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 that 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/minmax.h> 118 #include <linux/module.h> 119 #include <linux/moduleparam.h> 120 #include <linux/kernel.h> 121 #include <linux/mutex.h> 122 #include <linux/sched.h> 123 #include <linux/slab.h> 124 #include <linux/vmalloc.h> 125 #include <linux/unistd.h> 126 #include <linux/string.h> 127 #include <linux/ptrace.h> 128 #include <linux/errno.h> 129 #include <linux/ioport.h> 130 #include <linux/interrupt.h> 131 #include <linux/capability.h> 132 #include <linux/hrtimer.h> 133 #include <linux/freezer.h> 134 #include <linux/delay.h> 135 #include <linux/timer.h> 136 #include <linux/list.h> 137 #include <linux/init.h> 138 #include <linux/skbuff.h> 139 #include <linux/netdevice.h> 140 #include <linux/inet.h> 141 #include <linux/inetdevice.h> 142 #include <linux/rtnetlink.h> 143 #include <linux/if_arp.h> 144 #include <linux/if_vlan.h> 145 #include <linux/in.h> 146 #include <linux/ip.h> 147 #include <linux/ipv6.h> 148 #include <linux/udp.h> 149 #include <linux/proc_fs.h> 150 #include <linux/seq_file.h> 151 #include <linux/wait.h> 152 #include <linux/etherdevice.h> 153 #include <linux/kthread.h> 154 #include <linux/prefetch.h> 155 #include <linux/mmzone.h> 156 #include <net/net_namespace.h> 157 #include <net/checksum.h> 158 #include <net/ipv6.h> 159 #include <net/udp.h> 160 #include <net/ip6_checksum.h> 161 #include <net/addrconf.h> 162 #include <net/xfrm.h> 163 #include <net/netns/generic.h> 164 #include <asm/byteorder.h> 165 #include <linux/rcupdate.h> 166 #include <linux/bitops.h> 167 #include <linux/io.h> 168 #include <linux/timex.h> 169 #include <linux/uaccess.h> 170 #include <asm/dma.h> 171 #include <asm/div64.h> /* do_div */ 172 173 #define VERSION "2.75" 174 #define IP_NAME_SZ 32 175 #define MAX_MPLS_LABELS 16 /* This is the max label stack depth */ 176 #define MPLS_STACK_BOTTOM htonl(0x00000100) 177 /* Max number of internet mix entries that can be specified in imix_weights. */ 178 #define MAX_IMIX_ENTRIES 20 179 #define IMIX_PRECISION 100 /* Precision of IMIX distribution */ 180 181 #define func_enter() pr_debug("entering %s\n", __func__) 182 183 #define PKT_FLAGS \ 184 pf(IPV6) /* Interface in IPV6 Mode */ \ 185 pf(IPSRC_RND) /* IP-Src Random */ \ 186 pf(IPDST_RND) /* IP-Dst Random */ \ 187 pf(TXSIZE_RND) /* Transmit size is random */ \ 188 pf(UDPSRC_RND) /* UDP-Src Random */ \ 189 pf(UDPDST_RND) /* UDP-Dst Random */ \ 190 pf(UDPCSUM) /* Include UDP checksum */ \ 191 pf(NO_TIMESTAMP) /* Don't timestamp packets (default TS) */ \ 192 pf(MPLS_RND) /* Random MPLS labels */ \ 193 pf(QUEUE_MAP_RND) /* queue map Random */ \ 194 pf(QUEUE_MAP_CPU) /* queue map mirrors smp_processor_id() */ \ 195 pf(FLOW_SEQ) /* Sequential flows */ \ 196 pf(IPSEC) /* ipsec on for flows */ \ 197 pf(MACSRC_RND) /* MAC-Src Random */ \ 198 pf(MACDST_RND) /* MAC-Dst Random */ \ 199 pf(VID_RND) /* Random VLAN ID */ \ 200 pf(SVID_RND) /* Random SVLAN ID */ \ 201 pf(NODE) /* Node memory alloc*/ \ 202 pf(SHARED) /* Shared SKB */ \ 203 204 #define pf(flag) flag##_SHIFT, 205 enum pkt_flags { 206 PKT_FLAGS 207 }; 208 #undef pf 209 210 /* Device flag bits */ 211 #define pf(flag) static const __u32 F_##flag = (1<<flag##_SHIFT); 212 PKT_FLAGS 213 #undef pf 214 215 #define pf(flag) __stringify(flag), 216 static char *pkt_flag_names[] = { 217 PKT_FLAGS 218 }; 219 #undef pf 220 221 #define NR_PKT_FLAGS ARRAY_SIZE(pkt_flag_names) 222 223 /* Thread control flag bits */ 224 #define T_STOP (1<<0) /* Stop run */ 225 #define T_RUN (1<<1) /* Start run */ 226 #define T_REMDEVALL (1<<2) /* Remove all devs */ 227 #define T_REMDEV (1<<3) /* Remove one dev */ 228 229 /* Xmit modes */ 230 #define M_START_XMIT 0 /* Default normal TX */ 231 #define M_NETIF_RECEIVE 1 /* Inject packets into stack */ 232 #define M_QUEUE_XMIT 2 /* Inject packet into qdisc */ 233 234 /* If lock -- protects updating of if_list */ 235 #define if_lock(t) mutex_lock(&(t->if_lock)) 236 #define if_unlock(t) mutex_unlock(&(t->if_lock)) 237 238 /* Used to help with determining the pkts on receive */ 239 #define PKTGEN_MAGIC 0xbe9be955 240 #define PG_PROC_DIR "pktgen" 241 #define PGCTRL "pgctrl" 242 243 #define MAX_CFLOWS 65536 244 245 #define VLAN_TAG_SIZE(x) ((x)->vlan_id == 0xffff ? 0 : 4) 246 #define SVLAN_TAG_SIZE(x) ((x)->svlan_id == 0xffff ? 0 : 4) 247 248 struct imix_pkt { 249 u64 size; 250 u64 weight; 251 u64 count_so_far; 252 }; 253 254 struct flow_state { 255 __be32 cur_daddr; 256 int count; 257 #ifdef CONFIG_XFRM 258 struct xfrm_state *x; 259 #endif 260 __u32 flags; 261 }; 262 263 /* flow flag bits */ 264 #define F_INIT (1<<0) /* flow has been initialized */ 265 266 struct pktgen_dev { 267 /* 268 * Try to keep frequent/infrequent used vars. separated. 269 */ 270 struct proc_dir_entry *entry; /* proc file */ 271 struct pktgen_thread *pg_thread;/* the owner */ 272 struct list_head list; /* chaining in the thread's run-queue */ 273 struct rcu_head rcu; /* freed by RCU */ 274 275 int running; /* if false, the test will stop */ 276 277 /* If min != max, then we will either do a linear iteration, or 278 * we will do a random selection from within the range. 279 */ 280 __u32 flags; 281 int xmit_mode; 282 int min_pkt_size; 283 int max_pkt_size; 284 int pkt_overhead; /* overhead for MPLS, VLANs, IPSEC etc */ 285 int nfrags; 286 int removal_mark; /* non-zero => the device is marked for 287 * removal by worker thread 288 */ 289 290 struct page *page; 291 u64 delay; /* nano-seconds */ 292 293 __u64 count; /* Default No packets to send */ 294 __u64 sofar; /* How many pkts we've sent so far */ 295 __u64 tx_bytes; /* How many bytes we've transmitted */ 296 __u64 errors; /* Errors when trying to transmit, */ 297 298 /* runtime counters relating to clone_skb */ 299 300 __u32 clone_count; 301 int last_ok; /* Was last skb sent? 302 * Or a failed transmit of some sort? 303 * This will keep sequence numbers in order 304 */ 305 ktime_t next_tx; 306 ktime_t started_at; 307 ktime_t stopped_at; 308 u64 idle_acc; /* nano-seconds */ 309 310 __u32 seq_num; 311 312 int clone_skb; /* 313 * Use multiple SKBs during packet gen. 314 * If this number is greater than 1, then 315 * that many copies of the same packet will be 316 * sent before a new packet is allocated. 317 * If you want to send 1024 identical packets 318 * before creating a new packet, 319 * set clone_skb to 1024. 320 */ 321 322 char dst_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 323 char dst_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 324 char src_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 325 char src_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 326 327 struct in6_addr in6_saddr; 328 struct in6_addr in6_daddr; 329 struct in6_addr cur_in6_daddr; 330 struct in6_addr cur_in6_saddr; 331 /* For ranges */ 332 struct in6_addr min_in6_daddr; 333 struct in6_addr max_in6_daddr; 334 struct in6_addr min_in6_saddr; 335 struct in6_addr max_in6_saddr; 336 337 /* If we're doing ranges, random or incremental, then this 338 * defines the min/max for those ranges. 339 */ 340 __be32 saddr_min; /* inclusive, source IP address */ 341 __be32 saddr_max; /* exclusive, source IP address */ 342 __be32 daddr_min; /* inclusive, dest IP address */ 343 __be32 daddr_max; /* exclusive, dest IP address */ 344 345 __u16 udp_src_min; /* inclusive, source UDP port */ 346 __u16 udp_src_max; /* exclusive, source UDP port */ 347 __u16 udp_dst_min; /* inclusive, dest UDP port */ 348 __u16 udp_dst_max; /* exclusive, dest UDP port */ 349 350 /* DSCP + ECN */ 351 __u8 tos; /* six MSB of (former) IPv4 TOS 352 * are for dscp codepoint 353 */ 354 __u8 traffic_class; /* ditto for the (former) Traffic Class in IPv6 355 * (see RFC 3260, sec. 4) 356 */ 357 358 /* IMIX */ 359 unsigned int n_imix_entries; 360 struct imix_pkt imix_entries[MAX_IMIX_ENTRIES]; 361 /* Maps 0-IMIX_PRECISION range to imix_entry based on probability*/ 362 __u8 imix_distribution[IMIX_PRECISION]; 363 364 /* MPLS */ 365 unsigned int nr_labels; /* Depth of stack, 0 = no MPLS */ 366 __be32 labels[MAX_MPLS_LABELS]; 367 368 /* VLAN/SVLAN (802.1Q/Q-in-Q) */ 369 __u8 vlan_p; 370 __u8 vlan_cfi; 371 __u16 vlan_id; /* 0xffff means no vlan tag */ 372 373 __u8 svlan_p; 374 __u8 svlan_cfi; 375 __u16 svlan_id; /* 0xffff means no svlan tag */ 376 377 __u32 src_mac_count; /* How many MACs to iterate through */ 378 __u32 dst_mac_count; /* How many MACs to iterate through */ 379 380 unsigned char dst_mac[ETH_ALEN]; 381 unsigned char src_mac[ETH_ALEN]; 382 383 __u32 cur_dst_mac_offset; 384 __u32 cur_src_mac_offset; 385 __be32 cur_saddr; 386 __be32 cur_daddr; 387 __u16 ip_id; 388 __u16 cur_udp_dst; 389 __u16 cur_udp_src; 390 __u16 cur_queue_map; 391 __u32 cur_pkt_size; 392 __u32 last_pkt_size; 393 394 __u8 hh[14]; 395 /* = { 396 * 0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB, 397 * 398 * We fill in SRC address later 399 * 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 400 * 0x08, 0x00 401 * }; 402 */ 403 __u16 pad; /* pad out the hh struct to an even 16 bytes */ 404 405 struct sk_buff *skb; /* skb we are to transmit next, used for when we 406 * are transmitting the same one multiple times 407 */ 408 struct net_device *odev; /* The out-going device. 409 * Note that the device should have it's 410 * pg_info pointer pointing back to this 411 * device. 412 * Set when the user specifies the out-going 413 * device name (not when the inject is 414 * started as it used to do.) 415 */ 416 netdevice_tracker dev_tracker; 417 char odevname[32]; 418 struct flow_state *flows; 419 unsigned int cflows; /* Concurrent flows (config) */ 420 unsigned int lflow; /* Flow length (config) */ 421 unsigned int nflows; /* accumulated flows (stats) */ 422 unsigned int curfl; /* current sequenced flow (state)*/ 423 424 u16 queue_map_min; 425 u16 queue_map_max; 426 __u32 skb_priority; /* skb priority field */ 427 unsigned int burst; /* number of duplicated packets to burst */ 428 int node; /* Memory node */ 429 430 #ifdef CONFIG_XFRM 431 __u8 ipsmode; /* IPSEC mode (config) */ 432 __u8 ipsproto; /* IPSEC type (config) */ 433 __u32 spi; 434 struct xfrm_dst xdst; 435 struct dst_ops dstops; 436 #endif 437 char result[512]; 438 }; 439 440 struct pktgen_hdr { 441 __be32 pgh_magic; 442 __be32 seq_num; 443 __be32 tv_sec; 444 __be32 tv_usec; 445 }; 446 447 448 static unsigned int pg_net_id __read_mostly; 449 450 struct pktgen_net { 451 struct net *net; 452 struct proc_dir_entry *proc_dir; 453 struct list_head pktgen_threads; 454 bool pktgen_exiting; 455 }; 456 457 struct pktgen_thread { 458 struct mutex if_lock; /* for list of devices */ 459 struct list_head if_list; /* All device here */ 460 struct list_head th_list; 461 struct task_struct *tsk; 462 char result[512]; 463 464 /* Field for thread to receive "posted" events terminate, 465 * stop ifs etc. 466 */ 467 468 u32 control; 469 int cpu; 470 471 wait_queue_head_t queue; 472 struct completion start_done; 473 struct pktgen_net *net; 474 }; 475 476 #define REMOVE 1 477 #define FIND 0 478 479 static const char version[] = 480 "Packet Generator for packet performance testing. Version: " VERSION "\n"; 481 482 static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i); 483 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname); 484 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t, 485 const char *ifname, bool exact); 486 static int pktgen_device_event(struct notifier_block *, unsigned long, void *); 487 static void pktgen_run_all_threads(struct pktgen_net *pn); 488 static void pktgen_reset_all_threads(struct pktgen_net *pn); 489 static void pktgen_stop_all_threads(struct pktgen_net *pn); 490 491 static void pktgen_stop(struct pktgen_thread *t); 492 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev); 493 static void fill_imix_distribution(struct pktgen_dev *pkt_dev); 494 495 /* Module parameters, defaults. */ 496 static int pg_count_d __read_mostly = 1000; 497 static int pg_delay_d __read_mostly; 498 static int pg_clone_skb_d __read_mostly; 499 static int debug __read_mostly; 500 501 static DEFINE_MUTEX(pktgen_thread_lock); 502 503 static struct notifier_block pktgen_notifier_block = { 504 .notifier_call = pktgen_device_event, 505 }; 506 507 /* 508 * /proc handling functions 509 * 510 */ 511 512 static int pgctrl_show(struct seq_file *seq, void *v) 513 { 514 seq_puts(seq, version); 515 return 0; 516 } 517 518 static ssize_t pgctrl_write(struct file *file, const char __user *buf, 519 size_t count, loff_t *ppos) 520 { 521 char data[128]; 522 size_t max; 523 struct pktgen_net *pn = net_generic(current->nsproxy->net_ns, pg_net_id); 524 525 if (!capable(CAP_NET_ADMIN)) 526 return -EPERM; 527 528 if (count < 1) 529 return -EINVAL; 530 531 max = min(count, sizeof(data) - 1); 532 if (copy_from_user(data, buf, max)) 533 return -EFAULT; 534 535 if (data[max - 1] == '\n') 536 data[max - 1] = 0; /* strip trailing '\n', terminate string */ 537 else 538 data[max] = 0; /* terminate string */ 539 540 if (!strcmp(data, "stop")) 541 pktgen_stop_all_threads(pn); 542 else if (!strcmp(data, "start")) 543 pktgen_run_all_threads(pn); 544 else if (!strcmp(data, "reset")) 545 pktgen_reset_all_threads(pn); 546 else 547 return -EINVAL; 548 549 return count; 550 } 551 552 static int pgctrl_open(struct inode *inode, struct file *file) 553 { 554 return single_open(file, pgctrl_show, pde_data(inode)); 555 } 556 557 static const struct proc_ops pktgen_proc_ops = { 558 .proc_open = pgctrl_open, 559 .proc_read = seq_read, 560 .proc_lseek = seq_lseek, 561 .proc_write = pgctrl_write, 562 .proc_release = single_release, 563 }; 564 565 static int pktgen_if_show(struct seq_file *seq, void *v) 566 { 567 const struct pktgen_dev *pkt_dev = seq->private; 568 ktime_t stopped; 569 unsigned int i; 570 u64 idle; 571 572 seq_printf(seq, 573 "Params: count %llu min_pkt_size: %u max_pkt_size: %u\n", 574 (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size, 575 pkt_dev->max_pkt_size); 576 577 if (pkt_dev->n_imix_entries > 0) { 578 seq_puts(seq, " imix_weights: "); 579 for (i = 0; i < pkt_dev->n_imix_entries; i++) { 580 seq_printf(seq, "%llu,%llu ", 581 pkt_dev->imix_entries[i].size, 582 pkt_dev->imix_entries[i].weight); 583 } 584 seq_puts(seq, "\n"); 585 } 586 587 seq_printf(seq, 588 " frags: %d delay: %llu clone_skb: %d ifname: %s\n", 589 pkt_dev->nfrags, (unsigned long long) pkt_dev->delay, 590 pkt_dev->clone_skb, pkt_dev->odevname); 591 592 seq_printf(seq, " flows: %u flowlen: %u\n", pkt_dev->cflows, 593 pkt_dev->lflow); 594 595 seq_printf(seq, 596 " queue_map_min: %u queue_map_max: %u\n", 597 pkt_dev->queue_map_min, 598 pkt_dev->queue_map_max); 599 600 if (pkt_dev->skb_priority) 601 seq_printf(seq, " skb_priority: %u\n", 602 pkt_dev->skb_priority); 603 604 if (pkt_dev->flags & F_IPV6) { 605 seq_printf(seq, 606 " saddr: %pI6c min_saddr: %pI6c max_saddr: %pI6c\n" 607 " daddr: %pI6c min_daddr: %pI6c max_daddr: %pI6c\n", 608 &pkt_dev->in6_saddr, 609 &pkt_dev->min_in6_saddr, &pkt_dev->max_in6_saddr, 610 &pkt_dev->in6_daddr, 611 &pkt_dev->min_in6_daddr, &pkt_dev->max_in6_daddr); 612 } else { 613 seq_printf(seq, 614 " dst_min: %s dst_max: %s\n", 615 pkt_dev->dst_min, pkt_dev->dst_max); 616 seq_printf(seq, 617 " src_min: %s src_max: %s\n", 618 pkt_dev->src_min, pkt_dev->src_max); 619 } 620 621 seq_puts(seq, " src_mac: "); 622 623 seq_printf(seq, "%pM ", 624 is_zero_ether_addr(pkt_dev->src_mac) ? 