/net/core/pktgen.c
C | 3896 lines | 2946 code | 619 blank | 331 comment | 595 complexity | c59dadde243d10df992a44b07ef85506 MD5 | raw file
1// SPDX-License-Identifier: GPL-2.0-or-later 2/* 3 * Authors: 4 * Copyright 2001, 2002 by Robert Olsson <robert.olsson@its.uu.se> 5 * Uppsala University and 6 * Swedish University of Agricultural Sciences 7 * 8 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 9 * Ben Greear <greearb@candelatech.com> 10 * Jens Låås <jens.laas@data.slu.se> 11 * 12 * A tool for loading the network with preconfigurated packets. 13 * The tool is implemented as a linux module. Parameters are output 14 * device, delay (to hard_xmit), number of packets, and whether 15 * to use multiple SKBs or just the same one. 16 * pktgen uses the installed interface's output routine. 17 * 18 * Additional hacking by: 19 * 20 * Jens.Laas@data.slu.se 21 * Improved by ANK. 010120. 22 * Improved by ANK even more. 010212. 23 * MAC address typo fixed. 010417 --ro 24 * Integrated. 020301 --DaveM 25 * Added multiskb option 020301 --DaveM 26 * Scaling of results. 020417--sigurdur@linpro.no 27 * Significant re-work of the module: 28 * * Convert to threaded model to more efficiently be able to transmit 29 * and receive on multiple interfaces at once. 30 * * Converted many counters to __u64 to allow longer runs. 31 * * Allow configuration of ranges, like min/max IP address, MACs, 32 * and UDP-ports, for both source and destination, and can 33 * set to use a random distribution or sequentially walk the range. 34 * * Can now change most values after starting. 35 * * Place 12-byte packet in UDP payload with magic number, 36 * sequence number, and timestamp. 37 * * Add receiver code that detects dropped pkts, re-ordered pkts, and 38 * latencies (with micro-second) precision. 39 * * Add IOCTL interface to easily get counters & configuration. 40 * --Ben Greear <greearb@candelatech.com> 41 * 42 * Renamed multiskb to clone_skb and cleaned up sending core for two distinct 43 * skb modes. A clone_skb=0 mode for Ben "ranges" work and a clone_skb != 0 44 * as a "fastpath" with a configurable number of clones after alloc's. 45 * clone_skb=0 means all packets are allocated this also means ranges time 46 * stamps etc can be used. clone_skb=100 means 1 malloc is followed by 100 47 * clones. 48 * 49 * Also moved to /proc/net/pktgen/ 50 * --ro 51 * 52 * Sept 10: Fixed threading/locking. Lots of bone-headed and more clever 53 * mistakes. Also merged in DaveM's patch in the -pre6 patch. 54 * --Ben Greear <greearb@candelatech.com> 55 * 56 * Integrated to 2.5.x 021029 --Lucio Maciel (luciomaciel@zipmail.com.br) 57 * 58 * 021124 Finished major redesign and rewrite for new functionality. 59 * See Documentation/networking/pktgen.txt for how to use this. 60 * 61 * The new operation: 62 * For each CPU one thread/process is created at start. This process checks 63 * for running devices in the if_list and sends packets until count is 0 it 64 * also the thread checks the thread->control which is used for inter-process 65 * communication. controlling process "posts" operations to the threads this 66 * way. 67 * The if_list is RCU protected, and the if_lock remains to protect updating 68 * of if_list, from "add_device" as it invoked from userspace (via proc write). 69 * 70 * By design there should only be *one* "controlling" process. In practice 71 * multiple write accesses gives unpredictable result. Understood by "write" 72 * to /proc gives result code thats should be read be the "writer". 73 * For practical use this should be no problem. 74 * 75 * Note when adding devices to a specific CPU there good idea to also assign 76 * /proc/irq/XX/smp_affinity so TX-interrupts gets bound to the same CPU. 77 * --ro 78 * 79 * Fix refcount off by one if first packet fails, potential null deref, 80 * memleak 030710- KJP 81 * 82 * First "ranges" functionality for ipv6 030726 --ro 83 * 84 * Included flow support. 030802 ANK. 85 * 86 * Fixed unaligned access on IA-64 Grant Grundler <grundler@parisc-linux.org> 87 * 88 * Remove if fix from added Harald Welte <laforge@netfilter.org> 040419 89 * ia64 compilation fix from Aron Griffis <aron@hp.com> 040604 90 * 91 * New xmit() return, do_div and misc clean up by Stephen Hemminger 92 * <shemminger@osdl.org> 040923 93 * 94 * Randy Dunlap fixed u64 printk compiler warning 95 * 96 * Remove FCS from BW calculation. Lennert Buytenhek <buytenh@wantstofly.org> 97 * New time handling. Lennert Buytenhek <buytenh@wantstofly.org> 041213 98 * 99 * Corrections from Nikolai Malykh (nmalykh@bilim.com) 100 * Removed unused flags F_SET_SRCMAC & F_SET_SRCIP 041230 101 * 102 * interruptible_sleep_on_timeout() replaced Nishanth Aravamudan <nacc@us.ibm.com> 103 * 050103 104 * 105 * MPLS support by Steven Whitehouse <steve@chygwyn.com> 106 * 107 * 802.1Q/Q-in-Q support by Francesco Fondelli (FF) <francesco.fondelli@gmail.com> 108 * 109 * Fixed src_mac command to set source mac of packet to value specified in 110 * command by Adit Ranadive <adit.262@gmail.com> 111 */ 112 113#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 114 115#include <linux/sys.h> 116#include <linux/types.h> 117#include <linux/module.h> 118#include <linux/moduleparam.h> 119#include <linux/kernel.h> 120#include <linux/mutex.h> 121#include <linux/sched.h> 122#include <linux/slab.h> 123#include <linux/vmalloc.h> 124#include <linux/unistd.h> 125#include <linux/string.h> 126#include <linux/ptrace.h> 127#include <linux/errno.h> 128#include <linux/ioport.h> 129#include <linux/interrupt.h> 130#include <linux/capability.h> 131#include <linux/hrtimer.h> 132#include <linux/freezer.h> 133#include <linux/delay.h> 134#include <linux/timer.h> 135#include <linux/list.h> 136#include <linux/init.h> 137#include <linux/skbuff.h> 138#include <linux/netdevice.h> 139#include <linux/inet.h> 140#include <linux/inetdevice.h> 141#include <linux/rtnetlink.h> 142#include <linux/if_arp.h> 143#include <linux/if_vlan.h> 144#include <linux/in.h> 145#include <linux/ip.h> 146#include <linux/ipv6.h> 147#include <linux/udp.h> 148#include <linux/proc_fs.h> 149#include <linux/seq_file.h> 150#include <linux/wait.h> 151#include <linux/etherdevice.h> 152#include <linux/kthread.h> 153#include <linux/prefetch.h> 154#include <linux/mmzone.h> 155#include <net/net_namespace.h> 156#include <net/checksum.h> 157#include <net/ipv6.h> 158#include <net/udp.h> 159#include <net/ip6_checksum.h> 160#include <net/addrconf.h> 161#ifdef CONFIG_XFRM 162#include <net/xfrm.h> 163#endif 164#include <net/netns/generic.h> 165#include <asm/byteorder.h> 166#include <linux/rcupdate.h> 167#include <linux/bitops.h> 168#include <linux/io.h> 169#include <linux/timex.h> 170#include <linux/uaccess.h> 171#include <asm/dma.h> 172#include <asm/div64.h> /* do_div */ 173 174#define VERSION "2.75" 175#define IP_NAME_SZ 32 176#define MAX_MPLS_LABELS 16 /* This is the max label stack depth */ 177#define MPLS_STACK_BOTTOM htonl(0x00000100) 178 179#define func_enter() pr_debug("entering %s\n", __func__); 180 181#define PKT_FLAGS \ 182 pf(IPV6) /* Interface in IPV6 Mode */ \ 183 pf(IPSRC_RND) /* IP-Src Random */ \ 184 pf(IPDST_RND) /* IP-Dst Random */ \ 185 pf(TXSIZE_RND) /* Transmit size is random */ \ 186 pf(UDPSRC_RND) /* UDP-Src Random */ \ 187 pf(UDPDST_RND) /* UDP-Dst Random */ \ 188 pf(UDPCSUM) /* Include UDP checksum */ \ 189 pf(NO_TIMESTAMP) /* Don't timestamp packets (default TS) */ \ 190 pf(MPLS_RND) /* Random MPLS labels */ \ 191 pf(QUEUE_MAP_RND) /* queue map Random */ \ 192 pf(QUEUE_MAP_CPU) /* queue map mirrors smp_processor_id() */ \ 193 pf(FLOW_SEQ) /* Sequential flows */ \ 194 pf(IPSEC) /* ipsec on for flows */ \ 195 pf(MACSRC_RND) /* MAC-Src Random */ \ 196 pf(MACDST_RND) /* MAC-Dst Random */ \ 197 pf(VID_RND) /* Random VLAN ID */ \ 198 pf(SVID_RND) /* Random SVLAN ID */ \ 199 pf(NODE) /* Node memory alloc*/ \ 200 201#define pf(flag) flag##_SHIFT, 202enum pkt_flags { 203 PKT_FLAGS 204}; 205#undef pf 206 207/* Device flag bits */ 208#define pf(flag) static const __u32 F_##flag = (1<<flag##_SHIFT); 209PKT_FLAGS 210#undef pf 211 212#define pf(flag) __stringify(flag), 213static char *pkt_flag_names[] = { 214 PKT_FLAGS 215}; 216#undef pf 217 218#define NR_PKT_FLAGS ARRAY_SIZE(pkt_flag_names) 219 220/* Thread control flag bits */ 221#define T_STOP (1<<0) /* Stop run */ 222#define T_RUN (1<<1) /* Start run */ 223#define T_REMDEVALL (1<<2) /* Remove all devs */ 224#define T_REMDEV (1<<3) /* Remove one dev */ 225 226/* Xmit modes */ 227#define M_START_XMIT 0 /* Default normal TX */ 228#define M_NETIF_RECEIVE 1 /* Inject packets into stack */ 229#define M_QUEUE_XMIT 2 /* Inject packet into qdisc */ 230 231/* If lock -- protects updating of if_list */ 232#define if_lock(t) mutex_lock(&(t->if_lock)); 233#define if_unlock(t) mutex_unlock(&(t->if_lock)); 234 235/* Used to help with determining the pkts on receive */ 236#define PKTGEN_MAGIC 0xbe9be955 237#define PG_PROC_DIR "pktgen" 238#define PGCTRL "pgctrl" 239 240#define MAX_CFLOWS 65536 241 242#define VLAN_TAG_SIZE(x) ((x)->vlan_id == 0xffff ? 0 : 4) 243#define SVLAN_TAG_SIZE(x) ((x)->svlan_id == 0xffff ? 0 : 4) 244 245struct flow_state { 246 __be32 cur_daddr; 247 int count; 248#ifdef CONFIG_XFRM 249 struct xfrm_state *x; 250#endif 251 __u32 flags; 252}; 253 254/* flow flag bits */ 255#define F_INIT (1<<0) /* flow has been initialized */ 256 257struct pktgen_dev { 258 /* 259 * Try to keep frequent/infrequent used vars. separated. 260 */ 261 struct proc_dir_entry *entry; /* proc file */ 262 struct pktgen_thread *pg_thread;/* the owner */ 263 struct list_head list; /* chaining in the thread's run-queue */ 264 struct rcu_head rcu; /* freed by RCU */ 265 266 int running; /* if false, the test will stop */ 267 268 /* If min != max, then we will either do a linear iteration, or 269 * we will do a random selection from within the range. 270 */ 271 __u32 flags; 272 int xmit_mode; 273 int min_pkt_size; 274 int max_pkt_size; 275 int pkt_overhead; /* overhead for MPLS, VLANs, IPSEC etc */ 276 int nfrags; 277 int removal_mark; /* non-zero => the device is marked for 278 * removal by worker thread */ 279 280 struct page *page; 281 u64 delay; /* nano-seconds */ 282 283 __u64 count; /* Default No packets to send */ 284 __u64 sofar; /* How many pkts we've sent so far */ 285 __u64 tx_bytes; /* How many bytes we've transmitted */ 286 __u64 errors; /* Errors when trying to transmit, */ 287 288 /* runtime counters relating to clone_skb */ 289 290 __u32 clone_count; 291 int last_ok; /* Was last skb sent? 292 * Or a failed transmit of some sort? 293 * This will keep sequence numbers in order 294 */ 295 ktime_t next_tx; 296 ktime_t started_at; 297 ktime_t stopped_at; 298 u64 idle_acc; /* nano-seconds */ 299 300 __u32 seq_num; 301 302 int clone_skb; /* 303 * Use multiple SKBs during packet gen. 304 * If this number is greater than 1, then 305 * that many copies of the same packet will be 306 * sent before a new packet is allocated. 307 * If you want to send 1024 identical packets 308 * before creating a new packet, 309 * set clone_skb to 1024. 310 */ 311 312 char dst_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 313 char dst_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 314 char src_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 315 char src_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 316 317 struct in6_addr in6_saddr; 318 struct in6_addr in6_daddr; 319 struct in6_addr cur_in6_daddr; 320 struct in6_addr cur_in6_saddr; 321 /* For ranges */ 322 struct in6_addr min_in6_daddr; 323 struct in6_addr max_in6_daddr; 324 struct in6_addr min_in6_saddr; 325 struct in6_addr max_in6_saddr; 326 327 /* If we're doing ranges, random or incremental, then this 328 * defines the min/max for those ranges. 329 */ 330 __be32 saddr_min; /* inclusive, source IP address */ 331 __be32 saddr_max; /* exclusive, source IP address */ 332 __be32 daddr_min; /* inclusive, dest IP address */ 333 __be32 daddr_max; /* exclusive, dest IP address */ 334 335 __u16 udp_src_min; /* inclusive, source UDP port */ 336 __u16 udp_src_max; /* exclusive, source UDP port */ 337 __u16 udp_dst_min; /* inclusive, dest UDP port */ 338 __u16 udp_dst_max; /* exclusive, dest UDP port */ 339 340 /* DSCP + ECN */ 341 __u8 tos; /* six MSB of (former) IPv4 TOS 342 are for dscp codepoint */ 343 __u8 traffic_class; /* ditto for the (former) Traffic Class in IPv6 344 (see RFC 3260, sec. 4) */ 345 346 /* MPLS */ 347 unsigned int nr_labels; /* Depth of stack, 0 = no MPLS */ 348 __be32 labels[MAX_MPLS_LABELS]; 349 350 /* VLAN/SVLAN (802.1Q/Q-in-Q) */ 351 __u8 vlan_p; 352 __u8 vlan_cfi; 353 __u16 vlan_id; /* 0xffff means no vlan tag */ 354 355 __u8 svlan_p; 356 __u8 svlan_cfi; 357 __u16 svlan_id; /* 0xffff means no svlan tag */ 358 359 __u32 src_mac_count; /* How many MACs to iterate through */ 360 __u32 dst_mac_count; /* How many MACs to iterate through */ 361 362 unsigned char dst_mac[ETH_ALEN]; 363 unsigned char src_mac[ETH_ALEN]; 364 365 __u32 cur_dst_mac_offset; 366 __u32 cur_src_mac_offset; 367 __be32 cur_saddr; 368 __be32 cur_daddr; 369 __u16 ip_id; 370 __u16 cur_udp_dst; 371 __u16 cur_udp_src; 372 __u16 cur_queue_map; 373 __u32 cur_pkt_size; 374 __u32 last_pkt_size; 375 376 __u8 hh[14]; 377 /* = { 378 0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB, 379 380 We fill in SRC address later 381 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 382 0x08, 0x00 383 }; 384 */ 385 __u16 pad; /* pad out the hh struct to an even 16 bytes */ 386 387 struct sk_buff *skb; /* skb we are to transmit next, used for when we 388 * are transmitting the same one multiple times 389 */ 390 struct net_device *odev; /* The out-going device. 