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