter. */ 1480 void netmap_set_all_rings(struct netmap_adapter *, int stopped); 1481 /* convenience wrappers for netmap_set_all_rings */ 1482 void netmap_disable_all_rings(struct ifnet *); 1483 void netmap_enable_all_rings(struct ifnet *); 1484 1485 int netmap_buf_size_validate(const struct netmap_adapter *na, unsigned mtu); 1486 int netmap_do_regif(struct netmap_priv_d *priv, struct netmap_adapter *na, 1487 struct nmreq_header *); 1488 void netmap_do_unregif(struct netmap_priv_d *priv); 1489 1490 u_int nm_bound_var(u_int *v, u_int dflt, u_int lo, u_int hi, const char *msg); 1491 int netmap_get_na(struct nmreq_header *hdr, struct netmap_adapter **na, 1492 struct ifnet **ifp, struct netmap_mem_d *nmd, int create); 1493 void netmap_unget_na(struct netmap_adapter *na, struct ifnet *ifp); 1494 int netmap_get_hw_na(struct ifnet *ifp, 1495 struct netmap_mem_d *nmd, struct netmap_adapter **na); 1496 1497 #ifdef WITH_VALE 1498 uint32_t netmap_vale_learning(struct nm_bdg_fwd *ft, uint8_t *dst_ring, 1499 struct netmap_vp_adapter *, void *private_data); 1500 1501 /* these are redefined in case of no VALE support */ 1502 int netmap_get_vale_na(struct nmreq_header *hdr, struct netmap_adapter **na, 1503 struct netmap_mem_d *nmd, int create); 1504 void *netmap_vale_create(const char *bdg_name, int *return_status); 1505 int netmap_vale_destroy(const char *bdg_name, void *auth_token); 1506 1507 #else /* !WITH_VALE */ 1508 #define netmap_bdg_learning(_1, _2, _3, _4) 0 1509 #define netmap_get_vale_na(_1, _2, _3, _4) 0 1510 #define netmap_bdg_create(_1, _2) NULL 1511 #define netmap_bdg_destroy(_1, _2) 0 1512 #endif /* !WITH_VALE */ 1513 1514 #ifdef WITH_PIPES 1515 /* max number of pipes per device */ 1516 #define NM_MAXPIPES 64 /* XXX this should probably be a sysctl */ 1517 void netmap_pipe_dealloc(struct netmap_adapter *); 1518 int netmap_get_pipe_na(struct nmreq_header *hdr, struct netmap_adapter **na, 1519 struct netmap_mem_d *nmd, int create); 1520 #else /* !WITH_PIPES */ 1521 #define NM_MAXPIPES 0 1522 #define netmap_pipe_alloc(_1, _2) 0 1523 #define netmap_pipe_dealloc(_1) 1524 #define netmap_get_pipe_na(hdr, _2, _3, _4) \ 1525 ((strchr(hdr->nr_name, '{') != NULL || strchr(hdr->nr_name, '}') != NULL) ? EOPNOTSUPP : 0) 1526 #endif 1527 1528 #ifdef WITH_MONITOR 1529 int netmap_get_monitor_na(struct nmreq_header *hdr, struct netmap_adapter **na, 1530 struct netmap_mem_d *nmd, int create); 1531 void netmap_monitor_stop(struct netmap_adapter *na); 1532 #else 1533 #define netmap_get_monitor_na(hdr, _2, _3, _4) \ 1534 (((struct nmreq_register *)(uintptr_t)hdr->nr_body)->nr_flags & (NR_MONITOR_TX | NR_MONITOR_RX) ? EOPNOTSUPP : 0) 1535 #endif 1536 1537 #ifdef WITH_NMNULL 1538 int netmap_get_null_na(struct nmreq_header *hdr, struct netmap_adapter **na, 1539 struct netmap_mem_d *nmd, int create); 1540 #else /* !WITH_NMNULL */ 1541 #define netmap_get_null_na(hdr, _2, _3, _4) \ 1542 (((struct nmreq_register *)(uintptr_t)hdr->nr_body)->nr_flags & (NR_MONITOR_TX | NR_MONITOR_RX) ? EOPNOTSUPP : 0) 1543 #endif /* WITH_NMNULL */ 1544 1545 #ifdef CONFIG_NET_NS 1546 struct net *netmap_bns_get(void); 1547 void netmap_bns_put(struct net *); 1548 void netmap_bns_getbridges(struct nm_bridge **, u_int *); 1549 #else 1550 extern struct nm_bridge *nm_bridges; 1551 #define netmap_bns_get() 1552 #define netmap_bns_put(_1) 1553 #define netmap_bns_getbridges(b, n) \ 1554 do { *b = nm_bridges; *n = NM_BRIDGES; } while (0) 1555 #endif 1556 1557 /* Various prototypes */ 1558 int netmap_poll(struct netmap_priv_d *, int events, NM_SELRECORD_T *td); 1559 int netmap_init(void); 1560 void netmap_fini(void); 1561 int netmap_get_memory(struct netmap_priv_d* p); 1562 void netmap_dtor(void *data); 1563 1564 int netmap_ioctl(struct netmap_priv_d *priv, u_long cmd, caddr_t data, 1565 struct thread *, int nr_body_is_user); 1566 int netmap_ioctl_legacy(struct netmap_priv_d *priv, u_long cmd, caddr_t data, 1567 struct thread *td); 1568 size_t nmreq_size_by_type(uint16_t nr_reqtype); 1569 