391 * Note that the device should have it's 392 * pg_info pointer pointing back to this 393 * device. 394 * Set when the user specifies the out-going 395 * device name (not when the inject is 396 * started as it used to do.) 397 */ 398 char odevname[32]; 399 struct flow_state *flows; 400 unsigned int cflows; /* Concurrent flows (config) */ 401 unsigned int lflow; /* Flow length (config) */ 402 unsigned int nflows; /* accumulated flows (stats) */ 403 unsigned int curfl; /* current sequenced flow (state)*/ 404 405 u16 queue_map_min; 406 u16 queue_map_max; 407 __u32 skb_priority; /* skb priority field */ 408 unsigned int burst; /* number of duplicated packets to burst */ 409 int node; /* Memory node */ 410 411#ifdef CONFIG_XFRM 412 __u8 ipsmode; /* IPSEC mode (config) */ 413 __u8 ipsproto; /* IPSEC type (config) */ 414 __u32 spi; 415 struct xfrm_dst xdst; 416 struct dst_ops dstops; 417#endif 418 char result[512]; 419}; 420 421struct pktgen_hdr { 422 __be32 pgh_magic; 423 __be32 seq_num; 424 __be32 tv_sec; 425 __be32 tv_usec; 426}; 427 428 429static unsigned int pg_net_id __read_mostly; 430 431struct pktgen_net { 432 struct net *net; 433 struct proc_dir_entry *proc_dir; 434 struct list_head pktgen_threads; 435 bool pktgen_exiting; 436}; 437 438struct pktgen_thread { 439 struct mutex if_lock; /* for list of devices */ 440 struct list_head if_list; /* All device here */ 441 struct list_head th_list; 442 struct task_struct *tsk; 443 char result[512]; 444 445 /* Field for thread to receive "posted" events terminate, 446 stop ifs etc. */ 447 448 u32 control; 449 int cpu; 450 451 wait_queue_head_t queue; 452 struct completion start_done; 453 struct pktgen_net *net; 454}; 455 456#define REMOVE 1 457#define FIND 0 458 459static const char version[] = 460 "Packet Generator for packet performance testing. " 461 "Version: " VERSION "\n"; 462 463static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i); 464static int pktgen_add_device(struct pktgen_thread *t, const char *ifname); 465static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t, 466 const char *ifname, bool exact); 467static int pktgen_device_event(struct notifier_block *, unsigned long, void *); 468static void pktgen_run_all_threads(struct pktgen_net *pn); 469static void pktgen_reset_all_threads(struct pktgen_net *pn); 470static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn); 471 472static void pktgen_stop(struct pktgen_thread *t); 473static void pktgen_clear_counters(struct pktgen_dev *pkt_dev); 474 475/* Module parameters, defaults. */ 476static int pg_count_d __read_mostly = 1000; 477static int pg_delay_d __read_mostly; 478static int pg_clone_skb_d __read_mostly; 479static int debug __read_mostly; 480 481static DEFINE_MUTEX(pktgen_thread_lock); 482 483static struct notifier_block pktgen_notifier_block = { 484 .notifier_call = pktgen_device_event, 485}; 486 487/* 488 * /proc handling functions 489 * 490 */ 491 492static int pgctrl_show(struct seq_file *seq, void *v) 493{ 494 seq_puts(seq, version); 495 return 0; 496} 497 498static ssize_t pgctrl_write(struct file *file, const char __user *buf, 499 size_t count, loff_t *ppos) 500{ 501 char data[128]; 502 struct pktgen_net *pn = net_generic(current->nsproxy->net_ns, pg_net_id); 503 504 if (!capable(CAP_NET_ADMIN)) 505 return -EPERM; 506 507 if (count == 0) 508 return -EINVAL; 509 510 if (count > sizeof(data)) 511 count = sizeof(data); 512 513 if (copy_from_user(data, buf, count)) 514 return -EFAULT; 515 516 data[count - 1] = 0; /* Strip trailing '\n' and terminate string */ 517 518 if (!strcmp(data, "stop")) 519 pktgen_stop_all_threads_ifs(pn); 520 521 else if (!strcmp(data, "start")) 522 pktgen_run_all_threads(pn); 523 524 else if (!strcmp(data, "reset")) 525 pktgen_reset_all_threads(pn); 526 527 else 528 return -EINVAL; 529 530 return count; 531} 532 533static int pgctrl_open(struct inode *inode, struct file *file) 534{ 535 return single_open(file, pgctrl_show, PDE_DATA(inode)); 536} 537 538static const struct proc_ops pktgen_proc_ops = { 539 .proc_open = pgctrl_open, 540 .proc_read = seq_read, 541 .proc_lseek = seq_lseek, 542 .proc_write = pgctrl_write, 543 .proc_release = single_release, 544}; 545 546static int pktgen_if_show(struct seq_file *seq, void *v) 547{ 548 const struct pktgen_dev *pkt_dev = seq->private; 549 ktime_t stopped; 550 unsigned int i; 551 u64 idle; 552 553 seq_printf(seq, 554 "Params: count %llu min_pkt_size: %u max_pkt_size: %u\n", 555 (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size, 556 pkt_dev->max_pkt_size); 557 558 seq_printf(seq, 559 " frags: %d delay: %llu clone_skb: %d ifname: %s\n", 560 pkt_dev->nfrags, (unsigned long long) pkt_dev->delay, 561 pkt_dev->clone_skb, pkt_dev->odevname); 562 563 seq_printf(seq, " flows: %u flowlen: %u\n", pkt_dev->cflows, 564 pkt_dev->lflow); 565 566 seq_printf(seq, 567 " queue_map_min: %u queue_map_max: %u\n", 568 pkt_dev->queue_map_min, 569 pkt_dev->queue_map_max); 570 571 if (pkt_dev->skb_priority) 572 seq_printf(seq, " skb_priority: %u\n", 573 pkt_dev->skb_priority); 574 575 if (pkt_dev->flags & F_IPV6) { 576 seq_printf(seq, 577 " saddr: %pI6c min_saddr: %pI6c max_saddr: %pI6c\n" 578 " daddr: %pI6c min_daddr: %pI6c max_daddr: %pI6c\n", 579 &pkt_dev->in6_saddr, 580 &pkt_dev->min_in6_saddr, &pkt_dev->max_in6_saddr, 581 &pkt_dev->in6_daddr, 582 &pkt_dev->min_in6_daddr, &pkt_dev->max_in6_daddr); 583 } else { 584 seq_printf(seq, 585 " dst_min: %s dst_max: %s\n", 586 pkt_dev->dst_min, pkt_dev->dst_max); 587 seq_printf(seq, 588 " src_min: %s src_max: %s\n", 589 pkt_dev->src_min, pkt_dev->src_max); 590 } 591 592 seq_puts(seq, " src_mac: "); 593 594 seq_printf(seq, "%pM ", 595 is_zero_ether_addr(pkt_dev->src_mac) ? 596 pkt_dev->odev->dev_addr : pkt_dev->src_mac); 597 598 seq_puts(seq, "dst_mac: "); 599 seq_printf(seq, "%pM\n", pkt_dev->dst_mac); 600 601 seq_printf(seq, 602 " udp_src_min: %d udp_src_max: %d" 603 " udp_dst_min: %d udp_dst_max: %d\n", 604 pkt_dev->udp_src_min, pkt_dev->udp_src_max, 605 pkt_dev->udp_dst_min, pkt_dev->udp_dst_max); 606 607 seq_printf(seq, 608 " src_mac_count: %d dst_mac_count: %d\n", 609 pkt_dev->src_mac_count, pkt_dev->dst_mac_count); 610 611 if (pkt_dev->nr_labels) { 612 seq_puts(seq, " mpls: "); 613 for (i = 0; i < pkt_dev->nr_labels; i++) 614 seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]), 615 i == pkt_dev->nr_labels-1 ? "\n" : ", "); 616 } 617 618 if (pkt_dev->vlan_id != 0xffff) 619 seq_printf(seq, " vlan_id: %u vlan_p: %u vlan_cfi: %u\n", 620 pkt_dev->vlan_id, pkt_dev->vlan_p, 621 pkt_dev->vlan_cfi); 622 623 if (pkt_dev->svlan_id != 0xffff) 624 seq_printf(seq, " svlan_id: %u vlan_p: %u vlan_cfi: %u\n", 625 pkt_dev->svlan_id, pkt_dev->svlan_p, 626 pkt_dev->svlan_cfi); 627 628 if (pkt_dev->tos) 629 seq_printf(seq, " tos: 0x%02x\n", pkt_dev->tos); 630 631 if (pkt_dev->traffic_class) 632 seq_printf(seq, " traffic_class: 0x%02x\n", pkt_dev->traffic_class); 633 634 if (pkt_dev->burst > 1) 635 seq_printf(seq, " burst: %d\n", pkt_dev->burst); 636 637 if (pkt_dev->node >= 0) 638 seq_printf(seq, " node: %d\n", pkt_dev->node); 639 640 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) 641 seq_puts(seq, " xmit_mode: netif_receive\n"); 642 else if (pkt_dev->xmit_mode == M_QUEUE_XMIT) 643 seq_puts(seq, " xmit_mode: xmit_queue\n"); 644 645 seq_puts(seq, " Flags: "); 646 647 for (i = 0; i < NR_PKT_FLAGS; i++) { 648 if (i == F_FLOW_SEQ) 649 if (!pkt_dev->cflows) 650 continue; 651 652 if (pkt_dev->flags & (1 << i)) 653 seq_printf(seq, "%s ", pkt_flag_names[i]); 654 else if (i == F_FLOW_SEQ) 655 seq_puts(seq, "FLOW_RND "); 656 657#ifdef CONFIG_XFRM 658 if (i == F_IPSEC && pkt_dev->spi) 659 seq_printf(seq, "spi:%u", pkt_dev->spi); 660#endif 661 } 662 663 seq_puts(seq, "\n"); 664 665 /* not really stopped, more like last-running-at */ 666 stopped = pkt_dev->running ? ktime_get() : pkt_dev->stopped_at; 667 idle = pkt_dev->idle_acc; 668 do_div(idle, NSEC_PER_USEC); 669 670 seq_printf(seq, 671 "Current:\n pkts-sofar: %llu errors: %llu\n", 672 (unsigned long long)pkt_dev->sofar, 673 (unsigned long long)pkt_dev->errors); 674 675 seq_printf(seq, 676 " started: %lluus stopped: %lluus idle: %lluus\n", 677 (unsigned long long) ktime_to_us(pkt_dev->started_at), 678 (unsigned long long) ktime_to_us(stopped), 679 (unsigned long long) idle); 680 681 seq_printf(seq, 682 " seq_num: %d cur_dst_mac_offset: %d cur_src_mac_offset: %d\n", 683 pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset, 684 pkt_dev->cur_src_mac_offset); 685 686 if (pkt_dev->flags & F_IPV6) { 687 seq_printf(seq, " cur_saddr: %pI6c cur_daddr: %pI6c\n", 688 &pkt_dev->cur_in6_saddr, 689 &pkt_dev->cur_in6_daddr); 690 } else 691 seq_printf(seq, " cur_saddr: %pI4 cur_daddr: %pI4\n", 692 &pkt_dev->cur_saddr, &pkt_dev->cur_daddr); 693 694 seq_printf(seq, " cur_udp_dst: %d cur_udp_src: %d\n", 695 pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src); 696 697 seq_printf(seq, " cur_queue_map: %u\n", pkt_dev->cur_queue_map); 698 699 seq_printf(seq, " flows: %u\n", pkt_dev->nflows); 700 701 if (pkt_dev->result[0]) 702 seq_printf(seq, "Result: %s\n", pkt_dev->result); 703 else 704 seq_puts(seq, "Result: Idle\n"); 705 706 return 0; 707} 708 709 710static int hex32_arg(const char __user *user_buffer, unsigned long maxlen, 711 __u32 *num) 712{ 713 int i = 0; 714 *num = 0; 715 716 for (; i < maxlen; i++) { 717 int value; 718 char c; 719 *num <<= 4; 720 if (get_user(c, &user_buffer[i])) 721 return -EFAULT; 722 value = hex_to_bin(c); 723 if (value >= 0) 724 *num |= value; 725 else 726 break; 727 } 728 return i; 729} 730 731static int count_trail_chars(const char __user * user_buffer, 732 unsigned int maxlen) 733{ 734 int i; 735 736 for (i = 0; i < maxlen; i++) { 737 char c; 738 if (get_user(c, &user_buffer[i])) 739 return -EFAULT; 740 switch (c) { 741 case '\"': 742 case '\n': 743 case '\r': 744 case '\t': 745 case ' ': 746 case '=': 747 break; 748 default: 749 goto done; 750 } 751 } 752done: 753 return i; 754} 755 756static long num_arg(const char __user *user_buffer, unsigned long maxlen, 757 unsigned long *num) 758{ 759 int i; 760 *num = 0; 761 762 for (i = 0; i < maxlen; i++) { 763 char c; 764 if (get_user(c, &user_buffer[i])) 765 return -EFAULT; 766 if ((c >= '0') && (c <= '9')) { 767 *num *= 10; 768 *num += c - '0'; 769 } else 770 break; 771 } 772 return i; 773} 774 775static int strn_len(const char __user * user_buffer, unsigned int maxlen) 776{ 777 int i; 778 779 for (i = 0; i < maxlen; i++) { 780 char c; 781 if (get_user(c, &user_buffer[i])) 782 return -EFAULT; 783 switch (c) { 784 case '\"': 785 case '\n': 786 case '\r': 787 case '\t': 788 case ' ': 789 goto done_str; 790 default: 791 break; 792 } 793 } 794done_str: 795 return i; 796} 797 798static ssize_t get_labels(const char __user *buffer, struct pktgen_dev *pkt_dev) 799{ 800 unsigned int n = 0; 801 char c; 802 ssize_t i = 0; 803 int len; 804 805 pkt_dev->nr_labels = 0; 806 do { 807 __u32 tmp; 808 len = hex32_arg(&buffer[i], 8, &tmp); 809 if (len <= 0) 810 return len; 811 pkt_dev->labels[n] = htonl(tmp); 812 if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM) 813 pkt_dev->flags |= F_MPLS_RND; 814 i += len; 815 if (get_user(c, &buffer[i])) 816 return -EFAULT; 817 i++; 818 n++; 819 if (n >= MAX_MPLS_LABELS) 820 return -E2BIG; 821 } while (c == ','); 822 823 pkt_dev->nr_labels = n; 824 return i; 825} 826 827static __u32 pktgen_read_flag(const char *f, bool *disable) 828{ 829 __u32 i; 830 831 if (f[0] == '!') { 832 *disable = true; 833 f++; 834 } 835 836 for (i = 0; i < NR_PKT_FLAGS; i++) { 837 if (!IS_ENABLED(CONFIG_XFRM) && i == IPSEC_SHIFT) 838 continue; 839 840 /* allow only disabling ipv6 flag */ 841 if (!*disable && i == IPV6_SHIFT) 842 continue; 843 844 if (strcmp(f, pkt_flag_names[i]) == 0) 845 return 1 << i; 846 } 847 848 if (strcmp(f, "FLOW_RND") == 0) { 849 *disable = !