625 pkt_dev->odev->dev_addr : pkt_dev->src_mac); 626 627 seq_puts(seq, "dst_mac: "); 628 seq_printf(seq, "%pM\n", pkt_dev->dst_mac); 629 630 seq_printf(seq, 631 " udp_src_min: %d udp_src_max: %d udp_dst_min: %d udp_dst_max: %d\n", 632 pkt_dev->udp_src_min, pkt_dev->udp_src_max, 633 pkt_dev->udp_dst_min, pkt_dev->udp_dst_max); 634 635 seq_printf(seq, 636 " src_mac_count: %d dst_mac_count: %d\n", 637 pkt_dev->src_mac_count, pkt_dev->dst_mac_count); 638 639 if (pkt_dev->nr_labels) { 640 seq_puts(seq, " mpls: "); 641 for (i = 0; i < pkt_dev->nr_labels; i++) 642 seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]), 643 i == pkt_dev->nr_labels-1 ? "\n" : ", "); 644 } 645 646 if (pkt_dev->vlan_id != 0xffff) 647 seq_printf(seq, " vlan_id: %u vlan_p: %u vlan_cfi: %u\n", 648 pkt_dev->vlan_id, pkt_dev->vlan_p, 649 pkt_dev->vlan_cfi); 650 651 if (pkt_dev->svlan_id != 0xffff) 652 seq_printf(seq, " svlan_id: %u vlan_p: %u vlan_cfi: %u\n", 653 pkt_dev->svlan_id, pkt_dev->svlan_p, 654 pkt_dev->svlan_cfi); 655 656 if (pkt_dev->tos) 657 seq_printf(seq, " tos: 0x%02x\n", pkt_dev->tos); 658 659 if (pkt_dev->traffic_class) 660 seq_printf(seq, " traffic_class: 0x%02x\n", pkt_dev->traffic_class); 661 662 if (pkt_dev->burst > 1) 663 seq_printf(seq, " burst: %d\n", pkt_dev->burst); 664 665 if (pkt_dev->node >= 0) 666 seq_printf(seq, " node: %d\n", pkt_dev->node); 667 668 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) 669 seq_puts(seq, " xmit_mode: netif_receive\n"); 670 else if (pkt_dev->xmit_mode == M_QUEUE_XMIT) 671 seq_puts(seq, " xmit_mode: xmit_queue\n"); 672 673 seq_puts(seq, " Flags: "); 674 675 for (i = 0; i < NR_PKT_FLAGS; i++) { 676 if (i == FLOW_SEQ_SHIFT) 677 if (!pkt_dev->cflows) 678 continue; 679 680 if (pkt_dev->flags & (1 << i)) { 681 seq_printf(seq, "%s ", pkt_flag_names[i]); 682 #ifdef CONFIG_XFRM 683 if (i == IPSEC_SHIFT && pkt_dev->spi) 684 seq_printf(seq, "spi:%u ", pkt_dev->spi); 685 #endif 686 } else if (i == FLOW_SEQ_SHIFT) { 687 seq_puts(seq, "FLOW_RND "); 688 } 689 } 690 691 seq_puts(seq, "\n"); 692 693 /* not really stopped, more like last-running-at */ 694 stopped = pkt_dev->running ? ktime_get() : pkt_dev->stopped_at; 695 idle = pkt_dev->idle_acc; 696 do_div(idle, NSEC_PER_USEC); 697 698 seq_printf(seq, 699 "Current:\n pkts-sofar: %llu errors: %llu\n", 700 (unsigned long long)pkt_dev->sofar, 701 (unsigned long long)pkt_dev->errors); 702 703 if (pkt_dev->n_imix_entries > 0) { 704 int i; 705 706 seq_puts(seq, " imix_size_counts: "); 707 for (i = 0; i < pkt_dev->n_imix_entries; i++) { 708 seq_printf(seq, "%llu,%llu ", 709 pkt_dev->imix_entries[i].size, 710 pkt_dev->imix_entries[i].count_so_far); 711 } 712 seq_puts(seq, "\n"); 713 } 714 715 seq_printf(seq, 716 " started: %lluus stopped: %lluus idle: %lluus\n", 717 (unsigned long long) ktime_to_us(pkt_dev->started_at), 718 (unsigned long long) ktime_to_us(stopped), 719 (unsigned long long) idle); 720 721 seq_printf(seq, 722 " seq_num: %d cur_dst_mac_offset: %d cur_src_mac_offset: %d\n", 723 pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset, 724 pkt_dev->cur_src_mac_offset); 725 726 if (pkt_dev->flags & F_IPV6) { 727 seq_printf(seq, " cur_saddr: %pI6c cur_daddr: %pI6c\n", 728 &pkt_dev->cur_in6_saddr, 729 &pkt_dev->cur_in6_daddr); 730 } else 731 seq_printf(seq, " cur_saddr: %pI4 cur_daddr: %pI4\n", 732 &pkt_dev->cur_saddr, &pkt_dev->cur_daddr); 733 734 seq_printf(seq, " cur_udp_dst: %d cur_udp_src: %d\n", 735 pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src); 736 737 seq_printf(seq, " cur_queue_map: %u\n", pkt_dev->cur_queue_map); 738 739 seq_printf(seq, " flows: %u\n", pkt_dev->nflows); 740 741 if (pkt_dev->result[0]) 742 seq_printf(seq, "Result: %s\n", pkt_dev->result); 743 else 744 seq_puts(seq, "Result: Idle\n"); 745 746 return 0; 747 } 748 749 750 static ssize_t hex32_arg(const char __user *user_buffer, size_t maxlen, 751 __u32 *num) 752 { 753 size_t i = 0; 754 755 *num = 0; 756 757 for (; i < maxlen; i++) { 758 int value; 759 char c; 760 761 if (get_user(c, &user_buffer[i])) 762 return -EFAULT; 763 value = hex_to_bin(c); 764 if (value >= 0) { 765 *num <<= 4; 766 *num |= value; 767 } else { 768 break; 769 } 770 } 771 return i; 772 } 773 774 static ssize_t count_trail_chars(const char __user *user_buffer, size_t maxlen) 775 { 776 size_t i; 777 778 for (i = 0; i < maxlen; i++) { 779 char c; 780 781 if (get_user(c, &user_buffer[i])) 782 return -EFAULT; 783 switch (c) { 784 case '\"': 785 case '\n': 786 case '\r': 787 case '\t': 788 case ' ': 789 case '=': 790 break; 791 default: 792 goto done; 793 } 794 } 795 done: 796 return i; 797 } 798 799 static ssize_t num_arg(const char __user *user_buffer, size_t maxlen, 800 unsigned long *num) 801 { 802 size_t i; 803 *num = 0; 804 805 for (i = 0; i < maxlen; i++) { 806 char c; 807 808 if (get_user(c, &user_buffer[i])) 809 return -EFAULT; 810 if ((c >= '0') && (c <= '9')) { 811 *num *= 10; 812 *num += c - '0'; 813 } else 814 break; 815 } 816 return i; 817 } 818 819 static ssize_t strn_len(const char __user *user_buffer, size_t maxlen) 820 { 821 size_t i; 822 823 for (i = 0; i < maxlen; i++) { 824 char c; 825 826 if (get_user(c, &user_buffer[i])) 827 return -EFAULT; 828 switch (c) { 829 case '\"': 830 case '\n': 831 case '\r': 832 case '\t': 833 case ' ': 834 case '=': 835 goto done_str; 836 default: 837 break; 838 } 839 } 840 done_str: 841 return i; 842 } 843 844 /* Parses imix entries from user buffer. 845 * The user buffer should consist of imix entries separated by spaces 846 * where each entry consists of size and weight delimited by commas. 847 * "size1,weight_1 size2,weight_2 ... size_n,weight_n" for example. 848 */ 849 static ssize_t get_imix_entries(const char __user *buffer, 850 size_t maxlen, 851 struct pktgen_dev *pkt_dev) 852 { 853 size_t i = 0, max; 854 ssize_t len; 855 char c; 856 857 pkt_dev->n_imix_entries = 0; 858 859 do { 860 unsigned long weight; 861 unsigned long size; 862 863 if (pkt_dev->n_imix_entries >= MAX_IMIX_ENTRIES) 864 return -E2BIG; 865 866 if (i >= maxlen) 867 return -EINVAL; 868 869 max = min(10, maxlen - i); 870 len = num_arg(&buffer[i], max, &size); 871 if (len < 0) 872 return len; 873 i += len; 874 if (i >= maxlen) 875 return -EINVAL; 876 if (get_user(c, &buffer[i])) 877 return -EFAULT; 878 /* Check for comma between size_i and weight_i */ 879 if (c != ',') 880 return -EINVAL; 881 i++; 882 if (i >= maxlen) 883 return -EINVAL; 884 885 if (size < 14 + 20 + 8) 886 size = 14 + 20 + 8; 887 888 max = min(10, maxlen - i); 889 len = num_arg(&buffer[i], max, &weight); 890 if (len < 0) 891 return len; 892 if (weight <= 0) 893 return -EINVAL; 894 895 pkt_dev->imix_entries[pkt_dev->n_imix_entries].size = size; 896 pkt_dev->imix_entries[pkt_dev->n_imix_entries].weight = weight; 897 898 i += len; 899 pkt_dev->n_imix_entries++; 900 901 if (i >= maxlen) 902 break; 903 if (get_user(c, &buffer[i])) 904 return -EFAULT; 905 i++; 906 } while (c == ' '); 907 908 return i; 909 } 910 911 static ssize_t get_labels(const char __user *buffer, 912 size_t maxlen, struct pktgen_dev *pkt_dev) 913 { 914 unsigned int n = 0; 915 size_t i = 0, max; 916 ssize_t len; 917 char c; 918 919 pkt_dev->nr_labels = 0; 920 do { 921 __u32 tmp; 922 923 if (n >= MAX_MPLS_LABELS) 924 return -E2BIG; 925 926 if (i >= maxlen) 927 return -EINVAL; 928 929 max = min(8, maxlen - i); 930 len = hex32_arg(&buffer[i], max, &tmp); 931 if (len < 0) 932 return len; 933 934 /* return empty list in case of invalid input or zero value */ 935 if (len == 0 || tmp == 0) 936 return maxlen; 937 938 pkt_dev->labels[n] = htonl(tmp); 939 if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM) 940 pkt_dev->flags |= F_MPLS_RND; 941 i += len; 942 n++; 943 if (i >= maxlen) 944 break; 945 if (get_user(c, &buffer[i])) 946 return -EFAULT; 947 i++; 948 } while (c == ','); 949 950 pkt_dev->nr_labels = n; 951 return i; 952 } 953 954 static __u32 pktgen_read_flag(const char *f, bool *disable) 955 { 956 __u32 i; 957 958 if (f[0] == '!') { 959 *disable = true; 960 f++; 961 } 962 963 for (i = 0; i < NR_PKT_FLAGS; i++) { 964 if (!IS_ENABLED(CONFIG_XFRM) && i == IPSEC_SHIFT) 965 continue; 966 967 /* allow only disabling ipv6 flag */ 968 if (!*disable && i == IPV6_SHIFT) 969 continue; 970 971 if (strcmp(f, pkt_flag_names[i]) == 0) 972 return 1 << i; 973 } 974 975 if (strcmp(f, "FLOW_RND") == 0) { 976 *disable = !*disable; 977 return F_FLOW_SEQ; 978 } 979 980 return 0; 981 } 982 983 static ssize_t pktgen_if_write(struct file *file, 984 const char __user *user_buffer, size_t count, 985 loff_t *offset) 986 { 987 struct seq_file *seq = file->private_data; 988 struct pktgen_dev *pkt_dev = seq->private; 989 size_t i, max; 990 ssize_t len; 991 char name[16], valstr[32]; 992 unsigned long value = 0; 993 char *pg_result = NULL; 994 char buf[128]; 995 996 pg_result = &(pkt_dev->result[0]); 997 998 if (count < 1) { 999 pr_warn("wrong command format\n"); 1000 return -EINVAL; 1001 } 1002 1003 max = count; 1004 len = count_trail_chars(user_buffer, max); 1005 if (len < 0) { 1006 pr_warn("illegal format\n"); 1007 return len; 1008 } 1009 i = len; 1010 1011 /* Read variable name */ 1012 max = min(sizeof(name) - 1, count - i); 1013 len = strn_len(&user_buffer[i], max); 1014 if (len < 0) 1015 return len; 1016 1017 memset(name, 0, sizeof(name)); 1018 if (copy_from_user(name, &user_buffer[i], len)) 1019 return -EFAULT; 1020 i += len; 1021 1022 max = count - i; 1023 len = count_trail_chars(&user_buffer[i], max); 1024 if (len < 0) 1025 return len; 1026 1027 i += len; 1028 1029 if (debug) { 1030 size_t copy = min_t(size_t, count + 1, 1024); 1031 char *tp = strndup_user(user_buffer, copy); 1032 1033 if (IS_ERR(tp)) 1034 return PTR_ERR(tp); 1035 1036 pr_debug("%s,%zu buffer -:%s:-\n", name, count, tp); 1037 kfree(tp); 1038 } 1039 1040 if (!strcmp(name, "min_pkt_size")) { 1041 max = min(10, count - i); 1042 len = num_arg(&user_buffer[i], max, &value); 1043 if (len < 0) 1044 return len; 1045 1046 if (value < 14 + 20 + 8) 1047 value = 14 + 20 + 8; 1048 if (value != pkt_dev->min_pkt_size) { 1049 pkt_dev->min_pkt_size = value; 1050 pkt_dev->cur_pkt_size = value; 1051 } 1052 sprintf(pg_result, "OK: min_pkt_size=%d", 1053 pkt_dev->min_pkt_size); 1054 return count; 1055 } 1056 1057 if (!strcmp(name, "max_pkt_size")) { 1058 max = min(10, count - i); 1059 len = num_arg(&user_buffer[i], max, &value); 1060 if (len < 0) 1061 return len; 1062 1063 if (value < 14 + 20 + 8) 1064 value = 14 + 20 + 8; 1065 if (value != pkt_dev->max_pkt_size) { 1066 pkt_dev->max_pkt_size = value; 1067 pkt_dev->cur_pkt_size = value; 1068 } 1069 sprintf(pg_result, "OK: max_pkt_size=%d", 1070 pkt_dev->max_pkt_size); 1071 return count; 1072 } 1073 1074 /* Shortcut for min = max */ 1075 1076 if (!strcmp(name, "pkt_size")) { 1077 max = min(10, count - i); 1078 len = num_arg(&user_buffer[i], max, &value); 1079 if (len < 0) 1080 return len; 1081 1082 if (value < 14 + 20 + 8) 1083 value = 14 + 20 + 8; 1084 if (value != pkt_dev->min_pkt_size) { 1085 pkt_dev->min_pkt_size = value; 1086 pkt_dev->max_pkt_size = value; 1087 pkt_dev->cur_pkt_size = value; 1088 } 1089 sprintf(pg_result, "OK: pkt_size=%d", pkt_dev->min_pkt_size); 1090 return count; 1091 } 1092 1093 if (!strcmp(name, "imix_weights")) { 1094 if (pkt_dev->clone_skb > 0) 1095 return -EINVAL; 1096 1097 max = count - i; 1098 len = get_imix_entries(&user_buffer[i], max, pkt_dev); 1099 if (len < 0) 1100 return len; 1101 1102 fill_imix_distribution(pkt_dev); 1103 1104 return count; 1105 } 1106 1107 if (!strcmp(name, "debug")) { 1108 max = min(10, count - i); 1109 len = num_arg(&user_buffer[i], max, &value); 1110 if (len < 0) 1111 return len; 1112 1113 debug = value; 1114 sprintf(pg_result, "OK: debug=%u", debug); 1115 return count; 1116 } 1117 1118 if (!strcmp(name, "frags")) { 1119 max = min(10, count - i); 1120 len = num_arg(&user_buffer[i], max, &value); 1121 if (len < 0) 1122 return len; 1123 1124 pkt_dev->nfrags = value; 1125 sprintf(pg_result, "OK: frags=%d", pkt_dev->nfrags); 1126 return count; 1127 } 1128 if (!strcmp(name, "delay")) { 1129 max = min(10, count - i); 1130 len = num_arg(&user_buffer[i], max, &value); 1131 if (len < 0) 1132 return len; 1133 1134 if (value == 0x7FFFFFFF) 1135 pkt_dev->delay = ULLONG_MAX; 1136 else 1137 pkt_dev->delay = (u64)value; 1138 1139 sprintf(pg_result, "OK: delay=%llu", 1140 (unsigned long long) pkt_dev->delay); 