1570 /* netmap_adapter creation/destruction */ 1571 1572 // #define NM_DEBUG_PUTGET 1 1573 1574 #ifdef NM_DEBUG_PUTGET 1575 1576 #define NM_DBG(f) __##f 1577 1578 void __netmap_adapter_get(struct netmap_adapter *na); 1579 1580 #define netmap_adapter_get(na) \ 1581 do { \ 1582 struct netmap_adapter *__na = na; \ 1583 nm_prinf("getting %p:%s (%d)", __na, (__na)->name, (__na)->na_refcount); \ 1584 __netmap_adapter_get(__na); \ 1585 } while (0) 1586 1587 int __netmap_adapter_put(struct netmap_adapter *na); 1588 1589 #define netmap_adapter_put(na) \ 1590 ({ \ 1591 struct netmap_adapter *__na = na; \ 1592 nm_prinf("putting %p:%s (%d)", __na, (__na)->name, (__na)->na_refcount); \ 1593 __netmap_adapter_put(__na); \ 1594 }) 1595 1596 #else /* !NM_DEBUG_PUTGET */ 1597 1598 #define NM_DBG(f) f 1599 void netmap_adapter_get(struct netmap_adapter *na); 1600 int netmap_adapter_put(struct netmap_adapter *na); 1601 1602 #endif /* !NM_DEBUG_PUTGET */ 1603 1604 1605 /* 1606 * module variables 1607 */ 1608 #define NETMAP_BUF_BASE(_na) ((_na)->na_lut.lut[0].vaddr) 1609 #define NETMAP_BUF_SIZE(_na) ((_na)->na_lut.objsize) 1610 extern int netmap_no_pendintr; 1611 extern int netmap_verbose; 1612 #ifdef CONFIG_NETMAP_DEBUG 1613 extern int netmap_debug; /* for debugging */ 1614 #else /* !CONFIG_NETMAP_DEBUG */ 1615 #define netmap_debug (0) 1616 #endif /* !CONFIG_NETMAP_DEBUG */ 1617 enum { /* debug flags */ 1618 NM_DEBUG_ON = 1, /* generic debug messsages */ 1619 NM_DEBUG_HOST = 0x2, /* debug host stack */ 1620 NM_DEBUG_RXSYNC = 0x10, /* debug on rxsync/txsync */ 1621 NM_DEBUG_TXSYNC = 0x20, 1622 NM_DEBUG_RXINTR = 0x100, /* debug on rx/tx intr (driver) */ 1623 NM_DEBUG_TXINTR = 0x200, 1624 NM_DEBUG_NIC_RXSYNC = 0x1000, /* debug on rx/tx intr (driver) */ 1625 NM_DEBUG_NIC_TXSYNC = 0x2000, 1626 NM_DEBUG_MEM = 0x4000, /* verbose memory allocations/deallocations */ 1627 NM_DEBUG_VALE = 0x8000, /* debug messages from memory allocators */ 1628 NM_DEBUG_BDG = NM_DEBUG_VALE, 1629 }; 1630 1631 extern int netmap_txsync_retry; 1632 extern int netmap_generic_hwcsum; 1633 extern int netmap_generic_mit; 1634 extern int netmap_generic_ringsize; 1635 extern int netmap_generic_rings; 1636 #ifdef linux 1637 extern int netmap_generic_txqdisc; 1638 #endif 1639 1640 /* 1641 * NA returns a pointer to the struct netmap adapter from the ifp. 1642 * WNA is os-specific and must be defined in glue code. 1643 */ 1644 #define NA(_ifp) ((struct netmap_adapter *)WNA(_ifp)) 1645 1646 /* 1647 * we provide a default implementation of NM_ATTACH_NA/NM_DETACH_NA 1648 * based on the WNA field. 1649 * Glue code may override this by defining its own NM_ATTACH_NA 1650 */ 1651 #ifndef NM_ATTACH_NA 1652 /* 1653 * On old versions of FreeBSD, NA(ifp) is a pspare. On linux we 1654 * overload another pointer in the netdev. 1655 * 1656 * We check if NA(ifp) is set and its first element has a related 1657 * magic value. The capenable is within the struct netmap_adapter. 1658 */ 1659 #define NETMAP_MAGIC 0x52697a7a 1660 1661 #define NM_NA_VALID(ifp) (NA(ifp) && \ 1662 ((uint32_t)(uintptr_t)NA(ifp) ^ NA(ifp)->magic) == NETMAP_MAGIC ) 1663 1664 #define NM_ATTACH_NA(ifp, na) do { \ 1665 WNA(ifp) = na; \ 1666 if (NA(ifp)) \ 1667 NA(ifp)->magic = \ 1668 ((uint32_t)(uintptr_t)NA(ifp)) ^ NETMAP_MAGIC; \ 1669 } while(0) 1670 #define NM_RESTORE_NA(ifp, na) WNA(ifp) = na; 1671 1672 #define NM_DETACH_NA(ifp) do { WNA(ifp) = NULL; } while (0) 1673 #define NM_NA_CLASH(ifp) (NA(ifp) && !NM_NA_VALID(ifp)) 1674 #endif /* !NM_ATTACH_NA */ 1675 1676 1677 #define NM_IS_NATIVE(ifp) (NM_NA_VALID(ifp) && NA(ifp)->nm_dtor == netmap_hw_dtor) 1678 1679 #if defined(__FreeBSD__) 1680 1681 /* Assigns the device IOMMU domain to an allocator. 1682 * Returns -ENOMEM in case the domain is different */ 1683 #define nm_iommu_group_id(dev) (0) 1684 1685 /* Callback invoked by the dma machinery after a successful dmamap_load */ 1686 static void netmap_dmamap_cb(__unused void *arg, 1687 __unused bus_dma_segment_t * segs, __unused int nseg, __unused int error) 1688 { 1689 } 1690 1691 /* bus_dmamap_load wrapper: call aforementioned function if map != NULL. 1692 * XXX can we do it without a callback ? 