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/module.h> 118 #include <linux/moduleparam.h> 119 #include <linux/kernel.h> 120 #include <linux/mutex.h> 121 #include <linux/sched.h> 122 #include <linux/slab.h> 123 #include <linux/vmalloc.h> 124 #include <linux/unistd.h> 125 #include <linux/string.h> 126 #include <linux/ptrace.h> 127 #include <linux/errno.h> 128 #include <linux/ioport.h> 129 #include <linux/interrupt.h> 130 #include <linux/capability.h> 131 #include <linux/hrtimer.h> 132 #include <linux/freezer.h> 133 #include <linux/delay.h> 134 #include <linux/timer.h> 135 #include <linux/list.h> 136 #include <linux/init.h> 137 #include <linux/skbuff.h> 138 #include <linux/netdevice.h> 139 #include <linux/inet.h> 140 #include <linux/inetdevice.h> 141 #include <linux/rtnetlink.h> 142 #include <linux/if_arp.h> 143 #include <linux/if_vlan.h> 144 #include <linux/in.h> 145 #include <linux/ip.h> 146 #include <linux/ipv6.h> 147 #include <linux/udp.h> 148 #include <linux/proc_fs.h> 149 #include <linux/seq_file.h> 150 #include <linux/wait.h> 151 #include <linux/etherdevice.h> 152 #include <linux/kthread.h> 153 #include <linux/prefetch.h> 154 #include <linux/mmzone.h> 155 #include <net/net_namespace.h> 156 #include <net/checksum.h> 157 #include <net/ipv6.h> 158 #include <net/udp.h> 159 #include <net/ip6_checksum.h> 160 #include <net/addrconf.h> 161 #include <net/xfrm.h> 162 #include <net/netns/generic.h> 163 #include <asm/byteorder.h> 164 #include <linux/rcupdate.h> 165 #include <linux/bitops.h> 166 #include <linux/io.h> 167 #include <linux/timex.h> 168 #include <linux/uaccess.h> 169 #include <asm/dma.h> 170 #include <asm/div64.h> /* do_div */ 171 172 #define VERSION "2.75" 173 #define IP_NAME_SZ 32 174 #define MAX_MPLS_LABELS 16 /* This is the max label stack depth */ 175 #define MPLS_STACK_BOTTOM htonl(0x00000100) 176 /* Max number of internet mix entries that can be specified in imix_weights. */ 177 #define MAX_IMIX_ENTRIES 20 178 #define IMIX_PRECISION 100 /* Precision of IMIX distribution */ 179 180 #define func_enter() pr_debug("entering %s\n", __func__) 181 182 #define PKT_FLAGS \ 183 pf(IPV6) /* Interface in IPV6 Mode */ \ 184 pf(IPSRC_RND) /* IP-Src Random */ \ 185 pf(IPDST_RND) /* IP-Dst Random */ \ 186 pf(TXSIZE_RND) /* Transmit size is random */ \ 187 pf(UDPSRC_RND) /* UDP-Src Random */ \ 188 pf(UDPDST_RND) /* UDP-Dst Random */ \ 189 pf(UDPCSUM) /* Include UDP checksum */ \ 190 pf(NO_TIMESTAMP) /* Don't timestamp packets (default TS) */ \ 191 pf(MPLS_RND) /* Random MPLS labels */ \ 192 pf(QUEUE_MAP_RND) /* queue map Random */ \ 193 pf(QUEUE_MAP_CPU) /* queue map mirrors smp_processor_id() */ \ 194 pf(FLOW_SEQ) /* Sequential flows */ \ 195 pf(IPSEC) /* ipsec on for flows */ \ 196 pf(MACSRC_RND) /* MAC-Src Random */ \ 197 pf(MACDST_RND) /* MAC-Dst Random */ \ 198 pf(VID_RND) /* Random VLAN ID */ \ 199 pf(SVID_RND) /* Random SVLAN ID */ \ 200 pf(NODE) /* Node memory alloc*/ \ 201 pf(SHARED) /* Shared SKB */ \ 202 203 #define pf(flag) flag##_SHIFT, 204 enum pkt_flags { 205 PKT_FLAGS 206 }; 207 #undef pf 208 209 /* Device flag bits */ 210 #define pf(flag) static const __u32 F_##flag = (1<<flag##_SHIFT); 211 PKT_FLAGS 212 #undef pf 213 214 #define pf(flag) __stringify(flag), 215 static char *pkt_flag_names[] = { 216 PKT_FLAGS 217 }; 218 #undef pf 219 220 #define NR_PKT_FLAGS ARRAY_SIZE(pkt_flag_names) 221 222 /* Thread control flag bits */ 223 #define T_STOP (1<<0) /* Stop run */ 224 #define T_RUN (1<<1) /* Start run */ 225 #define T_REMDEVALL (1<<2) /* Remove all devs */ 226 #define T_REMDEV (1<<3) /* Remove one dev */ 227 228 /* Xmit modes */ 229 #define M_START_XMIT 0 /* Default normal TX */ 230 #define M_NETIF_RECEIVE 1 /* Inject packets into stack */ 231 #define M_QUEUE_XMIT 2 /* Inject packet into qdisc */ 232 233 /* If lock -- protects updating of if_list */ 234 #define if_lock(t) mutex_lock(&(t->if_lock)) 235 #define if_unlock(t) mutex_unlock(&(t->if_lock)) 236 237 /* Used to help with determining the pkts on receive */ 238 #define PKTGEN_MAGIC 0xbe9be955 239 #define PG_PROC_DIR "pktgen" 240 #define PGCTRL "pgctrl" 241 242 #define MAX_CFLOWS 65536 243 244 #define VLAN_TAG_SIZE(x) ((x)->vlan_id == 0xffff ? 0 : 4) 245 #define SVLAN_TAG_SIZE(x) ((x)->svlan_id == 0xffff ? 0 : 4) 246 247 struct imix_pkt { 248 u64 size; 249 u64 weight; 250 u64 count_so_far; 251 }; 252 253 struct flow_state { 254 __be32 cur_daddr; 255 int count; 256 #ifdef CONFIG_XFRM 257 struct xfrm_state *x; 258 #endif 259 __u32 flags; 260 }; 261 262 /* flow flag bits */ 263 #define F_INIT (1<<0) /* flow has been initialized */ 264 265 struct pktgen_dev { 266 /* 267 * Try to keep frequent/infrequent used vars. separated. 