1141 return count; 1142 } 1143 if (!strcmp(name, "rate")) { 1144 max = min(10, count - i); 1145 len = num_arg(&user_buffer[i], max, &value); 1146 if (len < 0) 1147 return len; 1148 1149 if (!value) 1150 return -EINVAL; 1151 pkt_dev->delay = pkt_dev->min_pkt_size*8*NSEC_PER_USEC/value; 1152 if (debug) 1153 pr_info("Delay set at: %llu ns\n", pkt_dev->delay); 1154 1155 sprintf(pg_result, "OK: rate=%lu", value); 1156 return count; 1157 } 1158 if (!strcmp(name, "ratep")) { 1159 max = min(10, count - i); 1160 len = num_arg(&user_buffer[i], max, &value); 1161 if (len < 0) 1162 return len; 1163 1164 if (!value) 1165 return -EINVAL; 1166 pkt_dev->delay = NSEC_PER_SEC/value; 1167 if (debug) 1168 pr_info("Delay set at: %llu ns\n", pkt_dev->delay); 1169 1170 sprintf(pg_result, "OK: rate=%lu", value); 1171 return count; 1172 } 1173 if (!strcmp(name, "udp_src_min")) { 1174 max = min(10, count - i); 1175 len = num_arg(&user_buffer[i], max, &value); 1176 if (len < 0) 1177 return len; 1178 1179 if (value != pkt_dev->udp_src_min) { 1180 pkt_dev->udp_src_min = value; 1181 pkt_dev->cur_udp_src = value; 1182 } 1183 sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min); 1184 return count; 1185 } 1186 if (!strcmp(name, "udp_dst_min")) { 1187 max = min(10, count - i); 1188 len = num_arg(&user_buffer[i], max, &value); 1189 if (len < 0) 1190 return len; 1191 1192 if (value != pkt_dev->udp_dst_min) { 1193 pkt_dev->udp_dst_min = value; 1194 pkt_dev->cur_udp_dst = value; 1195 } 1196 sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min); 1197 return count; 1198 } 1199 if (!strcmp(name, "udp_src_max")) { 1200 max = min(10, count - i); 1201 len = num_arg(&user_buffer[i], max, &value); 1202 if (len < 0) 1203 return len; 1204 1205 if (value != pkt_dev->udp_src_max) { 1206 pkt_dev->udp_src_max = value; 1207 pkt_dev->cur_udp_src = value; 1208 } 1209 sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max); 1210 return count; 1211 } 1212 if (!strcmp(name, "udp_dst_max")) { 1213 max = min(10, count - i); 1214 len = num_arg(&user_buffer[i], max, &value); 1215 if (len < 0) 1216 return len; 1217 1218 if (value != pkt_dev->udp_dst_max) { 1219 pkt_dev->udp_dst_max = value; 1220 pkt_dev->cur_udp_dst = value; 1221 } 1222 sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max); 1223 return count; 1224 } 1225 if (!strcmp(name, "clone_skb")) { 1226 max = min(10, count - i); 1227 len = num_arg(&user_buffer[i], max, &value); 1228 if (len < 0) 1229 return len; 1230 /* clone_skb is not supported for netif_receive xmit_mode and 1231 * IMIX mode. 1232 */ 1233 if ((value > 0) && 1234 ((pkt_dev->xmit_mode == M_NETIF_RECEIVE) || 1235 !(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING))) 1236 return -EOPNOTSUPP; 1237 if (value > 0 && (pkt_dev->n_imix_entries > 0 || 1238 !(pkt_dev->flags & F_SHARED))) 1239 return -EINVAL; 1240 1241 pkt_dev->clone_skb = value; 1242 1243 sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb); 1244 return count; 1245 } 1246 if (!strcmp(name, "count")) { 1247 max = min(10, count - i); 1248 len = num_arg(&user_buffer[i], max, &value); 1249 if (len < 0) 1250 return len; 1251 1252 pkt_dev->count = value; 1253 sprintf(pg_result, "OK: count=%llu", 1254 (unsigned long long)pkt_dev->count); 1255 return count; 1256 } 1257 if (!strcmp(name, "src_mac_count")) { 1258 max = min(10, count - i); 1259 len = num_arg(&user_buffer[i], max, &value); 1260 if (len < 0) 1261 return len; 1262 1263 if (pkt_dev->src_mac_count != value) { 1264 pkt_dev->src_mac_count = value; 1265 pkt_dev->cur_src_mac_offset = 0; 1266 } 1267 sprintf(pg_result, "OK: src_mac_count=%d", 1268 pkt_dev->src_mac_count); 1269 return count; 1270 } 1271 if (!strcmp(name, "dst_mac_count")) { 1272 max = min(10, count - i); 1273 len = num_arg(&user_buffer[i], max, &value); 1274 if (len < 0) 1275 return len; 1276 1277 if (pkt_dev->dst_mac_count != value) { 1278 pkt_dev->dst_mac_count = value; 1279 pkt_dev->cur_dst_mac_offset = 0; 1280 } 1281 sprintf(pg_result, "OK: dst_mac_count=%d", 1282 pkt_dev->dst_mac_count); 1283 return count; 1284 } 1285 if (!strcmp(name, "burst")) { 1286 max = min(10, count - i); 1287 len = num_arg(&user_buffer[i], max, &value); 1288 if (len < 0) 1289 return len; 1290 1291 if ((value > 1) && 1292 ((pkt_dev->xmit_mode == M_QUEUE_XMIT) || 1293 ((pkt_dev->xmit_mode == M_START_XMIT) && 1294 (!(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING))))) 1295 return -EOPNOTSUPP; 1296 1297 if (value > 1 && !(pkt_dev->flags & F_SHARED)) 1298 return -EINVAL; 1299 1300 pkt_dev->burst = value < 1 ? 1 : value; 1301 sprintf(pg_result, "OK: burst=%u", pkt_dev->burst); 1302 return count; 1303 } 1304 if (!strcmp(name, "node")) { 1305 max = min(10, count - i); 1306 len = num_arg(&user_buffer[i], max, &value); 1307 if (len < 0) 1308 return len; 1309 1310 if (node_possible(value)) { 1311 pkt_dev->node = value; 1312 sprintf(pg_result, "OK: node=%d", pkt_dev->node); 1313 if (pkt_dev->page) { 1314 put_page(pkt_dev->page); 1315 pkt_dev->page = NULL; 1316 } 1317 } else { 1318 sprintf(pg_result, "ERROR: node not possible"); 1319 } 1320 return count; 1321 } 1322 if (!strcmp(name, "xmit_mode")) { 1323 char f[32]; 1324 1325 max = min(sizeof(f) - 1, count - i); 1326 len = strn_len(&user_buffer[i], max); 1327 if (len < 0) 1328 return len; 1329 1330 memset(f, 0, sizeof(f)); 1331 if (copy_from_user(f, &user_buffer[i], len)) 1332 return -EFAULT; 1333 1334 if (strcmp(f, "start_xmit") == 0) { 1335 pkt_dev->xmit_mode = M_START_XMIT; 1336 } else if (strcmp(f, "netif_receive") == 0) { 1337 /* clone_skb set earlier, not supported in this mode */ 1338 if (pkt_dev->clone_skb > 0) 1339 return -EOPNOTSUPP; 1340 1341 pkt_dev->xmit_mode = M_NETIF_RECEIVE; 1342 1343 /* make sure new packet is allocated every time 1344 * pktgen_xmit() is called 1345 */ 1346 pkt_dev->last_ok = 1; 1347 } else if (strcmp(f, "queue_xmit") == 0) { 1348 pkt_dev->xmit_mode = M_QUEUE_XMIT; 1349 pkt_dev->last_ok = 1; 1350 } else { 1351 sprintf(pg_result, 1352 "xmit_mode -:%s:- unknown\nAvailable modes: %s", 1353 f, "start_xmit, netif_receive\n"); 1354 return count; 1355 } 1356 sprintf(pg_result, "OK: xmit_mode=%s", f); 1357 return count; 1358 } 1359 if (!strcmp(name, "flag")) { 1360 bool disable = false; 1361 __u32 flag; 1362 char f[32]; 1363 char *end; 1364 1365 max = min(sizeof(f) - 1, count - i); 1366 len = strn_len(&user_buffer[i], max); 1367 if (len < 0) 1368 return len; 1369 1370 memset(f, 0, 32); 1371 if (copy_from_user(f, &user_buffer[i], len)) 1372 return -EFAULT; 1373 1374 flag = pktgen_read_flag(f, &disable); 1375 if (flag) { 1376 if (disable) { 1377 /* If "clone_skb", or "burst" parameters are 1378 * configured, it means that the skb still 1379 * needs to be referenced by the pktgen, so 1380 * the skb must be shared. 1381 */ 1382 if (flag == F_SHARED && (pkt_dev->clone_skb || 1383 pkt_dev->burst > 1)) 1384 return -EINVAL; 1385 pkt_dev->flags &= ~flag; 1386 } else { 1387 pkt_dev->flags |= flag; 1388 } 1389 1390 sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags); 1391 return count; 1392 } 1393 1394 /* Unknown flag */ 1395 end = pkt_dev->result + sizeof(pkt_dev->result); 1396 pg_result += sprintf(pg_result, 1397 "Flag -:%s:- unknown\n" 1398 "Available flags, (prepend ! to un-set flag):\n", f); 1399 1400 for (int n = 0; n < NR_PKT_FLAGS && pg_result < end; n++) { 1401 if (!IS_ENABLED(CONFIG_XFRM) && n == IPSEC_SHIFT) 1402 continue; 1403 pg_result += snprintf(pg_result, end - pg_result, 1404 "%s, ", pkt_flag_names[n]); 1405 } 1406 if (!WARN_ON_ONCE(pg_result >= end)) { 1407 /* Remove the comma and whitespace at the end */ 1408 *(pg_result - 2) = '\0'; 1409 } 1410 1411 return count; 1412 } 1413 if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) { 1414 max = min(sizeof(pkt_dev->dst_min) - 1, count - i); 1415 len = strn_len(&user_buffer[i], max); 1416 if (len < 0) 1417 return len; 1418 1419 if (copy_from_user(buf, &user_buffer[i], len)) 1420 return -EFAULT; 1421 buf[len] = 0; 1422 if (strcmp(buf, pkt_dev->dst_min) != 0) { 1423 strscpy_pad(pkt_dev->dst_min, buf); 1424 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min); 1425 pkt_dev->cur_daddr = pkt_dev->daddr_min; 1426 } 1427 if (debug) 1428 pr_debug("dst_min set to: %s\n", pkt_dev->dst_min); 1429 1430 sprintf(pg_result, "OK: dst_min=%s", pkt_dev->dst_min); 1431 return count; 1432 } 1433 if (!strcmp(name, "dst_max")) { 1434 max = min(sizeof(pkt_dev->dst_max) - 1, count - i); 1435 len = strn_len(&user_buffer[i], max); 1436 if (len < 0) 1437 return len; 1438 1439 if (copy_from_user(buf, &user_buffer[i], len)) 1440 return -EFAULT; 1441 buf[len] = 0; 1442 if (strcmp(buf, pkt_dev->dst_max) != 0) { 1443 strscpy_pad(pkt_dev->dst_max, buf); 1444 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max); 1445 pkt_dev->cur_daddr = pkt_dev->daddr_max; 1446 } 1447 if (debug) 1448 pr_debug("dst_max set to: %s\n", pkt_dev->dst_max); 1449 1450 sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max); 1451 return count; 1452 } 1453 if (!strcmp(name, "dst6")) { 1454 max = min(sizeof(buf) - 1, count - i); 1455 len = strn_len(&user_buffer[i], max); 1456 if (len < 0) 1457 return len; 1458 1459 pkt_dev->flags |= F_IPV6; 1460 1461 if (copy_from_user(buf, &user_buffer[i], len)) 1462 return -EFAULT; 1463 buf[len] = 0; 1464 1465 in6_pton(buf, -1, pkt_dev->in6_daddr.s6_addr, -1, NULL); 1466 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_daddr); 1467 1468 pkt_dev->cur_in6_daddr = pkt_dev->in6_daddr; 1469 1470 if (debug) 1471 pr_debug("dst6 set to: %s\n", buf); 1472 1473 sprintf(pg_result, "OK: dst6=%s", buf); 1474 return count; 1475 } 1476 if (!strcmp(name, "dst6_min")) { 1477 max = min(sizeof(buf) - 1, count - i); 1478 len = strn_len(&user_buffer[i], max); 1479 if (len < 0) 1480 return len; 1481 1482 pkt_dev->flags |= F_IPV6; 1483 1484 if (copy_from_user(buf, &user_buffer[i], len)) 1485 return -EFAULT; 1486 buf[len] = 0; 1487 1488 in6_pton(buf, -1, pkt_dev->min_in6_daddr.s6_addr, -1, NULL); 1489 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->min_in6_daddr); 1490 1491 pkt_dev->cur_in6_daddr = pkt_dev->min_in6_daddr; 1492 if (debug) 1493 pr_debug("dst6_min set to: %s\n", buf); 1494 1495 sprintf(pg_result, "OK: dst6_min=%s", buf); 1496 return count; 1497 } 1498 if (!strcmp(name, "dst6_max")) { 1499 max = min(sizeof(buf) - 1, count - i); 1500 len = strn_len(&user_buffer[i], max); 1501 if (len < 0) 1502 return len; 1503 1504 pkt_dev->flags |= F_IPV6; 1505 1506 if (copy_from_user(buf, &user_buffer[i], len)) 1507 return -EFAULT; 1508 buf[len] = 0; 1509 1510 in6_pton(buf, -1, pkt_dev->max_in6_daddr.s6_addr, -1, NULL); 1511 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->max_in6_daddr); 1512 1513 if (debug) 1514 pr_debug("dst6_max set to: %s\n", buf); 1515 1516 sprintf(pg_result, "OK: dst6_max=%s", buf); 1517 return count; 1518 } 1519 if (!strcmp(name, "src6")) { 1520 max = min(sizeof(buf) - 1, count - i); 1521 len = strn_len(&user_buffer[i], max); 1522 if (len < 0) 1523 return len; 1524 1525 pkt_dev->flags |= F_IPV6; 1526 1527 if (copy_from_user(buf, &user_buffer[i], len)) 1528 return -EFAULT; 1529 buf[len] = 0; 1530 1531 in6_pton(buf, -1, pkt_dev->in6_saddr.s6_addr, -1, NULL); 1532 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_saddr); 1533 1534 pkt_dev->cur_in6_saddr = pkt_dev->in6_saddr; 1535 1536 if (debug) 1537 pr_debug("src6 set to: %s\n", buf); 1538 1539 sprintf(pg_result, "OK: src6=%s", buf); 1540 return count; 1541 } 1542 if (!strcmp(name, "src_min")) { 1543 max = min(sizeof(pkt_dev->src_min) - 1, count - i); 1544 len = strn_len(&user_buffer[i], max); 1545 if (len < 0) 1546 return len; 1547 1548 if (copy_from_user(buf, &user_buffer[i], len)) 1549 return -EFAULT; 1550 buf[len] = 0; 1551 if (strcmp(buf, pkt_dev->src_min) != 0) { 1552 strscpy_pad(pkt_dev->src_min, buf); 1553 pkt_dev->saddr_min = in_aton(pkt_dev->src_min); 1554 pkt_dev->cur_saddr = pkt_dev->saddr_min; 1555 } 1556 if (debug) 1557 pr_debug("src_min set to: %s\n", pkt_dev->src_min); 1558 1559 sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min); 1560 return count; 1561 } 1562 if (!strcmp(name, "src_max")) { 1563 max = min(sizeof(pkt_dev->src_max) - 1, count - i); 1564 len = strn_len(&user_buffer[i], max); 1565 if (len < 0) 1566 return len; 1567 1568 if (copy_from_user(buf, &user_buffer[i], len)) 1569 return -EFAULT; 1570 buf[len] = 0; 1571 if (strcmp(buf, pkt_dev->src_max) != 0) { 1572 strscpy_pad(pkt_dev->src_max, buf); 1573 pkt_dev->saddr_max = in_aton(pkt_dev->src_max); 1574 pkt_dev->cur_saddr = pkt_dev->saddr_max; 1575 } 1576 if (debug) 1577 pr_debug("src_max set to: %s\n", pkt_dev->src_max); 1578 1579 sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max); 1580 return count; 1581 } 1582 if (!strcmp(name, "dst_mac")) { 1583 max = min(sizeof(valstr) - 1, count - i); 1584 len = strn_len(&user_buffer[i], max); 1585 if (len < 0) 1586 return len; 1587 1588 memset(valstr, 0, sizeof(valstr)); 1589 if (copy_from_user(valstr, &user_buffer[i], len)) 1590 return -EFAULT; 1591 1592 if (!mac_pton(valstr, pkt_dev->dst_mac)) 1593 return -EINVAL; 1594 /* Set up Dest MAC */ 1595 ether_addr_copy(&pkt_dev->hh[0], pkt_dev->dst_mac); 1596 1597 sprintf(pg_result, "OK: dstmac %pM", pkt_dev->dst_mac); 1598 return count; 1599 } 1600 if (!strcmp(name, "src_mac")) { 1601 max = min(sizeof(valstr) - 1, count - i); 1602 len = strn_len(&user_buffer[i], max); 1603 if (len < 0) 1604 return len; 1605 1606 memset(valstr, 0, sizeof(valstr)); 1607 if (copy_from_user(valstr, &user_buffer[i], len)) 1608 return -EFAULT; 1609 1610 if (!mac_pton(valstr, pkt_dev->src_mac)) 1611 return -EINVAL; 1612 /* Set up Src MAC */ 1613 ether_addr_copy(&pkt_dev->hh[6], pkt_dev->src_mac); 1614 1615 sprintf(pg_result, "OK: srcmac %pM", pkt_dev->src_mac); 1616 return count; 1617 } 1618 1619 if (!strcmp(name, "clear_counters")) { 1620 pktgen_clear_counters(pkt_dev); 1621 sprintf(pg_result, "OK: Clearing counters.