1693 */ 1694 static inline int 1695 netmap_load_map(struct netmap_adapter *na, 1696 bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 1697 { 1698 if (map) 1699 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE(na), 1700 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT); 1701 return 0; 1702 } 1703 1704 static inline void 1705 netmap_unload_map(struct netmap_adapter *na, 1706 bus_dma_tag_t tag, bus_dmamap_t map) 1707 { 1708 if (map) 1709 bus_dmamap_unload(tag, map); 1710 } 1711 1712 #define netmap_sync_map(na, tag, map, sz, t) 1713 1714 /* update the map when a buffer changes. */ 1715 static inline void 1716 netmap_reload_map(struct netmap_adapter *na, 1717 bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 1718 { 1719 if (map) { 1720 bus_dmamap_unload(tag, map); 1721 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE(na), 1722 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT); 1723 } 1724 } 1725 1726 #elif defined(_WIN32) 1727 1728 #else /* linux */ 1729 1730 int nm_iommu_group_id(bus_dma_tag_t dev); 1731 #include <linux/dma-mapping.h> 1732 1733 /* 1734 * on linux we need 1735 * dma_map_single(&pdev->dev, virt_addr, len, direction) 1736 * dma_unmap_single(&adapter->pdev->dev, phys_addr, len, direction) 1737 */ 1738 #if 0 1739 struct e1000_buffer *buffer_info = &tx_ring->buffer_info[l]; 1740 /* set time_stamp *before* dma to help avoid a possible race */ 1741 buffer_info->time_stamp = jiffies; 1742 buffer_info->mapped_as_page = false; 1743 buffer_info->length = len; 1744 //buffer_info->next_to_watch = l; 1745 /* reload dma map */ 1746 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma, 1747 NETMAP_BUF_SIZE, DMA_TO_DEVICE); 1748 buffer_info->dma = dma_map_single(&adapter->pdev->dev, 1749 addr, NETMAP_BUF_SIZE, DMA_TO_DEVICE); 1750 1751 if (dma_mapping_error(&adapter->pdev->dev, buffer_info->dma)) { 1752 nm_prerr("dma mapping error"); 1753 /* goto dma_error; See e1000_put_txbuf() */ 1754 /* XXX reset */ 1755 } 1756 tx_desc->buffer_addr = htole64(buffer_info->dma); //XXX 1757 1758 #endif 1759 1760 static inline int 1761 netmap_load_map(struct netmap_adapter *na, 1762 bus_dma_tag_t tag, bus_dmamap_t map, void *buf, u_int size) 1763 { 1764 if (map) { 1765 *map = dma_map_single(na->pdev, buf, size, 1766 DMA_BIDIRECTIONAL); 1767 if (dma_mapping_error(na->pdev, *map)) { 1768 *map = 0; 1769 return ENOMEM; 1770 } 1771 } 1772 return 0; 1773 } 1774 1775 static inline void 1776 netmap_unload_map(struct netmap_adapter *na, 1777 bus_dma_tag_t tag, bus_dmamap_t map, u_int sz) 1778 { 1779 if (*map) { 1780 dma_unmap_single(na->pdev, *map, sz, 1781 DMA_BIDIRECTIONAL); 1782 } 1783 } 1784 1785 #ifdef NETMAP_LINUX_HAVE_DMASYNC 1786 static inline void 1787 netmap_sync_map_cpu(struct netmap_adapter *na, 1788 bus_dma_tag_t tag, bus_dmamap_t map, u_int sz, enum txrx t) 1789 { 1790 if (*map) { 1791 dma_sync_single_for_cpu(na->pdev, *map, sz, 1792 (t == NR_TX ? DMA_TO_DEVICE : DMA_FROM_DEVICE)); 1793 } 1794 } 1795 1796 static inline void 1797 netmap_sync_map_dev(struct netmap_adapter *na, 1798 bus_dma_tag_t tag, bus_dmamap_t map, u_int sz, enum txrx t) 1799 { 1800 if (*map) { 1801 dma_sync_single_for_device(na->pdev, *map, sz, 1802 (t == NR_TX ? DMA_TO_DEVICE : DMA_FROM_DEVICE)); 1803 } 1804 } 1805 1806 static inline void 1807 netmap_reload_map(struct netmap_adapter *na, 1808 bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 1809 { 1810 u_int sz = NETMAP_BUF_SIZE(na); 1811 1812 if (*map) { 1813 dma_unmap_single(na->pdev, *map, sz, 1814 DMA_BIDIRECTIONAL); 1815 } 1816 1817 *map = dma_map_single(na->pdev, buf, sz, 1818 DMA_BIDIRECTIONAL); 1819 } 1820 #else /* !NETMAP_LINUX_HAVE_DMASYNC */ 1821 #define netmap_sync_map_cpu(na, tag, map, sz, t) 1822 #define netmap_sync_map_dev(na, tag, map, sz, t) 1823 #endif /* NETMAP_LINUX_HAVE_DMASYNC */ 1824 1825 #endif /* linux */ 1826 1827 1828 /* 1829 * functions to map NIC to KRING indexes (n2k) and vice versa (k2n) 1830 */ 1831 static inline int 1832 netmap_idx_n2k(struct netmap_kring *kr, int idx) 1833 { 1834 int n = kr->nkr_num_slots; 1835 1836 if (likely(kr->nkr_hwofs == 0)) { 1837 return idx; 1838 } 1839 1840 idx += kr->nkr_hwofs; 1841 if (idx < 0) 1842 return idx + n; 1843 else if (idx < n) 1844 return idx; 1845 else 1846 return idx - n; 1847 } 1848 1849 1850 static inline int 1851 netmap_idx_k2n(struct netmap_kring *kr, int idx) 1852 { 1853 int n = kr->nkr_num_slots; 1854 1855 if (likely(kr->nkr_hwofs == 0)) { 1856 return idx; 1857 } 1858 1859 idx -= kr->nkr_hwofs; 1860 if (idx < 0) 1861 return idx + n; 1862 else if (idx < n) 1863 return idx; 1864 else 1865 return idx - n; 1866 } 1867 1868 1869 /* Entries of the look-up table. */ 1870 #ifdef __FreeBSD__ 1871 struct lut_entry { 1872 void *vaddr; /* virtual address. */ 1873 vm_paddr_t paddr; /* physical address. */ 1874 }; 1875 #else /* linux & _WIN32 */ 1876 /* dma-mapping in linux can assign a buffer a different address 1877 * depending on the device, so we need to have a separate 1878 * physical-address look-up table for each na. 1879 * We can still share the vaddrs, though, therefore we split 1880 * the lut_entry structure. 1881 */ 1882 struct lut_entry { 1883 void *vaddr; /* virtual address. */ 1884 }; 1885 1886 struct plut_entry { 1887 vm_paddr_t paddr; /* physical address. */ 1888 }; 1889 #endif /* linux & _WIN32 */ 1890 1891 struct netmap_obj_pool; 1892 1893 /* 1894 * NMB return the virtual address of a buffer (buffer 0 on bad index) 1895 * PNMB also fills the physical address 1896 */ 1897 static inline void * 1898 NMB(struct netmap_adapter *na, struct netmap_slot *slot) 1899 { 1900 struct lut_entry *lut = na->na_lut.lut; 1901 uint32_t i = slot->buf_idx; 1902 return (unlikely(i >= na->na_lut.objtotal)) ? 1903 lut[0].vaddr : lut[i].vaddr; 1904 } 1905 1906 static inline void * 1907 PNMB(struct netmap_adapter *na, struct netmap_slot *slot, uint64_t *pp) 1908 { 1909 uint32_t i = slot->buf_idx; 1910 struct lut_entry *lut = na->na_lut.lut; 1911 struct plut_entry *plut = na->na_lut.plut; 1912 void *ret = (i >= na->na_lut.objtotal) ? lut[0].vaddr : lut[i].vaddr; 1913 1914 #ifdef _WIN32 1915 *pp = (i >= na->na_lut.objtotal) ? (uint64_t)plut[0].paddr.QuadPart : (uint64_t)plut[i].paddr.QuadPart; 1916 #else 1917 *pp = (i >= na->na_lut.objtotal) ? plut[0].paddr : plut[i].paddr; 1918 #endif 1919 return ret; 1920 } 1921 1922 1923 /* 1924 * Structure associated to each netmap file descriptor. 1925 * It is created on open and left unbound (np_nifp == NULL). 1926 * A successful NIOCREGIF will set np_nifp and the first few fields; 1927 * this is protected by a global lock (NMG_LOCK) due to low contention. 1928 * 1929 * np_refs counts the number of references to the structure: one for the fd, 1930 * plus (on FreeBSD) one for each active mmap which we track ourselves 1931 * (linux automatically tracks them, but FreeBSD does not). 1932 * np_refs is protected by NMG_LOCK. 1933 * 1934 * Read access to the structure is lock free, because ni_nifp once set 1935 * can only go to 0 when nobody is using the entry anymore. Readers 1936 * must check that np_nifp != NULL before using the other fields. 1937 */ 1938 struct netmap_priv_d { 1939 struct netmap_if * volatile np_nifp; /* netmap if descriptor. */ 1940 1941 struct netmap_adapter *np_na; 1942 struct ifnet *np_ifp; 1943 uint32_t np_flags; /* from the ioctl */ 1944 u_int np_qfirst[NR_TXRX], 1945 np_qlast[NR_TXRX]; /* range of tx/rx rings to scan */ 1946 uint16_t np_txpoll; 1947 uint16_t np_kloop_state; /* use with NMG_LOCK held */ 1948 #define NM_SYNC_KLOOP_RUNNING (1 << 0) 1949 #define NM_SYNC_KLOOP_STOPPING (1 << 1) 1950 int np_sync_flags; /* to be passed to nm_sync */ 1951 1952 int np_refs; /* use with NMG_LOCK held */ 1953 1954 /* pointers to the selinfo to be used for selrecord. 1955 * Either the local or the global one depending on the 1956 * number of rings. 