268 */ 269 struct proc_dir_entry *entry; /* proc file */ 270 struct pktgen_thread *pg_thread;/* the owner */ 271 struct list_head list; /* chaining in the thread's run-queue */ 272 struct rcu_head rcu; /* freed by RCU */ 273 274 int running; /* if false, the test will stop */ 275 276 /* If min != max, then we will either do a linear iteration, or 277 * we will do a random selection from within the range. 278 */ 279 __u32 flags; 280 int xmit_mode; 281 int min_pkt_size; 282 int max_pkt_size; 283 int pkt_overhead; /* overhead for MPLS, VLANs, IPSEC etc */ 284 int nfrags; 285 int removal_mark; /* non-zero => the device is marked for 286 * removal by worker thread 287 */ 288 289 struct page *page; 290 u64 delay; /* nano-seconds */ 291 292 __u64 count; /* Default No packets to send */ 293 __u64 sofar; /* How many pkts we've sent so far */ 294 __u64 tx_bytes; /* How many bytes we've transmitted */ 295 __u64 errors; /* Errors when trying to transmit, */ 296 297 /* runtime counters relating to clone_skb */ 298 299 __u32 clone_count; 300 int last_ok; /* Was last skb sent? 301 * Or a failed transmit of some sort? 302 * This will keep sequence numbers in order 303 */ 304 ktime_t next_tx; 305 ktime_t started_at; 306 ktime_t stopped_at; 307 u64 idle_acc; /* nano-seconds */ 308 309 __u32 seq_num; 310 311 int clone_skb; /* 312 * Use multiple SKBs during packet gen. 313 * If this number is greater than 1, then 314 * that many copies of the same packet will be 315 * sent before a new packet is allocated. 316 * If you want to send 1024 identical packets 317 * before creating a new packet, 318 * set clone_skb to 1024. 319 */ 320 321 char dst_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 322 char dst_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 323 char src_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 324 char src_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 325 326 struct in6_addr in6_saddr; 327 struct in6_addr in6_daddr; 328 struct in6_addr cur_in6_daddr; 329 struct in6_addr cur_in6_saddr; 330 /* For ranges */ 331 struct in6_addr min_in6_daddr; 332 struct in6_addr max_in6_daddr; 333 struct in6_addr min_in6_saddr; 334 struct in6_addr max_in6_saddr; 335 336 /* If we're doing ranges, random or incremental, then this 337 * defines the min/max for those ranges. 338 */ 339 __be32 saddr_min; /* inclusive, source IP address */ 340 __be32 saddr_max; /* exclusive, source IP address */ 341 __be32 daddr_min; /* inclusive, dest IP address */ 342 __be32 daddr_max; /* exclusive, dest IP address */ 343 344 __u16 udp_src_min; /* inclusive, source UDP port */ 345 __u16 udp_src_max; /* exclusive, source UDP port */ 346 __u16 udp_dst_min; /* inclusive, dest UDP port */ 347 __u16 udp_dst_max; /* exclusive, dest UDP port */ 348 349 /* DSCP + ECN */ 350 __u8 tos; /* six MSB of (former) IPv4 TOS 351 * are for dscp codepoint 352 */ 353 __u8 traffic_class; /* ditto for the (former) Traffic Class in IPv6 354 * (see RFC 3260, sec. 4) 355 */ 356 357 /* IMIX */ 358 unsigned int n_imix_entries; 359 struct imix_pkt imix_entries[MAX_IMIX_ENTRIES]; 360 /* Maps 0-IMIX_PRECISION range to imix_entry based on probability*/ 361 __u8 imix_distribution[IMIX_PRECISION]; 362 363 /* MPLS */ 364 unsigned int nr_labels; /* Depth of stack, 0 = no MPLS */ 365 __be32 labels[MAX_MPLS_LABELS]; 366 367 /* VLAN/SVLAN (802.1Q/Q-in-Q) */ 368 __u8 vlan_p; 369 __u8 vlan_cfi; 370 __u16 vlan_id; /* 0xffff means no vlan tag */ 371 372 __u8 svlan_p; 373 __u8 svlan_cfi; 374 __u16 svlan_id; /* 0xffff means no svlan tag */ 375 376 __u32 src_mac_count; /* How many MACs to iterate through */ 377 __u32 dst_mac_count; /* How many MACs to iterate through */ 378 379 unsigned char dst_mac[ETH_ALEN]; 380 unsigned char src_mac[ETH_ALEN]; 381 382 __u32 cur_dst_mac_offset; 383 __u32 cur_src_mac_offset; 384 __be32 cur_saddr; 385 __be32 cur_daddr; 386 __u16 ip_id; 387 __u16 cur_udp_dst; 388 __u16 cur_udp_src; 389 __u16 cur_queue_map; 390 __u32 cur_pkt_size; 391 __u32 last_pkt_size; 392 393 __u8 hh[14]; 394 /* = { 395 * 0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB, 396 * 397 * We fill in SRC address later 398 * 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 399 * 0x08, 0x00 400 * }; 401 */ 402 __u16 pad; /* pad out the hh struct to an even 16 bytes */ 403 404 struct sk_buff *skb; /* skb we are to transmit next, used for when we 405 * are transmitting the same one multiple times 406 */ 407 struct net_device *odev; /* The out-going device. 408 * Note that the device should have it's 409 * pg_info pointer pointing back to this 410 * device. 