*disable; 850 return F_FLOW_SEQ; 851 } 852 853 return 0; 854} 855 856static ssize_t pktgen_if_write(struct file *file, 857 const char __user * user_buffer, size_t count, 858 loff_t * offset) 859{ 860 struct seq_file *seq = file->private_data; 861 struct pktgen_dev *pkt_dev = seq->private; 862 int i, max, len; 863 char name[16], valstr[32]; 864 unsigned long value = 0; 865 char *pg_result = NULL; 866 int tmp = 0; 867 char buf[128]; 868 869 pg_result = &(pkt_dev->result[0]); 870 871 if (count < 1) { 872 pr_warn("wrong command format\n"); 873 return -EINVAL; 874 } 875 876 max = count; 877 tmp = count_trail_chars(user_buffer, max); 878 if (tmp < 0) { 879 pr_warn("illegal format\n"); 880 return tmp; 881 } 882 i = tmp; 883 884 /* Read variable name */ 885 886 len = strn_len(&user_buffer[i], sizeof(name) - 1); 887 if (len < 0) 888 return len; 889 890 memset(name, 0, sizeof(name)); 891 if (copy_from_user(name, &user_buffer[i], len)) 892 return -EFAULT; 893 i += len; 894 895 max = count - i; 896 len = count_trail_chars(&user_buffer[i], max); 897 if (len < 0) 898 return len; 899 900 i += len; 901 902 if (debug) { 903 size_t copy = min_t(size_t, count + 1, 1024); 904 char *tp = strndup_user(user_buffer, copy); 905 906 if (IS_ERR(tp)) 907 return PTR_ERR(tp); 908 909 pr_debug("%s,%zu buffer -:%s:-\n", name, count, tp); 910 kfree(tp); 911 } 912 913 if (!strcmp(name, "min_pkt_size")) { 914 len = num_arg(&user_buffer[i], 10, &value); 915 if (len < 0) 916 return len; 917 918 i += len; 919 if (value < 14 + 20 + 8) 920 value = 14 + 20 + 8; 921 if (value != pkt_dev->min_pkt_size) { 922 pkt_dev->min_pkt_size = value; 923 pkt_dev->cur_pkt_size = value; 924 } 925 sprintf(pg_result, "OK: min_pkt_size=%u", 926 pkt_dev->min_pkt_size); 927 return count; 928 } 929 930 if (!strcmp(name, "max_pkt_size")) { 931 len = num_arg(&user_buffer[i], 10, &value); 932 if (len < 0) 933 return len; 934 935 i += len; 936 if (value < 14 + 20 + 8) 937 value = 14 + 20 + 8; 938 if (value != pkt_dev->max_pkt_size) { 939 pkt_dev->max_pkt_size = value; 940 pkt_dev->cur_pkt_size = value; 941 } 942 sprintf(pg_result, "OK: max_pkt_size=%u", 943 pkt_dev->max_pkt_size); 944 return count; 945 } 946 947 /* Shortcut for min = max */ 948 949 if (!strcmp(name, "pkt_size")) { 950 len = num_arg(&user_buffer[i], 10, &value); 951 if (len < 0) 952 return len; 953 954 i += len; 955 if (value < 14 + 20 + 8) 956 value = 14 + 20 + 8; 957 if (value != pkt_dev->min_pkt_size) { 958 pkt_dev->min_pkt_size = value; 959 pkt_dev->max_pkt_size = value; 960 pkt_dev->cur_pkt_size = value; 961 } 962 sprintf(pg_result, "OK: pkt_size=%u", pkt_dev->min_pkt_size); 963 return count; 964 } 965 966 if (!strcmp(name, "debug")) { 967 len = num_arg(&user_buffer[i], 10, &value); 968 if (len < 0) 969 return len; 970 971 i += len; 972 debug = value; 973 sprintf(pg_result, "OK: debug=%u", debug); 974 return count; 975 } 976 977 if (!strcmp(name, "frags")) { 978 len = num_arg(&user_buffer[i], 10, &value); 979 if (len < 0) 980 return len; 981 982 i += len; 983 pkt_dev->nfrags = value; 984 sprintf(pg_result, "OK: frags=%u", pkt_dev->nfrags); 985 return count; 986 } 987 if (!strcmp(name, "delay")) { 988 len = num_arg(&user_buffer[i], 10, &value); 989 if (len < 0) 990 return len; 991 992 i += len; 993 if (value == 0x7FFFFFFF) 994 pkt_dev->delay = ULLONG_MAX; 995 else 996 pkt_dev->delay = (u64)value; 997 998 sprintf(pg_result, "OK: delay=%llu", 999 (unsigned long long) pkt_dev->delay); 1000 return count; 1001 } 1002 if (!strcmp(name, "rate")) { 1003 len = num_arg(&user_buffer[i], 10, &value); 1004 if (len < 0) 1005 return len; 1006 1007 i += len; 1008 if (!value) 1009 return len; 1010 pkt_dev->delay = pkt_dev->min_pkt_size*8*NSEC_PER_USEC/value; 1011 if (debug) 1012 pr_info("Delay set at: %llu ns\n", pkt_dev->delay); 1013 1014 sprintf(pg_result, "OK: rate=%lu", value); 1015 return count; 1016 } 1017 if (!strcmp(name, "ratep")) { 1018 len = num_arg(&user_buffer[i], 10, &value); 1019 if (len < 0) 1020 return len; 1021 1022 i += len; 1023 if (!value) 1024 return len; 1025 pkt_dev->delay = NSEC_PER_SEC/value; 1026 if (debug) 1027 pr_info("Delay set at: %llu ns\n", pkt_dev->delay); 1028 1029 sprintf(pg_result, "OK: rate=%lu", value); 1030 return count; 1031 } 1032 if (!strcmp(name, "udp_src_min")) { 1033 len = num_arg(&user_buffer[i], 10, &value); 1034 if (len < 0) 1035 return len; 1036 1037 i += len; 1038 if (value != pkt_dev->udp_src_min) { 1039 pkt_dev->udp_src_min = value; 1040 pkt_dev->cur_udp_src = value; 1041 } 1042 sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min); 1043 return count; 1044 } 1045 if (!strcmp(name, "udp_dst_min")) { 1046 len = num_arg(&user_buffer[i], 10, &value); 1047 if (len < 0) 1048 return len; 1049 1050 i += len; 1051 if (value != pkt_dev->udp_dst_min) { 1052 pkt_dev->udp_dst_min = value; 1053 pkt_dev->cur_udp_dst = value; 1054 } 1055 sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min); 1056 return count; 1057 } 1058 if (!strcmp(name, "udp_src_max")) { 1059 len = num_arg(&user_buffer[i], 10, &value); 1060 if (len < 0) 1061 return len; 1062 1063 i += len; 1064 if (value != pkt_dev->udp_src_max) { 1065 pkt_dev->udp_src_max = value; 1066 pkt_dev->cur_udp_src = value; 1067 } 1068 sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max); 1069 return count; 1070 } 1071 if (!strcmp(name, "udp_dst_max")) { 1072 len = num_arg(&user_buffer[i], 10, &value); 1073 if (len < 0) 1074 return len; 1075 1076 i += len; 1077 if (value != pkt_dev->udp_dst_max) { 1078 pkt_dev->udp_dst_max = value; 1079 pkt_dev->cur_udp_dst = value; 1080 } 1081 sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max); 1082 return count; 1083 } 1084 if (!strcmp(name, "clone_skb")) { 1085 len = num_arg(&user_buffer[i], 10, &value); 1086 if (len < 0) 1087 return len; 1088 if ((value > 0) && 1089 ((pkt_dev->xmit_mode == M_NETIF_RECEIVE) || 1090 !(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING))) 1091 return -ENOTSUPP; 1092 i += len; 1093 pkt_dev->clone_skb = value; 1094 1095 sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb); 1096 return count; 1097 } 1098 if (!strcmp(name, "count")) { 1099 len = num_arg(&user_buffer[i], 10, &value); 1100 if (len < 0) 1101 return len; 1102 1103 i += len; 1104 pkt_dev->count = value; 1105 sprintf(pg_result, "OK: count=%llu", 1106 (unsigned long long)pkt_dev->count); 1107 return count; 1108 } 1109 if (!strcmp(name, "src_mac_count")) { 1110 len = num_arg(&user_buffer[i], 10, &value); 1111 if (len < 0) 1112 return len; 1113 1114 i += len; 1115 if (pkt_dev->src_mac_count != value) { 1116 pkt_dev->src_mac_count = value; 1117 pkt_dev->cur_src_mac_offset = 0; 1118 } 1119 sprintf(pg_result, "OK: src_mac_count=%d", 1120 pkt_dev->src_mac_count); 1121 return count; 1122 } 1123 if (!strcmp(name, "dst_mac_count")) { 1124 len = num_arg(&user_buffer[i], 10, &value); 1125 if (len < 0) 1126 return len; 1127 1128 i += len; 1129 if (pkt_dev->dst_mac_count != value) { 1130 pkt_dev->dst_mac_count = value; 1131 pkt_dev->cur_dst_mac_offset = 0; 1132 } 1133 sprintf(pg_result, "OK: dst_mac_count=%d", 1134 pkt_dev->dst_mac_count); 1135 return count; 1136 } 1137 if (!strcmp(name, "burst")) { 1138 len = num_arg(&user_buffer[i], 10, &value); 1139 if (len < 0) 1140 return len; 1141 1142 i += len; 1143 if ((value > 1) && 1144 ((pkt_dev->xmit_mode == M_QUEUE_XMIT) || 1145 ((pkt_dev->xmit_mode == M_START_XMIT) && 1146 (!(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING))))) 1147 return -ENOTSUPP; 1148 pkt_dev->burst = value < 1 ? 1 : value; 1149 sprintf(pg_result, "OK: burst=%d", pkt_dev->burst); 1150 return count; 1151 } 1152 if (!strcmp(name, "node")) { 1153 len = num_arg(&user_buffer[i], 10, &value); 1154 if (len < 0) 1155 return len; 1156 1157 i += len; 1158 1159 if (node_possible(value)) { 1160 pkt_dev->node = value; 1161 sprintf(pg_result, "OK: node=%d", pkt_dev->node); 1162 if (pkt_dev->page) { 1163 put_page(pkt_dev->page); 1164 pkt_dev->page = NULL; 1165 } 1166 } 1167 else 1168 sprintf(pg_result, "ERROR: node not possible"); 1169 return count; 1170 } 1171 if (!strcmp(name, "xmit_mode")) { 1172 char f[32]; 1173 1174 memset(f, 0, 32); 1175 len = strn_len(&user_buffer[i], sizeof(f) - 1); 1176 if (len < 0) 1177 return len; 1178 1179 if (copy_from_user(f, &user_buffer[i], len)) 1180 return -EFAULT; 1181 i += len; 1182 1183 if (strcmp(f, "start_xmit") == 0) { 1184 pkt_dev->xmit_mode = M_START_XMIT; 1185 } else if (strcmp(f, "netif_receive") == 0) { 1186 /* clone_skb set earlier, not supported in this mode */ 1187 if (pkt_dev->clone_skb > 0) 1188 return -ENOTSUPP; 1189 1190 pkt_dev->xmit_mode = M_NETIF_RECEIVE; 1191 1192 /* make sure new packet is allocated every time 1193 * pktgen_xmit() is called 1194 */ 1195 pkt_dev->last_ok = 1; 1196 1197 /* override clone_skb if user passed default value 1198 * at module loading time 1199 */ 1200 pkt_dev->clone_skb = 0; 1201 } else if (strcmp(f, "queue_xmit") == 0) { 1202 pkt_dev->xmit_mode = M_QUEUE_XMIT; 1203 pkt_dev->last_ok = 1; 1204 } else { 1205 sprintf(pg_result, 1206 "xmit_mode -:%s:- unknown\nAvailable modes: %s", 1207 f, "start_xmit, netif_receive\n"); 1208 return count; 1209 } 1210 sprintf(pg_result, "OK: xmit_mode=%s", f); 1211 return count; 1212 } 1213 if (!strcmp(name, "flag")) { 1214 __u32 flag; 1215 char f[32]; 1216 bool disable = false; 1217 1218 memset(f, 0, 32); 1219 len = strn_len(&user_buffer[i], sizeof(f) - 1); 1220 if (len < 0) 1221 return len; 1222 1223 if (copy_from_user(f, &user_buffer[i], len)) 1224 return -EFAULT; 1225 i += len; 1226 1227 flag = pktgen_read_flag(f, &disable); 1228 1229 if (flag) { 1230 if (disable) 1231 pkt_dev->flags &= ~flag; 1232 else 1233 pkt_dev->flags |= flag; 1234 } else { 1235 sprintf(pg_result, 1236 "Flag -:%s:- unknown\nAvailable flags, (prepend ! to un-set flag):\n%s", 1237 f, 1238 "IPSRC_RND, IPDST_RND, UDPSRC_RND, UDPDST_RND, " 1239 "MACSRC_RND, MACDST_RND, TXSIZE_RND, IPV6, " 1240 "MPLS_RND, VID_RND, SVID_RND, FLOW_SEQ, " 1241 "QUEUE_MAP_RND, QUEUE_MAP_CPU, UDPCSUM, " 1242 "NO_TIMESTAMP, " 1243#ifdef CONFIG_XFRM 1244 "IPSEC, " 1245#endif 1246 "NODE_ALLOC\n"); 1247 return count; 1248 } 1249 sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags); 1250 return count; 1251 } 1252 if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) { 1253 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_min) - 1); 1254 if (len < 0) 1255 return len; 1256 1257 if (copy_from_user(buf, &user_buffer[i], len)) 1258 return -EFAULT; 1259 buf[len] = 0; 1260 if (strcmp(buf, pkt_dev->dst_min) != 0) { 1261 memset(pkt_dev->dst_min, 0, sizeof(pkt_dev->dst_min)); 1262 strcpy(pkt_dev->dst_min, buf); 1263 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min); 1264 pkt_dev->cur_daddr = pkt_dev->daddr_min; 1265 } 1266 if (debug) 1267 pr_debug("dst_min set to: %s\n", pkt_dev->dst_min); 1268 i += len; 1269 sprintf(pg_result, "OK: dst_min=%s", pkt_dev->dst_min); 1270 return count; 1271 } 1272 if (!strcmp(name, "dst_max")) { 1273 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_max) - 1); 1274 if (len < 0) 1275 return len; 1276 1277 if (copy_from_user(buf, &user_buffer[i], len)) 1278 return -EFAULT; 1279 buf[len] = 0; 1280 if (strcmp(buf, pkt_dev->dst_max) != 0) { 1281 memset(pkt_dev->dst_max, 0, sizeof(pkt_dev->dst_max)); 1282 strcpy(pkt_dev->dst_max, buf); 1283 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max); 1284 pkt_dev->cur_daddr = pkt_dev->daddr_max; 1285 } 1286 if (debug) 1287 pr_debug("dst_max set to: %s\n", pkt_dev->dst_max); 1288 i += len; 1289 sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max); 1290 return count; 1291 } 1292 if (!strcmp(name, "dst6")) { 1293 len = strn_len(&user_buffer[i], sizeof(buf) - 1); 1294 if (len < 0) 1295 return len; 1296 1297 pkt_dev->flags |= F_IPV6; 1298 1299 if (copy_from_user(buf, &user_buffer[i], len)) 1300 return -EFAULT; 1301 buf[len] = 0; 1302 1303 in6_pton(buf, -1, pkt_dev->in6_daddr.s6_addr, -1, NULL); 1304 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_daddr); 1305 1306 pkt_dev->cur_in6_daddr = pkt_dev->in6_daddr; 1307 1308 if (debug) 1309 pr_debug("dst6 set to: %s\n", buf); 1310 1311 i += len; 1312 sprintf(pg_result, "OK: dst6=%s", buf); 1313 return count; 1314 } 1315 if (!strcmp(name, "dst6_min")) { 1316 len = strn_len(&user_buffer[i], sizeof(buf) - 1); 1317 if (len < 0) 1318 return len; 1319 1320 pkt_dev->flags |= F_IPV6; 1321 1322 if (copy_from_user(buf, &user_buffer[i], len)) 1323 return -EFAULT; 1324 buf[len] = 0; 1325 1326 in6_pton(buf, -1, pkt_dev->min_in6_daddr.s6_addr, -1, NULL); 1327 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->min_in6_daddr); 1328 1329 pkt_dev->cur_in6_daddr = pkt_dev->min_in6_daddr; 1330 if (debug) 1331 pr_debug("dst6_min set to: %s\n", buf); 1332 1333 i += len; 1334 sprintf(pg_result, "OK: dst6_min=%s", buf); 1335 return count; 1336 } 1337 if (!strcmp(name, "dst6_max")) { 1338 len = strn_len(&user_buffer[i], sizeof(buf) - 1); 1339 if (len < 0) 1340 return len; 1341 1342 pkt_dev->flags |= F_IPV6; 1343 1344 if (copy_from_user(buf, &user_buffer[i], len)) 1345 return -EFAULT; 1346 buf[len] = 0; 1347 1348 in6_pton(buf, -1, pkt_dev->max_in6_daddr.s6_addr, -1, NULL); 1349 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->max_in6_daddr); 1350 1351 if (debug) 1352 pr_debug("dst6_max set to: %s\n", buf); 1353 1354 i += len; 1355 sprintf(pg_result, "OK: dst6_max=%s", buf); 1356 return count; 1357 } 1358 if (!strcmp(name, "src6")) { 1359 len = strn_len(&user_buffer[i], sizeof(buf) - 1); 1360 if (len < 0) 1361 return len; 1362 1363 pkt_dev->flags |= F_IPV6; 1364 1365 if (copy_from_user(buf, &user_buffer[i], len)) 1366 return -EFAULT; 1367 buf[len] = 0; 1368 1369 in6_pton(buf, -1, pkt_dev->in6_saddr.s6_addr, -1, NULL); 1370 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_saddr); 1371 1372 pkt_dev->cur_in6_saddr = pkt_dev->in6_saddr; 1373 1374 if (debug) 1375 pr_debug("src6 set to: %s\n", buf); 1376 1377 i += len; 1378 sprintf(pg_result, "OK: src6=%s", buf); 1379 return count; 1380 } 1381 if (!strcmp(name, "src_min")) { 1382 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_min) - 1); 1383 if (len < 0) 1384 return len; 1385 1386 if (copy_from_user(buf, &user_buffer[i], len)) 1387 return -EFAULT; 1388 buf[len] = 0; 1389 if (strcmp(buf, pkt_dev->src_min) != 0) { 1390 memset(pkt_dev->src_min, 0, sizeof(pkt_dev->src_min)); 1391 strcpy(pkt_dev->src_min, buf); 1392 pkt_dev->saddr_min = in_aton(pkt_dev->src_min); 1393 pkt_dev->cur_saddr = pkt_dev->saddr_min; 1394 } 1395 if (debug) 1396 pr_debug("src_min set to: %s\n", pkt_dev->src_min); 1397 i += len; 1398 sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min); 1399 return count; 1400 } 1401 if (!strcmp(name, "src_max")) { 1402 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_max) - 1); 1403 if (len < 0) 1404 return len; 1405 1406 if (copy_from_user(buf, &user_buffer[i], len)) 1407 return -EFAULT; 1408 buf[len] = 0; 1409 if (strcmp(buf, pkt_dev->src_max) != 0) { 1410 memset(pkt_dev->src_max, 0, sizeof(pkt_dev->src_max)); 1411 strcpy(pkt_dev->src_max, buf); 1412 pkt_dev->saddr_max = in_aton(pkt_dev->src_max); 1413 pkt_dev->cur_saddr = pkt_dev->saddr_max; 1414 } 1415 if (debug) 1416 pr_debug("src_max set to: %s\n", pkt_dev->src_max); 1417 i += len; 1418 sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max); 1419 return count; 1420 } 1421 if (!strcmp(name, "dst_mac")) { 1422 len = strn_len(&user_buffer[i], sizeof(valstr) - 1); 1423 if (len < 0) 1424 return len; 1425 1426 memset(valstr, 0, sizeof(valstr)); 1427 if (copy_from_user(valstr, &user_buffer[i], len)) 1428 return -EFAULT; 1429 1430 if (!mac_pton(valstr, pkt_dev->dst_mac)) 1431 return -EINVAL; 1432 /* Set up Dest MAC */ 1433 ether_addr_copy(&pkt_dev->hh[0], pkt_dev->dst_mac); 1434 1435 sprintf(pg_result, "OK: dstmac %pM", pkt_dev->dst_mac); 1436 return count; 1437 } 1438 if (!strcmp(name, "src_mac")) { 1439 len = strn_len(&user_buffer[i], sizeof(valstr) - 1); 1440 if (len < 0) 1441 return len; 1442 1443 memset(valstr, 0, sizeof(valstr)); 1444 if (copy_from_user(valstr, &user_buffer[i], len)) 1445 return -EFAULT; 1446 1447 if (!mac_pton(valstr, pkt_dev->src_mac)) 1448 return -EINVAL; 1449 /* Set up Src MAC */ 1450 ether_addr_copy(&pkt_dev->hh[6], pkt_dev->src_mac); 1451 1452 sprintf(pg_result, "OK: srcmac %pM", pkt_dev->src_mac); 1453 return count; 1454 } 1455 1456 if (!strcmp(name, "clear_counters")) { 1457 pktgen_clear_counters(pkt_dev); 1458 sprintf(pg_result, "OK: Clearing counters.