\n"); 1622 return count; 1623 } 1624 1625 if (!strcmp(name, "flows")) { 1626 max = min(10, count - i); 1627 len = num_arg(&user_buffer[i], max, &value); 1628 if (len < 0) 1629 return len; 1630 1631 if (value > MAX_CFLOWS) 1632 value = MAX_CFLOWS; 1633 1634 pkt_dev->cflows = value; 1635 sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows); 1636 return count; 1637 } 1638 #ifdef CONFIG_XFRM 1639 if (!strcmp(name, "spi")) { 1640 max = min(10, count - i); 1641 len = num_arg(&user_buffer[i], max, &value); 1642 if (len < 0) 1643 return len; 1644 1645 pkt_dev->spi = value; 1646 sprintf(pg_result, "OK: spi=%u", pkt_dev->spi); 1647 return count; 1648 } 1649 #endif 1650 if (!strcmp(name, "flowlen")) { 1651 max = min(10, count - i); 1652 len = num_arg(&user_buffer[i], max, &value); 1653 if (len < 0) 1654 return len; 1655 1656 pkt_dev->lflow = value; 1657 sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow); 1658 return count; 1659 } 1660 1661 if (!strcmp(name, "queue_map_min")) { 1662 max = min(5, count - i); 1663 len = num_arg(&user_buffer[i], max, &value); 1664 if (len < 0) 1665 return len; 1666 1667 pkt_dev->queue_map_min = value; 1668 sprintf(pg_result, "OK: queue_map_min=%u", pkt_dev->queue_map_min); 1669 return count; 1670 } 1671 1672 if (!strcmp(name, "queue_map_max")) { 1673 max = min(5, count - i); 1674 len = num_arg(&user_buffer[i], max, &value); 1675 if (len < 0) 1676 return len; 1677 1678 pkt_dev->queue_map_max = value; 1679 sprintf(pg_result, "OK: queue_map_max=%u", pkt_dev->queue_map_max); 1680 return count; 1681 } 1682 1683 if (!strcmp(name, "mpls")) { 1684 unsigned int n, cnt; 1685 1686 max = count - i; 1687 len = get_labels(&user_buffer[i], max, pkt_dev); 1688 if (len < 0) 1689 return len; 1690 1691 cnt = sprintf(pg_result, "OK: mpls="); 1692 for (n = 0; n < pkt_dev->nr_labels; n++) 1693 cnt += sprintf(pg_result + cnt, 1694 "%08x%s", ntohl(pkt_dev->labels[n]), 1695 n == pkt_dev->nr_labels-1 ? "" : ","); 1696 1697 if (pkt_dev->nr_labels && pkt_dev->vlan_id != 0xffff) { 1698 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1699 pkt_dev->svlan_id = 0xffff; 1700 1701 if (debug) 1702 pr_debug("VLAN/SVLAN auto turned off\n"); 1703 } 1704 return count; 1705 } 1706 1707 if (!strcmp(name, "vlan_id")) { 1708 max = min(4, count - i); 1709 len = num_arg(&user_buffer[i], max, &value); 1710 if (len < 0) 1711 return len; 1712 1713 if (value <= 4095) { 1714 pkt_dev->vlan_id = value; /* turn on VLAN */ 1715 1716 if (debug) 1717 pr_debug("VLAN turned on\n"); 1718 1719 if (debug && pkt_dev->nr_labels) 1720 pr_debug("MPLS auto turned off\n"); 1721 1722 pkt_dev->nr_labels = 0; /* turn off MPLS */ 1723 sprintf(pg_result, "OK: vlan_id=%u", pkt_dev->vlan_id); 1724 } else { 1725 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1726 pkt_dev->svlan_id = 0xffff; 1727 1728 if (debug) 1729 pr_debug("VLAN/SVLAN turned off\n"); 1730 } 1731 return count; 1732 } 1733 1734 if (!strcmp(name, "vlan_p")) { 1735 max = min(1, count - i); 1736 len = num_arg(&user_buffer[i], max, &value); 1737 if (len < 0) 1738 return len; 1739 1740 if ((value <= 7) && (pkt_dev->vlan_id != 0xffff)) { 1741 pkt_dev->vlan_p = value; 1742 sprintf(pg_result, "OK: vlan_p=%u", pkt_dev->vlan_p); 1743 } else { 1744 sprintf(pg_result, "ERROR: vlan_p must be 0-7"); 1745 } 1746 return count; 1747 } 1748 1749 if (!strcmp(name, "vlan_cfi")) { 1750 max = min(1, count - i); 1751 len = num_arg(&user_buffer[i], max, &value); 1752 if (len < 0) 1753 return len; 1754 1755 if ((value <= 1) && (pkt_dev->vlan_id != 0xffff)) { 1756 pkt_dev->vlan_cfi = value; 1757 sprintf(pg_result, "OK: vlan_cfi=%u", pkt_dev->vlan_cfi); 1758 } else { 1759 sprintf(pg_result, "ERROR: vlan_cfi must be 0-1"); 1760 } 1761 return count; 1762 } 1763 1764 if (!strcmp(name, "svlan_id")) { 1765 max = min(4, count - i); 1766 len = num_arg(&user_buffer[i], max, &value); 1767 if (len < 0) 1768 return len; 1769 1770 if ((value <= 4095) && ((pkt_dev->vlan_id != 0xffff))) { 1771 pkt_dev->svlan_id = value; /* turn on SVLAN */ 1772 1773 if (debug) 1774 pr_debug("SVLAN turned on\n"); 1775 1776 if (debug && pkt_dev->nr_labels) 1777 pr_debug("MPLS auto turned off\n"); 1778 1779 pkt_dev->nr_labels = 0; /* turn off MPLS */ 1780 sprintf(pg_result, "OK: svlan_id=%u", pkt_dev->svlan_id); 1781 } else { 1782 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1783 pkt_dev->svlan_id = 0xffff; 1784 1785 if (debug) 1786 pr_debug("VLAN/SVLAN turned off\n"); 1787 } 1788 return count; 1789 } 1790 1791 if (!strcmp(name, "svlan_p")) { 1792 max = min(1, count - i); 1793 len = num_arg(&user_buffer[i], max, &value); 1794 if (len < 0) 1795 return len; 1796 1797 if ((value <= 7) && (pkt_dev->svlan_id != 0xffff)) { 1798 pkt_dev->svlan_p = value; 1799 sprintf(pg_result, "OK: svlan_p=%u", pkt_dev->svlan_p); 1800 } else { 1801 sprintf(pg_result, "ERROR: svlan_p must be 0-7"); 1802 } 1803 return count; 1804 } 1805 1806 if (!strcmp(name, "svlan_cfi")) { 1807 max = min(1, count - i); 1808 len = num_arg(&user_buffer[i], max, &value); 1809 if (len < 0) 1810 return len; 1811 1812 if ((value <= 1) && (pkt_dev->svlan_id != 0xffff)) { 1813 pkt_dev->svlan_cfi = value; 1814 sprintf(pg_result, "OK: svlan_cfi=%u", pkt_dev->svlan_cfi); 1815 } else { 1816 sprintf(pg_result, "ERROR: svlan_cfi must be 0-1"); 1817 } 1818 return count; 1819 } 1820 1821 if (!strcmp(name, "tos")) { 1822 __u32 tmp_value; 1823 1824 max = min(2, count - i); 1825 len = hex32_arg(&user_buffer[i], max, &tmp_value); 1826 if (len < 0) 1827 return len; 1828 1829 if (len == 2) { 1830 pkt_dev->tos = tmp_value; 1831 sprintf(pg_result, "OK: tos=0x%02x", pkt_dev->tos); 1832 } else { 1833 sprintf(pg_result, "ERROR: tos must be 00-ff"); 1834 } 1835 return count; 1836 } 1837 1838 if (!strcmp(name, "traffic_class")) { 1839 __u32 tmp_value; 1840 1841 max = min(2, count - i); 1842 len = hex32_arg(&user_buffer[i], max, &tmp_value); 1843 if (len < 0) 1844 return len; 1845 1846 if (len == 2) { 1847 pkt_dev->traffic_class = tmp_value; 1848 sprintf(pg_result, "OK: traffic_class=0x%02x", pkt_dev->traffic_class); 1849 } else { 1850 sprintf(pg_result, "ERROR: traffic_class must be 00-ff"); 1851 } 1852 return count; 1853 } 1854 1855 if (!strcmp(name, "skb_priority")) { 1856 max = min(9, count - i); 1857 len = num_arg(&user_buffer[i], max, &value); 1858 if (len < 0) 1859 return len; 1860 1861 pkt_dev->skb_priority = value; 1862 sprintf(pg_result, "OK: skb_priority=%i", 1863 pkt_dev->skb_priority); 1864 return count; 1865 } 1866 1867 sprintf(pkt_dev->result, "No such parameter \"%s\"", name); 1868 return -EINVAL; 1869 } 1870 1871 static int pktgen_if_open(struct inode *inode, struct file *file) 1872 { 1873 return single_open(file, pktgen_if_show, pde_data(inode)); 1874 } 1875 1876 static const struct proc_ops pktgen_if_proc_ops = { 1877 .proc_open = pktgen_if_open, 1878 .proc_read = seq_read, 1879 .proc_lseek = seq_lseek, 1880 .proc_write = pktgen_if_write, 1881 .proc_release = single_release, 1882 }; 1883 1884 static int pktgen_thread_show(struct seq_file *seq, void *v) 1885 { 1886 struct pktgen_thread *t = seq->private; 1887 const struct pktgen_dev *pkt_dev; 1888 1889 BUG_ON(!t); 1890 1891 seq_puts(seq, "Running: "); 1892 1893 rcu_read_lock(); 1894 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 1895 if (pkt_dev->running) 1896 seq_printf(seq, "%s ", pkt_dev->odevname); 1897 1898 seq_puts(seq, "\nStopped: "); 1899 1900 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 1901 if (!pkt_dev->running) 1902 seq_printf(seq, "%s ", pkt_dev->odevname); 1903 1904 if (t->result[0]) 1905 seq_printf(seq, "\nResult: %s\n", t->result); 1906 else 1907 seq_puts(seq, "\nResult: NA\n"); 1908 1909 rcu_read_unlock(); 1910 1911 return 0; 1912 } 1913 1914 static ssize_t pktgen_thread_write(struct file *file, 1915 const char __user *user_buffer, 1916 size_t count, loff_t *offset) 1917 { 1918 struct seq_file *seq = file->private_data; 1919 struct pktgen_thread *t = seq->private; 1920 size_t i, max; 1921 ssize_t len, ret; 1922 char name[40]; 1923 char *pg_result; 1924 1925 if (count < 1) { 1926 // sprintf(pg_result, "Wrong command format"); 1927 return -EINVAL; 1928 } 1929 1930 max = count; 1931 len = count_trail_chars(user_buffer, max); 1932 if (len < 0) 1933 return len; 1934 1935 i = len; 1936 1937 /* Read variable name */ 1938 max = min(sizeof(name) - 1, count - i); 1939 len = strn_len(&user_buffer[i], max); 1940 if (len < 0) 1941 return len; 1942 1943 memset(name, 0, sizeof(name)); 1944 if (copy_from_user(name, &user_buffer[i], len)) 1945 return -EFAULT; 1946 i += len; 1947 1948 max = count - i; 1949 len = count_trail_chars(&user_buffer[i], max); 1950 if (len < 0) 1951 return len; 1952 1953 i += len; 1954 1955 if (debug) 1956 pr_debug("t=%s, count=%lu\n", name, (unsigned long)count); 1957 1958 if (!t) { 1959 pr_err("ERROR: No thread\n"); 1960 ret = -EINVAL; 1961 goto out; 1962 } 1963 1964 pg_result = &(t->result[0]); 1965 1966 if (!strcmp(name, "add_device")) { 1967 char f[32]; 1968 1969 memset(f, 0, 32); 1970 max = min(sizeof(f) - 1, count - i); 1971 len = strn_len(&user_buffer[i], max); 1972 if (len < 0) { 1973 ret = len; 1974 goto out; 1975 } 1976 if (copy_from_user(f, &user_buffer[i], len)) 1977 return -EFAULT; 1978 1979 mutex_lock(&pktgen_thread_lock); 1980 ret = pktgen_add_device(t, f); 1981 mutex_unlock(&pktgen_thread_lock); 1982 if (!ret) { 1983 ret = count; 1984 sprintf(pg_result, "OK: add_device=%s", f); 1985 } else 1986 sprintf(pg_result, "ERROR: can not add device %s", f); 1987 goto out; 1988 } 1989 1990 if (!strcmp(name, "rem_device_all")) { 1991 mutex_lock(&pktgen_thread_lock); 1992 t->control |= T_REMDEVALL; 1993 mutex_unlock(&pktgen_thread_lock); 1994 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */ 1995 ret = count; 1996 sprintf(pg_result, "OK: rem_device_all"); 1997 goto out; 1998 } 1999 2000 if (!strcmp(name, "max_before_softirq")) { 2001 sprintf(pg_result, "OK: Note! max_before_softirq is obsoleted -- Do not use"); 2002 ret = count; 2003 goto out; 2004 } 2005 2006 ret = -EINVAL; 2007 out: 2008 return ret; 2009 } 2010 2011 static int pktgen_thread_open(struct inode *inode, struct file *file) 2012 { 2013 return single_open(file, pktgen_thread_show, pde_data(inode)); 2014 } 2015 2016 static const struct proc_ops pktgen_thread_proc_ops = { 2017 .proc_open = pktgen_thread_open, 2018 .proc_read = seq_read, 2019 .proc_lseek = seq_lseek, 2020 .proc_write = pktgen_thread_write, 2021 .proc_release = single_release, 2022 }; 2023 2024 /* Think find or remove for NN */ 2025 static struct pktgen_dev *__pktgen_NN_threads(const struct pktgen_net *pn, 2026 const char *ifname, int remove) 2027 { 2028 struct pktgen_thread *t; 2029 struct pktgen_dev *pkt_dev = NULL; 2030 bool exact = (remove == FIND); 2031 2032 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 2033 pkt_dev = pktgen_find_dev(t, ifname, exact); 2034 if (pkt_dev) { 2035 if (remove) { 2036 pkt_dev->removal_mark = 1; 2037 t->control |= T_REMDEV; 2038 } 2039 break; 2040 } 2041 } 2042 return pkt_dev; 2043 } 2044 2045 /* 2046 * mark a device for removal 2047 */ 2048 static void pktgen_mark_device(const struct pktgen_net *pn, const char *ifname) 2049 { 2050 struct pktgen_dev *pkt_dev = NULL; 2051 const int max_tries = 10, msec_per_try = 125; 2052 int i = 0; 2053 2054 mutex_lock(&pktgen_thread_lock); 2055 pr_debug("%s: marking %s for removal\n", __func__, ifname); 2056 2057 while (1) { 2058 2059 pkt_dev = __pktgen_NN_threads(pn, ifname, REMOVE); 2060 if (pkt_dev == NULL) 2061 break; /* success */ 2062 2063 mutex_unlock(&pktgen_thread_lock); 2064 pr_debug("%s: waiting for %s to disappear....\n", 2065 __func__, ifname); 2066 schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try)); 2067 mutex_lock(&pktgen_thread_lock); 2068 2069 if (++i >= max_tries) { 2070 pr_err("%s: timed out after waiting %d msec for device %s to be removed\n", 2071 __func__, msec_per_try * i, ifname); 2072 break; 2073 } 2074 2075 } 2076 2077 mutex_unlock(&pktgen_thread_lock); 2078 } 2079 2080 static void pktgen_change_name(const struct pktgen_net *pn, struct net_device *dev) 2081 { 2082 struct pktgen_thread *t; 2083 2084 mutex_lock(&pktgen_thread_lock); 2085 2086 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 2087 struct pktgen_dev *pkt_dev; 2088 2089 if_lock(t); 2090 list_for_each_entry(pkt_dev, &t->if_list, list) { 2091 if (pkt_dev->odev != dev) 2092 continue; 2093 2094 proc_remove(pkt_dev->entry); 2095 2096 pkt_dev->entry = proc_create_data(dev->name, 0600, 2097 pn->proc_dir, 2098 &pktgen_if_proc_ops, 2099 pkt_dev); 2100 if (!pkt_dev->entry) 2101 pr_err("can't move proc entry for '%s'\n", 2102 dev->name); 2103 break; 2104 } 2105 if_unlock(t); 2106 } 2107 mutex_unlock(&pktgen_thread_lock); 2108 } 2109 2110 static int pktgen_device_event(struct notifier_block *unused, 2111 unsigned long event, void *ptr) 2112 { 2113 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 2114 struct pktgen_net *pn = net_generic(dev_net(dev), pg_net_id); 2115 2116 if (pn->pktgen_exiting) 2117 return NOTIFY_DONE; 2118 2119 /* It is OK that we do not hold the group lock right now, 2120 * as we run under the RTNL lock. 2121 */ 2122 2123 switch (event) { 2124 case NETDEV_CHANGENAME: 2125 pktgen_change_name(pn, dev); 2126 break; 2127 2128 case NETDEV_UNREGISTER: 2129 pktgen_mark_device(pn, dev->name); 2130 break; 2131 } 2132 2133 return NOTIFY_DONE; 2134 } 2135 2136 static struct net_device *pktgen_dev_get_by_name(const struct pktgen_net *pn, 2137 struct pktgen_dev *pkt_dev, 2138 const char *ifname) 2139 { 2140 char b[IFNAMSIZ+5]; 2141 int i; 2142 2143 for (i = 0; ifname[i] != '@'; i++) { 2144 if (i == IFNAMSIZ) 2145 break; 2146 2147 b[i] = ifname[i]; 2148 } 2149 b[i] = 0; 2150 2151 return dev_get_by_name(pn->net, b); 2152 } 2153 2154 2155 /* Associate pktgen_dev with a device. */ 2156 2157 static int pktgen_setup_dev(const struct pktgen_net *pn, 2158 struct pktgen_dev *pkt_dev, const char *ifname) 2159 { 2160 struct net_device *odev; 2161 int err; 2162 2163 /* Clean old setups */ 2164 if (pkt_dev->odev) { 2165 netdev_put(pkt_dev->odev, &pkt_dev->dev_tracker); 2166 pkt_dev->odev = NULL; 2167 } 2168 2169 odev = pktgen_dev_get_by_name(pn, pkt_dev, ifname); 2170 if (!odev) { 2171 pr_err("no such netdevice: \"%s\"\n", ifname); 2172 return -ENODEV; 2173 } 2174 2175 if (odev->type != ARPHRD_ETHER && odev->type != ARPHRD_LOOPBACK) { 2176 pr_err("not an ethernet or loopback device: \"%s\"\n", ifname); 2177 err = -EINVAL; 2178 } else if (!netif_running(odev)) { 2179 pr_err("device is down: \"%s\"\n", ifname); 2180 err = -ENETDOWN; 2181 } else { 2182 pkt_dev->odev = odev; 2183 netdev_tracker_alloc(odev, &pkt_dev->dev_tracker, GFP_KERNEL); 2184 return 0; 2185 } 2186 2187 dev_put(odev); 2188 return err; 2189 } 2190 2191 /* Read pkt_dev from the interface and set up internal pktgen_dev 2192 * structure to have the right information to create/send packets 2193 */ 2194 static void pktgen_setup_inject(struct pktgen_dev *pkt_dev) 2195 { 2196 int ntxq; 2197 2198 if (!pkt_dev->odev) { 2199 pr_err("ERROR: pkt_dev->odev == NULL in setup_inject\n"); 2200 sprintf(pkt_dev->result, 2201 "ERROR: pkt_dev->odev == NULL in setup_inject.