411 * Set when the user specifies the out-going 412 * device name (not when the inject is 413 * started as it used to do.) 414 */ 415 netdevice_tracker dev_tracker; 416 char odevname[32]; 417 struct flow_state *flows; 418 unsigned int cflows; /* Concurrent flows (config) */ 419 unsigned int lflow; /* Flow length (config) */ 420 unsigned int nflows; /* accumulated flows (stats) */ 421 unsigned int curfl; /* current sequenced flow (state)*/ 422 423 u16 queue_map_min; 424 u16 queue_map_max; 425 __u32 skb_priority; /* skb priority field */ 426 unsigned int burst; /* number of duplicated packets to burst */ 427 int node; /* Memory node */ 428 429 #ifdef CONFIG_XFRM 430 __u8 ipsmode; /* IPSEC mode (config) */ 431 __u8 ipsproto; /* IPSEC type (config) */ 432 __u32 spi; 433 struct xfrm_dst xdst; 434 struct dst_ops dstops; 435 #endif 436 char result[512]; 437 }; 438 439 struct pktgen_hdr { 440 __be32 pgh_magic; 441 __be32 seq_num; 442 __be32 tv_sec; 443 __be32 tv_usec; 444 }; 445 446 447 static unsigned int pg_net_id __read_mostly; 448 449 struct pktgen_net { 450 struct net *net; 451 struct proc_dir_entry *proc_dir; 452 struct list_head pktgen_threads; 453 bool pktgen_exiting; 454 }; 455 456 struct pktgen_thread { 457 struct mutex if_lock; /* for list of devices */ 458 struct list_head if_list; /* All device here */ 459 struct list_head th_list; 460 struct task_struct *tsk; 461 char result[512]; 462 463 /* Field for thread to receive "posted" events terminate, 464 * stop ifs etc. 465 */ 466 467 u32 control; 468 int cpu; 469 470 wait_queue_head_t queue; 471 struct completion start_done; 472 struct pktgen_net *net; 473 }; 474 475 #define REMOVE 1 476 #define FIND 0 477 478 static const char version[] = 479 "Packet Generator for packet performance testing. Version: " VERSION "\n"; 480 481 static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i); 482 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname); 483 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t, 484 const char *ifname, bool exact); 485 static int pktgen_device_event(struct notifier_block *, unsigned long, void *); 486 static void pktgen_run_all_threads(struct pktgen_net *pn); 487 static void pktgen_reset_all_threads(struct pktgen_net *pn); 488 static void pktgen_stop_all_threads(struct pktgen_net *pn); 489 490 static void pktgen_stop(struct pktgen_thread *t); 491 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev); 492 static void fill_imix_distribution(struct pktgen_dev *pkt_dev); 493 494 /* Module parameters, defaults. */ 495 static int pg_count_d __read_mostly = 1000; 496 static int pg_delay_d __read_mostly; 497 static int pg_clone_skb_d __read_mostly; 498 static int debug __read_mostly; 499 500 static DEFINE_MUTEX(pktgen_thread_lock); 501 502 static struct notifier_block pktgen_notifier_block = { 503 .notifier_call = pktgen_device_event, 504 }; 505 506 /* 507 * /proc handling functions 508 * 509 */ 510 511 static int pgctrl_show(struct seq_file *seq, void *v) 512 { 513 seq_puts(seq, version); 514 return 0; 515 } 516 517 static ssize_t pgctrl_write(struct file *file, const char __user *buf, 518 size_t count, loff_t *ppos) 519 { 520 char data[128]; 521 size_t max; 522 struct pktgen_net *pn = net_generic(current->nsproxy->net_ns, pg_net_id); 523 524 if (!capable(CAP_NET_ADMIN)) 525 return -EPERM; 526 527 if (count < 1) 528 return -EINVAL; 529 530 max = min(count, sizeof(data) - 1); 531 if (copy_from_user(data, buf, max)) 532 return -EFAULT; 533 534 if (data[max - 1] == '\n') 535 data[max - 1] = 0; /* strip trailing '\n', terminate string */ 536 else 537 data[max] = 0; /* terminate string */ 538 539 if (!strcmp(data, "stop")) 540 pktgen_stop_all_threads(pn); 541 else if (!strcmp(data, "start")) 542 pktgen_run_all_threads(pn); 543 else if (!strcmp(data, "reset")) 544 pktgen_reset_all_threads(pn); 545 else 546 return -EINVAL; 547 548 return count; 549 } 550 551 static int pgctrl_open(struct inode *inode, struct file *file) 552 { 553 return single_open(file, pgctrl_show, pde_data(inode)); 554 } 555 556 static const struct proc_ops pktgen_proc_ops = { 557 .proc_open = pgctrl_open, 558 .proc_read = seq_read, 559 .proc_lseek = seq_lseek, 560 .proc_write = pgctrl_write, 561 .proc_release = single_release, 562 }; 563 564 static int pktgen_if_show(struct seq_file *seq, void *v) 565 { 566 const struct pktgen_dev *pkt_dev = seq->private; 567 ktime_t stopped; 568 unsigned int i; 569 u64 idle; 570 571 seq_printf(seq, 572 "Params: count %llu min_pkt_size: %u max_pkt_size: %u\n", 573 (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size, 574 pkt_dev->max_pkt_size); 575 576 if (pkt_dev->n_imix_entries > 0) { 577 seq_puts(seq, " imix_weights: "); 578 for (i = 0; i < pkt_dev->n_imix_entries; i++) { 579 seq_printf(seq, "%llu,%llu ", 580 pkt_dev->imix_entries[i].size, 581 pkt_dev->imix_entries[i].weight); 582 } 583 seq_puts(seq, "\n"); 584 } 585 586 seq_printf(seq, 587 " frags: %d delay: %llu