\n"); 1459 return count; 1460 } 1461 1462 if (!strcmp(name, "flows")) { 1463 len = num_arg(&user_buffer[i], 10, &value); 1464 if (len < 0) 1465 return len; 1466 1467 i += len; 1468 if (value > MAX_CFLOWS) 1469 value = MAX_CFLOWS; 1470 1471 pkt_dev->cflows = value; 1472 sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows); 1473 return count; 1474 } 1475#ifdef CONFIG_XFRM 1476 if (!strcmp(name, "spi")) { 1477 len = num_arg(&user_buffer[i], 10, &value); 1478 if (len < 0) 1479 return len; 1480 1481 i += len; 1482 pkt_dev->spi = value; 1483 sprintf(pg_result, "OK: spi=%u", pkt_dev->spi); 1484 return count; 1485 } 1486#endif 1487 if (!strcmp(name, "flowlen")) { 1488 len = num_arg(&user_buffer[i], 10, &value); 1489 if (len < 0) 1490 return len; 1491 1492 i += len; 1493 pkt_dev->lflow = value; 1494 sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow); 1495 return count; 1496 } 1497 1498 if (!strcmp(name, "queue_map_min")) { 1499 len = num_arg(&user_buffer[i], 5, &value); 1500 if (len < 0) 1501 return len; 1502 1503 i += len; 1504 pkt_dev->queue_map_min = value; 1505 sprintf(pg_result, "OK: queue_map_min=%u", pkt_dev->queue_map_min); 1506 return count; 1507 } 1508 1509 if (!strcmp(name, "queue_map_max")) { 1510 len = num_arg(&user_buffer[i], 5, &value); 1511 if (len < 0) 1512 return len; 1513 1514 i += len; 1515 pkt_dev->queue_map_max = value; 1516 sprintf(pg_result, "OK: queue_map_max=%u", pkt_dev->queue_map_max); 1517 return count; 1518 } 1519 1520 if (!strcmp(name, "mpls")) { 1521 unsigned int n, cnt; 1522 1523 len = get_labels(&user_buffer[i], pkt_dev); 1524 if (len < 0) 1525 return len; 1526 i += len; 1527 cnt = sprintf(pg_result, "OK: mpls="); 1528 for (n = 0; n < pkt_dev->nr_labels; n++) 1529 cnt += sprintf(pg_result + cnt, 1530 "%08x%s", ntohl(pkt_dev->labels[n]), 1531 n == pkt_dev->nr_labels-1 ? "" : ","); 1532 1533 if (pkt_dev->nr_labels && pkt_dev->vlan_id != 0xffff) { 1534 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1535 pkt_dev->svlan_id = 0xffff; 1536 1537 if (debug) 1538 pr_debug("VLAN/SVLAN auto turned off\n"); 1539 } 1540 return count; 1541 } 1542 1543 if (!strcmp(name, "vlan_id")) { 1544 len = num_arg(&user_buffer[i], 4, &value); 1545 if (len < 0) 1546 return len; 1547 1548 i += len; 1549 if (value <= 4095) { 1550 pkt_dev->vlan_id = value; /* turn on VLAN */ 1551 1552 if (debug) 1553 pr_debug("VLAN turned on\n"); 1554 1555 if (debug && pkt_dev->nr_labels) 1556 pr_debug("MPLS auto turned off\n"); 1557 1558 pkt_dev->nr_labels = 0; /* turn off MPLS */ 1559 sprintf(pg_result, "OK: vlan_id=%u", pkt_dev->vlan_id); 1560 } else { 1561 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1562 pkt_dev->svlan_id = 0xffff; 1563 1564 if (debug) 1565 pr_debug("VLAN/SVLAN turned off\n"); 1566 } 1567 return count; 1568 } 1569 1570 if (!strcmp(name, "vlan_p")) { 1571 len = num_arg(&user_buffer[i], 1, &value); 1572 if (len < 0) 1573 return len; 1574 1575 i += len; 1576 if ((value <= 7) && (pkt_dev->vlan_id != 0xffff)) { 1577 pkt_dev->vlan_p = value; 1578 sprintf(pg_result, "OK: vlan_p=%u", pkt_dev->vlan_p); 1579 } else { 1580 sprintf(pg_result, "ERROR: vlan_p must be 0-7"); 1581 } 1582 return count; 1583 } 1584 1585 if (!strcmp(name, "vlan_cfi")) { 1586 len = num_arg(&user_buffer[i], 1, &value); 1587 if (len < 0) 1588 return len; 1589 1590 i += len; 1591 if ((value <= 1) && (pkt_dev->vlan_id != 0xffff)) { 1592 pkt_dev->vlan_cfi = value; 1593 sprintf(pg_result, "OK: vlan_cfi=%u", pkt_dev->vlan_cfi); 1594 } else { 1595 sprintf(pg_result, "ERROR: vlan_cfi must be 0-1"); 1596 } 1597 return count; 1598 } 1599 1600 if (!strcmp(name, "svlan_id")) { 1601 len = num_arg(&user_buffer[i], 4, &value); 1602 if (len < 0) 1603 return len; 1604 1605 i += len; 1606 if ((value <= 4095) && ((pkt_dev->vlan_id != 0xffff))) { 1607 pkt_dev->svlan_id = value; /* turn on SVLAN */ 1608 1609 if (debug) 1610 pr_debug("SVLAN turned on\n"); 1611 1612 if (debug && pkt_dev->nr_labels) 1613 pr_debug("MPLS auto turned off\n"); 1614 1615 pkt_dev->nr_labels = 0; /* turn off MPLS */ 1616 sprintf(pg_result, "OK: svlan_id=%u", pkt_dev->svlan_id); 1617 } else { 1618 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1619 pkt_dev->svlan_id = 0xffff; 1620 1621 if (debug) 1622 pr_debug("VLAN/SVLAN turned off\n"); 1623 } 1624 return count; 1625 } 1626 1627 if (!strcmp(name, "svlan_p")) { 1628 len = num_arg(&user_buffer[i], 1, &value); 1629 if (len < 0) 1630 return len; 1631 1632 i += len; 1633 if ((value <= 7) && (pkt_dev->svlan_id != 0xffff)) { 1634 pkt_dev->svlan_p = value; 1635 sprintf(pg_result, "OK: svlan_p=%u", pkt_dev->svlan_p); 1636 } else { 1637 sprintf(pg_result, "ERROR: svlan_p must be 0-7"); 1638 } 1639 return count; 1640 } 1641 1642 if (!strcmp(name, "svlan_cfi")) { 1643 len = num_arg(&user_buffer[i], 1, &value); 1644 if (len < 0) 1645 return len; 1646 1647 i += len; 1648 if ((value <= 1) && (pkt_dev->svlan_id != 0xffff)) { 1649 pkt_dev->svlan_cfi = value; 1650 sprintf(pg_result, "OK: svlan_cfi=%u", pkt_dev->svlan_cfi); 1651 } else { 1652 sprintf(pg_result, "ERROR: svlan_cfi must be 0-1"); 1653 } 1654 return count; 1655 } 1656 1657 if (!strcmp(name, "tos")) { 1658 __u32 tmp_value = 0; 1659 len = hex32_arg(&user_buffer[i], 2, &tmp_value); 1660 if (len < 0) 1661 return len; 1662 1663 i += len; 1664 if (len == 2) { 1665 pkt_dev->tos = tmp_value; 1666 sprintf(pg_result, "OK: tos=0x%02x", pkt_dev->tos); 1667 } else { 1668 sprintf(pg_result, "ERROR: tos must be 00-ff"); 1669 } 1670 return count; 1671 } 1672 1673 if (!strcmp(name, "traffic_class")) { 1674 __u32 tmp_value = 0; 1675 len = hex32_arg(&user_buffer[i], 2, &tmp_value); 1676 if (len < 0) 1677 return len; 1678 1679 i += len; 1680 if (len == 2) { 1681 pkt_dev->traffic_class = tmp_value; 1682 sprintf(pg_result, "OK: traffic_class=0x%02x", pkt_dev->traffic_class); 1683 } else { 1684 sprintf(pg_result, "ERROR: traffic_class must be 00-ff"); 1685 } 1686 return count; 1687 } 1688 1689 if (!strcmp(name, "skb_priority")) { 1690 len = num_arg(&user_buffer[i], 9, &value); 1691 if (len < 0) 1692 return len; 1693 1694 i += len; 1695 pkt_dev->skb_priority = value; 1696 sprintf(pg_result, "OK: skb_priority=%i", 1697 pkt_dev->skb_priority); 1698 return count; 1699 } 1700 1701 sprintf(pkt_dev->result, "No such parameter \"%s\"", name); 1702 return -EINVAL; 1703} 1704 1705static int pktgen_if_open(struct inode *inode, struct file *file) 1706{ 1707 return single_open(file, pktgen_if_show, PDE_DATA(inode)); 1708} 1709 1710static const struct proc_ops pktgen_if_proc_ops = { 1711 .proc_open = pktgen_if_open, 1712 .proc_read = seq_read, 1713 .proc_lseek = seq_lseek, 1714 .proc_write = pktgen_if_write, 1715 .proc_release = single_release, 1716}; 1717 1718static int pktgen_thread_show(struct seq_file *seq, void *v) 1719{ 1720 struct pktgen_thread *t = seq->private; 1721 const struct pktgen_dev *pkt_dev; 1722 1723 BUG_ON(!t); 1724 1725 seq_puts(seq, "Running: "); 1726 1727 rcu_read_lock(); 1728 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 1729 if (pkt_dev->running) 1730 seq_printf(seq, "%s ", pkt_dev->odevname); 1731 1732 seq_puts(seq, "\nStopped: "); 1733 1734 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 1735 if (!pkt_dev->running) 1736 seq_printf(seq, "%s ", pkt_dev->odevname); 1737 1738 if (t->result[0]) 1739 seq_printf(seq, "\nResult: %s\n", t->result); 1740 else 1741 seq_puts(seq, "\nResult: NA\n"); 1742 1743 rcu_read_unlock(); 1744 1745 return 0; 1746} 1747 1748static ssize_t pktgen_thread_write(struct file *file, 1749 const char __user * user_buffer, 1750 size_t count, loff_t * offset) 1751{ 1752 struct seq_file *seq = file->private_data; 1753 struct pktgen_thread *t = seq->private; 1754 int i, max, len, ret; 1755 char name[40]; 1756 char *pg_result; 1757 1758 if (count < 1) { 1759 // sprintf(pg_result, "Wrong command format"); 1760 return -EINVAL; 1761 } 1762 1763 max = count; 1764 len = count_trail_chars(user_buffer, max); 1765 if (len < 0) 1766 return len; 1767 1768 i = len; 1769 1770 /* Read variable name */ 1771 1772 len = strn_len(&user_buffer[i], sizeof(name) - 1); 1773 if (len < 0) 1774 return len; 1775 1776 memset(name, 0, sizeof(name)); 1777 if (copy_from_user(name, &user_buffer[i], len)) 1778 return -EFAULT; 1779 i += len; 1780 1781 max = count - i; 1782 len = count_trail_chars(&user_buffer[i], max); 1783 if (len < 0) 1784 return len; 1785 1786 i += len; 1787 1788 if (debug) 1789 pr_debug("t=%s, count=%lu\n", name, (unsigned long)count); 1790 1791 if (!t) { 1792 pr_err("ERROR: No thread\n"); 1793 ret = -EINVAL; 1794 goto out; 1795 } 1796 1797 pg_result = &(t->result[0]); 1798 1799 if (!strcmp(name, "add_device")) { 1800 char f[32]; 1801 memset(f, 0, 32); 1802 len = strn_len(&user_buffer[i], sizeof(f) - 1); 1803 if (len < 0) { 1804 ret = len; 1805 goto out; 1806 } 1807 if (copy_from_user(f, &user_buffer[i], len)) 1808 return -EFAULT; 1809 i += len; 1810 mutex_lock(&pktgen_thread_lock); 1811 ret = pktgen_add_device(t, f); 1812 mutex_unlock(&pktgen_thread_lock); 1813 if (!ret) { 1814 ret = count; 1815 sprintf(pg_result, "OK: add_device=%s", f); 1816 } else 1817 sprintf(pg_result, "ERROR: can not add device %s", f); 1818 goto out; 1819 } 1820 1821 if (!strcmp(name, "rem_device_all")) { 1822 mutex_lock(&pktgen_thread_lock); 1823 t->control |= T_REMDEVALL; 1824 mutex_unlock(&pktgen_thread_lock); 1825 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */ 1826 ret = count; 1827 sprintf(pg_result, "OK: rem_device_all"); 1828 goto out; 1829 } 1830 1831 if (!strcmp(name, "max_before_softirq")) { 1832 sprintf(pg_result, "OK: Note! max_before_softirq is obsoleted -- Do not use"); 1833 ret = count; 1834 goto out; 1835 } 1836 1837 ret = -EINVAL; 1838out: 1839 return ret; 1840} 1841 1842static int pktgen_thread_open(struct inode *inode, struct file *file) 1843{ 1844 return single_open(file, pktgen_thread_show, PDE_DATA(inode)); 1845} 1846 1847static const struct proc_ops pktgen_thread_proc_ops = { 1848 .proc_open = pktgen_thread_open, 1849 .proc_read = seq_read, 1850 .proc_lseek = seq_lseek, 1851 .proc_write = pktgen_thread_write, 1852 .proc_release = single_release, 1853}; 1854 1855/* Think find or remove for NN */ 1856static struct pktgen_dev *__pktgen_NN_threads(const struct pktgen_net *pn, 1857 const char *ifname, int remove) 1858{ 1859 struct pktgen_thread *t; 1860 struct pktgen_dev *pkt_dev = NULL; 1861 bool exact = (remove == FIND); 1862 1863 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 1864 pkt_dev = pktgen_find_dev(t, ifname, exact); 1865 if (pkt_dev) { 1866 if (remove) { 1867 pkt_dev->removal_mark = 1; 1868 t->control |= T_REMDEV; 1869 } 1870 break; 1871 } 1872 } 1873 return pkt_dev; 1874} 1875 1876/* 1877 * mark a device for removal 1878 */ 1879static void pktgen_mark_device(const struct pktgen_net *pn, const char *ifname) 1880{ 1881 struct pktgen_dev *pkt_dev = NULL; 1882 const int max_tries = 10, msec_per_try = 125; 1883 int i = 0; 1884 1885 mutex_lock(&pktgen_thread_lock); 1886 pr_debug("%s: marking %s for removal\n", __func__, ifname); 1887 1888 while (1) { 1889 1890 pkt_dev = __pktgen_NN_threads(pn, ifname, REMOVE); 1891 if (pkt_dev == NULL) 1892 break; /* success */ 1893 1894 mutex_unlock(&pktgen_thread_lock); 1895 pr_debug("%s: waiting for %s to disappear....