\n"); 2202 return; 2203 } 2204 2205 /* make sure that we don't pick a non-existing transmit queue */ 2206 ntxq = pkt_dev->odev->real_num_tx_queues; 2207 2208 if (ntxq <= pkt_dev->queue_map_min) { 2209 pr_warn("WARNING: Requested queue_map_min (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n", 2210 pkt_dev->queue_map_min, (ntxq ?: 1) - 1, ntxq, 2211 pkt_dev->odevname); 2212 pkt_dev->queue_map_min = (ntxq ?: 1) - 1; 2213 } 2214 if (pkt_dev->queue_map_max >= ntxq) { 2215 pr_warn("WARNING: Requested queue_map_max (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n", 2216 pkt_dev->queue_map_max, (ntxq ?: 1) - 1, ntxq, 2217 pkt_dev->odevname); 2218 pkt_dev->queue_map_max = (ntxq ?: 1) - 1; 2219 } 2220 2221 /* Default to the interface's mac if not explicitly set. */ 2222 2223 if (is_zero_ether_addr(pkt_dev->src_mac)) 2224 ether_addr_copy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr); 2225 2226 /* Set up Dest MAC */ 2227 ether_addr_copy(&(pkt_dev->hh[0]), pkt_dev->dst_mac); 2228 2229 if (pkt_dev->flags & F_IPV6) { 2230 int i, set = 0, err = 1; 2231 struct inet6_dev *idev; 2232 2233 if (pkt_dev->min_pkt_size == 0) { 2234 pkt_dev->min_pkt_size = 14 + sizeof(struct ipv6hdr) 2235 + sizeof(struct udphdr) 2236 + sizeof(struct pktgen_hdr) 2237 + pkt_dev->pkt_overhead; 2238 } 2239 2240 for (i = 0; i < sizeof(struct in6_addr); i++) 2241 if (pkt_dev->cur_in6_saddr.s6_addr[i]) { 2242 set = 1; 2243 break; 2244 } 2245 2246 if (!set) { 2247 2248 /* 2249 * Use linklevel address if unconfigured. 2250 * 2251 * use ipv6_get_lladdr if/when it's get exported 2252 */ 2253 2254 rcu_read_lock(); 2255 idev = __in6_dev_get(pkt_dev->odev); 2256 if (idev) { 2257 struct inet6_ifaddr *ifp; 2258 2259 read_lock_bh(&idev->lock); 2260 list_for_each_entry(ifp, &idev->addr_list, if_list) { 2261 if ((ifp->scope & IFA_LINK) && 2262 !(ifp->flags & IFA_F_TENTATIVE)) { 2263 pkt_dev->cur_in6_saddr = ifp->addr; 2264 err = 0; 2265 break; 2266 } 2267 } 2268 read_unlock_bh(&idev->lock); 2269 } 2270 rcu_read_unlock(); 2271 if (err) 2272 pr_err("ERROR: IPv6 link address not available\n"); 2273 } 2274 } else { 2275 if (pkt_dev->min_pkt_size == 0) { 2276 pkt_dev->min_pkt_size = 14 + sizeof(struct iphdr) 2277 + sizeof(struct udphdr) 2278 + sizeof(struct pktgen_hdr) 2279 + pkt_dev->pkt_overhead; 2280 } 2281 2282 pkt_dev->saddr_min = 0; 2283 pkt_dev->saddr_max = 0; 2284 if (strlen(pkt_dev->src_min) == 0) { 2285 2286 struct in_device *in_dev; 2287 2288 rcu_read_lock(); 2289 in_dev = __in_dev_get_rcu(pkt_dev->odev); 2290 if (in_dev) { 2291 const struct in_ifaddr *ifa; 2292 2293 ifa = rcu_dereference(in_dev->ifa_list); 2294 if (ifa) { 2295 pkt_dev->saddr_min = ifa->ifa_address; 2296 pkt_dev->saddr_max = pkt_dev->saddr_min; 2297 } 2298 } 2299 rcu_read_unlock(); 2300 } else { 2301 pkt_dev->saddr_min = in_aton(pkt_dev->src_min); 2302 pkt_dev->saddr_max = in_aton(pkt_dev->src_max); 2303 } 2304 2305 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min); 2306 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max); 2307 } 2308 /* Initialize current values. */ 2309 pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size; 2310 if (pkt_dev->min_pkt_size > pkt_dev->max_pkt_size) 2311 pkt_dev->max_pkt_size = pkt_dev->min_pkt_size; 2312 2313 pkt_dev->cur_dst_mac_offset = 0; 2314 pkt_dev->cur_src_mac_offset = 0; 2315 pkt_dev->cur_saddr = pkt_dev->saddr_min; 2316 pkt_dev->cur_daddr = pkt_dev->daddr_min; 2317 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min; 2318 pkt_dev->cur_udp_src = pkt_dev->udp_src_min; 2319 pkt_dev->nflows = 0; 2320 } 2321 2322 2323 static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until) 2324 { 2325 ktime_t start_time, end_time; 2326 s64 remaining; 2327 struct hrtimer_sleeper t; 2328 2329 hrtimer_setup_sleeper_on_stack(&t, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); 2330 hrtimer_set_expires(&t.timer, spin_until); 2331 2332 remaining = ktime_to_ns(hrtimer_expires_remaining(&t.timer)); 2333 if (remaining <= 0) 2334 goto out; 2335 2336 start_time = ktime_get(); 2337 if (remaining < 100000) { 2338 /* for small delays (<100us), just loop until limit is reached */ 2339 do { 2340 end_time = ktime_get(); 2341 } while (ktime_compare(end_time, spin_until) < 0); 2342 } else { 2343 do { 2344 set_current_state(TASK_INTERRUPTIBLE); 2345 hrtimer_sleeper_start_expires(&t, HRTIMER_MODE_ABS); 2346 2347 if (likely(t.task)) 2348 schedule(); 2349 2350 hrtimer_cancel(&t.timer); 2351 } while (t.task && pkt_dev->running && !signal_pending(current)); 2352 __set_current_state(TASK_RUNNING); 2353 end_time = ktime_get(); 2354 } 2355 2356 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(end_time, start_time)); 2357 out: 2358 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay); 2359 destroy_hrtimer_on_stack(&t.timer); 2360 } 2361 2362 static inline void set_pkt_overhead(struct pktgen_dev *pkt_dev) 2363 { 2364 pkt_dev->pkt_overhead = 0; 2365 pkt_dev->pkt_overhead += pkt_dev->nr_labels*sizeof(u32); 2366 pkt_dev->pkt_overhead += VLAN_TAG_SIZE(pkt_dev); 2367 pkt_dev->pkt_overhead += SVLAN_TAG_SIZE(pkt_dev); 2368 } 2369 2370 static inline int f_seen(const struct pktgen_dev *pkt_dev, int flow) 2371 { 2372 return !!(pkt_dev->flows[flow].flags & F_INIT); 2373 } 2374 2375 static inline int f_pick(struct pktgen_dev *pkt_dev) 2376 { 2377 int flow = pkt_dev->curfl; 2378 2379 if (pkt_dev->flags & F_FLOW_SEQ) { 2380 if (pkt_dev->flows[flow].count >= pkt_dev->lflow) { 2381 /* reset time */ 2382 pkt_dev->flows[flow].count = 0; 2383 pkt_dev->flows[flow].flags = 0; 2384 pkt_dev->curfl += 1; 2385 if (pkt_dev->curfl >= pkt_dev->cflows) 2386 pkt_dev->curfl = 0; /*reset */ 2387 } 2388 } else { 2389 flow = get_random_u32_below(pkt_dev->cflows); 2390 pkt_dev->curfl = flow; 2391 2392 if (pkt_dev->flows[flow].count > pkt_dev->lflow) { 2393 pkt_dev->flows[flow].count = 0; 2394 pkt_dev->flows[flow].flags = 0; 2395 } 2396 } 2397 2398 return pkt_dev->curfl; 2399 } 2400 2401 2402 /* If there was already an IPSEC SA, we keep it as is, else 2403 * we go look for it ... 2404 */ 2405 #define DUMMY_MARK 0 2406 static void get_ipsec_sa(struct pktgen_dev *pkt_dev, int flow) 2407 { 2408 #ifdef CONFIG_XFRM 2409 struct xfrm_state *x = pkt_dev->flows[flow].x; 2410 struct pktgen_net *pn = net_generic(dev_net(pkt_dev->odev), pg_net_id); 2411 2412 if (!x) { 2413 2414 if (pkt_dev->spi) { 2415 /* We need as quick as possible to find the right SA 2416 * Searching with minimum criteria to achieve, this. 2417 */ 2418 x = xfrm_state_lookup_byspi(pn->net, htonl(pkt_dev->spi), AF_INET); 2419 } else { 2420 /* slow path: we don't already have xfrm_state */ 2421 x = xfrm_stateonly_find(pn->net, DUMMY_MARK, 0, 2422 (xfrm_address_t *)&pkt_dev->cur_daddr, 2423 (xfrm_address_t *)&pkt_dev->cur_saddr, 2424 AF_INET, 2425 pkt_dev->ipsmode, 2426 pkt_dev->ipsproto, 0); 2427 } 2428 if (x) { 2429 pkt_dev->flows[flow].x = x; 2430 set_pkt_overhead(pkt_dev); 2431 pkt_dev->pkt_overhead += x->props.header_len; 2432 } 2433 2434 } 2435 #endif 2436 } 2437 static void set_cur_queue_map(struct pktgen_dev *pkt_dev) 2438 { 2439 if (pkt_dev->flags & F_QUEUE_MAP_CPU) 2440 pkt_dev->cur_queue_map = smp_processor_id(); 2441 2442 else if (pkt_dev->queue_map_min <= pkt_dev->queue_map_max) { 2443 __u16 t; 2444 2445 if (pkt_dev->flags & F_QUEUE_MAP_RND) { 2446 t = get_random_u32_inclusive(pkt_dev->queue_map_min, 2447 pkt_dev->queue_map_max); 2448 } else { 2449 t = pkt_dev->cur_queue_map + 1; 2450 if (t > pkt_dev->queue_map_max) 2451 t = pkt_dev->queue_map_min; 2452 } 2453 pkt_dev->cur_queue_map = t; 2454 } 2455 pkt_dev->cur_queue_map = pkt_dev->cur_queue_map % pkt_dev->odev->real_num_tx_queues; 2456 } 2457 2458 /* Increment/randomize headers according to flags and current values 2459 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst 2460 */ 2461 static void mod_cur_headers(struct pktgen_dev *pkt_dev) 2462 { 2463 __u32 imn; 2464 __u32 imx; 2465 int flow = 0; 2466 2467 if (pkt_dev->cflows) 2468 flow = f_pick(pkt_dev); 2469 2470 /* Deal with source MAC */ 2471 if (pkt_dev->src_mac_count > 1) { 2472 __u32 mc; 2473 __u32 tmp; 2474 2475 if (pkt_dev->flags & F_MACSRC_RND) 2476 mc = get_random_u32_below(pkt_dev->src_mac_count); 2477 else { 2478 mc = pkt_dev->cur_src_mac_offset++; 2479 if (pkt_dev->cur_src_mac_offset >= 2480 pkt_dev->src_mac_count) 2481 pkt_dev->cur_src_mac_offset = 0; 2482 } 2483 2484 tmp = pkt_dev->src_mac[5] + (mc & 0xFF); 2485 pkt_dev->hh[11] = tmp; 2486 tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8)); 2487 pkt_dev->hh[10] = tmp; 2488 tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8)); 2489 pkt_dev->hh[9] = tmp; 2490 tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8)); 2491 pkt_dev->hh[8] = tmp; 2492 tmp = (pkt_dev->src_mac[1] + (tmp >> 8)); 2493 pkt_dev->hh[7] = tmp; 2494 } 2495 2496 /* Deal with Destination MAC */ 2497 if (pkt_dev->dst_mac_count > 1) { 2498 __u32 mc; 2499 __u32 tmp; 2500 2501 if (pkt_dev->flags & F_MACDST_RND) 2502 mc = get_random_u32_below(pkt_dev->dst_mac_count); 2503 2504 else { 2505 mc = pkt_dev->cur_dst_mac_offset++; 2506 if (pkt_dev->cur_dst_mac_offset >= 2507 pkt_dev->dst_mac_count) { 2508 pkt_dev->cur_dst_mac_offset = 0; 2509 } 2510 } 2511 2512 tmp = pkt_dev->dst_mac[5] + (mc & 0xFF); 2513 pkt_dev->hh[5] = tmp; 2514 tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8)); 2515 pkt_dev->hh[4] = tmp; 2516 tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8)); 2517 pkt_dev->hh[3] = tmp; 2518 tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8)); 2519 pkt_dev->hh[2] = tmp; 2520 tmp = (pkt_dev->dst_mac[1] + (tmp >> 8)); 2521 pkt_dev->hh[1] = tmp; 2522 } 2523 2524 if (pkt_dev->flags & F_MPLS_RND) { 2525 unsigned int i; 2526 2527 for (i = 0; i < pkt_dev->nr_labels; i++) 2528 if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM) 2529 pkt_dev->labels[i] = MPLS_STACK_BOTTOM | 2530 ((__force __be32)get_random_u32() & 2531 htonl(0x000fffff)); 2532 } 2533 2534 if ((pkt_dev->flags & F_VID_RND) && (pkt_dev->vlan_id != 0xffff)) { 2535 pkt_dev->vlan_id = get_random_u32_below(4096); 2536 } 2537 2538 if ((pkt_dev->flags & F_SVID_RND) && (pkt_dev->svlan_id != 0xffff)) { 2539 pkt_dev->svlan_id = get_random_u32_below(4096); 2540 } 2541 2542 if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) { 2543 if (pkt_dev->flags & F_UDPSRC_RND) 2544 pkt_dev->cur_udp_src = get_random_u32_inclusive(pkt_dev->udp_src_min, 2545 pkt_dev->udp_src_max - 1); 2546 2547 else { 2548 pkt_dev->cur_udp_src++; 2549 if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max) 2550 pkt_dev->cur_udp_src = pkt_dev->udp_src_min; 2551 } 2552 } 2553 2554 if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) { 2555 if (pkt_dev->flags & F_UDPDST_RND) { 2556 pkt_dev->cur_udp_dst = get_random_u32_inclusive(pkt_dev->udp_dst_min, 2557 pkt_dev->udp_dst_max - 1); 2558 } else { 2559 pkt_dev->cur_udp_dst++; 2560 if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max) 2561 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min; 2562 } 2563 } 2564 2565 if (!