clone_skb: %d ifname: %s\n", 588 pkt_dev->nfrags, (unsigned long long) pkt_dev->delay, 589 pkt_dev->clone_skb, pkt_dev->odevname); 590 591 seq_printf(seq, " flows: %u flowlen: %u\n", pkt_dev->cflows, 592 pkt_dev->lflow); 593 594 seq_printf(seq, 595 " queue_map_min: %u queue_map_max: %u\n", 596 pkt_dev->queue_map_min, 597 pkt_dev->queue_map_max); 598 599 if (pkt_dev->skb_priority) 600 seq_printf(seq, " skb_priority: %u\n", 601 pkt_dev->skb_priority); 602 603 if (pkt_dev->flags & F_IPV6) { 604 seq_printf(seq, 605 " saddr: %pI6c min_saddr: %pI6c max_saddr: %pI6c\n" 606 " daddr: %pI6c min_daddr: %pI6c max_daddr: %pI6c\n", 607 &pkt_dev->in6_saddr, 608 &pkt_dev->min_in6_saddr, &pkt_dev->max_in6_saddr, 609 &pkt_dev->in6_daddr, 610 &pkt_dev->min_in6_daddr, &pkt_dev->max_in6_daddr); 611 } else { 612 seq_printf(seq, 613 " dst_min: %s dst_max: %s\n", 614 pkt_dev->dst_min, pkt_dev->dst_max); 615 seq_printf(seq, 616 " src_min: %s src_max: %s\n", 617 pkt_dev->src_min, pkt_dev->src_max); 618 } 619 620 seq_puts(seq, " src_mac: "); 621 622 seq_printf(seq, "%pM ", 623 is_zero_ether_addr(pkt_dev->src_mac) ? 624 pkt_dev->odev->dev_addr : pkt_dev->src_mac); 625 626 seq_puts(seq, "dst_mac: "); 627 seq_printf(seq, "%pM\n", pkt_dev->dst_mac); 628 629 seq_printf(seq, 630 " udp_src_min: %d udp_src_max: %d udp_dst_min: %d udp_dst_max: %d\n", 631 pkt_dev->udp_src_min, pkt_dev->udp_src_max, 632 pkt_dev->udp_dst_min, pkt_dev->udp_dst_max); 633 634 seq_printf(seq, 635 " src_mac_count: %d dst_mac_count: %d\n", 636 pkt_dev->src_mac_count, pkt_dev->dst_mac_count); 637 638 if (pkt_dev->nr_labels) { 639 seq_puts(seq, " mpls: "); 640 for (i = 0; i < pkt_dev->nr_labels; i++) 641 seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]), 642 i == pkt_dev->nr_labels-1 ? "\n" : ", "); 643 } 644 645 if (pkt_dev->vlan_id != 0xffff) 646 seq_printf(seq, " vlan_id: %u vlan_p: %u vlan_cfi: %u\n", 647 pkt_dev->vlan_id, pkt_dev->vlan_p, 648 pkt_dev->vlan_cfi); 649 650 if (pkt_dev->svlan_id != 0xffff) 651 seq_printf(seq, " svlan_id: %u vlan_p: %u vlan_cfi: %u\n", 652 pkt_dev->svlan_id, pkt_dev->svlan_p, 653 pkt_dev->svlan_cfi); 654 655 if (pkt_dev->tos) 656 seq_printf(seq, " tos: 0x%02x\n", pkt_dev->tos); 657 658 if (pkt_dev->traffic_class) 659 seq_printf(seq, " traffic_class: 0x%02x\n", pkt_dev->traffic_class); 660 661 if (pkt_dev->burst > 1) 662 seq_printf(seq, " burst: %d\n", pkt_dev->burst); 663 664 if (pkt_dev->node >= 0) 665 seq_printf(seq, " node: %d\n", pkt_dev->node); 666 667 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) 668 seq_puts(seq, " xmit_mode: netif_receive\n"); 669 else if (pkt_dev->xmit_mode == M_QUEUE_XMIT) 670 seq_puts(seq, " xmit_mode: xmit_queue\n"); 671 672 seq_puts(seq, " Flags: "); 673 674 for (i = 0; i < NR_PKT_FLAGS; i++) { 675 if (i == FLOW_SEQ_SHIFT) 676 if (!pkt_dev->cflows) 677 continue; 678 679 if (pkt_dev->flags & (1 << i)) { 680 seq_printf(seq, "%s ", pkt_flag_names[i]); 681 #ifdef CONFIG_XFRM 682 if (i == IPSEC_SHIFT && pkt_dev->spi) 683 seq_printf(seq, "spi:%u ", pkt_dev->spi); 684 #endif 685 } else if (i == FLOW_SEQ_SHIFT) { 686 seq_puts(seq, "FLOW_RND "); 687 } 688 } 689 690 seq_puts(seq, "\n"); 691 692 /* not really stopped, more like last-running-at */ 693 stopped = pkt_dev->running ? ktime_get() : pkt_dev->stopped_at; 694 idle = pkt_dev->idle_acc; 695 do_div(idle, NSEC_PER_USEC); 696 697 seq_printf(seq, 698 "Current:\n pkts-sofar: %llu errors: %llu\n", 699 (unsigned long long)pkt_dev->sofar, 700 (unsigned long long)pkt_dev->errors); 701 702 if (pkt_dev->n_imix_entries > 0) { 703 int i; 704 705 seq_puts(seq, " imix_size_counts: "); 706 for (i = 0; i < pkt_dev->n_imix_entries; i++) { 707 seq_printf(seq, "%llu,%llu ", 708 pkt_dev->imix_entries[i].size, 709 pkt_dev->imix_entries[i].count_so_far); 710 } 711 seq_puts(seq, "\n"); 712 } 713 714 seq_printf(seq, 715 " started: %lluus stopped: %lluus idle: %lluus\n", 716 (unsigned long long) ktime_to_us(pkt_dev->started_at), 717 (unsigned long long) ktime_to_us(stopped), 718 (unsigned long long) idle); 719 720 seq_printf(seq, 721 " seq_num: %d cur_dst_mac_offset: %d cur_src_mac_offset: %d\n", 722 pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset, 723 pkt_dev->cur_src_mac_offset); 724 725 if (pkt_dev->flags & F_IPV6) { 726 seq_printf(seq, " cur_saddr: %pI6c cur_daddr: %pI6c\n", 727 &pkt_dev->cur_in6_saddr, 728 &pkt_dev->cur_in6_daddr); 729 } else 730 seq_printf(seq, " cur_saddr: %pI4 cur_daddr: %pI4\n", 731 &pkt_dev->cur_saddr, &pkt_dev->cur_daddr); 732 733 seq_printf(seq, " cur_udp_dst: %d cur_udp_src: %d\n", 734 pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src); 735 736 seq_printf(seq, " cur_queue_map: %u\n", pkt_dev->cur_queue_map); 737 738 seq_printf(seq, " flows: %u\n", pkt_dev->nflows); 739 