1957 */ 1958 NM_SELINFO_T *np_si[NR_TXRX]; 1959 1960 /* In the optional CSB mode, the user must specify the start address 1961 * of two arrays of Communication Status Block (CSB) entries, for the 1962 * two directions (kernel read application write, and kernel write 1963 * application read). 1964 * The number of entries must agree with the number of rings bound to 1965 * the netmap file descriptor. The entries corresponding to the TX 1966 * rings are laid out before the ones corresponding to the RX rings. 1967 * 1968 * Array of CSB entries for application --> kernel communication 1969 * (N entries). */ 1970 struct nm_csb_atok *np_csb_atok_base; 1971 /* Array of CSB entries for kernel --> application communication 1972 * (N entries). */ 1973 struct nm_csb_ktoa *np_csb_ktoa_base; 1974 1975 #ifdef linux 1976 struct file *np_filp; /* used by sync kloop */ 1977 #endif /* linux */ 1978 }; 1979 1980 struct netmap_priv_d *netmap_priv_new(void); 1981 void netmap_priv_delete(struct netmap_priv_d *); 1982 1983 static inline int nm_kring_pending(struct netmap_priv_d *np) 1984 { 1985 struct netmap_adapter *na = np->np_na; 1986 enum txrx t; 1987 int i; 1988 1989 for_rx_tx(t) { 1990 for (i = np->np_qfirst[t]; i < np->np_qlast[t]; i++) { 1991 struct netmap_kring *kring = NMR(na, t)[i]; 1992 if (kring->nr_mode != kring->nr_pending_mode) { 1993 return 1; 1994 } 1995 } 1996 } 1997 return 0; 1998 } 1999 2000 /* call with NMG_LOCK held */ 2001 static __inline int 2002 nm_si_user(struct netmap_priv_d *priv, enum txrx t) 2003 { 2004 return (priv->np_na != NULL && 2005 (priv->np_qlast[t] - priv->np_qfirst[t] > 1)); 2006 } 2007 2008 #ifdef WITH_PIPES 2009 int netmap_pipe_txsync(struct netmap_kring *txkring, int flags); 2010 int netmap_pipe_rxsync(struct netmap_kring *rxkring, int flags); 2011 int netmap_pipe_krings_create_both(struct netmap_adapter *na, 2012 struct netmap_adapter *ona); 2013 void netmap_pipe_krings_delete_both(struct netmap_adapter *na, 2014 struct netmap_adapter *ona); 2015 int netmap_pipe_reg_both(struct netmap_adapter *na, 2016 struct netmap_adapter *ona); 2017 #endif /* WITH_PIPES */ 2018 2019 #ifdef WITH_MONITOR 2020 2021 struct netmap_monitor_adapter { 2022 struct netmap_adapter up; 2023 2024 struct netmap_priv_d priv; 2025 uint32_t flags; 2026 }; 2027 2028 #endif /* WITH_MONITOR */ 2029 2030 2031 #ifdef WITH_GENERIC 2032 /* 2033 * generic netmap emulation for devices that do not have 2034 * native netmap support. 2035 */ 2036 int generic_netmap_attach(struct ifnet *ifp); 2037 int generic_rx_handler(struct ifnet *ifp, struct mbuf *m);; 2038 2039 int nm_os_catch_rx(struct netmap_generic_adapter *gna, int intercept); 2040 int nm_os_catch_tx(struct netmap_generic_adapter *gna, int intercept); 2041 2042 int na_is_generic(struct netmap_adapter *na); 2043 2044 /* 2045 * the generic transmit routine is passed a structure to optionally 2046 * build a queue of descriptors, in an OS-specific way. 2047 * The payload is at addr, if non-null, and the routine should send or queue 2048 * the packet, returning 0 if successful, 1 on failure. 2049 * 2050 * At the end, if head is non-null, there will be an additional call 2051 * to the function with addr = NULL; this should tell the OS-specific 2052 * routine to send the queue and free any resources. Failure is ignored. 2053 */ 2054 struct nm_os_gen_arg { 2055 struct ifnet *ifp; 2056 void *m; /* os-specific mbuf-like object */ 2057 void *head, *tail; /* tailq, if the OS-specific routine needs to build one */ 2058 void *addr; /* payload of current packet */ 2059 u_int len; /* packet length */ 2060 u_int ring_nr; /* packet length */ 2061 u_int qevent; /* in txqdisc mode, place an event on this mbuf */ 2062 }; 2063 2064 int nm_os_generic_xmit_frame(struct nm_os_gen_arg *); 2065 int nm_os_generic_find_num_desc(struct ifnet *ifp, u_int *tx, u_int *rx); 2066 void nm_os_generic_find_num_queues(struct ifnet *ifp, u_int *txq, u_int *rxq); 2067 void nm_os_generic_set_features(struct netmap_generic_adapter *gna); 2068 2069 static inline struct ifnet* 2070 netmap_generic_getifp(struct netmap_generic_adapter *gna) 2071 { 2072 if (gna->prev) 2073 return gna->prev->ifp; 2074 2075 return gna->up.up.ifp; 2076 } 2077 2078 void netmap_generic_irq(struct netmap_adapter *na, u_int q, u_int *work_done); 2079 2080 //#define RATE_GENERIC /* Enables communication statistics for generic. */ 2081 #ifdef RATE_GENERIC 2082 void generic_rate(int txp, int txs, int txi, int rxp, int rxs, int rxi); 2083 #else 2084 #define generic_rate(txp, txs, txi, rxp, rxs, rxi) 2085 #endif 2086 2087 /* 2088 * netmap_mitigation API. This is used by the generic adapter 2089 * to reduce the number of interrupt requests/selwakeup 2090 * to clients on incoming packets. 2091 */ 2092 void nm_os_mitigation_init(struct nm_generic_mit *mit, int idx, 2093 struct netmap_adapter *na); 2094 void nm_os_mitigation_start(struct nm_generic_mit *mit); 2095 void nm_os_mitigation_restart(struct nm_generic_mit *mit); 2096 int nm_os_mitigation_active(struct nm_generic_mit *mit); 2097 void nm_os_mitigation_cleanup(struct nm_generic_mit *mit); 2098 #else /* !WITH_GENERIC */ 2099 #define generic_netmap_attach(ifp) (EOPNOTSUPP) 2100 #define na_is_generic(na) (0) 2101 #endif /* WITH_GENERIC */ 2102 2103 /* Shared declarations for the VALE switch. */ 2104 2105 /* 2106 * Each transmit queue accumulates a batch of packets into 2107 * a structure before forwarding. Packets to the same 2108 * destination are put in a list using ft_next as a link field. 2109 * ft_frags and ft_next are valid only on the first fragment. 2110 */ 2111 struct nm_bdg_fwd { /* forwarding entry for a bridge */ 2112 void *ft_buf; /* netmap or indirect buffer */ 2113 uint8_t ft_frags; /* how many fragments (only on 1st frag) */ 2114 uint16_t ft_offset; /* dst port (unused) */ 2115 uint16_t ft_flags; /* flags, e.g. indirect */ 2116 uint16_t ft_len; /* src fragment len */ 2117 uint16_t ft_next; /* next packet to same destination */ 2118 }; 2119 2120 /* struct 'virtio_net_hdr' from linux. */ 2121 struct nm_vnet_hdr { 2122 #define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /* Use csum_start, csum_offset */ 2123 #define VIRTIO_NET_HDR_F_DATA_VALID 2 /* Csum is valid */ 2124 uint8_t flags; 2125 #define VIRTIO_NET_HDR_GSO_NONE 0 /* Not a GSO frame */ 2126 #define VIRTIO_NET_HDR_GSO_TCPV4 1 /* GSO frame, IPv4 TCP (TSO) */ 2127 #define VIRTIO_NET_HDR_GSO_UDP 3 /* GSO frame, IPv4 UDP (UFO) */ 2128 #define VIRTIO_NET_HDR_GSO_TCPV6 4 /* GSO frame, IPv6 TCP */ 2129 #define VIRTIO_NET_HDR_GSO_ECN 0x80 /* TCP has ECN set */ 2130 uint8_t gso_type; 2131 uint16_t hdr_len; 2132 uint16_t gso_size; 2133 uint16_t csum_start; 2134 uint16_t csum_offset; 2135 }; 2136 2137 #define WORST_CASE_GSO_HEADER (14+40+60) /* IPv6 + TCP */ 2138 2139 /* Private definitions for IPv4, IPv6, UDP and TCP headers. */ 2140 2141 struct nm_iphdr { 2142 uint8_t version_ihl; 2143 uint8_t tos; 2144 uint16_t tot_len; 2145 uint16_t id; 2146 uint16_t frag_off; 2147 uint8_t ttl; 2148 uint8_t protocol; 2149 uint16_t check; 2150 uint32_t saddr; 2151 uint32_t daddr; 2152 /*The options start here. */ 2153 }; 2154 2155 struct nm_tcphdr { 2156 uint16_t source; 2157 uint16_t dest; 2158 uint32_t seq; 2159 uint32_t ack_seq; 2160 uint8_t doff; /* Data offset + Reserved */ 2161 uint8_t flags; 2162 uint16_t window; 2163 uint16_t check; 2164 uint16_t urg_ptr; 2165 }; 2166 2167 struct nm_udphdr { 2168 uint16_t source; 2169 uint16_t dest; 2170 uint16_t len; 2171 uint16_t check; 2172 }; 2173 2174 struct nm_ipv6hdr { 2175 uint8_t priority_version; 2176 uint8_t flow_lbl[3]; 2177 2178 uint16_t payload_len; 2179 uint8_t nexthdr; 2180 uint8_t hop_limit; 2181 2182 uint8_t saddr[16]; 2183 uint8_t daddr[16]; 2184 }; 2185 2186 /* Type used to store a checksum (in host byte order) that hasn't been 2187 * folded yet. 2188 */ 2189 #define rawsum_t uint32_t 2190 2191 rawsum_t nm_os_csum_raw(uint8_t *data, size_t len, rawsum_t cur_sum); 2192 uint16_t nm_os_csum_ipv4(struct nm_iphdr *iph); 