(pkt_dev->flags & F_IPV6)) { 2566 2567 imn = ntohl(pkt_dev->saddr_min); 2568 imx = ntohl(pkt_dev->saddr_max); 2569 if (imn < imx) { 2570 __u32 t; 2571 2572 if (pkt_dev->flags & F_IPSRC_RND) 2573 t = get_random_u32_inclusive(imn, imx - 1); 2574 else { 2575 t = ntohl(pkt_dev->cur_saddr); 2576 t++; 2577 if (t > imx) 2578 t = imn; 2579 2580 } 2581 pkt_dev->cur_saddr = htonl(t); 2582 } 2583 2584 if (pkt_dev->cflows && f_seen(pkt_dev, flow)) { 2585 pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr; 2586 } else { 2587 imn = ntohl(pkt_dev->daddr_min); 2588 imx = ntohl(pkt_dev->daddr_max); 2589 if (imn < imx) { 2590 __u32 t; 2591 __be32 s; 2592 2593 if (pkt_dev->flags & F_IPDST_RND) { 2594 2595 do { 2596 t = get_random_u32_inclusive(imn, imx - 1); 2597 s = htonl(t); 2598 } while (ipv4_is_loopback(s) || 2599 ipv4_is_multicast(s) || 2600 ipv4_is_lbcast(s) || 2601 ipv4_is_zeronet(s) || 2602 ipv4_is_local_multicast(s)); 2603 pkt_dev->cur_daddr = s; 2604 } else { 2605 t = ntohl(pkt_dev->cur_daddr); 2606 t++; 2607 if (t > imx) { 2608 t = imn; 2609 } 2610 pkt_dev->cur_daddr = htonl(t); 2611 } 2612 } 2613 if (pkt_dev->cflows) { 2614 pkt_dev->flows[flow].flags |= F_INIT; 2615 pkt_dev->flows[flow].cur_daddr = 2616 pkt_dev->cur_daddr; 2617 if (pkt_dev->flags & F_IPSEC) 2618 get_ipsec_sa(pkt_dev, flow); 2619 pkt_dev->nflows++; 2620 } 2621 } 2622 } else { /* IPV6 * */ 2623 2624 if (!ipv6_addr_any(&pkt_dev->min_in6_daddr)) { 2625 int i; 2626 2627 /* Only random destinations yet */ 2628 2629 for (i = 0; i < 4; i++) { 2630 pkt_dev->cur_in6_daddr.s6_addr32[i] = 2631 (((__force __be32)get_random_u32() | 2632 pkt_dev->min_in6_daddr.s6_addr32[i]) & 2633 pkt_dev->max_in6_daddr.s6_addr32[i]); 2634 } 2635 } 2636 } 2637 2638 if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) { 2639 __u32 t; 2640 2641 if (pkt_dev->flags & F_TXSIZE_RND) { 2642 t = get_random_u32_inclusive(pkt_dev->min_pkt_size, 2643 pkt_dev->max_pkt_size - 1); 2644 } else { 2645 t = pkt_dev->cur_pkt_size + 1; 2646 if (t > pkt_dev->max_pkt_size) 2647 t = pkt_dev->min_pkt_size; 2648 } 2649 pkt_dev->cur_pkt_size = t; 2650 } else if (pkt_dev->n_imix_entries > 0) { 2651 struct imix_pkt *entry; 2652 __u32 t = get_random_u32_below(IMIX_PRECISION); 2653 __u8 entry_index = pkt_dev->imix_distribution[t]; 2654 2655 entry = &pkt_dev->imix_entries[entry_index]; 2656 entry->count_so_far++; 2657 pkt_dev->cur_pkt_size = entry->size; 2658 } 2659 2660 set_cur_queue_map(pkt_dev); 2661 2662 pkt_dev->flows[flow].count++; 2663 } 2664 2665 static void fill_imix_distribution(struct pktgen_dev *pkt_dev) 2666 { 2667 int cumulative_probabilites[MAX_IMIX_ENTRIES]; 2668 int j = 0; 2669 __u64 cumulative_prob = 0; 2670 __u64 total_weight = 0; 2671 int i = 0; 2672 2673 for (i = 0; i < pkt_dev->n_imix_entries; i++) 2674 total_weight += pkt_dev->imix_entries[i].weight; 2675 2676 /* Fill cumulative_probabilites with sum of normalized probabilities */ 2677 for (i = 0; i < pkt_dev->n_imix_entries - 1; i++) { 2678 cumulative_prob += div64_u64(pkt_dev->imix_entries[i].weight * 2679 IMIX_PRECISION, 2680 total_weight); 2681 cumulative_probabilites[i] = cumulative_prob; 2682 } 2683 cumulative_probabilites[pkt_dev->n_imix_entries - 1] = 100; 2684 2685 for (i = 0; i < IMIX_PRECISION; i++) { 2686 if (i == cumulative_probabilites[j]) 2687 j++; 2688 pkt_dev->imix_distribution[i] = j; 2689 } 2690 } 2691 2692 #ifdef CONFIG_XFRM 2693 static u32 pktgen_dst_metrics[RTAX_MAX + 1] = { 2694 2695 [RTAX_HOPLIMIT] = 0x5, /* Set a static hoplimit */ 2696 }; 2697 2698 static int pktgen_output_ipsec(struct sk_buff *skb, struct pktgen_dev *pkt_dev) 2699 { 2700 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x; 2701 int err = 0; 2702 struct net *net = dev_net(pkt_dev->odev); 2703 2704 if (!x) 2705 return 0; 2706 /* XXX: we dont support tunnel mode for now until 2707 * we resolve the dst issue 2708 */ 2709 if ((x->props.mode != XFRM_MODE_TRANSPORT) && (pkt_dev->spi == 0)) 2710 return 0; 2711 2712 /* But when user specify an valid SPI, transformation 2713 * supports both transport/tunnel mode + ESP/AH type. 2714 */ 2715 if ((x->props.mode == XFRM_MODE_TUNNEL) && (pkt_dev->spi != 0)) 2716 skb->_skb_refdst = (unsigned long)&pkt_dev->xdst.u.dst | SKB_DST_NOREF; 2717 2718 rcu_read_lock_bh(); 2719 err = pktgen_xfrm_outer_mode_output(x, skb); 2720 rcu_read_unlock_bh(); 2721 if (err) { 2722 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR); 2723 goto error; 2724 } 2725 err = x->type->output(x, skb); 2726 if (err) { 2727 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR); 2728 goto error; 2729 } 2730 spin_lock_bh(&x->lock); 2731 x->curlft.bytes += skb->len; 2732 x->curlft.packets++; 2733 spin_unlock_bh(&x->lock); 2734 error: 2735 return err; 2736 } 2737 2738 static void free_SAs(struct pktgen_dev *pkt_dev) 2739 { 2740 if (pkt_dev->cflows) { 2741 /* let go of the SAs if we have them */ 2742 int i; 2743 2744 for (i = 0; i < pkt_dev->cflows; i++) { 2745 struct xfrm_state *x = pkt_dev->flows[i].x; 2746 2747 if (x) { 2748 xfrm_state_put(x); 2749 pkt_dev->flows[i].x = NULL; 2750 } 2751 } 2752 } 2753 } 2754 2755 static int process_ipsec(struct pktgen_dev *pkt_dev, 2756 struct sk_buff *skb, __be16 protocol) 2757 { 2758 if (pkt_dev->flags & F_IPSEC) { 2759 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x; 2760 int nhead = 0; 2761 2762 if (x) { 2763 struct ethhdr *eth; 2764 struct iphdr *iph; 2765 int ret; 2766 2767 nhead = x->props.header_len - skb_headroom(skb); 2768 if (nhead > 0) { 2769 ret = pskb_expand_head(skb, nhead, 0, GFP_ATOMIC); 2770 if (ret < 0) { 2771 pr_err("Error expanding ipsec packet %d\n", 2772 ret); 2773 goto err; 2774 } 2775 } 2776 2777 /* ipsec is not expecting ll header */ 2778 skb_pull(skb, ETH_HLEN); 2779 ret = pktgen_output_ipsec(skb, pkt_dev); 2780 if (ret) { 2781 pr_err("Error creating ipsec packet %d\n", ret); 2782 goto err; 2783 } 2784 /* restore ll */ 2785 eth = skb_push(skb, ETH_HLEN); 2786 memcpy(eth, pkt_dev->hh, 2 * ETH_ALEN); 2787 eth->h_proto = protocol; 2788 2789 /* Update IPv4 header len as well as checksum value */ 2790 iph = ip_hdr(skb); 2791 iph->tot_len = htons(skb->len - ETH_HLEN); 2792 ip_send_check(iph); 2793 } 2794 } 2795 return 1; 2796 err: 2797 kfree_skb(skb); 2798 return 0; 2799 } 2800 #endif 2801 2802 static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev) 2803 { 2804 unsigned int i; 2805 2806 for (i = 0; i < pkt_dev->nr_labels; i++) 2807 *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM; 2808 2809 mpls--; 2810 *mpls |= MPLS_STACK_BOTTOM; 2811 } 2812 2813 static inline __be16 build_tci(unsigned int id, unsigned int cfi, 2814 unsigned int prio) 2815 { 2816 return htons(id | (cfi << 12) | (prio << 13)); 2817 } 2818 2819 static void pktgen_finalize_skb(struct pktgen_dev *pkt_dev, struct sk_buff *skb, 2820 int datalen) 2821 { 2822 struct timespec64 timestamp; 2823 struct pktgen_hdr *pgh; 2824 2825 pgh = skb_put(skb, sizeof(*pgh)); 2826 datalen -= sizeof(*pgh); 2827 2828 if (pkt_dev->nfrags <= 0) { 2829 skb_put_zero(skb, datalen); 2830 } else { 2831 int frags = pkt_dev->nfrags; 2832 int i, len; 2833 int frag_len; 2834 2835 2836 if (frags > MAX_SKB_FRAGS) 2837 frags = MAX_SKB_FRAGS; 2838 len = datalen - frags * PAGE_SIZE; 2839 if (len > 0) { 2840 skb_put_zero(skb, len); 2841 datalen = frags * PAGE_SIZE; 2842 } 2843 2844 i = 0; 2845 frag_len = min_t(int, datalen / frags, PAGE_SIZE); 2846 while (datalen > 0) { 2847 if (unlikely(!pkt_dev->page)) { 2848 int node = numa_node_id(); 2849 2850 if (pkt_dev->node >= 0 && (pkt_dev->flags & F_NODE)) 2851 node = pkt_dev->node; 2852 pkt_dev->page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0); 2853 if (!pkt_dev->page) 2854 break; 2855 } 2856 get_page(pkt_dev->page); 2857 2858 /*last fragment, fill rest of data*/ 2859 if (i == (frags - 1)) 2860 skb_frag_fill_page_desc(&skb_shinfo(skb)->frags[i], 2861 pkt_dev->page, 0, 2862 min(datalen, PAGE_SIZE)); 2863 else 2864 skb_frag_fill_page_desc(&skb_shinfo(skb)->frags[i], 2865 pkt_dev->page, 0, frag_len); 2866 2867 datalen -= skb_frag_size(&skb_shinfo(skb)->frags[i]); 2868 skb->len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2869 skb->data_len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2870 i++; 2871 skb_shinfo(skb)->nr_frags = i; 2872 } 2873 } 2874 2875 /* Stamp the time, and sequence number, 2876 * convert them to network byte order 2877 */ 2878 pgh->pgh_magic = htonl(PKTGEN_MAGIC); 2879 pgh->seq_num = htonl(pkt_dev->seq_num); 2880 2881 if (pkt_dev->flags & F_NO_TIMESTAMP) { 2882 pgh->tv_sec = 0; 2883 pgh->tv_usec = 0; 2884 } else { 2885 /* 2886 * pgh->tv_sec wraps in y2106 when interpreted as unsigned 2887 * as done by wireshark, or y2038 when interpreted as signed. 2888 * This is probably harmless, but if anyone wants to improve 2889 * it, we could introduce a variant that puts 64-bit nanoseconds 2890 * into the respective header bytes. 2891 * This would also be slightly faster to read. 2892 */ 2893 ktime_get_real_ts64(×tamp); 2894 pgh->tv_sec = htonl(timestamp.tv_sec); 2895 pgh->tv_usec = htonl(timestamp.tv_nsec / NSEC_PER_USEC); 2896 } 2897 } 2898 2899 static struct sk_buff *pktgen_alloc_skb(struct net_device *dev, 2900 struct pktgen_dev *pkt_dev) 2901 { 2902 unsigned int extralen = LL_RESERVED_SPACE(dev); 2903 struct sk_buff *skb = NULL; 2904 unsigned int size; 2905 2906 size = pkt_dev->cur_pkt_size + 64 + extralen + pkt_dev->pkt_overhead; 2907 if (pkt_dev->flags & F_NODE) { 2908 int node = pkt_dev->node >= 0 ? pkt_dev->node : numa_node_id(); 2909 2910 skb = __alloc_skb(NET_SKB_PAD + size, GFP_NOWAIT, 0, node); 2911 if (likely(skb)) { 2912 skb_reserve(skb, NET_SKB_PAD); 2913 skb->dev = dev; 2914 } 2915 } else { 2916 skb = __netdev_alloc_skb(dev, size, GFP_NOWAIT); 2917 } 2918 2919 /* the caller pre-fetches from skb->data and reserves for the mac hdr */ 2920 if (likely(skb)) 2921 skb_reserve(skb, extralen - 16); 2922 2923 return skb; 2924 } 2925 2926 static struct sk_buff *fill_packet_ipv4(struct net_device *odev, 2927 struct pktgen_dev *pkt_dev) 2928 { 2929 struct sk_buff *skb = NULL; 2930 __u8 *eth; 2931 struct udphdr *udph; 2932 int datalen, iplen; 2933 struct iphdr *iph; 2934 __be16 protocol = htons(ETH_P_IP); 2935 __be32 *mpls; 2936 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */ 2937 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */ 2938 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */ 2939 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */ 2940 u16 queue_map; 2941 2942 if (pkt_dev->nr_labels) 2943 protocol = htons(ETH_P_MPLS_UC); 2944 2945 if (pkt_dev->vlan_id != 0xffff) 2946 protocol = htons(ETH_P_8021Q); 2947 2948 /* Update any of the values, used when we're incrementing various 2949 * fields. 2950 */ 2951 mod_cur_headers(pkt_dev); 2952 queue_map = pkt_dev->cur_queue_map; 2953 2954 skb = pktgen_alloc_skb(odev, pkt_dev); 2955 if (!skb) { 2956 sprintf(pkt_dev->result, "No memory"); 2957 return NULL; 2958 } 2959 2960 prefetchw(skb->data); 2961 skb_reserve(skb, 16); 2962 2963 /* Reserve for ethernet and IP header */ 2964 eth = skb_push(skb, 14); 2965 mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32)); 2966 if (pkt_dev->nr_labels) 2967 mpls_push(mpls, pkt_dev); 2968 2969 if (pkt_dev->vlan_id != 0xffff) { 2970 if (pkt_dev->svlan_id != 0xffff) { 2971 svlan_tci = skb_put(skb, sizeof(__be16)); 2972 *svlan_tci = build_tci(pkt_dev->svlan_id, 2973 pkt_dev->svlan_cfi, 2974 pkt_dev->svlan_p); 2975 svlan_encapsulated_proto = skb_put(skb, 2976 sizeof(__be16)); 2977 *svlan_encapsulated_proto = htons(ETH_P_8021Q); 2978 } 2979 vlan_tci = skb_put(skb, sizeof(__be16)); 2980 *vlan_tci = build_tci(pkt_dev->vlan_id, 2981 pkt_dev->vlan_cfi, 2982 pkt_dev->vlan_p); 2983 vlan_encapsulated_proto = skb_put(skb, sizeof(__be16)); 2984 *vlan_encapsulated_proto = htons(ETH_P_IP); 2985 } 2986 2987 skb_reset_mac_header(skb); 2988 skb_set_network_header(skb, skb->len); 2989 iph = skb_put(skb, sizeof(struct iphdr)); 2990 2991 skb_set_transport_header(skb, skb->len); 2992 udph = skb_put(skb, sizeof(struct udphdr)); 2993 skb_set_queue_mapping(skb, queue_map); 2994 skb->priority = pkt_dev->skb_priority; 2995 2996 memcpy(eth, pkt_dev->hh, 12); 2997 *(__be16 *)ð[12] = protocol; 2998 2999 /* Eth + IPh + UDPh + mpls */ 3000 datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 - 3001 pkt_dev->pkt_overhead; 3002 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) 3003 datalen = sizeof(struct pktgen_hdr); 3004 3005 udph->source = htons(pkt_dev->cur_udp_src); 3006 udph->dest = htons(pkt_dev->cur_udp_dst); 3007 udph->len = htons(datalen + 8); /* DATA + udphdr */ 3008 udph->check = 0; 3009 3010 iph->ihl = 5; 3011 iph->version = 4; 3012 iph->ttl = 32; 3013 iph->tos = pkt_dev->tos; 3014 iph->protocol = IPPROTO_UDP; /* UDP */ 3015 iph->saddr = pkt_dev->cur_saddr; 3016 iph->daddr = pkt_dev->cur_daddr; 3017 iph->id = htons(pkt_dev->ip_id); 3018 pkt_dev->ip_id++; 3019 iph->frag_off = 0; 3020 iplen = 20 + 8 + datalen; 3021 iph->tot_len = htons(iplen); 3022 ip_send_check(iph); 3023 skb->protocol = protocol; 3024 skb->dev = odev; 3025 skb->pkt_type = PACKET_HOST; 3026 3027 pktgen_finalize_skb(pkt_dev, skb, datalen); 3028 3029 if (!(pkt_dev->flags & F_UDPCSUM)) { 3030 skb->ip_summed = CHECKSUM_NONE; 3031 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM)) { 3032 skb->ip_summed = CHECKSUM_PARTIAL; 3033 skb->csum = 0; 3034 udp4_hwcsum(skb, iph->saddr, iph->daddr); 3035 } else { 3036 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), datalen + 8, 0); 3037 3038 /* add protocol-dependent pseudo-header */ 3039 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr, 3040 datalen + 8, IPPROTO_UDP, csum); 3041 3042 if (udph->check == 0) 3043 udph->check = CSUM_MANGLED_0; 3044 } 3045 3046 #ifdef CONFIG_XFRM 3047 if (!process_ipsec(pkt_dev, skb, protocol)) 3048 return NULL; 3049 #endif 3050 3051 return skb; 3052 } 3053 3054 static struct sk_buff *fill_packet_ipv6(struct net_device *odev, 3055 struct pktgen_dev *pkt_dev) 3056 { 3057 struct sk_buff *skb = NULL; 3058 __u8 *eth; 3059 struct udphdr *udph; 3060 int datalen, udplen; 3061 struct ipv6hdr *iph; 3062 __be16 protocol = htons(ETH_P_IPV6); 3063 __be32 *mpls; 3064 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */ 3065 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */ 3066 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */ 3067 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */ 3068 u16 queue_map; 3069 3070 if (pkt_dev->nr_labels) 3071 protocol = htons(ETH_P_MPLS_UC); 3072 3073 if (pkt_dev->vlan_id != 0xffff) 3074 protocol = htons(ETH_P_8021Q); 3075 3076 /* Update any of the values, used when we're incrementing various 3077 * fields. 