740 if (pkt_dev->result[0]) 741 seq_printf(seq, "Result: %s\n", pkt_dev->result); 742 else 743 seq_puts(seq, "Result: Idle\n"); 744 745 return 0; 746 } 747 748 749 static ssize_t hex32_arg(const char __user *user_buffer, size_t maxlen, 750 __u32 *num) 751 { 752 size_t i = 0; 753 754 *num = 0; 755 756 for (; i < maxlen; i++) { 757 int value; 758 char c; 759 760 if (get_user(c, &user_buffer[i])) 761 return -EFAULT; 762 value = hex_to_bin(c); 763 if (value >= 0) { 764 *num <<= 4; 765 *num |= value; 766 } else { 767 break; 768 } 769 } 770 return i; 771 } 772 773 static ssize_t count_trail_chars(const char __user *user_buffer, size_t maxlen) 774 { 775 size_t i; 776 777 for (i = 0; i < maxlen; i++) { 778 char c; 779 780 if (get_user(c, &user_buffer[i])) 781 return -EFAULT; 782 switch (c) { 783 case '\"': 784 case '\n': 785 case '\r': 786 case '\t': 787 case ' ': 788 case '=': 789 break; 790 default: 791 goto done; 792 } 793 } 794 done: 795 return i; 796 } 797 798 static ssize_t num_arg(const char __user *user_buffer, size_t maxlen, 799 unsigned long *num) 800 { 801 size_t i; 802 *num = 0; 803 804 for (i = 0; i < maxlen; i++) { 805 char c; 806 807 if (get_user(c, &user_buffer[i])) 808 return -EFAULT; 809 if ((c >= '0') && (c <= '9')) { 810 *num *= 10; 811 *num += c - '0'; 812 } else 813 break; 814 } 815 return i; 816 } 817 818 static ssize_t strn_len(const char __user *user_buffer, size_t maxlen) 819 { 820 size_t i; 821 822 for (i = 0; i < maxlen; i++) { 823 char c; 824 825 if (get_user(c, &user_buffer[i])) 826 return -EFAULT; 827 switch (c) { 828 case '\"': 829 case '\n': 830 case '\r': 831 case '\t': 832 case ' ': 833 case '=': 834 goto done_str; 835 default: 836 break; 837 } 838 } 839 done_str: 840 return i; 841 } 842 843 /* Parses imix entries from user buffer. 844 * The user buffer should consist of imix entries separated by spaces 845 * where each entry consists of size and weight delimited by commas. 846 * "size1,weight_1 size2,weight_2 ... size_n,weight_n" for example. 847 */ 848 static ssize_t get_imix_entries(const char __user *buffer, 849 size_t maxlen, 850 struct pktgen_dev *pkt_dev) 851 { 852 size_t i = 0, max; 853 ssize_t len; 854 char c; 855 856 pkt_dev->n_imix_entries = 0; 857 858 do { 859 unsigned long weight; 860 unsigned long size; 861 862 if (pkt_dev->n_imix_entries >= MAX_IMIX_ENTRIES) 863 return -E2BIG; 864 865 if (i >= maxlen) 866 return -EINVAL; 867 868 max = min(10, maxlen - i); 869 len = num_arg(&buffer[i], max, &size); 870 if (len < 0) 871 return len; 872 i += len; 873 if (i >= maxlen) 874 return -EINVAL; 875 if (get_user(c, &buffer[i])) 876 return -EFAULT; 877 /* Check for comma between size_i and weight_i */ 878 if (c != ',') 879 return -EINVAL; 880 i++; 881 if (i >= maxlen) 882 return -EINVAL; 883 884 if (size < 14 + 20 + 8) 885 size = 14 + 20 + 8; 886 887 max = min(10, maxlen - i); 888 len = num_arg(&buffer[i], max, &weight); 889 if (len < 0) 890 return len; 891 if (weight <= 0) 892 return -EINVAL; 893 894 pkt_dev->imix_entries[pkt_dev->n_imix_entries].size = size; 895 pkt_dev->imix_entries[pkt_dev->n_imix_entries].weight = weight; 896 897 i += len; 898 pkt_dev->n_imix_entries++; 899 900 if (i >= maxlen) 901 break; 902 if (get_user(c, &buffer[i])) 903 return -EFAULT; 904 i++; 905 } while (c == ' '); 906 907 return i; 908 } 909 910 static ssize_t get_labels(const char __user *buffer, 911 size_t maxlen, struct pktgen_dev *pkt_dev) 912 { 913 unsigned int n = 0; 914 size_t i = 0, max; 915 ssize_t len; 916 char c; 917 918 pkt_dev->nr_labels = 0; 919 do { 920 __u32 tmp; 921 922 if (n >= MAX_MPLS_LABELS) 923 return -E2BIG; 924 925 if (i >= maxlen) 926 return -EINVAL; 927 928 max = min(8, maxlen - i); 929 len = hex32_arg(&buffer[i], max, &tmp); 930 if (len < 0) 931 return len; 932 933 /* return empty list in case of invalid input or zero value */ 934 if (len == 0 || tmp == 0) 935 return maxlen; 936 937 pkt_dev->labels[n] = htonl(tmp); 938 if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM) 939 pkt_dev->flags |= F_MPLS_RND; 940 i += len; 941 n++; 942 if (i >= maxlen) 943 break; 944 if (get_user(c, &buffer[i])) 945 return -EFAULT; 946 i++; 947 } while (c == ','); 948 949 pkt_dev->nr_labels = n; 950 return i; 951 } 952 953 static __u32 pktgen_read_flag(const char *f, bool *disable) 954 { 955 __u32 i; 956 957 if (f[0] == '!') { 958 *disable = true; 959 f++; 960 } 961 962 for (i = 0; i < NR_PKT_FLAGS; i++) { 963 if (!IS_ENABLED(CONFIG_XFRM) && i == IPSEC_SHIFT) 964 continue; 965 966 /* allow only disabling ipv6 flag */ 967 if (!*disable && i == IPV6_SHIFT) 968 continue; 969 970 if (strcmp(f, pkt_flag_names[i]) == 0) 971 return 1 << i; 972 } 973 974 if (strcmp(f, "FLOW_RND") == 0) { 975 *disable = !