2193 void nm_os_csum_tcpudp_ipv4(struct nm_iphdr *iph, void *data, 2194 size_t datalen, uint16_t *check); 2195 void nm_os_csum_tcpudp_ipv6(struct nm_ipv6hdr *ip6h, void *data, 2196 size_t datalen, uint16_t *check); 2197 uint16_t nm_os_csum_fold(rawsum_t cur_sum); 2198 2199 void bdg_mismatch_datapath(struct netmap_vp_adapter *na, 2200 struct netmap_vp_adapter *dst_na, 2201 const struct nm_bdg_fwd *ft_p, 2202 struct netmap_ring *dst_ring, 2203 u_int *j, u_int lim, u_int *howmany); 2204 2205 /* persistent virtual port routines */ 2206 int nm_os_vi_persist(const char *, struct ifnet **); 2207 void nm_os_vi_detach(struct ifnet *); 2208 void nm_os_vi_init_index(void); 2209 2210 /* 2211 * kernel thread routines 2212 */ 2213 struct nm_kctx; /* OS-specific kernel context - opaque */ 2214 typedef void (*nm_kctx_worker_fn_t)(void *data); 2215 2216 /* kthread configuration */ 2217 struct nm_kctx_cfg { 2218 long type; /* kthread type/identifier */ 2219 nm_kctx_worker_fn_t worker_fn; /* worker function */ 2220 void *worker_private;/* worker parameter */ 2221 int attach_user; /* attach kthread to user process */ 2222 }; 2223 /* kthread configuration */ 2224 struct nm_kctx *nm_os_kctx_create(struct nm_kctx_cfg *cfg, 2225 void *opaque); 2226 int nm_os_kctx_worker_start(struct nm_kctx *); 2227 void nm_os_kctx_worker_stop(struct nm_kctx *); 2228 void nm_os_kctx_destroy(struct nm_kctx *); 2229 void nm_os_kctx_worker_setaff(struct nm_kctx *, int); 2230 u_int nm_os_ncpus(void); 2231 2232 int netmap_sync_kloop(struct netmap_priv_d *priv, 2233 struct nmreq_header *hdr); 2234 int netmap_sync_kloop_stop(struct netmap_priv_d *priv); 2235 2236 #ifdef WITH_PTNETMAP 2237 /* ptnetmap guest routines */ 2238 2239 /* 2240 * ptnetmap_memdev routines used to talk with ptnetmap_memdev device driver 2241 */ 2242 struct ptnetmap_memdev; 2243 int nm_os_pt_memdev_iomap(struct ptnetmap_memdev *, vm_paddr_t *, void **, 2244 uint64_t *); 2245 void nm_os_pt_memdev_iounmap(struct ptnetmap_memdev *); 2246 uint32_t nm_os_pt_memdev_ioread(struct ptnetmap_memdev *, unsigned int); 2247 2248 /* 2249 * netmap adapter for guest ptnetmap ports 2250 */ 2251 struct netmap_pt_guest_adapter { 2252 /* The netmap adapter to be used by netmap applications. 2253 * This field must be the first, to allow upcast. */ 2254 struct netmap_hw_adapter hwup; 2255 2256 /* The netmap adapter to be used by the driver. */ 2257 struct netmap_hw_adapter dr; 2258 2259 /* Reference counter to track users of backend netmap port: the 2260 * network stack and netmap clients. 2261 * Used to decide when we need (de)allocate krings/rings and 2262 * start (stop) ptnetmap kthreads. */ 2263 int backend_users; 2264 2265 }; 2266 2267 int netmap_pt_guest_attach(struct netmap_adapter *na, 2268 unsigned int nifp_offset, 2269 unsigned int memid); 2270 bool netmap_pt_guest_txsync(struct nm_csb_atok *atok, 2271 struct nm_csb_ktoa *ktoa, 2272 struct netmap_kring *kring, int flags); 2273 bool netmap_pt_guest_rxsync(struct nm_csb_atok *atok, 2274 struct nm_csb_ktoa *ktoa, 2275 struct netmap_kring *kring, int flags); 2276 int ptnet_nm_krings_create(struct netmap_adapter *na); 2277 void ptnet_nm_krings_delete(struct netmap_adapter *na); 2278 void ptnet_nm_dtor(struct netmap_adapter *na); 2279 2280 /* Helper function wrapping nm_sync_kloop_appl_read(). */ 2281 static inline void 2282 ptnet_sync_tail(struct nm_csb_ktoa *ktoa, struct netmap_kring *kring) 2283 { 2284 struct netmap_ring *ring = kring->ring; 2285 2286 /* Update hwcur and hwtail as known by the host. */ 2287 nm_sync_kloop_appl_read(ktoa, &kring->nr_hwtail, &kring->nr_hwcur); 2288 2289 /* nm_sync_finalize */ 2290 ring->tail = kring->rtail = kring->nr_hwtail; 2291 } 2292 #endif /* WITH_PTNETMAP */ 2293 2294 #ifdef __FreeBSD__ 2295 /* 2296 * FreeBSD mbuf allocator/deallocator in emulation mode: 2297 */ 2298 #if __FreeBSD_version < 1100000 2299 2300 /* 2301 * For older versions of FreeBSD: 2302 * 2303 * We allocate EXT_PACKET mbuf+clusters, but need to set M_NOFREE 2304 * so that the destructor, if invoked, will not free the packet. 