3078 */ 3079 mod_cur_headers(pkt_dev); 3080 queue_map = pkt_dev->cur_queue_map; 3081 3082 skb = pktgen_alloc_skb(odev, pkt_dev); 3083 if (!skb) { 3084 sprintf(pkt_dev->result, "No memory"); 3085 return NULL; 3086 } 3087 3088 prefetchw(skb->data); 3089 skb_reserve(skb, 16); 3090 3091 /* Reserve for ethernet and IP header */ 3092 eth = skb_push(skb, 14); 3093 mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32)); 3094 if (pkt_dev->nr_labels) 3095 mpls_push(mpls, pkt_dev); 3096 3097 if (pkt_dev->vlan_id != 0xffff) { 3098 if (pkt_dev->svlan_id != 0xffff) { 3099 svlan_tci = skb_put(skb, sizeof(__be16)); 3100 *svlan_tci = build_tci(pkt_dev->svlan_id, 3101 pkt_dev->svlan_cfi, 3102 pkt_dev->svlan_p); 3103 svlan_encapsulated_proto = skb_put(skb, 3104 sizeof(__be16)); 3105 *svlan_encapsulated_proto = htons(ETH_P_8021Q); 3106 } 3107 vlan_tci = skb_put(skb, sizeof(__be16)); 3108 *vlan_tci = build_tci(pkt_dev->vlan_id, 3109 pkt_dev->vlan_cfi, 3110 pkt_dev->vlan_p); 3111 vlan_encapsulated_proto = skb_put(skb, sizeof(__be16)); 3112 *vlan_encapsulated_proto = htons(ETH_P_IPV6); 3113 } 3114 3115 skb_reset_mac_header(skb); 3116 skb_set_network_header(skb, skb->len); 3117 iph = skb_put(skb, sizeof(struct ipv6hdr)); 3118 3119 skb_set_transport_header(skb, skb->len); 3120 udph = skb_put(skb, sizeof(struct udphdr)); 3121 skb_set_queue_mapping(skb, queue_map); 3122 skb->priority = pkt_dev->skb_priority; 3123 3124 memcpy(eth, pkt_dev->hh, 12); 3125 *(__be16 *) ð[12] = protocol; 3126 3127 /* Eth + IPh + UDPh + mpls */ 3128 datalen = pkt_dev->cur_pkt_size - 14 - 3129 sizeof(struct ipv6hdr) - sizeof(struct udphdr) - 3130 pkt_dev->pkt_overhead; 3131 3132 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) { 3133 datalen = sizeof(struct pktgen_hdr); 3134 net_info_ratelimited("increased datalen to %d\n", datalen); 3135 } 3136 3137 udplen = datalen + sizeof(struct udphdr); 3138 udph->source = htons(pkt_dev->cur_udp_src); 3139 udph->dest = htons(pkt_dev->cur_udp_dst); 3140 udph->len = htons(udplen); 3141 udph->check = 0; 3142 3143 *(__be32 *) iph = htonl(0x60000000); /* Version + flow */ 3144 3145 if (pkt_dev->traffic_class) { 3146 /* Version + traffic class + flow (0) */ 3147 *(__be32 *)iph |= htonl(0x60000000 | (pkt_dev->traffic_class << 20)); 3148 } 3149 3150 iph->hop_limit = 32; 3151 3152 iph->payload_len = htons(udplen); 3153 iph->nexthdr = IPPROTO_UDP; 3154 3155 iph->daddr = pkt_dev->cur_in6_daddr; 3156 iph->saddr = pkt_dev->cur_in6_saddr; 3157 3158 skb->protocol = protocol; 3159 skb->dev = odev; 3160 skb->pkt_type = PACKET_HOST; 3161 3162 pktgen_finalize_skb(pkt_dev, skb, datalen); 3163 3164 if (!(pkt_dev->flags & F_UDPCSUM)) { 3165 skb->ip_summed = CHECKSUM_NONE; 3166 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM)) { 3167 skb->ip_summed = CHECKSUM_PARTIAL; 3168 skb->csum_start = skb_transport_header(skb) - skb->head; 3169 skb->csum_offset = offsetof(struct udphdr, check); 3170 udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0); 3171 } else { 3172 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), udplen, 0); 3173 3174 /* add protocol-dependent pseudo-header */ 3175 udph->check = csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, csum); 3176 3177 if (udph->check == 0) 3178 udph->check = CSUM_MANGLED_0; 3179 } 3180 3181 return skb; 3182 } 3183 3184 static struct sk_buff *fill_packet(struct net_device *odev, 3185 struct pktgen_dev *pkt_dev) 3186 { 3187 if (pkt_dev->flags & F_IPV6) 3188 return fill_packet_ipv6(odev, pkt_dev); 3189 else 3190 return fill_packet_ipv4(odev, pkt_dev); 3191 } 3192 3193 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev) 3194 { 3195 pkt_dev->seq_num = 1; 3196 pkt_dev->idle_acc = 0; 3197 pkt_dev->sofar = 0; 3198 pkt_dev->tx_bytes = 0; 3199 pkt_dev->errors = 0; 3200 } 3201 3202 /* Set up structure for sending pkts, clear counters */ 3203 3204 static void pktgen_run(struct pktgen_thread *t) 3205 { 3206 struct pktgen_dev *pkt_dev; 3207 int started = 0; 3208 3209 func_enter(); 3210 3211 rcu_read_lock(); 3212 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3213 3214 /* 3215 * setup odev and create initial packet. 3216 */ 3217 pktgen_setup_inject(pkt_dev); 3218 3219 if (pkt_dev->odev) { 3220 pktgen_clear_counters(pkt_dev); 3221 pkt_dev->skb = NULL; 3222 pkt_dev->started_at = pkt_dev->next_tx = ktime_get(); 3223 3224 set_pkt_overhead(pkt_dev); 3225 3226 strscpy(pkt_dev->result, "Starting"); 3227 pkt_dev->running = 1; /* Cranke yeself! */ 3228 started++; 3229 } else 3230 strscpy(pkt_dev->result, "Error starting"); 3231 } 3232 rcu_read_unlock(); 3233 if (started) 3234 t->control &= ~(T_STOP); 3235 } 3236 3237 static void pktgen_handle_all_threads(struct pktgen_net *pn, u32 flags) 3238 { 3239 struct pktgen_thread *t; 3240 3241 mutex_lock(&pktgen_thread_lock); 3242 3243 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3244 t->control |= (flags); 3245 3246 mutex_unlock(&pktgen_thread_lock); 3247 } 3248 3249 static void pktgen_stop_all_threads(struct pktgen_net *pn) 3250 { 3251 func_enter(); 3252 3253 pktgen_handle_all_threads(pn, T_STOP); 3254 } 3255 3256 static int thread_is_running(const struct pktgen_thread *t) 3257 { 3258 const struct pktgen_dev *pkt_dev; 3259 3260 rcu_read_lock(); 3261 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 3262 if (pkt_dev->running) { 3263 rcu_read_unlock(); 3264 return 1; 3265 } 3266 rcu_read_unlock(); 3267 return 0; 3268 } 3269 3270 static int pktgen_wait_thread_run(struct pktgen_thread *t) 3271 { 3272 while (thread_is_running(t)) { 3273 3274 /* note: 't' will still be around even after the unlock/lock 3275 * cycle because pktgen_thread threads are only cleared at 3276 * net exit 3277 */ 3278 mutex_unlock(&pktgen_thread_lock); 3279 msleep_interruptible(100); 3280 mutex_lock(&pktgen_thread_lock); 3281 3282 if (signal_pending(current)) 3283 goto signal; 3284 } 3285 return 1; 3286 signal: 3287 return 0; 3288 } 3289 3290 static int pktgen_wait_all_threads_run(struct pktgen_net *pn) 3291 { 3292 struct pktgen_thread *t; 3293 int sig = 1; 3294 3295 /* prevent from racing with rmmod */ 3296 if (!try_module_get(THIS_MODULE)) 3297 return sig; 3298 3299 mutex_lock(&pktgen_thread_lock); 3300 3301 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 3302 sig = pktgen_wait_thread_run(t); 3303 if (sig == 0) 3304 break; 3305 } 3306 3307 if (sig == 0) 3308 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3309 t->control |= (T_STOP); 3310 3311 mutex_unlock(&pktgen_thread_lock); 3312 module_put(THIS_MODULE); 3313 return sig; 3314 } 3315 3316 static void pktgen_run_all_threads(struct pktgen_net *pn) 3317 { 3318 func_enter(); 3319 3320 pktgen_handle_all_threads(pn, T_RUN); 3321 3322 /* Propagate thread->control */ 3323 schedule_timeout_interruptible(msecs_to_jiffies(125)); 3324 3325 pktgen_wait_all_threads_run(pn); 3326 } 3327 3328 static void pktgen_reset_all_threads(struct pktgen_net *pn) 3329 { 3330 func_enter(); 3331 3332 pktgen_handle_all_threads(pn, T_REMDEVALL); 3333 3334 /* Propagate thread->control */ 3335 schedule_timeout_interruptible(msecs_to_jiffies(125)); 3336 3337 pktgen_wait_all_threads_run(pn); 3338 } 3339 3340 static void show_results(struct pktgen_dev *pkt_dev, int nr_frags) 3341 { 3342 __u64 bps, mbps, pps; 3343 char *p = pkt_dev->result; 3344 ktime_t elapsed = ktime_sub(pkt_dev->stopped_at, 3345 pkt_dev->started_at); 3346 ktime_t idle = ns_to_ktime(pkt_dev->idle_acc); 3347 3348 p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n", 3349 (unsigned long long)ktime_to_us(elapsed), 3350 (unsigned long long)ktime_to_us(ktime_sub(elapsed, idle)), 3351 (unsigned long long)ktime_to_us(idle), 3352 (unsigned long long)pkt_dev->sofar, 3353 pkt_dev->cur_pkt_size, nr_frags); 3354 3355 pps = div64_u64(pkt_dev->sofar * NSEC_PER_SEC, 3356 ktime_to_ns(elapsed)); 3357 3358 if (pkt_dev->n_imix_entries > 0) { 3359 int i; 3360 struct imix_pkt *entry; 3361 3362 bps = 0; 3363 for (i = 0; i < pkt_dev->n_imix_entries; i++) { 3364 entry = &pkt_dev->imix_entries[i]; 3365 bps += entry->size * entry->count_so_far; 3366 } 3367 bps = div64_u64(bps * 8 * NSEC_PER_SEC, ktime_to_ns(elapsed)); 3368 } else { 3369 bps = pps * 8 * pkt_dev->cur_pkt_size; 3370 } 3371 3372 mbps = bps; 3373 do_div(mbps, 1000000); 3374 p += sprintf(p, " %llupps %lluMb/sec (%llubps) errors: %llu", 3375 (unsigned long long)pps, 3376 (unsigned long long)mbps, 3377 (unsigned long long)bps, 3378 (unsigned long long)pkt_dev->errors); 3379 } 3380 3381 /* Set stopped-at timer, remove from running list, do counters & statistics */ 3382 static int pktgen_stop_device(struct pktgen_dev *pkt_dev) 3383 { 3384 int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1; 3385 3386 if (!pkt_dev->running) { 3387 pr_warn("interface: %s is already stopped\n", 3388 pkt_dev->odevname); 3389 return -EINVAL; 3390 } 3391 3392 pkt_dev->running = 0; 3393 kfree_skb(pkt_dev->skb); 3394 pkt_dev->skb = NULL; 3395 pkt_dev->stopped_at = ktime_get(); 3396 3397 show_results(pkt_dev, nr_frags); 3398 3399 return 0; 3400 } 3401 3402 static struct pktgen_dev *next_to_run(struct pktgen_thread *t) 3403 { 3404 struct pktgen_dev *pkt_dev, *best = NULL; 3405 3406 rcu_read_lock(); 3407 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3408 if (!pkt_dev->running) 3409 continue; 3410 if (best == NULL) 3411 best = pkt_dev; 3412 else if (ktime_compare(pkt_dev->next_tx, best->next_tx) < 0) 3413 best = pkt_dev; 3414 } 3415 rcu_read_unlock(); 3416 3417 return best; 3418 } 3419 3420 static void pktgen_stop(struct pktgen_thread *t) 3421 { 3422 struct pktgen_dev *pkt_dev; 3423 3424 func_enter(); 3425 3426 rcu_read_lock(); 3427 3428 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3429 pktgen_stop_device(pkt_dev); 3430 } 3431 3432 rcu_read_unlock(); 3433 } 3434 3435 /* 3436 * one of our devices needs to be removed - find it 3437 * and remove it 3438 */ 3439 static void pktgen_rem_one_if(struct pktgen_thread *t) 3440 { 3441 struct list_head *q, *n; 3442 struct pktgen_dev *cur; 3443 3444 func_enter(); 3445 3446 list_for_each_safe(q, n, &t->if_list) { 3447 cur = list_entry(q, struct pktgen_dev, list); 3448 3449 if (!cur->removal_mark) 3450 continue; 3451 3452 kfree_skb(cur->skb); 3453 cur->skb = NULL; 3454 3455 pktgen_remove_device(t, cur); 3456 3457 break; 3458 } 3459 } 3460 3461 static void pktgen_rem_all_ifs(struct pktgen_thread *t) 3462 { 3463 struct list_head *q, *n; 3464 struct pktgen_dev *cur; 3465 3466 func_enter(); 3467 3468 /* Remove all devices, free mem */ 3469 3470 list_for_each_safe(q, n, &t->if_list) { 3471 cur = list_entry(q, struct pktgen_dev, list); 3472 3473 kfree_skb(cur->skb); 3474 cur->skb = NULL; 3475 3476 pktgen_remove_device(t, cur); 3477 } 3478 } 3479 3480 static void pktgen_rem_thread(struct pktgen_thread *t) 3481 { 3482 /* Remove from the thread list */ 3483 remove_proc_entry(t->tsk->comm, t->net->proc_dir); 3484 } 3485 3486 static void pktgen_resched(struct pktgen_dev *pkt_dev) 3487 { 3488 ktime_t idle_start = ktime_get(); 3489 3490 schedule(); 3491 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start)); 3492 } 3493 3494 static void pktgen_wait_for_skb(struct pktgen_dev *pkt_dev) 3495 { 3496 ktime_t idle_start = ktime_get(); 3497 3498 while (refcount_read(&(pkt_dev->skb->users)) != 1) { 3499 if (signal_pending(current)) 3500 break; 3501 3502 if (need_resched()) 3503 pktgen_resched(pkt_dev); 3504 else 3505 cpu_relax(); 3506 } 3507 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start)); 3508 } 3509 3510 static void pktgen_xmit(struct pktgen_dev *pkt_dev) 3511 { 3512 bool skb_shared = !!(READ_ONCE(pkt_dev->flags) & F_SHARED); 3513 struct net_device *odev = pkt_dev->odev; 3514 struct netdev_queue *txq; 3515 unsigned int burst = 1; 3516 struct sk_buff *skb; 3517 int clone_skb = 0; 3518 int ret; 3519 3520 /* If 'skb_shared' is false, the read of possible 3521 * new values (if any) for 'burst' and 'clone_skb' will be skipped to 3522 * prevent some concurrent changes from slipping in. And the stabilized 3523 * config will be read in during the next run of pktgen_xmit. 3524 */ 3525 if (skb_shared) { 3526 burst = READ_ONCE(pkt_dev->burst); 3527 clone_skb = READ_ONCE(pkt_dev->clone_skb); 3528 } 3529 3530 /* If device is offline, then don't send */ 3531 if (unlikely(!netif_running(odev) || !netif_carrier_ok(odev))) { 3532 pktgen_stop_device(pkt_dev); 3533 return; 3534 } 3535 3536 /* This is max DELAY, this has special meaning of 3537 * "never transmit" 3538 */ 3539 if (unlikely(pkt_dev->delay == ULLONG_MAX)) { 3540 pkt_dev->next_tx = ktime_add_ns(ktime_get(), ULONG_MAX); 3541 return; 3542 } 3543 3544 /* If no skb or clone count exhausted then get new one */ 3545 if (!pkt_dev->skb || (pkt_dev->last_ok && 3546 ++pkt_dev->clone_count >= clone_skb)) { 3547 /* build a new pkt */ 3548 kfree_skb(pkt_dev->skb); 3549 3550 pkt_dev->skb = fill_packet(odev, pkt_dev); 3551 if (pkt_dev->skb == NULL) { 3552 pr_err("ERROR: couldn't allocate skb in fill_packet\n"); 3553 schedule(); 3554 pkt_dev->clone_count--; /* back out increment, OOM */ 3555 return; 3556 } 3557 pkt_dev->last_pkt_size = pkt_dev->skb->len; 3558 pkt_dev->clone_count = 0; /* reset counter */ 3559 } 3560 3561 if (pkt_dev->delay && pkt_dev->last_ok) 3562 spin(pkt_dev, pkt_dev->next_tx); 3563 3564 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) { 3565 skb = pkt_dev->skb; 3566 skb->protocol = eth_type_trans(skb, skb->dev); 3567 if (skb_shared) 3568 refcount_add(burst, &skb->users); 3569 local_bh_disable(); 3570 do { 3571 ret = netif_receive_skb(skb); 3572 if (ret == NET_RX_DROP) 3573 pkt_dev->errors++; 3574 pkt_dev->sofar++; 3575 pkt_dev->seq_num++; 3576 if (unlikely(!skb_shared)) { 3577 pkt_dev->skb = NULL; 3578 break; 3579 } 3580 if (refcount_read(&skb->users) != burst) { 3581 /* skb was queued by rps/rfs or taps, 3582 * so cannot reuse this skb 3583 */ 3584 WARN_ON(refcount_sub_and_test(burst - 1, &skb->users)); 3585 /* get out of the loop and wait 3586 * until skb is consumed 3587 */ 3588 break; 3589 } 3590 /* skb was 'freed' by stack, so clean few 3591 * bits and reuse it 3592 */ 3593 skb_reset_redirect(skb); 3594 } while (--burst > 0); 3595 goto out; /* Skips xmit_mode M_START_XMIT */ 3596 } else if (pkt_dev->xmit_mode == M_QUEUE_XMIT) { 3597 local_bh_disable(); 3598 if (skb_shared) 3599 refcount_inc(&pkt_dev->skb->users); 3600 3601 ret = dev_queue_xmit(pkt_dev->skb); 3602 3603 if (!skb_shared && dev_xmit_complete(ret)) 3604 pkt_dev->skb = NULL; 3605 3606 switch (ret) { 3607 case NET_XMIT_SUCCESS: 3608 pkt_dev->sofar++; 3609 pkt_dev->seq_num++; 3610 pkt_dev->tx_bytes += pkt_dev->last_pkt_size; 3611 break; 3612 case NET_XMIT_DROP: 3613 case NET_XMIT_CN: 3614 /* These are all valid return codes for a qdisc but 3615 * indicate packets are being dropped or will likely 3616 * be dropped soon. 3617 */ 3618 case NETDEV_TX_BUSY: 3619 /* qdisc may call dev_hard_start_xmit directly in cases 3620 * where no queues exist e.g. loopback device, virtual 3621 * devices, etc. In this case we need to handle 3622 * NETDEV_TX_ codes. 