*disable; 976 return F_FLOW_SEQ; 977 } 978 979 return 0; 980 } 981 982 static ssize_t pktgen_if_write(struct file *file, 983 const char __user *user_buffer, size_t count, 984 loff_t *offset) 985 { 986 struct seq_file *seq = file->private_data; 987 struct pktgen_dev *pkt_dev = seq->private; 988 size_t i, max; 989 ssize_t len; 990 char name[16], valstr[32]; 991 unsigned long value = 0; 992 char *pg_result = NULL; 993 char buf[128]; 994 995 pg_result = &(pkt_dev->result[0]); 996 997 if (count < 1) { 998 pr_warn("wrong command format\n"); 999 return -EINVAL; 1000 } 1001 1002 max = count; 1003 len = count_trail_chars(user_buffer, max); 1004 if (len < 0) { 1005 pr_warn("illegal format\n"); 1006 return len; 1007 } 1008 i = len; 1009 1010 /* Read variable name */ 1011 max = min(sizeof(name) - 1, count - i); 1012 len = strn_len(&user_buffer[i], max); 1013 if (len < 0) 1014 return len; 1015 1016 memset(name, 0, sizeof(name)); 1017 if (copy_from_user(name, &user_buffer[i], len)) 1018 return -EFAULT; 1019 i += len; 1020 1021 max = count - i; 1022 len = count_trail_chars(&user_buffer[i], max); 1023 if (len < 0) 1024 return len; 1025 1026 i += len; 1027 1028 if (debug) { 1029 size_t copy = min_t(size_t, count + 1, 1024); 1030 char *tp = strndup_user(user_buffer, copy); 1031 1032 if (IS_ERR(tp)) 1033 return PTR_ERR(tp); 1034 1035 pr_debug("%s,%zu buffer -:%s:-\n", name, count, tp); 1036 kfree(tp); 1037 } 1038 1039 if (!strcmp(name, "min_pkt_size")) { 1040 max = min(10, count - i); 1041 len = num_arg(&user_buffer[i], max, &value); 1042 if (len < 0) 1043 return len; 1044 1045 if (value < 14 + 20 + 8) 1046 value = 14 + 20 + 8; 1047 if (value != pkt_dev->min_pkt_size) { 1048 pkt_dev->min_pkt_size = value; 1049 pkt_dev->cur_pkt_size = value; 1050 } 1051 sprintf(pg_result, "OK: min_pkt_size=%d", 1052 pkt_dev->min_pkt_size); 1053 return count; 1054 } 1055 1056 if (!strcmp(name, "max_pkt_size")) { 1057 max = min(10, count - i); 1058 len = num_arg(&user_buffer[i], max, &value); 1059 if (len < 0) 1060 return len; 1061 1062 if (value < 14 + 20 + 8) 1063 value = 14 + 20 + 8; 1064 if (value != pkt_dev->max_pkt_size) { 1065 pkt_dev->max_pkt_size = value; 1066 pkt_dev->cur_pkt_size = value; 1067 } 1068 sprintf(pg_result, "OK: max_pkt_size=%d", 1069 pkt_dev->max_pkt_size); 1070 return count; 1071 } 1072 1073 /* Shortcut for min = max */ 1074 1075 if (!strcmp(name, "pkt_size")) { 1076 max = min(10, count - i); 1077 len = num_arg(&user_buffer[i], max, &value); 1078 if (len < 0) 1079 return len; 1080 1081 if (value < 14 + 20 + 8) 1082 value = 14 + 20 + 8; 1083 if (value != pkt_dev->min_pkt_size) { 1084 pkt_dev->min_pkt_size = value; 1085 pkt_dev->max_pkt_size = value; 1086 pkt_dev->cur_pkt_size = value; 1087 } 1088 sprintf(pg_result, "OK: pkt_size=%d", pkt_dev->min_pkt_size); 1089 return count; 1090 } 1091 1092 if (!strcmp(name, "imix_weights")) { 1093 if (pkt_dev->clone_skb > 0) 1094 return -EINVAL; 1095 1096 max = count - i; 1097 len = get_imix_entries(&user_buffer[i], max, pkt_dev); 1098 if (len < 0) 1099 return len; 1100 1101 fill_imix_distribution(pkt_dev); 1102 1103 return count; 1104 } 1105 1106 if (!strcmp(name, "debug")) { 1107 max = min(10, count - i); 1108 len = num_arg(&user_buffer[i], max, &value); 1109 if (len < 0) 1110 return len; 1111 1112 debug = value; 1113 sprintf(pg_result, "OK: debug=%u", debug); 1114 return count; 1115 } 1116 1117 if (!strcmp(name, "frags")) { 1118 max = min(10, count - i); 1119 len = num_arg(&user_buffer[i], max, &value); 1120 if (len < 0) 1121 return len; 1122 1123 pkt_dev->nfrags = value; 1124 sprintf(pg_result, "OK: frags=%d", pkt_dev->nfrags); 1125 return count; 1126 } 1127 if (!strcmp(name, "delay")) { 1128 max = min(10, count - i); 1129 len = num_arg(&user_buffer[i], max, &value); 1130 if (len < 0) 1131 return len; 1132 1133 if (value == 0x7FFFFFFF) 1134 pkt_dev->delay = ULLONG_MAX; 1135 else 1136 pkt_dev->delay = (u64)value; 1137 1138 sprintf(pg_result, "OK: delay=%llu", 1139 (unsigned long long) pkt_dev->delay); 1140 return count; 1141 } 1142 if (!strcmp(name, "rate")) { 1143 max = min(10, count - i); 1144 len = num_arg(&user_buffer[i], max, &value); 1145 if (len < 0) 1146 return len; 1147 1148 if (!value) 1149 return -EINVAL; 1150 pkt_dev->delay = pkt_dev->min_pkt_size*8*NSEC_PER_USEC/value; 1151 if (debug) 1152 pr_info("Delay set at: %llu ns\n", pkt_dev->delay); 1153 1154 sprintf(pg_result, "OK: rate=%lu", value); 1155 return count; 1156 } 1157 if (!strcmp(name, "ratep")) { 1158 max = min(10, count - i); 1159 len = num_arg(&user_buffer[i], max, &value); 1160 if (len < 0) 1161 return len; 1162 1163 if (!value) 1164 return -EINVAL; 1165 pkt_dev->delay = NSEC_PER_SEC/value; 1166 if (debug) 1167 pr_info("Delay set at: %llu ns\n", pkt_dev->delay); 1168 1169 