2305 * In principle we should set the destructor only on demand, 2306 * but since there might be a race we better do it on allocation. 2307 * As a consequence, we also need to set the destructor or we 2308 * would leak buffers. 2309 */ 2310 2311 /* mbuf destructor, also need to change the type to EXT_EXTREF, 2312 * add an M_NOFREE flag, and then clear the flag and 2313 * chain into uma_zfree(zone_pack, mf) 2314 * (or reinstall the buffer ?) 2315 */ 2316 #define SET_MBUF_DESTRUCTOR(m, fn) do { \ 2317 (m)->m_ext.ext_free = (void *)fn; \ 2318 (m)->m_ext.ext_type = EXT_EXTREF; \ 2319 } while (0) 2320 2321 static int 2322 void_mbuf_dtor(struct mbuf *m, void *arg1, void *arg2) 2323 { 2324 /* restore original mbuf */ 2325 m->m_ext.ext_buf = m->m_data = m->m_ext.ext_arg1; 2326 m->m_ext.ext_arg1 = NULL; 2327 m->m_ext.ext_type = EXT_PACKET; 2328 m->m_ext.ext_free = NULL; 2329 if (MBUF_REFCNT(m) == 0) 2330 SET_MBUF_REFCNT(m, 1); 2331 uma_zfree(zone_pack, m); 2332 2333 return 0; 2334 } 2335 2336 static inline struct mbuf * 2337 nm_os_get_mbuf(struct ifnet *ifp, int len) 2338 { 2339 struct mbuf *m; 2340 2341 (void)ifp; 2342 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2343 if (m) { 2344 /* m_getcl() (mb_ctor_mbuf) has an assert that checks that 2345 * M_NOFREE flag is not specified as third argument, 2346 * so we have to set M_NOFREE after m_getcl(). */ 2347 m->m_flags |= M_NOFREE; 2348 m->m_ext.ext_arg1 = m->m_ext.ext_buf; // XXX save 2349 m->m_ext.ext_free = (void *)void_mbuf_dtor; 2350 m->m_ext.ext_type = EXT_EXTREF; 2351 nm_prdis(5, "create m %p refcnt %d", m, MBUF_REFCNT(m)); 2352 } 2353 return m; 2354 } 2355 2356 #else /* __FreeBSD_version >= 1100000 */ 2357 2358 /* 2359 * Newer versions of FreeBSD, using a straightforward scheme. 2360 * 2361 * We allocate mbufs with m_gethdr(), since the mbuf header is needed 2362 * by the driver. We also attach a customly-provided external storage, 2363 * which in this case is a netmap buffer. When calling m_extadd(), however 2364 * we pass a NULL address, since the real address (and length) will be 2365 * filled in by nm_os_generic_xmit_frame() right before calling 2366 * if_transmit(). 2367 * 2368 * The dtor function does nothing, however we need it since mb_free_ext() 2369 * has a KASSERT(), checking that the mbuf dtor function is not NULL. 2370 */ 2371 2372 #if __FreeBSD_version <= 1200050 2373 static void void_mbuf_dtor(struct mbuf *m, void *arg1, void *arg2) { } 2374 #else /* __FreeBSD_version >= 1200051 */ 2375 /* The arg1 and arg2 pointers argument were removed by r324446, which 2376 * in included since version 1200051. */ 2377 static void void_mbuf_dtor(struct mbuf *m) { } 2378 #endif /* __FreeBSD_version >= 1200051 */ 2379 2380 #define SET_MBUF_DESTRUCTOR(m, fn) do { \ 2381 (m)->m_ext.ext_free = (fn != NULL) ? \ 2382 (void *)fn : (void *)void_mbuf_dtor; \ 2383 } while (0) 2384 2385 static inline struct mbuf * 2386 nm_os_get_mbuf(struct ifnet *ifp, int len) 2387 { 2388 struct mbuf *m; 2389 2390 (void)ifp; 2391 (void)len; 2392 2393 m = m_gethdr(M_NOWAIT, MT_DATA); 2394 if (m == NULL) { 2395 return m; 2396 } 2397 2398 m_extadd(m, NULL /* buf */, 0 /* size */, void_mbuf_dtor, 2399 NULL, NULL, 0, EXT_NET_DRV); 2400 2401 return m; 2402 } 2403 2404 #endif /* __FreeBSD_version >= 1100000 */ 2405 #endif /* __FreeBSD__ */ 2406 2407 struct nmreq_option * nmreq_getoption(struct nmreq_header *, uint16_t); 2408 2409 int netmap_init_bridges(void); 2410 void netmap_uninit_bridges(void); 2411 2412 /* Functions to read and write CSB fields from the kernel. */ 2413 #if defined (linux) 2414 #define CSB_READ(csb, field, r) (get_user(r, &csb->field)) 2415 #define CSB_WRITE(csb, field, v) (put_user(v, &csb->field)) 2416 #else /* ! linux */ 2417 #define CSB_READ(csb, field, r) (r = fuword32(&csb->field)) 2418 #define CSB_WRITE(csb, field, v) (suword32(&csb->field, v)) 2419 #endif /* ! linux */ 2420 2421 #endif /* _NET_NETMAP_KERN_H_ */ 2422