3623 */ 3624 default: 3625 pkt_dev->errors++; 3626 net_info_ratelimited("%s xmit error: %d\n", 3627 pkt_dev->odevname, ret); 3628 break; 3629 } 3630 goto out; 3631 } 3632 3633 txq = skb_get_tx_queue(odev, pkt_dev->skb); 3634 3635 local_bh_disable(); 3636 3637 HARD_TX_LOCK(odev, txq, smp_processor_id()); 3638 3639 if (unlikely(netif_xmit_frozen_or_drv_stopped(txq))) { 3640 pkt_dev->last_ok = 0; 3641 goto unlock; 3642 } 3643 if (skb_shared) 3644 refcount_add(burst, &pkt_dev->skb->users); 3645 3646 xmit_more: 3647 ret = netdev_start_xmit(pkt_dev->skb, odev, txq, --burst > 0); 3648 3649 if (!skb_shared && dev_xmit_complete(ret)) 3650 pkt_dev->skb = NULL; 3651 3652 switch (ret) { 3653 case NETDEV_TX_OK: 3654 pkt_dev->last_ok = 1; 3655 pkt_dev->sofar++; 3656 pkt_dev->seq_num++; 3657 pkt_dev->tx_bytes += pkt_dev->last_pkt_size; 3658 if (burst > 0 && !netif_xmit_frozen_or_drv_stopped(txq)) 3659 goto xmit_more; 3660 break; 3661 case NET_XMIT_DROP: 3662 case NET_XMIT_CN: 3663 /* skb has been consumed */ 3664 pkt_dev->errors++; 3665 break; 3666 default: /* Drivers are not supposed to return other values! */ 3667 net_info_ratelimited("%s xmit error: %d\n", 3668 pkt_dev->odevname, ret); 3669 pkt_dev->errors++; 3670 fallthrough; 3671 case NETDEV_TX_BUSY: 3672 /* Retry it next time */ 3673 if (skb_shared) 3674 refcount_dec(&pkt_dev->skb->users); 3675 pkt_dev->last_ok = 0; 3676 } 3677 if (unlikely(burst)) 3678 WARN_ON(refcount_sub_and_test(burst, &pkt_dev->skb->users)); 3679 unlock: 3680 HARD_TX_UNLOCK(odev, txq); 3681 3682 out: 3683 local_bh_enable(); 3684 3685 /* If pkt_dev->count is zero, then run forever */ 3686 if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) { 3687 if (pkt_dev->skb) 3688 pktgen_wait_for_skb(pkt_dev); 3689 3690 /* Done with this */ 3691 pktgen_stop_device(pkt_dev); 3692 } 3693 } 3694 3695 /* 3696 * Main loop of the thread goes here 3697 */ 3698 3699 static int pktgen_thread_worker(void *arg) 3700 { 3701 struct pktgen_thread *t = arg; 3702 struct pktgen_dev *pkt_dev = NULL; 3703 int cpu = t->cpu; 3704 3705 WARN_ON_ONCE(smp_processor_id() != cpu); 3706 3707 init_waitqueue_head(&t->queue); 3708 complete(&t->start_done); 3709 3710 pr_debug("starting pktgen/%d: pid=%d\n", cpu, task_pid_nr(current)); 3711 3712 set_freezable(); 3713 3714 while (!kthread_should_stop()) { 3715 pkt_dev = next_to_run(t); 3716 3717 if (unlikely(!pkt_dev && t->control == 0)) { 3718 if (t->net->pktgen_exiting) 3719 break; 3720 wait_event_freezable_timeout(t->queue, 3721 t->control != 0, HZ / 10); 3722 continue; 3723 } 3724 3725 if (likely(pkt_dev)) { 3726 pktgen_xmit(pkt_dev); 3727 3728 if (need_resched()) 3729 pktgen_resched(pkt_dev); 3730 else 3731 cpu_relax(); 3732 } 3733 3734 if (t->control & T_STOP) { 3735 pktgen_stop(t); 3736 t->control &= ~(T_STOP); 3737 } 3738 3739 if (t->control & T_RUN) { 3740 pktgen_run(t); 3741 t->control &= ~(T_RUN); 3742 } 3743 3744 if (t->control & T_REMDEVALL) { 3745 pktgen_rem_all_ifs(t); 3746 t->control &= ~(T_REMDEVALL); 3747 } 3748 3749 if (t->control & T_REMDEV) { 3750 pktgen_rem_one_if(t); 3751 t->control &= ~(T_REMDEV); 3752 } 3753 3754 try_to_freeze(); 3755 } 3756 3757 pr_debug("%s stopping all device\n", t->tsk->comm); 3758 pktgen_stop(t); 3759 3760 pr_debug("%s removing all device\n", t->tsk->comm); 3761 pktgen_rem_all_ifs(t); 3762 3763 pr_debug("%s removing thread\n", t->tsk->comm); 3764 pktgen_rem_thread(t); 3765 3766 return 0; 3767 } 3768 3769 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t, 3770 const char *ifname, bool exact) 3771 { 3772 struct pktgen_dev *p, *pkt_dev = NULL; 3773 size_t len = strlen(ifname); 3774 3775 rcu_read_lock(); 3776 list_for_each_entry_rcu(p, &t->if_list, list) 3777 if (strncmp(p->odevname, ifname, len) == 0) { 3778 if (p->odevname[len]) { 3779 if (exact || p->odevname[len] != '@') 3780 continue; 3781 } 3782 pkt_dev = p; 3783 break; 3784 } 3785 3786 rcu_read_unlock(); 3787 pr_debug("find_dev(%s) returning %p\n", ifname, pkt_dev); 3788 return pkt_dev; 3789 } 3790 3791 /* 3792 * Adds a dev at front of if_list. 3793 */ 3794 3795 static int add_dev_to_thread(struct pktgen_thread *t, 3796 struct pktgen_dev *pkt_dev) 3797 { 3798 int rv = 0; 3799 3800 /* This function cannot be called concurrently, as its called 3801 * under pktgen_thread_lock mutex, but it can run from 3802 * userspace on another CPU than the kthread. The if_lock() 3803 * is used here to sync with concurrent instances of 3804 * _rem_dev_from_if_list() invoked via kthread, which is also 3805 * updating the if_list 3806 */ 3807 if_lock(t); 3808 3809 if (pkt_dev->pg_thread) { 3810 pr_err("ERROR: already assigned to a thread\n"); 3811 rv = -EBUSY; 3812 goto out; 3813 } 3814 3815 pkt_dev->running = 0; 3816 pkt_dev->pg_thread = t; 3817 list_add_rcu(&pkt_dev->list, &t->if_list); 3818 3819 out: 3820 if_unlock(t); 3821 return rv; 3822 } 3823 3824 /* Called under thread lock */ 3825 3826 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname) 3827 { 3828 struct pktgen_dev *pkt_dev; 3829 int err; 3830 int node = cpu_to_node(t->cpu); 3831 3832 /* We don't allow a device to be on several threads */ 3833 3834 pkt_dev = __pktgen_NN_threads(t->net, ifname, FIND); 3835 if (pkt_dev) { 3836 pr_err("ERROR: interface already used\n"); 3837 return -EBUSY; 3838 } 3839 3840 pkt_dev = kzalloc_node(sizeof(struct pktgen_dev), GFP_KERNEL, node); 3841 if (!pkt_dev) 3842 return -ENOMEM; 3843 3844 strscpy(pkt_dev->odevname, ifname); 3845 pkt_dev->flows = vzalloc_node(array_size(MAX_CFLOWS, 3846 sizeof(struct flow_state)), 3847 node); 3848 if (pkt_dev->flows == NULL) { 3849 kfree(pkt_dev); 3850 return -ENOMEM; 3851 } 3852 3853 pkt_dev->removal_mark = 0; 3854 pkt_dev->nfrags = 0; 3855 pkt_dev->delay = pg_delay_d; 3856 pkt_dev->count = pg_count_d; 3857 pkt_dev->sofar = 0; 3858 pkt_dev->udp_src_min = 9; /* sink port */ 3859 pkt_dev->udp_src_max = 9; 3860 pkt_dev->udp_dst_min = 9; 3861 pkt_dev->udp_dst_max = 9; 3862 pkt_dev->vlan_p = 0; 3863 pkt_dev->vlan_cfi = 0; 3864 pkt_dev->vlan_id = 0xffff; 3865 pkt_dev->svlan_p = 0; 3866 pkt_dev->svlan_cfi = 0; 3867 pkt_dev->svlan_id = 0xffff; 3868 pkt_dev->burst = 1; 3869 pkt_dev->node = NUMA_NO_NODE; 3870 pkt_dev->flags = F_SHARED; /* SKB shared by default */ 3871 3872 err = pktgen_setup_dev(t->net, pkt_dev, ifname); 3873 if (err) 3874 goto out1; 3875 if (pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING) 3876 pkt_dev->clone_skb = pg_clone_skb_d; 3877 3878 pkt_dev->entry = proc_create_data(ifname, 0600, t->net->proc_dir, 3879 &pktgen_if_proc_ops, pkt_dev); 3880 if (!pkt_dev->entry) { 3881 pr_err("cannot create %s/%s procfs entry\n", 3882 PG_PROC_DIR, ifname); 3883 err = -EINVAL; 3884 goto out2; 3885 } 3886 #ifdef CONFIG_XFRM 3887 pkt_dev->ipsmode = XFRM_MODE_TRANSPORT; 3888 pkt_dev->ipsproto = IPPROTO_ESP; 3889 3890 /* xfrm tunnel mode needs additional dst to extract outer 3891 * ip header protocol/ttl/id field, here create a phony one. 3892 * instead of looking for a valid rt, which definitely hurting 3893 * performance under such circumstance. 3894 */ 3895 pkt_dev->dstops.family = AF_INET; 3896 pkt_dev->xdst.u.dst.dev = pkt_dev->odev; 3897 dst_init_metrics(&pkt_dev->xdst.u.dst, pktgen_dst_metrics, false); 3898 pkt_dev->xdst.child = &pkt_dev->xdst.u.dst; 3899 pkt_dev->xdst.u.dst.ops = &pkt_dev->dstops; 3900 #endif 3901 3902 return add_dev_to_thread(t, pkt_dev); 3903 out2: 3904 netdev_put(pkt_dev->odev, &pkt_dev->dev_tracker); 3905 out1: 3906 #ifdef CONFIG_XFRM 3907 free_SAs(pkt_dev); 3908 #endif 3909 vfree(pkt_dev->flows); 3910 kfree(pkt_dev); 3911 return err; 3912 } 3913 3914 static int __net_init pktgen_create_thread(int cpu, struct pktgen_net *pn) 3915 { 3916 struct pktgen_thread *t; 3917 struct proc_dir_entry *pe; 3918 struct task_struct *p; 3919 3920 t = kzalloc_node(sizeof(struct pktgen_thread), GFP_KERNEL, 3921 cpu_to_node(cpu)); 3922 if (!t) { 3923 pr_err("ERROR: out of memory, can't create new thread\n"); 3924 return -ENOMEM; 3925 } 3926 3927 mutex_init(&t->if_lock); 3928 t->cpu = cpu; 3929 3930 INIT_LIST_HEAD(&t->if_list); 3931 3932 list_add_tail(&t->th_list, &pn->pktgen_threads); 3933 init_completion(&t->start_done); 3934 3935 p = kthread_create_on_cpu(pktgen_thread_worker, t, cpu, "kpktgend_%d"); 3936 if (IS_ERR(p)) { 3937 pr_err("kthread_create_on_node() failed for cpu %d\n", t->cpu); 3938 list_del(&t->th_list); 3939 kfree(t); 3940 return PTR_ERR(p); 3941 } 3942 3943 t->tsk = p; 3944 3945 pe = proc_create_data(t->tsk->comm, 0600, pn->proc_dir, 3946 &pktgen_thread_proc_ops, t); 3947 if (!pe) { 3948 pr_err("cannot create %s/%s procfs entry\n", 3949 PG_PROC_DIR, t->tsk->comm); 3950 kthread_stop(p); 3951 list_del(&t->th_list); 3952 kfree(t); 3953 return -EINVAL; 3954 } 3955 3956 t->net = pn; 3957 get_task_struct(p); 3958 wake_up_process(p); 3959 wait_for_completion(&t->start_done); 3960 3961 return 0; 3962 } 3963 3964 /* 3965 * Removes a device from the thread if_list. 3966 */ 3967 static void _rem_dev_from_if_list(struct pktgen_thread *t, 3968 struct pktgen_dev *pkt_dev) 3969 { 3970 struct list_head *q, *n; 3971 struct pktgen_dev *p; 3972 3973 if_lock(t); 3974 list_for_each_safe(q, n, &t->if_list) { 3975 p = list_entry(q, struct pktgen_dev, list); 3976 if (p == pkt_dev) 3977 list_del_rcu(&p->list); 3978 } 3979 if_unlock(t); 3980 } 3981 3982 static int pktgen_remove_device(struct pktgen_thread *t, 3983 struct pktgen_dev *pkt_dev) 3984 { 3985 pr_debug("remove_device pkt_dev=%p\n", pkt_dev); 3986 3987 if (pkt_dev->running) { 3988 pr_warn("WARNING: trying to remove a running interface, stopping it now\n"); 3989 pktgen_stop_device(pkt_dev); 3990 } 3991 3992 /* Dis-associate from the interface */ 3993 3994 if (pkt_dev->odev) { 3995 netdev_put(pkt_dev->odev, &pkt_dev->dev_tracker); 3996 pkt_dev->odev = NULL; 3997 } 3998 3999 /* Remove proc before if_list entry, because add_device uses 4000 * list to determine if interface already exist, avoid race 4001 * with proc_create_data() 4002 */ 4003 proc_remove(pkt_dev->entry); 4004 4005 /* And update the thread if_list */ 4006 _rem_dev_from_if_list(t, pkt_dev); 4007 4008 #ifdef CONFIG_XFRM 4009 free_SAs(pkt_dev); 4010 #endif 4011 vfree(pkt_dev->flows); 4012 if (pkt_dev->page) 4013 put_page(pkt_dev->page); 4014 kfree_rcu(pkt_dev, rcu); 4015 return 0; 4016 } 4017 4018 static int __net_init pg_net_init(struct net *net) 4019 { 4020 struct pktgen_net *pn = net_generic(net, pg_net_id); 4021 struct proc_dir_entry *pe; 4022 int cpu, ret = 0; 4023 4024 pn->net = net; 4025 INIT_LIST_HEAD(&pn->pktgen_threads); 4026 pn->pktgen_exiting = false; 4027 pn->proc_dir = proc_mkdir(PG_PROC_DIR, pn->net->proc_net); 4028 if (!pn->proc_dir) { 4029 pr_warn("cannot create /proc/net/%s\n", PG_PROC_DIR); 4030 return -ENODEV; 4031 } 4032 pe = proc_create(PGCTRL, 0600, pn->proc_dir, &pktgen_proc_ops); 4033 if (pe == NULL) { 4034 pr_err("cannot create %s procfs entry\n", PGCTRL); 4035 ret = -EINVAL; 4036 goto remove; 4037 } 4038 4039 cpus_read_lock(); 4040 for_each_online_cpu(cpu) { 4041 int err; 4042 4043 err = pktgen_create_thread(cpu, pn); 4044 if (err) 4045 pr_warn("Cannot create thread for cpu %d (%d)\n", 4046 cpu, err); 4047 } 4048 cpus_read_unlock(); 4049 4050 if (list_empty(&pn->pktgen_threads)) { 4051 pr_err("Initialization failed for all threads\n"); 4052 ret = -ENODEV; 4053 goto remove_entry; 4054 } 4055 4056 return 0; 4057 4058 remove_entry: 4059 remove_proc_entry(PGCTRL, pn->proc_dir); 4060 remove: 4061 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net); 4062 return ret; 4063 } 4064 4065 static void __net_exit pg_net_exit(struct net *net) 4066 { 4067 struct pktgen_net *pn = net_generic(net, pg_net_id); 4068 struct pktgen_thread *t; 4069 struct list_head *q, *n; 4070 LIST_HEAD(list); 4071 4072 /* Stop all interfaces & threads */ 4073 pn->pktgen_exiting = true; 4074 4075 mutex_lock(&pktgen_thread_lock); 4076 list_splice_init(&pn->pktgen_threads, &list); 4077 mutex_unlock(&pktgen_thread_lock); 4078 4079 list_for_each_safe(q, n, &list) { 4080 t = list_entry(q, struct pktgen_thread, th_list); 4081 list_del(&t->th_list); 4082 kthread_stop_put(t->tsk); 4083 kfree(t); 4084 } 4085 4086 remove_proc_entry(PGCTRL, pn->proc_dir); 4087 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net); 4088 } 4089 4090 static struct pernet_operations pg_net_ops = { 4091 .init = pg_net_init, 4092 .exit = pg_net_exit, 4093 .id = &pg_net_id, 4094 .size = sizeof(struct pktgen_net), 4095 }; 4096 4097 static int __init pg_init(void) 4098 { 4099 int ret = 0; 4100 4101 pr_info("%s", version); 4102 ret = register_pernet_subsys(&pg_net_ops); 4103 if (ret) 4104 return ret; 4105 ret = register_netdevice_notifier(&pktgen_notifier_block); 4106 if (ret) 4107 unregister_pernet_subsys(&pg_net_ops); 4108 4109 return ret; 4110 } 4111 4112 static void __exit pg_cleanup(void) 4113 { 4114 unregister_netdevice_notifier(&pktgen_notifier_block); 4115 unregister_pernet_subsys(&pg_net_ops); 4116 /* Don't need rcu_barrier() due to use of kfree_rcu() */ 4117 } 4118 4119 module_init(pg_init); 4120 module_exit(pg_cleanup); 4121 4122 MODULE_AUTHOR("Robert Olsson <robert.olsson@its.uu.se>"); 4123 MODULE_DESCRIPTION("Packet Generator tool"); 4124 MODULE_LICENSE("GPL"); 4125 MODULE_VERSION(VERSION); 4126 module_param(pg_count_d, int, 0); 4127 MODULE_PARM_DESC(pg_count_d, "Default number of packets to inject"); 4128 module_param(pg_delay_d, int, 0); 4129 MODULE_PARM_DESC(pg_delay_d, "Default delay between packets (nanoseconds)"); 4130 module_param(pg_clone_skb_d, int, 0); 4131 MODULE_PARM_DESC(pg_clone_skb_d, "Default number of copies of the same packet"); 4132 module_param(debug, int, 0); 4133 MODULE_PARM_DESC(debug, "Enable debugging of pktgen module"); 4134