sprintf(pg_result, "OK: rate=%lu", value); 1170 return count; 1171 } 1172 if (!strcmp(name, "udp_src_min")) { 1173 max = min(10, count - i); 1174 len = num_arg(&user_buffer[i], max, &value); 1175 if (len < 0) 1176 return len; 1177 1178 if (value != pkt_dev->udp_src_min) { 1179 pkt_dev->udp_src_min = value; 1180 pkt_dev->cur_udp_src = value; 1181 } 1182 sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min); 1183 return count; 1184 } 1185 if (!strcmp(name, "udp_dst_min")) { 1186 max = min(10, count - i); 1187 len = num_arg(&user_buffer[i], max, &value); 1188 if (len < 0) 1189 return len; 1190 1191 if (value != pkt_dev->udp_dst_min) { 1192 pkt_dev->udp_dst_min = value; 1193 pkt_dev->cur_udp_dst = value; 1194 } 1195 sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min); 1196 return count; 1197 } 1198 if (!strcmp(name, "udp_src_max")) { 1199 max = min(10, count - i); 1200 len = num_arg(&user_buffer[i], max, &value); 1201 if (len < 0) 1202 return len; 1203 1204 if (value != pkt_dev->udp_src_max) { 1205 pkt_dev->udp_src_max = value; 1206 pkt_dev->cur_udp_src = value; 1207 } 1208 sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max); 1209 return count; 1210 } 1211 if (!strcmp(name, "udp_dst_max")) { 1212 max = min(10, count - i); 1213 len = num_arg(&user_buffer[i], max, &value); 1214 if (len < 0) 1215 return len; 1216 1217 if (value != pkt_dev->udp_dst_max) { 1218 pkt_dev->udp_dst_max = value; 1219 pkt_dev->cur_udp_dst = value; 1220 } 1221 sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max); 1222 return count; 1223 } 1224 if (!strcmp(name, "clone_skb")) { 1225 max = min(10, count - i); 1226 len = num_arg(&user_buffer[i], max, &value); 1227 if (len < 0) 1228 return len; 1229 /* clone_skb is not supported for netif_receive xmit_mode and 1230 * IMIX mode. 1231 */ 1232 if ((value > 0) && 1233 ((pkt_dev->xmit_mode == M_NETIF_RECEIVE) || 1234 !(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING))) 1235 return -EOPNOTSUPP; 1236 if (value > 0 && (pkt_dev->n_imix_entries > 0 || 1237 !(pkt_dev->flags & F_SHARED))) 1238 return -EINVAL; 1239 1240 pkt_dev->clone_skb = value; 1241 1242 sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb); 1243 return count; 1244 } 1245 if (!strcmp(name, "count")) { 1246 max = min(10, count - i); 1247 len = num_arg(&user_buffer[i], max, &value); 1248 if (len < 0) 1249 return len; 1250 1251 pkt_dev->count = value; 1252 sprintf(pg_result, "OK: count=%llu", 1253 (unsigned long long)pkt_dev->count); 1254 return count; 1255 } 1256 if (!strcmp(name, "src_mac_count")) { 1257 max = min(10, count - i); 1258 len = num_arg(&user_buffer[i], max, &value); 1259 if (len < 0) 1260 return len; 1261 1262 if (pkt_dev->src_mac_count != value) { 1263 pkt_dev->src_mac_count = value; 1264 pkt_dev->cur_src_mac_offset = 0; 1265 } 1266 sprintf(pg_result, "OK: src_mac_count=%d", 1267 pkt_dev->src_mac_count); 1268 return count; 1269 } 1270 if (!strcmp(name, "dst_mac_count")) { 1271 max = min(10, count - i); 1272 len = num_arg(&user_buffer[i], max, &value); 1273 if (len < 0) 1274 return len; 1275 1276 if (pkt_dev->dst_mac_count != value) { 1277 pkt_dev->dst_mac_count = value; 1278 pkt_dev->cur_dst_mac_offset = 0; 1279 } 1280 sprintf(pg_result, "OK: dst_mac_count=%d", 1281 pkt_dev->dst_mac_count); 1282 return count; 1283 } 1284 if (!strcmp(name, "burst")) { 1285 max = min(10, count - i); 1286 len = num_arg(&user_buffer[i], max, &value); 1287 if (len < 0) 1288 return len; 1289 1290 if ((value > 1) && 1291 ((pkt_dev->xmit_mode == M_QUEUE_XMIT) || 1292 ((pkt_dev->xmit_mode == M_START_XMIT) && 1293 (!(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING))))) 1294 return -EOPNOTSUPP; 1295 1296 if (value > 1 && !(pkt_dev->flags & F_SHARED)) 1297 return -EINVAL; 1298 1299 pkt_dev->burst = value < 1 ? 1 : value; 1300 sprintf(pg_result, "OK: burst=%u", pkt_dev->burst); 1301 return count; 1302 } 1303 if (!strcmp(name, "node")) { 1304 max = min(10, count - i); 1305 len = num_arg(&user_buffer[i], max, &value); 1306 if (len < 0) 1307 return len; 1308 1309 if (node_possible(value)) { 1310 pkt_dev->node = value; 1311 sprintf(pg_result, "OK: node=%d", pkt_dev->node); 1312 if (pkt_dev->page) { 1313 put_page(pkt_dev->page); 1314 pkt_dev->page = NULL; 1315 } 1316 } 1317 else 1318 sprintf(pg_result, "ERROR: node not possible"); 1319 return count; 1320 } 1321 if (!strcmp(name, "xmit_mode")) { 1322 char f[32]; 1323 1324 max = min(sizeof(f) - 1, count - i); 1325 len = strn_len(&user_buffer[i], max); 1326 if (len < 0) 1327 return len; 1328 1329 memset(f, 0, sizeof(f)); 1330 if (copy_from_user(f, &user_buffer[i], len)) 1331 return -EFAULT; 1332 1333 if (strcmp(f, "start_xmit") == 0) { 1334 pkt_dev->xmit_mode = M_START_XMIT; 1335 } else if (strcmp(f, "netif_receive") == 