\n", 1896 __func__, ifname); 1897 schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try)); 1898 mutex_lock(&pktgen_thread_lock); 1899 1900 if (++i >= max_tries) { 1901 pr_err("%s: timed out after waiting %d msec for device %s to be removed\n", 1902 __func__, msec_per_try * i, ifname); 1903 break; 1904 } 1905 1906 } 1907 1908 mutex_unlock(&pktgen_thread_lock); 1909} 1910 1911static void pktgen_change_name(const struct pktgen_net *pn, struct net_device *dev) 1912{ 1913 struct pktgen_thread *t; 1914 1915 mutex_lock(&pktgen_thread_lock); 1916 1917 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 1918 struct pktgen_dev *pkt_dev; 1919 1920 if_lock(t); 1921 list_for_each_entry(pkt_dev, &t->if_list, list) { 1922 if (pkt_dev->odev != dev) 1923 continue; 1924 1925 proc_remove(pkt_dev->entry); 1926 1927 pkt_dev->entry = proc_create_data(dev->name, 0600, 1928 pn->proc_dir, 1929 &pktgen_if_proc_ops, 1930 pkt_dev); 1931 if (!pkt_dev->entry) 1932 pr_err("can't move proc entry for '%s'\n", 1933 dev->name); 1934 break; 1935 } 1936 if_unlock(t); 1937 } 1938 mutex_unlock(&pktgen_thread_lock); 1939} 1940 1941static int pktgen_device_event(struct notifier_block *unused, 1942 unsigned long event, void *ptr) 1943{ 1944 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1945 struct pktgen_net *pn = net_generic(dev_net(dev), pg_net_id); 1946 1947 if (pn->pktgen_exiting) 1948 return NOTIFY_DONE; 1949 1950 /* It is OK that we do not hold the group lock right now, 1951 * as we run under the RTNL lock. 1952 */ 1953 1954 switch (event) { 1955 case NETDEV_CHANGENAME: 1956 pktgen_change_name(pn, dev); 1957 break; 1958 1959 case NETDEV_UNREGISTER: 1960 pktgen_mark_device(pn, dev->name); 1961 break; 1962 } 1963 1964 return NOTIFY_DONE; 1965} 1966 1967static struct net_device *pktgen_dev_get_by_name(const struct pktgen_net *pn, 1968 struct pktgen_dev *pkt_dev, 1969 const char *ifname) 1970{ 1971 char b[IFNAMSIZ+5]; 1972 int i; 1973 1974 for (i = 0; ifname[i] != '@'; i++) { 1975 if (i == IFNAMSIZ) 1976 break; 1977 1978 b[i] = ifname[i]; 1979 } 1980 b[i] = 0; 1981 1982 return dev_get_by_name(pn->net, b); 1983} 1984 1985 1986/* Associate pktgen_dev with a device. */ 1987 1988static int pktgen_setup_dev(const struct pktgen_net *pn, 1989 struct pktgen_dev *pkt_dev, const char *ifname) 1990{ 1991 struct net_device *odev; 1992 int err; 1993 1994 /* Clean old setups */ 1995 if (pkt_dev->odev) { 1996 dev_put(pkt_dev->odev); 1997 pkt_dev->odev = NULL; 1998 } 1999 2000 odev = pktgen_dev_get_by_name(pn, pkt_dev, ifname); 2001 if (!odev) { 2002 pr_err("no such netdevice: \"%s\"\n", ifname); 2003 return -ENODEV; 2004 } 2005 2006 if (odev->type != ARPHRD_ETHER && odev->type != ARPHRD_LOOPBACK) { 2007 pr_err("not an ethernet or loopback device: \"%s\"\n", ifname); 2008 err = -EINVAL; 2009 } else if (!netif_running(odev)) { 2010 pr_err("device is down: \"%s\"\n", ifname); 2011 err = -ENETDOWN; 2012 } else { 2013 pkt_dev->odev = odev; 2014 return 0; 2015 } 2016 2017 dev_put(odev); 2018 return err; 2019} 2020 2021/* Read pkt_dev from the interface and set up internal pktgen_dev 2022 * structure to have the right information to create/send packets 2023 */ 2024static void pktgen_setup_inject(struct pktgen_dev *pkt_dev) 2025{ 2026 int ntxq; 2027 2028 if (!pkt_dev->odev) { 2029 pr_err("ERROR: pkt_dev->odev == NULL in setup_inject\n"); 2030 sprintf(pkt_dev->result, 2031 "ERROR: pkt_dev->odev == NULL in setup_inject.\n"); 2032 return; 2033 } 2034 2035 /* make sure that we don't pick a non-existing transmit queue */ 2036 ntxq = pkt_dev->odev->real_num_tx_queues; 2037 2038 if (ntxq <= pkt_dev->queue_map_min) { 2039 pr_warn("WARNING: Requested queue_map_min (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n", 2040 pkt_dev->queue_map_min, (ntxq ?: 1) - 1, ntxq, 2041 pkt_dev->odevname); 2042 pkt_dev->queue_map_min = (ntxq ?: 1) - 1; 2043 } 2044 if (pkt_dev->queue_map_max >= ntxq) { 2045 pr_warn("WARNING: Requested queue_map_max (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n", 2046 pkt_dev->queue_map_max, (ntxq ?: 1) - 1, ntxq, 2047 pkt_dev->odevname); 2048 pkt_dev->queue_map_max = (ntxq ?: 1) - 1; 2049 } 2050 2051 /* Default to the interface's mac if not explicitly set. */ 2052 2053 if (is_zero_ether_addr(pkt_dev->src_mac)) 2054 ether_addr_copy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr); 2055 2056 /* Set up Dest MAC */ 2057 ether_addr_copy(&(pkt_dev->hh[0]), pkt_dev->dst_mac); 2058 2059 if (pkt_dev->flags & F_IPV6) { 2060 int i, set = 0, err = 1; 2061 struct inet6_dev *idev; 2062 2063 if (pkt_dev->min_pkt_size == 0) { 2064 pkt_dev->min_pkt_size = 14 + sizeof(struct ipv6hdr) 2065 + sizeof(struct udphdr) 2066 + sizeof(struct pktgen_hdr) 2067 + pkt_dev->pkt_overhead; 2068 } 2069 2070 for (i = 0; i < sizeof(struct in6_addr); i++) 2071 if (pkt_dev->cur_in6_saddr.s6_addr[i]) { 2072 set = 1; 2073 break; 2074 } 2075 2076 if (!set) { 2077 2078 /* 2079 * Use linklevel address if unconfigured. 2080 * 2081 * use ipv6_get_lladdr if/when it's get exported 2082 */ 2083 2084 rcu_read_lock(); 2085 idev = __in6_dev_get(pkt_dev->odev); 2086 if (idev) { 2087 struct inet6_ifaddr *ifp; 2088 2089 read_lock_bh(&idev->lock); 2090 list_for_each_entry(ifp, &idev->addr_list, if_list) { 2091 if ((ifp->scope & IFA_LINK) && 2092 !(ifp->flags & IFA_F_TENTATIVE)) { 2093 pkt_dev->cur_in6_saddr = ifp->addr; 2094 err = 0; 2095 break; 2096 } 2097 } 2098 read_unlock_bh(&idev->lock); 2099 } 2100 rcu_read_unlock(); 2101 if (err) 2102 pr_err("ERROR: IPv6 link address not available\n"); 2103 } 2104 } else { 2105 if (pkt_dev->min_pkt_size == 0) { 2106 pkt_dev->min_pkt_size = 14 + sizeof(struct iphdr) 2107 + sizeof(struct udphdr) 2108 + sizeof(struct pktgen_hdr) 2109 + pkt_dev->pkt_overhead; 2110 } 2111 2112 pkt_dev->saddr_min = 0; 2113 pkt_dev->saddr_max = 0; 2114 if (strlen(pkt_dev->src_min) == 0) { 2115 2116 struct in_device *in_dev; 2117 2118 rcu_read_lock(); 2119 in_dev = __in_dev_get_rcu(pkt_dev->odev); 2120 if (in_dev) { 2121 const struct in_ifaddr *ifa; 2122 2123 ifa = rcu_dereference(in_dev->ifa_list); 2124 if (ifa) { 2125 pkt_dev->saddr_min = ifa->ifa_address; 2126 pkt_dev->saddr_max = pkt_dev->saddr_min; 2127 } 2128 } 2129 rcu_read_unlock(); 2130 } else { 2131 pkt_dev->saddr_min = in_aton(pkt_dev->src_min); 2132 pkt_dev->saddr_max = in_aton(pkt_dev->src_max); 2133 } 2134 2135 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min); 2136 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max); 2137 } 2138 /* Initialize current values. */ 2139 pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size; 2140 if (pkt_dev->min_pkt_size > pkt_dev->max_pkt_size) 2141 pkt_dev->max_pkt_size = pkt_dev->min_pkt_size; 2142 2143 pkt_dev->cur_dst_mac_offset = 0; 2144 pkt_dev->cur_src_mac_offset = 0; 2145 pkt_dev->cur_saddr = pkt_dev->saddr_min; 2146 pkt_dev->cur_daddr = pkt_dev->daddr_min; 2147 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min; 2148 pkt_dev->cur_udp_src = pkt_dev->udp_src_min; 2149 pkt_dev->nflows = 0; 2150} 2151 2152 2153static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until) 2154{ 2155 ktime_t start_time, end_time; 2156 s64 remaining; 2157 struct hrtimer_sleeper t; 2158 2159 hrtimer_init_sleeper_on_stack(&t, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); 2160 hrtimer_set_expires(&t.timer, spin_until); 2161 2162 remaining = ktime_to_ns(hrtimer_expires_remaining(&t.timer)); 2163 if (remaining <= 0) 2164 goto out; 2165 2166 start_time = ktime_get(); 2167 if (remaining < 100000) { 2168 /* for small delays (<100us), just loop until limit is reached */ 2169 do { 2170 end_time = ktime_get(); 2171 } while (ktime_compare(end_time, spin_until) < 0); 2172 } else { 2173 do { 2174 set_current_state(TASK_INTERRUPTIBLE); 2175 hrtimer_sleeper_start_expires(&t, HRTIMER_MODE_ABS); 2176 2177 if (likely(t.task)) 2178 schedule(); 2179 2180 hrtimer_cancel(&t.timer); 2181 } while (t.task && pkt_dev->running && !signal_pending(current)); 2182 __set_current_state(TASK_RUNNING); 2183 end_time = ktime_get(); 2184 } 2185 2186 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(end_time, start_time)); 2187out: 2188 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay); 2189 destroy_hrtimer_on_stack(&t.timer); 2190} 2191 2192static inline void set_pkt_overhead(struct pktgen_dev *pkt_dev) 2193{ 2194 pkt_dev->pkt_overhead = 0; 2195 pkt_dev->pkt_overhead += pkt_dev->nr_labels*sizeof(u32); 2196 pkt_dev->pkt_overhead += VLAN_TAG_SIZE(pkt_dev); 2197 pkt_dev->pkt_overhead += SVLAN_TAG_SIZE(pkt_dev); 2198} 2199 2200static inline int f_seen(const struct pktgen_dev *pkt_dev, int flow) 2201{ 2202 return !!(pkt_dev->flows[flow].flags & F_INIT); 2203} 2204 2205static inline int f_pick(struct pktgen_dev *pkt_dev) 2206{ 2207 int flow = pkt_dev->curfl; 2208 2209 if (pkt_dev->flags & F_FLOW_SEQ) { 2210 if (pkt_dev->flows[flow].count >= pkt_dev->lflow) { 2211 /* reset time */ 2212 pkt_dev->flows[flow].count = 0; 2213 pkt_dev->flows[flow].flags = 0; 2214 pkt_dev->curfl += 1; 2215 if (pkt_dev->curfl >= pkt_dev->cflows) 2216 pkt_dev->curfl = 0; /*reset */ 2217 } 2218 } else { 2219 flow = prandom_u32() % pkt_dev->cflows; 2220 pkt_dev->curfl = flow; 2221 2222 if (pkt_dev->flows[flow].count > pkt_dev->lflow) { 2223 pkt_dev->flows[flow].count = 0; 2224 pkt_dev->flows[flow].flags = 0; 2225 } 2226 } 2227 2228 return pkt_dev->curfl; 2229} 2230 2231 2232#ifdef CONFIG_XFRM 2233/* If there was already an IPSEC SA, we keep it as is, else 2234 * we go look for it ... 2235*/ 2236#define DUMMY_MARK 0 2237static void get_ipsec_sa(struct pktgen_dev *pkt_dev, int flow) 2238{ 2239 struct xfrm_state *x = pkt_dev->flows[flow].x; 2240 struct pktgen_net *pn = net_generic(dev_net(pkt_dev->odev), pg_net_id); 2241 if (!x) { 2242 2243 if (pkt_dev->spi) { 2244 /* We need as quick as possible to find the right SA 2245 * Searching with minimum criteria to archieve this. 2246 */ 2247 x = xfrm_state_lookup_byspi(pn->net, htonl(pkt_dev->spi), AF_INET); 2248 } else { 2249 /* slow path: we dont already have xfrm_state */ 2250 x = xfrm_stateonly_find(pn->net, DUMMY_MARK, 0, 2251 (xfrm_address_t *)&pkt_dev->cur_daddr, 2252 (xfrm_address_t *)&pkt_dev->cur_saddr, 2253 AF_INET, 2254 pkt_dev->ipsmode, 2255 pkt_dev->ipsproto, 0); 2256 } 2257 if (x) { 2258 pkt_dev->flows[flow].x = x; 2259 set_pkt_overhead(pkt_dev); 2260 pkt_dev->pkt_overhead += x->props.header_len; 2261 } 2262 2263 } 2264} 2265#endif 2266static void set_cur_queue_map(struct pktgen_dev *pkt_dev) 2267{ 2268 2269 if (pkt_dev->flags & F_QUEUE_MAP_CPU) 2270 pkt_dev->cur_queue_map = smp_processor_id(); 2271 2272 else if (pkt_dev->queue_map_min <= pkt_dev->queue_map_max) { 2273 __u16 t; 2274 if (pkt_dev->flags & F_QUEUE_MAP_RND) { 2275 t = prandom_u32() % 2276 (pkt_dev->queue_map_max - 2277 pkt_dev->queue_map_min + 1) 2278 + pkt_dev->queue_map_min; 2279 } else { 2280 t = pkt_dev->cur_queue_map + 1; 2281 if (t > pkt_dev->queue_map_max) 2282 t = pkt_dev->queue_map_min; 2283 } 2284 pkt_dev->cur_queue_map = t; 2285 } 2286 pkt_dev->cur_queue_map = pkt_dev->cur_queue_map % pkt_dev->odev->real_num_tx_queues; 2287} 2288 2289/* Increment/randomize headers according to flags and current values 2290 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst 2291 */ 2292static void mod_cur_headers(struct pktgen_dev *pkt_dev) 2293{ 2294 __u32 imn; 2295 __u32 imx; 2296 int flow = 0; 2297 2298 if (pkt_dev->cflows) 2299 flow = f_pick(pkt_dev); 2300 2301 /* Deal with source MAC */ 2302 if (pkt_dev->src_mac_count > 1) { 2303 __u32 mc; 2304 __u32 tmp; 2305 2306 if (pkt_dev->flags & F_MACSRC_RND) 2307 mc = prandom_u32() % pkt_dev->src_mac_count; 2308 else { 2309 mc = pkt_dev->cur_src_mac_offset++; 2310 if (pkt_dev->cur_src_mac_offset >= 2311 pkt_dev->src_mac_count) 2312 pkt_dev->cur_src_mac_offset = 0; 2313 } 2314 2315 tmp = pkt_dev->src_mac[5] + (mc & 0xFF); 2316 pkt_dev->hh[11] = tmp; 2317 tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8)); 2318 pkt_dev->hh[10] = tmp; 2319 tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8)); 2320 pkt_dev->hh[9] = tmp; 2321 tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8)); 2322 pkt_dev->hh[8] = tmp; 2323 tmp = (pkt_dev->src_mac[1] + (tmp >> 8)); 2324 pkt_dev->hh[7] = tmp; 2325 } 2326 2327 /* Deal with Destination MAC */ 2328 if (pkt_dev->dst_mac_count > 1) { 2329 __u32 mc; 2330 __u32 tmp; 2331 2332 if (pkt_dev->flags & F_MACDST_RND) 2333 mc = prandom_u32() % pkt_dev->dst_mac_count; 2334 2335 else { 2336 mc = pkt_dev->cur_dst_mac_offset++; 2337 if (pkt_dev->cur_dst_mac_offset >= 2338 pkt_dev->dst_mac_count) { 2339 pkt_dev->cur_dst_mac_offset = 0; 2340 } 2341 } 2342 2343 tmp = pkt_dev->dst_mac[5] + (mc & 0xFF); 2344 pkt_dev->hh[5] = tmp; 2345 tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8)); 2346 pkt_dev->hh[4] = tmp; 2347 tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8)); 2348 pkt_dev->hh[3] = tmp; 2349 tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8)); 2350 pkt_dev->hh[2] = tmp; 2351 tmp = (pkt_dev->dst_mac[1] + (tmp >> 8)); 2352 pkt_dev->hh[1] = tmp; 2353 } 2354 2355 if (pkt_dev->flags & F_MPLS_RND) { 2356 unsigned int i; 2357 for (i = 0; i < pkt_dev->nr_labels; i++) 2358 if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM) 2359 pkt_dev->labels[i] = MPLS_STACK_BOTTOM | 2360 ((__force __be32)prandom_u32() & 2361 htonl(0x000fffff)); 2362 } 2363 2364 if ((pkt_dev->flags & F_VID_RND) && (pkt_dev->vlan_id != 0xffff)) { 2365 pkt_dev->vlan_id = prandom_u32() & (4096 - 1); 2366 } 2367 2368 if ((pkt_dev->flags & F_SVID_RND) && (pkt_dev->svlan_id != 0xffff)) { 2369 pkt_dev->svlan_id = prandom_u32() & (4096 - 1); 2370 } 2371 2372 if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) { 2373 if (pkt_dev->flags & F_UDPSRC_RND) 2374 pkt_dev->cur_udp_src = prandom_u32() % 2375 (pkt_dev->udp_src_max - pkt_dev->udp_src_min) 2376 + pkt_dev->udp_src_min; 2377 2378 else { 2379 pkt_dev->cur_udp_src++; 2380 if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max) 2381 pkt_dev->cur_udp_src = pkt_dev->udp_src_min; 2382 } 2383 } 2384 2385 if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) { 2386 if (pkt_dev->flags & F_UDPDST_RND) { 2387 pkt_dev->cur_udp_dst = prandom_u32() % 2388 (pkt_dev->udp_dst_max - pkt_dev->udp_dst_min) 2389 + pkt_dev->udp_dst_min; 2390 } else { 2391 pkt_dev->cur_udp_dst++; 2392 if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max) 2393 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min; 2394 } 2395 } 2396 2397 if (!(pkt_dev->flags & F_IPV6)) { 2398 2399 imn = ntohl(pkt_dev->saddr_min); 2400 imx = ntohl(pkt_dev->saddr_max); 2401 if (imn < imx) { 2402 __u32 t; 2403 if (pkt_dev->flags & F_IPSRC_RND) 2404 t = prandom_u32() % (imx - imn) + imn; 2405 else { 2406 t = ntohl(pkt_dev->cur_saddr); 2407 t++; 2408 if (t > imx) 2409 t = imn; 2410 2411 } 2412 pkt_dev->cur_saddr = htonl(t); 2413 } 2414 2415 if (pkt_dev->cflows && f_seen(pkt_dev, flow)) { 2416 pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr; 2417 } else { 2418 imn = ntohl(pkt_dev->daddr_min); 2419 imx = ntohl(pkt_dev->daddr_max); 2420 if (imn < imx) { 2421 __u32 t; 2422 __be32 s; 2423 if (pkt_dev->flags & F_IPDST_RND) { 2424 2425 do { 2426 t = prandom_u32() % 2427 (imx - imn) + imn; 2428 s = htonl(t); 2429 } while (ipv4_is_loopback(s) || 2430 ipv4_is_multicast(s) || 2431 ipv4_is_lbcast(s) || 2432 ipv4_is_zeronet(s) || 2433 ipv4_is_local_multicast(s)); 2434 pkt_dev->cur_daddr = s; 2435 } else { 2436 t = ntohl(pkt_dev->cur_daddr); 2437 t++; 2438 if (t > imx) { 2439 t = imn; 2440 } 2441 pkt_dev->cur_daddr = htonl(t); 2442 } 2443 } 2444 if (pkt_dev->cflows) { 2445 pkt_dev->flows[flow].flags |= F_INIT; 2446 pkt_dev->flows[flow].cur_daddr = 2447 pkt_dev->cur_daddr; 2448#ifdef CONFIG_XFRM 2449 if (pkt_dev->flags & F_IPSEC) 2450 get_ipsec_sa(pkt_dev, flow); 2451#endif 2452 pkt_dev->nflows++; 2453 } 2454 } 2455 } else { /* IPV6 * */ 2456 2457 if (!ipv6_addr_any(&pkt_dev->min_in6_daddr)) { 2458 int i; 2459 2460 /* Only random destinations yet */ 2461 2462 for (i = 0; i < 4; i++) { 2463 pkt_dev->cur_in6_daddr.s6_addr32[i] = 2464 (((__force __be32)prandom_u32() | 2465 pkt_dev->min_in6_daddr.s6_addr32[i]) & 2466 pkt_dev->max_in6_daddr.s6_addr32[i]); 2467 } 2468 } 2469 } 2470 2471 if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) { 2472 __u32 t; 2473 if (pkt_dev->flags & F_TXSIZE_RND) { 2474 t = prandom_u32() % 2475 (pkt_dev->max_pkt_size - pkt_dev->min_pkt_size) 2476 + pkt_dev->min_pkt_size; 2477 } else { 2478 t = pkt_dev->cur_pkt_size + 1; 2479 if (t > pkt_dev->max_pkt_size) 2480 t = pkt_dev->min_pkt_size; 2481 } 2482 pkt_dev->cur_pkt_size = t; 2483 } 2484 2485 set_cur_queue_map(pkt_dev); 2486 2487 pkt_dev->flows[flow].count++; 2488} 2489 2490 2491#ifdef CONFIG_XFRM 2492static u32 pktgen_dst_metrics[RTAX_MAX + 1] = { 2493 2494 [RTAX_HOPLIMIT] = 0x5, /* Set a static hoplimit */ 2495}; 2496 2497static int pktgen_output_ipsec(struct sk_buff *skb, struct pktgen_dev *pkt_dev) 2498{ 2499 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x; 2500 int err = 0; 2501 struct net *net = dev_net(pkt_dev->odev); 2502 2503 if (!x) 2504 return 0; 2505 /* XXX: we dont support tunnel mode for now until 2506 * we resolve the dst issue */ 2507 if ((x->props.mode != XFRM_MODE_TRANSPORT) && (pkt_dev->spi == 0)) 2508 return 0; 2509 2510 /* But when user specify an valid SPI, transformation 2511 * supports both transport/tunnel mode + ESP/AH type. 2512 */ 2513 if ((x->props.mode == XFRM_MODE_TUNNEL) && (pkt_dev->spi != 0)) 2514 skb->_skb_refdst = (unsigned long)&pkt_dev->xdst.u.dst | SKB_DST_NOREF; 2515 2516 rcu_read_lock_bh(); 2517 err = pktgen_xfrm_outer_mode_output(x, skb); 2518 rcu_read_unlock_bh(); 2519 if (err) { 2520 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR); 2521 goto error; 2522 } 2523 err = x->type->output(x, skb); 2524 if (err) { 2525 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR); 2526 goto error; 2527 } 2528 spin_lock_bh(&x->lock); 2529 x->curlft.bytes += skb->len; 2530 x->curlft.packets++; 2531 spin_unlock_bh(&x->lock); 2532error: 2533 return err; 2534} 2535 2536static void free_SAs(struct pktgen_dev *pkt_dev) 2537{ 2538 if (pkt_dev->cflows) { 2539 /* let go of the SAs if we have them */ 2540 int i; 2541 for (i = 0; i < pkt_dev->cflows; i++) { 2542 struct xfrm_state *x = pkt_dev->flows[i].x; 2543 if (x) { 2544 xfrm_state_put(x); 2545 pkt_dev->flows[i].x = NULL; 2546 } 2547 } 2548 } 2549} 2550 2551static int process_ipsec(struct pktgen_dev *pkt_dev, 2552 struct sk_buff *skb, __be16 protocol) 2553{ 2554 if (pkt_dev->flags & F_IPSEC) { 2555 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x; 2556 int nhead = 0; 2557 if (x) { 2558 struct ethhdr *eth; 2559 struct iphdr *iph; 2560 int ret; 2561 2562 nhead = x->props.header_len - skb_headroom(skb); 2563 if (nhead > 0) { 2564 ret = pskb_expand_head(skb, nhead, 0, GFP_ATOMIC); 2565 if (ret < 0) { 2566 pr_err("Error expanding ipsec packet %d\n", 2567 ret); 2568 goto err; 2569 } 2570 } 2571 2572 /* ipsec is not expecting ll header */ 2573 skb_pull(skb, ETH_HLEN); 2574 ret = pktgen_output_ipsec(skb, pkt_dev); 2575 if (ret) { 2576 pr_err("Error creating ipsec packet %d\n", ret); 2577 goto err; 2578 } 2579 /* restore ll */ 2580 eth = skb_push(skb, ETH_HLEN); 2581 memcpy(eth, pkt_dev->hh, 2 * ETH_ALEN); 2582 eth->h_proto = protocol; 2583 2584 /* Update IPv4 header len as well as checksum value */ 2585 iph = ip_hdr(skb); 2586 iph->tot_len = htons(skb->len - ETH_HLEN); 2587 ip_send_check(iph); 2588 } 2589 } 2590 return 1; 2591err: 2592 kfree_skb(skb); 2593 return 0; 2594} 2595#endif 2596 2597static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev) 2598{ 2599 unsigned int i; 2600 for (i = 0; i < pkt_dev->nr_labels; i++) 2601 *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM; 2602 2603 mpls--; 2604 *mpls |= MPLS_STACK_BOTTOM; 2605} 2606 2607static inline __be16 build_tci(unsigned int id, unsigned int cfi, 2608 unsigned int prio) 2609{ 2610 return htons(id | (cfi << 12) | (prio << 13)); 2611} 2612 2613static void pktgen_finalize_skb(struct pktgen_dev *pkt_dev, struct sk_buff *skb, 2614 int datalen) 2615{ 2616 struct timespec64 timestamp; 2617 struct pktgen_hdr *pgh; 2618 2619 pgh = skb_put(skb, sizeof(*pgh)); 2620 datalen -= sizeof(*pgh); 2621 2622 if (pkt_dev->nfrags <= 0) { 2623 skb_put_zero(skb, datalen); 2624 } else { 2625 int frags = pkt_dev->nfrags; 2626 int i, len; 2627 int frag_len; 2628 2629 2630 if (frags > MAX_SKB_FRAGS) 2631 frags = MAX_SKB_FRAGS; 2632 len = datalen - frags * PAGE_SIZE; 2633 if (len > 0) { 2634 skb_put_zero(skb, len); 2635 datalen = frags * PAGE_SIZE; 2636 } 2637 2638 i = 0; 2639 frag_len = (datalen/frags) < PAGE_SIZE ? 2640 (datalen/frags) : PAGE_SIZE; 2641 while (datalen > 0) { 2642 if (unlikely(!pkt_dev->page)) { 2643 int node = numa_node_id(); 2644 2645 if (pkt_dev->node >= 0 && (pkt_dev->flags & F_NODE)) 2646 node = pkt_dev->node; 2647 pkt_dev->page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0); 2648 if (!pkt_dev->page) 2649 break; 2650 } 2651 get_page(pkt_dev->page); 2652 skb_frag_set_page(skb, i, pkt_dev->page); 2653 skb_frag_off_set(&skb_shinfo(skb)->frags[i], 0); 2654 /*last fragment, fill rest of data*/ 2655 if (i == (frags - 1)) 2656 skb_frag_size_set(&skb_shinfo(skb)->frags[i], 2657 (datalen < PAGE_SIZE ? datalen : PAGE_SIZE)); 2658 else 2659 skb_frag_size_set(&skb_shinfo(skb)->frags[i], frag_len); 2660 datalen -= skb_frag_size(&skb_shinfo(skb)->frags[i]); 2661 skb->len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2662 skb->data_len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2663 i++; 2664 skb_shinfo(skb)->nr_frags = i; 2665 } 2666 } 2667 2668 /* Stamp the time, and sequence number, 2669 * convert them to network byte order 2670 */ 2671 pgh->pgh_magic = htonl(PKTGEN_MAGIC); 2672 pgh->seq_num = htonl(pkt_dev->seq_num); 2673 2674 if (pkt_dev->flags & F_NO_TIMESTAMP) { 2675 pgh->tv_sec = 0; 2676 pgh->tv_usec = 0; 2677 } else { 2678 /* 2679 * pgh->tv_sec wraps in y2106 when interpreted as unsigned 2680 * as done by wireshark, or y2038 when interpreted as signed. 2681 * This is probably harmless, but if anyone wants to improve 2682 * it, we could introduce a variant that puts 64-bit nanoseconds 2683 * into the respective header bytes. 2684 * This would also be slightly faster to read. 2685 */ 2686 ktime_get_real_ts64(×tamp); 2687 pgh->tv_sec = htonl(timestamp.tv_sec); 2688 pgh->tv_usec = htonl(timestamp.tv_nsec / NSEC_PER_USEC); 2689 } 2690} 2691 2692static struct sk_buff *pktgen_alloc_skb(struct net_device *dev, 2693 struct pktgen_dev *pkt_dev) 2694{ 2695 unsigned int extralen = LL_RESERVED_SPACE(dev); 2696 struct sk_buff *skb = NULL; 2697 unsigned int size; 2698 2699 size = pkt_dev->cur_pkt_size + 64 + extralen + pkt_dev->pkt_overhead; 2700 if (pkt_dev->flags & F_NODE) { 2701 int node = pkt_dev->node >= 0 ? pkt_dev->node : numa_node_id(); 2702 2703 skb = __alloc_skb(NET_SKB_PAD + size, GFP_NOWAIT, 0, node); 2704 if (likely(skb)) { 2705 skb_reserve(skb, NET_SKB_PAD); 2706 skb->dev = dev; 2707 } 2708 } else { 2709 skb = __netdev_alloc_skb(dev, size, GFP_NOWAIT); 2710 } 2711 2712 /* the caller pre-fetches from skb->data and reserves for the mac hdr */ 2713 if (likely(skb)) 2714 skb_reserve(skb, extralen - 16); 2715 2716 return skb; 2717} 2718 2719static struct sk_buff *fill_packet_ipv4(struct net_device *odev, 2720 struct pktgen_dev *pkt_dev) 2721{ 2722 struct sk_buff *skb = NULL; 2723 __u8 *eth; 2724 struct udphdr *udph; 2725 int datalen, iplen; 2726 struct iphdr *iph; 2727 __be16 protocol = htons(ETH_P_IP); 2728 __be32 *mpls; 2729 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */ 2730 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */ 2731 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */ 2732 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */ 2733 u16 queue_map; 2734 2735 if (pkt_dev->nr_labels) 2736 protocol = htons(ETH_P_MPLS_UC); 2737 2738 if (pkt_dev->vlan_id != 0xffff) 2739 protocol = htons(ETH_P_8021Q); 2740 2741 /* Update any of the values, used when we're incrementing various 2742 * fields. 2743 */ 2744 mod_cur_headers(pkt_dev); 2745 queue_map = pkt_dev->cur_queue_map; 2746 2747 skb = pktgen_alloc_skb(odev, pkt_dev); 2748 if (!skb) { 2749 sprintf(pkt_dev->result, "No memory"); 2750 return NULL; 2751 } 2752 2753 prefetchw(skb->data); 2754 skb_reserve(skb, 16); 2755 2756 /* Reserve for ethernet and IP header */ 2757 eth = skb_push(skb, 14); 2758 mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32)); 2759 if (pkt_dev->nr_labels) 2760 mpls_push(mpls, pkt_dev); 2761 2762 if (pkt_dev->vlan_id != 0xffff) { 2763 if (pkt_dev->svlan_id != 0xffff) { 2764 svlan_tci = skb_put(skb, sizeof(__be16)); 2765 *svlan_tci = build_tci(pkt_dev->svlan_id, 2766 pkt_dev->svlan_cfi, 2767 pkt_dev->svlan_p); 2768 svlan_encapsulated_proto = skb_put(skb, 2769 sizeof(__be16)); 2770 *svlan_encapsulated_proto = htons(ETH_P_8021Q); 2771 } 2772 vlan_tci = skb_put(skb, sizeof(__be16)); 2773 *vlan_tci = build_tci(pkt_dev->vlan_id, 2774 pkt_dev->vlan_cfi, 2775 pkt_dev->vlan_p); 2776 vlan_encapsulated_proto = skb_put(skb, sizeof(__be16)); 2777 *vlan_encapsulated_proto = htons(ETH_P_IP); 2778 } 2779 2780 skb_reset_mac_header(skb); 2781 skb_set_network_header(skb, skb->len); 2782 iph = skb_put(skb, sizeof(struct iphdr)); 2783 2784 skb_set_transport_header(skb, skb->len); 2785 udph = skb_put(skb, sizeof(struct udphdr)); 2786 skb_set_queue_mapping(skb, queue_map); 2787 skb->priority = pkt_dev->skb_priority; 2788 2789 memcpy(eth, pkt_dev->hh, 12); 2790 *(__be16 *) & eth[12] = protocol; 2791 2792 /* Eth + IPh + UDPh + mpls */ 2793 datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 - 2794 pkt_dev->pkt_overhead; 2795 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) 2796 datalen = sizeof(struct pktgen_hdr); 2797 2798 udph->source = htons(pkt_dev->cur_udp_src); 2799 udph->dest = htons(pkt_dev->cur_udp_dst); 2800 udph->len = htons(datalen + 8); /* DATA + udphdr */ 2801 udph->check = 0; 2802 2803 iph->ihl = 5; 2804 iph->version = 4; 2805 iph->ttl = 32; 2806 iph->tos = pkt_dev->tos; 2807 iph->protocol = IPPROTO_UDP; /* UDP */ 2808 iph->saddr = pkt_dev->cur_saddr; 2809 iph->daddr = pkt_dev->cur_daddr; 2810 iph->id = htons(pkt_dev->ip_id); 2811 pkt_dev->ip_id++; 2812 iph->frag_off = 0; 2813 iplen = 20 + 8 + datalen; 2814 iph->tot_len = htons(iplen); 2815 ip_send_check(iph); 2816 skb->protocol = protocol; 2817 skb->dev = odev; 2818 skb->pkt_type = PACKET_HOST; 2819 2820 pktgen_finalize_skb(pkt_dev, skb, datalen); 2821 2822 if (!(pkt_dev->flags & F_UDPCSUM)) { 2823 skb->ip_summed = CHECKSUM_NONE; 2824 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM)) { 2825 skb->ip_summed = CHECKSUM_PARTIAL; 2826 skb->csum = 0; 2827 udp4_hwcsum(skb, iph->saddr, iph->daddr); 2828 } else { 2829 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), datalen + 8, 0); 2830 2831 /* add protocol-dependent pseudo-header */ 2832 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr, 2833 datalen + 8, IPPROTO_UDP, csum); 2834 2835 if (udph->check == 0) 2836 udph->check = CSUM_MANGLED_0; 2837 } 2838 2839#ifdef CONFIG_XFRM 2840 if (!process_ipsec(pkt_dev, skb, protocol)) 2841 return NULL; 2842#endif 2843 2844 return skb; 2845} 2846 2847static struct sk_buff *fill_packet_ipv6(struct net_device *odev, 2848 struct pktgen_dev *pkt_dev) 2849{ 2850 struct sk_buff *skb = NULL; 2851 __u8 *eth; 2852 struct udphdr *udph; 2853 int datalen, udplen; 2854 struct ipv6hdr *iph; 2855 __be16 protocol = htons(ETH_P_IPV6); 2856 __be32 *mpls; 2857 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */ 2858 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */ 2859 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */ 2860 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */ 2861 u16 queue_map; 2862 2863 if (pkt_dev->nr_labels) 2864 protocol = htons(ETH_P_MPLS_UC); 2865 2866 if (pkt_dev->vlan_id != 0xffff) 2867 protocol = htons(ETH_P_8021Q); 2868 2869 /* Update any of the values, used when we're incrementing various 2870 * fields. 2871 */ 2872 mod_cur_headers(pkt_dev); 2873 queue_map = pkt_dev->cur_queue_map; 2874 2875 skb = pktgen_alloc_skb(odev, pkt_dev); 2876 if (!skb) { 2877 sprintf(pkt_dev->result, "No memory"); 2878 return NULL; 2879 } 2880 2881 prefetchw(skb->data); 2882 skb_reserve(skb, 16); 2883 2884 /* Reserve for ethernet and IP header */ 2885 eth = skb_push(skb, 14); 2886 mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32)); 2887 if (pkt_dev->nr_labels) 2888 mpls_push(mpls, pkt_dev); 2889 2890 if (pkt_dev->vlan_id != 0xffff) { 2891 if (pkt_dev->svlan_id != 0xffff) { 2892 svlan_tci = skb_put(skb, sizeof(__be16)); 2893 *svlan_tci = build_tci(pkt_dev->svlan_id, 2894 pkt_dev->svlan_cfi, 2895 pkt_dev->svlan_p); 2896 svlan_encapsulated_proto = skb_put(skb, 2897 sizeof(__be16)); 2898 *svlan_encapsulated_proto = htons(ETH_P_8021Q); 2899 } 2900 vlan_tci = skb_put(skb, sizeof(__be16)); 2901 *vlan_tci = build_tci(pkt_dev->vlan_id, 2902 pkt_dev->vlan_cfi, 2903 pkt_dev->vlan_p); 2904 vlan_encapsulated_proto = skb_put(skb, sizeof(__be16)); 2905 *vlan_encapsulated_proto = htons(ETH_P_IPV6); 2906 } 2907 2908 skb_reset_mac_header(skb); 2909 skb_set_network_header(skb, skb->len); 2910 iph = skb_put(skb, sizeof(struct ipv6hdr)); 2911 2912 skb_set_transport_header(skb, skb->len); 2913 udph = skb_put(skb, sizeof(struct udphdr)); 2914 skb_set_queue_mapping(skb, queue_map); 2915 skb->priority = pkt_dev->skb_priority; 2916 2917 memcpy(eth, pkt_dev->hh, 12); 2918 *(__be16 *) ð[12] = protocol; 2919 2920 /* Eth + IPh + UDPh + mpls */ 2921 datalen = pkt_dev->cur_pkt_size - 14 - 2922 sizeof(struct ipv6hdr) - sizeof(struct udphdr) - 2923 pkt_dev->pkt_overhead; 2924 2925 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) { 2926 datalen = sizeof(struct pktgen_hdr); 2927 net_info_ratelimited("increased datalen to %d\n", datalen); 2928 } 2929 2930 udplen = datalen + sizeof(struct udphdr); 2931 udph->source = htons(pkt_dev->cur_udp_src); 2932 udph->dest = htons(pkt_dev->cur_udp_dst); 2933 udph->len = htons(udplen); 2934 udph->check = 0; 2935 2936 *(__be32 *) iph = htonl(0x60000000); /* Version + flow */ 2937 2938 if (pkt_dev->traffic_class) { 2939 /* Version + traffic class + flow (0) */ 2940 *(__be32 *)iph |= htonl(0x60000000 | (pkt_dev->traffic_class << 20)); 2941 } 2942 2943 iph->hop_limit = 32; 2944 2945 iph->payload_len = htons(udplen); 2946 iph->nexthdr = IPPROTO_UDP; 2947 2948 iph->daddr = pkt_dev->cur_in6_daddr; 2949 iph->saddr = pkt_dev->cur_in6_saddr; 2950 2951 skb->protocol = protocol; 2952 skb->dev = odev; 2953 skb->pkt_type = PACKET_HOST; 2954 2955 pktgen_finalize_skb(pkt_dev, skb, datalen); 2956 2957 if (!(pkt_dev->flags & F_UDPCSUM)) { 2958 skb->ip_summed = CHECKSUM_NONE; 2959 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM)) { 2960 skb->ip_summed = CHECKSUM_PARTIAL; 2961 skb->csum_start = skb_transport_header(skb) - skb->head; 2962 skb->csum_offset = offsetof(struct udphdr, check); 2963 udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0); 2964 } else { 2965 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), udplen, 0); 2966 2967 /* add protocol-dependent pseudo-header */ 2968 udph->check = csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, csum); 2969 2970 if (udph->check == 0) 2971 udph->check = CSUM_MANGLED_0; 2972 } 2973 2974 return skb; 2975} 2976 2977static struct sk_buff *fill_packet(struct net_device *odev, 2978 struct pktgen_dev *pkt_dev) 2979{ 2980 if (pkt_dev->flags & F_IPV6) 2981 return fill_packet_ipv6(odev, pkt_dev); 2982 else 2983 return fill_packet_ipv4(odev, pkt_dev); 2984} 2985 2986static void pktgen_clear_counters(struct pktgen_dev *pkt_dev) 2987{ 2988 pkt_dev->seq_num = 1; 2989 pkt_dev->idle_acc = 0; 2990 pkt_dev->sofar = 0; 2991 pkt_dev->tx_bytes = 0; 2992 pkt_dev->errors = 0; 2993} 2994 2995/* Set up structure for sending pkts, clear counters */ 2996 2997static void pktgen_run(struct pktgen_thread *t) 2998{ 2999 struct pktgen_dev *pkt_dev; 3000 int started = 0; 3001 3002 func_enter(); 3003 3004 rcu_read_lock(); 3005 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3006 3007 /* 3008 * setup odev and create initial packet. 3009 */ 3010 pktgen_setup_inject(pkt_dev); 3011 3012 if (pkt_dev->odev) { 3013 pktgen_clear_counters(pkt_dev); 3014 pkt_dev->skb = NULL; 3015 pkt_dev->started_at = pkt_dev->next_tx = ktime_get(); 3016 3017 set_pkt_overhead(pkt_dev); 3018 3019 strcpy(pkt_dev->result, "Starting"); 3020 pkt_dev->running = 1; /* Cranke yeself! */ 3021 started++; 3022 } else 3023 strcpy(pkt_dev->result, "Error starting"); 3024 } 3025 rcu_read_unlock(); 3026 if (started) 3027 t->control &= ~(T_STOP); 3028} 3029 3030static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn) 3031{ 3032 struct pktgen_thread *t; 3033 3034 func_enter(); 3035 3036 mutex_lock(&pktgen_thread_lock); 3037 3038 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3039 t->control |= T_STOP; 3040 3041 mutex_unlock(&pktgen_thread_lock); 3042} 3043 3044static int thread_is_running(const struct pktgen_thread *t) 3045{ 3046 const struct pktgen_dev *pkt_dev; 3047 3048 rcu_read_lock(); 3049 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 3050 if (pkt_dev->running) { 3051 rcu_read_unlock(); 3052 return 1; 3053 } 3054 rcu_read_unlock(); 3055 return 0; 3056} 3057 3058static int pktgen_wait_thread_run(struct pktgen_thread *t) 3059{ 3060 while (thread_is_running(t)) { 3061 3062 /* note: 't' will still be around even after the unlock/lock 3063 * cycle because pktgen_thread threads are only cleared at 3064 * net exit 3065 */ 3066 mutex_unlock(&pktgen_thread_lock); 3067 msleep_interruptible(100); 3068 mutex_lock(&pktgen_thread_lock); 3069 3070 if (signal_pending(current)) 3071 goto signal; 3072 } 3073 return 1; 3074signal: 3075 return 0; 3076} 3077 3078static int pktgen_wait_all_threads_run(struct pktgen_net *pn) 3079{ 3080 struct pktgen_thread *t; 3081 int sig = 1; 3082 3083 /* prevent from racing with rmmod */ 3084 if (!try_module_get(THIS_MODULE)) 3085 return sig; 3086 3087 mutex_lock(&pktgen_thread_lock); 3088 3089 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 3090 sig = pktgen_wait_thread_run(t); 3091 if (sig == 0) 3092 break; 3093 } 3094 3095 if (sig == 0) 3096 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3097 t->control |= (T_STOP); 3098 3099 mutex_unlock(&pktgen_thread_lock); 3100 module_put(THIS_MODULE); 3101 return sig; 3102} 3103 3104static void pktgen_run_all_threads(struct pktgen_net *pn) 3105{ 3106 struct pktgen_thread *t; 3107 3108 func_enter(); 3109 3110 mutex_lock(&pktgen_thread_lock); 3111 3112 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3113 t->control |= (T_RUN); 3114 3115 mutex_unlock(&pktgen_thread_lock); 3116 3117 /* Propagate thread->control */ 3118 schedule_timeout_interruptible(msecs_to_jiffies(125)); 3119 3120 pktgen_wait_all_threads_run(pn); 3121} 3122 3123static void pktgen_reset_all_threads(struct pktgen_net *pn) 3124{ 3125 struct pktgen_thread *t; 3126 3127 func_enter(); 3128 3129 mutex_lock(&pktgen_thread_lock); 3130 3131 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3132 t->control |= (T_REMDEVALL); 3133 3134 mutex_unlock(&pktgen_thread_lock); 3135 3136 /* Propagate thread->control */ 3137 schedule_timeout_interruptible(msecs_to_jiffies(125)); 3138 3139 pktgen_wait_all_threads_run(pn); 3140} 3141 3142static void show_results(struct pktgen_dev *pkt_dev, int nr_frags) 3143{ 3144 __u64 bps, mbps, pps; 3145 char *p = pkt_dev->result; 3146 ktime_t elapsed = ktime_sub(pkt_dev->stopped_at, 3147 pkt_dev->started_at); 3148 ktime_t idle = ns_to_ktime(pkt_dev->idle_acc); 3149 3150 p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n", 3151 (unsigned long long)ktime_to_us(elapsed), 3152 (unsigned long long)ktime_to_us(ktime_sub(elapsed, idle)), 3153 (unsigned long long)ktime_to_us(idle), 3154 (unsigned long long)pkt_dev->sofar, 3155 pkt_dev->cur_pkt_size, nr_frags); 3156 3157 pps = div64_u64(pkt_dev->sofar * NSEC_PER_SEC, 3158 ktime_to_ns(elapsed)); 3159 3160 bps = pps * 8 * pkt_dev->cur_pkt_size; 3161 3162 mbps = bps; 3163 do_div(mbps, 1000000); 3164 p += sprintf(p, " %llupps %lluMb/sec (%llubps) errors: %llu", 3165 (unsigned long long)pps, 3166 (unsigned long long)mbps, 3167 (unsigned long long)bps, 3168 (unsigned long long)pkt_dev->errors); 3169} 3170 3171/* Set stopped-at timer, remove from running list, do counters & statistics */ 3172static int pktgen_stop_device(struct pktgen_dev *pkt_dev) 3173{ 3174 int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1; 3175 3176 if (!pkt_dev->running) { 3177 pr_warn("interface: %s is already stopped\n", 3178 pkt_dev->odevname); 3179 return -EINVAL; 3180 } 3181 3182 pkt_dev->running = 0; 3183 kfree_skb(pkt_dev->skb); 3184 pkt_dev->skb = NULL; 3185 pkt_dev->stopped_at = ktime_get(); 3186 3187 show_results(pkt_dev, nr_frags); 3188 3189 return 0; 3190} 3191 3192static struct pktgen_dev *next_to_run(struct pktgen_thread *t) 3193{ 3194 struct pktgen_dev *pkt_dev, *best = NULL; 3195 3196 rcu_read_lock(); 3197 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3198 if (!pkt_dev->running) 3199 continue; 3200 if (best == NULL) 3201 best = pkt_dev; 3202 else if (ktime_compare(pkt_dev->next_tx, best->next_tx) < 0) 3203 best = pkt_dev; 3204 } 3205 rcu_read_unlock(); 3206 3207 return best; 3208} 3209 3210static void pktgen_stop(struct pktgen_thread *t) 3211{ 3212 struct pktgen_dev *pkt_dev; 3213 3214 func_enter(); 3215 3216 rcu_read_lock(); 3217 3218 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3219 pktgen_stop_device(pkt_dev); 3220 } 3221 3222 rcu_read_unlock(); 3223} 3224 3225/* 3226 * one of our devices needs to be removed - find it 3227 * and remove it 3228 */ 3229static void pktgen_rem_one_if(struct pktgen_thread *t) 3230{ 3231 struct list_head *q, *n; 3232 struct pktgen_dev *cur; 3233 3234 func_enter(); 3235 3236 list_for_each_safe(q, n, &t->if_list) { 3237 cur = list_entry(q, struct pktgen_dev, list); 3238 3239 if (!cur->removal_mark) 3240 continue; 3241 3242 kfree_skb(cur->skb); 3243 cur->skb = NULL; 3244 3245 pktgen_remove_device(t, cur); 3246 3247 break; 3248 } 3249} 3250 3251static void pktgen_rem_all_ifs(struct pktgen_thread *t) 3252{ 3253 struct list_head *q, *n; 3254 struct pktgen_dev *cur; 3255 3256 func_enter(); 3257 3258 /* Remove all devices, free mem */ 3259 3260 list_for_each_safe(q, n, &t->if_list) { 3261 cur = list_entry(q, struct pktgen_dev, list); 3262 3263 kfree_skb(cur->skb); 3264 cur->skb = NULL; 3265 3266 pktgen_remove_device(t, cur); 3267 } 3268} 3269 3270static void pktgen_rem_thread(struct pktgen_thread *t) 3271{ 3272 /* Remove from the thread list */ 3273 remove_proc_entry(t->tsk->comm, t->net->proc_dir); 3274} 3275 3276static void pktgen_resched(struct pktgen_dev *pkt_dev) 3277{ 3278 ktime_t idle_start = ktime_get(); 3279 schedule(); 3280 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start)); 3281} 3282 3283static void pktgen_wait_for_skb(struct pktgen_dev *pkt_dev) 3284{ 3285 ktime_t idle_start = ktime_get(); 3286 3287 while (refcount_read(&(pkt_dev->skb->users)) != 1) { 3288 if (signal_pending(current)) 3289 break; 3290 3291 if (need_resched()) 3292 pktgen_resched(pkt_dev); 3293 else 3294 cpu_relax(); 3295 } 3296 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start)); 3297} 3298 3299static void pktgen_xmit(struct pktgen_dev *pkt_dev) 3300{ 3301 unsigned int burst = READ_ONCE(pkt_dev->burst); 3302 struct net_device *odev = pkt_dev->odev; 3303 struct netdev_queue *txq; 3304 struct sk_buff *skb; 3305 int ret; 3306 3307 /* If device is offline, then don't send */ 3308 if (unlikely(!netif_running(odev) || !netif_carrier_ok(odev))) { 3309 pktgen_stop_device(pkt_dev); 3310 return; 3311 } 3312 3313 /* This is max DELAY, this has special meaning of 3314 * "never transmit" 3315 */ 3316 if (unlikely(pkt_dev->delay == ULLONG_MAX)) { 3317 pkt_dev->next_tx = ktime_add_ns(ktime_get(), ULONG_MAX); 3318 return; 3319 } 3320 3321 /* If no skb or clone count exhausted then get new one */ 3322 if (!pkt_dev->skb || (pkt_dev->last_ok && 3323 ++pkt_dev->clone_count >= pkt_dev->clone_skb)) { 3324 /* build a new pkt */ 3325 kfree_skb(pkt_dev->skb); 3326 3327 pkt_dev->skb = fill_packet(odev, pkt_dev); 3328 if (pkt_dev->skb == NULL) { 3329 pr_err("ERROR: couldn't allocate skb in fill_packet\n"); 3330 schedule(); 3331 pkt_dev->clone_count--; /* back out increment, OOM */ 3332 return; 3333 } 3334 pkt_dev->last_pkt_size = pkt_dev->skb->len; 3335 pkt_dev->clone_count = 0; /* reset counter */ 3336 } 3337 3338 if (pkt_dev->delay && pkt_dev->last_ok) 3339 spin(pkt_dev, pkt_dev->next_tx); 3340 3341 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) { 3342 skb = pkt_dev->skb; 3343 skb->protocol = eth_type_trans(skb, skb->dev); 3344 refcount_add(burst, &skb->users); 3345 local_bh_disable(); 3346 do { 3347 ret = netif_receive_skb(skb); 3348 if (ret == NET_RX_DROP) 3349 pkt_dev->errors++; 3350 pkt_dev->sofar++; 3351 pkt_dev->seq_num++; 3352 if (refcount_read(&skb->users) != burst) { 3353 /* skb was queued by rps/rfs or taps, 3354 * so cannot reuse this skb 3355 */ 3356 WARN_ON(refcount_sub_and_test(burst - 1, &skb->users)); 3357 /* get out of the loop and wait 3358 * until skb is consumed 3359 */ 3360 break; 3361 } 3362 /* skb was 'freed' by stack, so clean few 3363 * bits and reuse it 3364 */ 3365 skb_reset_redirect(skb); 3366 } while (--burst > 0); 3367 goto out; /* Skips xmit_mode M_START_XMIT */ 3368 } else if (pkt_dev->xmit_mode == M_QUEUE_XMIT) { 3369 local_bh_disable(); 3370 refcount_inc(&pkt_dev->skb->users); 3371 3372 ret = dev_queue_xmit(pkt_dev->skb); 3373 switch (ret) { 3374 case NET_XMIT_SUCCESS: 3375 pkt_dev->sofar++; 3376 pkt_dev->seq_num++; 3377 pkt_dev->tx_bytes += pkt_dev->last_pkt_size; 3378 break; 3379 case NET_XMIT_DROP: 3380 case NET_XMIT_CN: 3381 /* These are all valid return codes for a qdisc but 3382 * indicate packets are being dropped or will likely 3383 * be dropped soon. 3384 */ 3385 case NETDEV_TX_BUSY: 3386 /* qdisc may call dev_hard_start_xmit directly in cases 3387 * where no queues exist e.g. loopback device, virtual 3388 * devices, etc. In this case we need to handle 3389 * NETDEV_TX_ codes. 3390 */ 3391 default: 3392 pkt_dev->errors++; 3393 net_info_ratelimited("%s xmit error: %d\n", 3394 pkt_dev->odevname, ret); 3395 break; 3396 } 3397 goto out; 3398 } 3399 3400 txq = skb_get_tx_queue(odev, pkt_dev->skb); 3401 3402 local_bh_disable(); 3403 3404 HARD_TX_LOCK(odev, txq, smp_processor_id()); 3405 3406 if (unlikely(netif_xmit_frozen_or_drv_stopped(txq))) { 3407 pkt_dev->last_ok = 0; 3408 goto unlock; 3409 } 3410 refcount_add(burst, &pkt_dev->skb->users); 3411 3412xmit_more: 3413 ret = netdev_start_xmit(pkt_dev->skb, odev, txq, --burst > 0); 3414 3415 switch (ret) { 3416 case NETDEV_TX_OK: 3417 pkt_dev->last_ok = 1; 3418 pkt_dev->sofar++; 3419 pkt_dev->seq_num++; 3420 pkt_dev->tx_bytes += pkt_dev->last_pkt_size; 3421 if (burst > 0 && !netif_xmit_frozen_or_drv_stopped(txq)) 3422 goto xmit_more; 3423 break; 3424 case NET_XMIT_DROP: 3425 case NET_XMIT_CN: 3426 /* skb has been consumed */ 3427 pkt_dev->errors++; 3428 break; 3429 default: /* Drivers are not supposed to return other values! */ 3430 net_info_ratelimited("%s xmit error: %d\n", 3431 pkt_dev->odevname, ret); 3432 pkt_dev->errors++; 3433 /* fall through */ 3434 case NETDEV_TX_BUSY: 3435 /* Retry it next time */ 3436 refcount_dec(&(pkt_dev->skb->users)); 3437 pkt_dev->last_ok = 0; 3438 } 3439 if (unlikely(burst)) 3440 WARN_ON(refcount_sub_and_test(burst, &pkt_dev->skb->users)); 3441unlock: 3442 HARD_TX_UNLOCK(odev, txq); 3443 3444out: 3445 local_bh_enable(); 3446 3447 /* If pkt_dev->count is zero, then run forever */ 3448 if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) { 3449 pktgen_wait_for_skb(pkt_dev); 3450 3451 /* Done with this */ 3452 pktgen_stop_device(pkt_dev); 3453 } 3454} 3455 3456/* 3457 * Main loop of the thread goes here 3458 */ 3459 3460static int pktgen_thread_worker(void *arg) 3461{ 3462 DEFINE_WAIT(wait); 3463 struct pktgen_thread *t = arg; 3464 struct pktgen_dev *pkt_dev = NULL; 3465 int cpu = t->cpu; 3466 3467 BUG_ON(smp_processor_id() != cpu); 3468 3469 init_waitqueue_head(&t->queue); 3470 complete(&t->start_done); 3471 3472 pr_debug("starting pktgen/%d: pid=%d\n", cpu, task_pid_nr(current)); 3473 3474 set_freezable(); 3475 3476 while (!kthread_should_stop()) { 3477 pkt_dev = next_to_run(t); 3478 3479 if (unlikely(!pkt_dev && t->control == 0)) { 3480 if (t->net->pktgen_exiting) 3481 break; 3482 wait_event_interruptible_timeout(t->queue, 3483 t->control != 0, 3484 HZ/10); 3485 try_to_freeze(); 3486 continue; 3487 } 3488 3489 if (likely(pkt_dev)) { 3490 pktgen_xmit(pkt_dev); 3491 3492 if (need_resched()) 3493 pktgen_resched(pkt_dev); 3494 else 3495 cpu_relax(); 3496 } 3497 3498 if (t->control & T_STOP) { 3499 pktgen_stop(t); 3500 t->control &= ~(T_STOP); 3501 } 3502 3503 if (t->control & T_RUN) { 3504 pktgen_run(t); 3505 t->control &= ~(T_RUN); 3506 } 3507 3508 if (t->control & T_REMDEVALL) { 3509 pktgen_rem_all_ifs(t); 3510 t->control &= ~(T_REMDEVALL); 3511 } 3512 3513 if (t->control & T_REMDEV) { 3514 pktgen_rem_one_if(t); 3515 t->control &= ~(T_REMDEV); 3516 } 3517 3518 try_to_freeze(); 3519 } 3520 3521 pr_debug("%s stopping all device\n", t->tsk->comm); 3522 pktgen_stop(t); 3523 3524 pr_debug("%s removing all device\n", t->tsk->comm); 3525 pktgen_rem_all_ifs(t); 3526 3527 pr_debug("%s removing thread\n", t->tsk->comm); 3528 pktgen_rem_thread(t); 3529 3530 return 0; 3531} 3532 3533static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t, 3534 const char *ifname, bool exact) 3535{ 3536 struct pktgen_dev *p, *pkt_dev = NULL; 3537 size_t len = strlen(ifname); 3538 3539 rcu_read_lock(); 3540 list_for_each_entry_rcu(p, &t->if_list, list) 3541 if (strncmp(p->odevname, ifname, len) == 0) { 3542 if (p->odevname[len]) { 3543 if (exact || p->odevname[len] != '@') 3544 continue; 3545 } 3546 pkt_dev = p; 3547 break; 3548 } 3549 3550 rcu_read_unlock(); 3551 pr_debug("find_dev(%s) returning %p\n", ifname, pkt_dev); 3552 return pkt_dev; 3553} 3554 3555/* 3556 * Adds a dev at front of if_list. 3557 */ 3558 3559static int add_dev_to_thread(struct pktgen_thread *t, 3560 struct pktgen_dev *pkt_dev) 3561{ 3562 int rv = 0; 3563 3564 /* This function cannot be called concurrently, as its called 3565 * under pktgen_thread_lock mutex, but it can run from 3566 * userspace on another CPU than the kthread. The if_lock() 3567 * is used here to sync with concurrent instances of 3568 * _rem_dev_from_if_list() invoked via kthread, which is also 3569 * updating the if_list */ 3570 if_lock(t); 3571 3572 if (pkt_dev->pg_thread) { 3573 pr_err("ERROR: already assigned to a thread\n"); 3574 rv = -EBUSY; 3575 goto out; 3576 } 3577 3578 pkt_dev->running = 0; 3579 pkt_dev->pg_thread = t; 3580 list_add_rcu(&pkt_dev->list, &t->if_list); 3581 3582out: 3583 if_unlock(t); 3584 return rv; 3585} 3586 3587/* Called under thread lock */ 3588 3589static int pktgen_add_device(struct pktgen_thread *t, const char *ifname) 3590{ 3591 struct pktgen_dev *pkt_dev; 3592 int err; 3593 int node = cpu_to_node(t->cpu); 3594 3595 /* We don't allow a device to be on several threads */ 3596 3597 pkt_dev = __pktgen_NN_threads(t->net, ifname, FIND); 3598 if (pkt_dev) { 3599 pr_err("ERROR: interface already used\n"); 3600 return -EBUSY; 3601 } 3602 3603 pkt_dev = kzalloc_node(sizeof(struct pktgen_dev), GFP_KERNEL, node); 3604 if (!pkt_dev) 3605 return -ENOMEM; 3606 3607 strcpy(pkt_dev->odevname, ifname); 3608 pkt_dev->flows = vzalloc_node(array_size(MAX_CFLOWS, 3609 sizeof(struct flow_state)), 3610 node); 3611 if (pkt_dev->flows == NULL) { 3612 kfree(pkt_dev); 3613 return -ENOMEM; 3614 } 3615 3616 pkt_dev->removal_mark = 0; 3617 pkt_dev->nfrags = 0; 3618 pkt_dev->delay = pg_delay_d; 3619 pkt_dev->count = pg_count_d; 3620 pkt_dev->sofar = 0; 3621 pkt_dev->udp_src_min = 9; /* sink port */ 3622 pkt_dev->udp_src_max = 9; 3623 pkt_dev->udp_dst_min = 9; 3624 pkt_dev->udp_dst_max = 9; 3625 pkt_dev->vlan_p = 0; 3626 pkt_dev->vlan_cfi = 0; 3627 pkt_dev->vlan_id = 0xffff; 3628 pkt_dev->svlan_p = 0; 3629 pkt_dev->svlan_cfi = 0; 3630 pkt_dev->svlan_id = 0xffff; 3631 pkt_dev->burst = 1; 3632 pkt_dev->node = NUMA_NO_NODE; 3633 3634 err = pktgen_setup_dev(t->net, pkt_dev, ifname); 3635 if (err) 3636 goto out1; 3637 if (pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING) 3638 pkt_dev->clone_skb = pg_clone_skb_d; 3639 3640 pkt_dev->entry = proc_create_data(ifname, 0600, t->net->proc_dir, 3641 &pktgen_if_proc_ops, pkt_dev); 3642 if (!pkt_dev->entry) { 3643 pr_err("cannot create %s/%s procfs entry\n", 3644 PG_PROC_DIR, ifname); 3645 err = -EINVAL; 3646 goto out2; 3647 } 3648#ifdef CONFIG_XFRM 3649 pkt_dev->ipsmode = XFRM_MODE_TRANSPORT; 3650 pkt_dev->ipsproto = IPPROTO_ESP; 3651 3652 /* xfrm tunnel mode needs additional dst to extract outter 3653 * ip header protocol/ttl/id field, here creat a phony one. 3654 * instead of looking for a valid rt, which definitely hurting 3655 * performance under such circumstance. 3656 */ 3657 pkt_dev->dstops.family = AF_INET; 3658 pkt_dev->xdst.u.dst.dev = pkt_dev->odev; 3659 dst_init_metrics(&pkt_dev->xdst.u.dst, pktgen_dst_metrics, false); 3660 pkt_dev->xdst.child = &pkt_dev->xdst.u.dst; 3661 pkt_dev->xdst.u.dst.ops = &pkt_dev->dstops; 3662#endif 3663 3664 return add_dev_to_thread(t, pkt_dev); 3665out2: 3666 dev_put(pkt_dev->odev); 3667out1: 3668#ifdef CONFIG_XFRM 3669 free_SAs(pkt_dev); 3670#endif 3671 vfree(pkt_dev->flows); 3672 kfree(pkt_dev); 3673 return err; 3674} 3675 3676static int __net_init pktgen_create_thread(int cpu, struct pktgen_net *pn) 3677{ 3678 struct pktgen_thread *t; 3679 struct proc_dir_entry *pe; 3680 struct task_struct *p; 3681 3682 t = kzalloc_node(sizeof(struct pktgen_thread), GFP_KERNEL, 3683 cpu_to_node(cpu)); 3684 if (!t) { 3685 pr_err("ERROR: out of memory, can't create new thread\n"); 3686 return -ENOMEM; 3687 } 3688 3689 mutex_init(&t->if_lock); 3690 t->cpu = cpu; 3691 3692 INIT_LIST_HEAD(&t->if_list); 3693 3694 list_add_tail(&t->th_list, &pn->pktgen_threads); 3695 init_completion(&t->start_done); 3696 3697 p = kthread_create_on_node(pktgen_thread_worker, 3698 t, 3699 cpu_to_node(cpu), 3700 "kpktgend_%d", cpu); 3701 if (IS_ERR(p)) { 3702 pr_err("kernel_thread() failed for cpu %d\n", t->cpu); 3703 list_del(&t->th_list); 3704 kfree(t); 3705 return PTR_ERR(p); 3706 } 3707 kthread_bind(p, cpu); 3708 t->tsk = p; 3709 3710 pe = proc_create_data(t->tsk->comm, 0600, pn->proc_dir, 3711 &pktgen_thread_proc_ops, t); 3712 if (!pe) { 3713 pr_err("cannot create %s/%s procfs entry\n", 3714 PG_PROC_DIR, t->tsk->comm); 3715 kthread_stop(p); 3716 list_del(&t->th_list); 3717 kfree(t); 3718 return -EINVAL; 3719 } 3720 3721 t->net = pn; 3722 get_task_struct(p); 3723 wake_up_process(p); 3724 wait_for_completion(&t->start_done); 3725 3726 return 0; 3727} 3728 3729/* 3730 * Removes a device from the thread if_list. 3731 */ 3732static void _rem_dev_from_if_list(struct pktgen_thread *t, 3733 struct pktgen_dev *pkt_dev) 3734{ 3735 struct list_head *q, *n; 3736 struct pktgen_dev *p; 3737 3738 if_lock(t); 3739 list_for_each_safe(q, n, &t->if_list) { 3740 p = list_entry(q, struct pktgen_dev, list); 3741 if (p == pkt_dev) 3742 list_del_rcu(&p->list); 3743 } 3744 if_unlock(t); 3745} 3746 3747static int pktgen_remove_device(struct pktgen_thread *t, 3748 struct pktgen_dev *pkt_dev) 3749{ 3750 pr_debug("remove_device pkt_dev=%p\n", pkt_dev); 3751 3752 if (pkt_dev->running) { 3753 pr_warn("WARNING: trying to remove a running interface, stopping it now\n"); 3754 pktgen_stop_device(pkt_dev); 3755 } 3756 3757 /* Dis-associate from the interface */ 3758 3759 if (pkt_dev->odev) { 3760 dev_put(pkt_dev->odev); 3761 pkt_dev->odev = NULL; 3762 } 3763 3764 /* Remove proc before if_list entry, because add_device uses 3765 * list to determine if interface already exist, avoid race 3766 * with proc_create_data() */ 3767 proc_remove(pkt_dev->entry); 3768 3769 /* And update the thread if_list */ 3770 _rem_dev_from_if_list(t, pkt_dev); 3771 3772#ifdef CONFIG_XFRM 3773 free_SAs(pkt_dev); 3774#endif 3775 vfree(pkt_dev->flows); 3776 if (pkt_dev->page) 3777 put_page(pkt_dev->page); 3778 kfree_rcu(pkt_dev, rcu); 3779 return 0; 3780} 3781 3782static int __net_init pg_net_init(struct net *net) 3783{ 3784 struct pktgen_net *pn = net_generic(net, pg_net_id); 3785 struct proc_dir_entry *pe; 3786 int cpu, ret = 0; 3787 3788 pn->net = net; 3789 INIT_LIST_HEAD(&pn->pktgen_threads); 3790 pn->pktgen_exiting = false; 3791 pn->proc_dir = proc_mkdir(PG_PROC_DIR, pn->net->proc_net); 3792 if (!pn->proc_dir) { 3793 pr_warn("cannot create /proc/net/%s\n", PG_PROC_DIR); 3794 return -ENODEV; 3795 } 3796 pe = proc_create(PGCTRL, 0600, pn->proc_dir, &pktgen_proc_ops); 3797 if (pe == NULL) { 3798 pr_err("cannot create %s procfs entry\n", PGCTRL); 3799 ret = -EINVAL; 3800 goto remove; 3801 } 3802 3803 for_each_online_cpu(cpu) { 3804 int err; 3805 3806 err = pktgen_create_thread(cpu, pn); 3807 if (err) 3808 pr_warn("Cannot create thread for cpu %d (%d)\n", 3809 cpu, err); 3810 } 3811 3812 if (list_empty(&pn->pktgen_threads)) { 3813 pr_err("Initialization failed for all threads\n"); 3814 ret = -ENODEV; 3815 goto remove_entry; 3816 } 3817 3818 return 0; 3819 3820remove_entry: 3821 remove_proc_entry(PGCTRL, pn->proc_dir); 3822remove: 3823 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net); 3824 return ret; 3825} 3826 3827static void __net_exit pg_net_exit(struct net *net) 3828{ 3829 struct pktgen_net *pn = net_generic(net, pg_net_id); 3830 struct pktgen_thread *t; 3831 struct list_head *q, *n; 3832 LIST_HEAD(list); 3833 3834 /* Stop all interfaces & threads */ 3835 pn->pktgen_exiting = true; 3836 3837 mutex_lock(&pktgen_thread_lock); 3838 list_splice_init(&pn->pktgen_threads, &list); 3839 mutex_unlock(&pktgen_thread_lock); 3840 3841 list_for_each_safe(q, n, &list) { 3842 t = list_entry(q, struct pktgen_thread, th_list); 3843 list_del(&t->th_list); 3844 kthread_stop(t->tsk); 3845 put_task_struct(t->tsk); 3846 kfree(t); 3847 } 3848 3849 remove_proc_entry(PGCTRL, pn->proc_dir); 3850 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net); 3851} 3852 3853static struct pernet_operations pg_net_ops = { 3854 .init = pg_net_init, 3855 .exit = pg_net_exit, 3856 .id = &pg_net_id, 3857 .size = sizeof(struct pktgen_net), 3858}; 3859 3860static int __init pg_init(void) 3861{ 3862 int ret = 0; 3863 3864 pr_info("%s", version); 3865 ret = register_pernet_subsys(&pg_net_ops); 3866 if (ret) 3867 return ret; 3868 ret = register_netdevice_notifier(&pktgen_notifier_block); 3869 if (ret) 3870 unregister_pernet_subsys(&pg_net_ops); 3871 3872 return ret; 3873} 3874 3875static void __exit pg_cleanup(void) 3876{ 3877 unregister_netdevice_notifier(&pktgen_notifier_block); 3878 unregister_pernet_subsys(&pg_net_ops); 3879 /* Don't need rcu_barrier() due to use of kfree_rcu() */ 3880} 3881 3882module_init(pg_init); 3883module_exit(pg_cleanup); 3884 3885MODULE_AUTHOR("Robert Olsson <robert.olsson@its.uu.se>"); 3886MODULE_DESCRIPTION("Packet Generator tool"); 3887MODULE_LICENSE("GPL"); 3888MODULE_VERSION(VERSION); 3889module_param(pg_count_d, int, 0); 3890MODULE_PARM_DESC(pg_count_d, "Default number of packets to inject"); 3891module_param(pg_delay_d, int, 0); 3892MODULE_PARM_DESC(pg_delay_d, "Default delay between packets (nanoseconds)"); 3893module_param(pg_clone_skb_d, int, 0); 3894MODULE_PARM_DESC(pg_clone_skb_d, "Default number of copies of the same packet"); 3895module_param(debug, int, 0); 3896MODULE_PARM_DESC(debug, "Enable debugging of pktgen module");