0) { 1336 /* clone_skb set earlier, not supported in this mode */ 1337 if (pkt_dev->clone_skb > 0) 1338 return -EOPNOTSUPP; 1339 1340 pkt_dev->xmit_mode = M_NETIF_RECEIVE; 1341 1342 /* make sure new packet is allocated every time 1343 * pktgen_xmit() is called 1344 */ 1345 pkt_dev->last_ok = 1; 1346 } else if (strcmp(f, "queue_xmit") == 0) { 1347 pkt_dev->xmit_mode = M_QUEUE_XMIT; 1348 pkt_dev->last_ok = 1; 1349 } else { 1350 sprintf(pg_result, 1351 "xmit_mode -:%s:- unknown\nAvailable modes: %s", 1352 f, "start_xmit, netif_receive\n"); 1353 return count; 1354 } 1355 sprintf(pg_result, "OK: xmit_mode=%s", f); 1356 return count; 1357 } 1358 if (!strcmp(name, "flag")) { 1359 bool disable = false; 1360 __u32 flag; 1361 char f[32]; 1362 char *end; 1363 1364 max = min(sizeof(f) - 1, count - i); 1365 len = strn_len(&user_buffer[i], max); 1366 if (len < 0) 1367 return len; 1368 1369 memset(f, 0, 32); 1370 if (copy_from_user(f, &user_buffer[i], len)) 1371 return -EFAULT; 1372 1373 flag = pktgen_read_flag(f, &disable); 1374 if (flag) { 1375 if (disable) { 1376 /* If "clone_skb", or "burst" parameters are 1377 * configured, it means that the skb still 1378 * needs to be referenced by the pktgen, so 1379 * the skb must be shared. 1380 */ 1381 if (flag == F_SHARED && (pkt_dev->clone_skb || 1382 pkt_dev->burst > 1)) 1383 return -EINVAL; 1384 pkt_dev->flags &= ~flag; 1385 } else { 1386 pkt_dev->flags |= flag; 1387 } 1388 1389 sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags); 1390 return count; 1391 } 1392 1393 /* Unknown flag */ 1394 end = pkt_dev->result + sizeof(pkt_dev->result); 1395 pg_result += sprintf(pg_result, 1396 "Flag -:%s:- unknown\n" 1397 "Available flags, (prepend ! to un-set flag):\n", f); 1398 1399 for (int n = 0; n < NR_PKT_FLAGS && pg_result < end; n++) { 1400 if (!IS_ENABLED(CONFIG_XFRM) && n == IPSEC_SHIFT) 1401 continue; 1402 pg_result += snprintf(pg_result, end - pg_result, 1403 "%s, ", pkt_flag_names[n]); 1404 } 1405 if (!WARN_ON_ONCE(pg_result >= end)) { 1406 /* Remove the comma and whitespace at the end */ 1407 *(pg_result - 2) = '\0'; 1408 } 1409 1410 return count; 1411 } 1412 if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) { 1413 max = min(sizeof(pkt_dev->dst_min) - 1, count - i); 1414 len = strn_len(&user_buffer[i], max); 1415 if (len < 0) 1416 return len; 1417 1418 if (copy_from_user(buf, &user_buffer[i], len)) 1419 return -EFAULT; 1420 buf[len] = 0; 1421 if (strcmp(buf, pkt_dev->dst_min) != 0) { 1422 memset(pkt_dev->dst_min, 0, sizeof(pkt_dev->dst_min)); 1423 strcpy(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 memset(pkt_dev->dst_max, 0, sizeof(pkt_dev->dst_max)); 1444 strcpy(pkt_dev->dst_max, buf); 1445 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max); 1446 pkt_dev->cur_daddr = pkt_dev->daddr_max; 1447 } 1448 if (debug) 1449 pr_debug("dst_max set to: %s\n", pkt_dev->dst_max); 1450 1451 sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max); 1452 return count; 1453 } 1454 if (!strcmp(name, "dst6")) { 1455 max = min(sizeof(buf) - 1, count - i); 1456 len = strn_len(&user_buffer[i], max); 1457 if (len < 0) 1458 return len; 1459 1460 pkt_dev->flags |= F_IPV6; 1461 1462 if (copy_from_user(buf, &user_buffer[i], len)) 1463 return -EFAULT; 1464 buf[len] = 0; 1465 1466 in6_pton(buf, -1, pkt_dev->in6_daddr.s6_addr, -1, NULL); 1467 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_daddr); 1468 1469 pkt_dev->cur_in6_daddr = pkt_dev->in6_daddr; 1470 1471 if (debug) 1472 pr_debug("dst6 set to: %s\n", buf); 1473 1474 sprintf(pg_result, "OK: dst6=%s", buf); 1475 return count; 1476 } 1477 if (!strcmp(name, "dst6_min")) { 1478 max = min(sizeof(buf) - 1, count - i); 1479 len = strn_len(&user_buffer[i], max); 1480 if (len < 0) 1481 return len; 1482 1483 pkt_dev->flags |= F_IPV6; 1484 1485 if (copy_from_user(buf, &user_buffer[i], len)) 1486 return -EFAULT; 1487 buf[len] = 0; 1488 1489 in6_pton(buf, -1, pkt_dev->min_in6_daddr.s6_addr, -1, NULL); 1490 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->min_in6_daddr); 1491 1492 pkt_dev->cur_in6_daddr = pkt_dev->min_in6_daddr; 1493 if (debug) 1494 pr_debug("dst6_min set to: %s\n", buf); 1495 1496 sprintf(pg_result, "OK: dst6_min=%s", buf); 1497 return count; 1498 } 1499 if (!strcmp(name, "dst6_max")) { 1500 max = min(sizeof(buf) - 1, count - i); 1501 len = strn_len(&user_buffer[i], max); 1502 if (len < 0) 1503 return len; 1504 1505 pkt_dev->flags |= F_IPV6; 1506 1507 if (copy_from_user(buf, &user_buffer[i], len)) 1508 return -EFAULT; 1509 buf[len] = 0; 1510 1511 in6_pton(buf, -1, pkt_dev->max_in6_daddr.s6_addr, -1, NULL); 1512 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->max_in6_daddr); 1513 1514 if (debug) 1515 pr_debug("dst6_max set to: %s\n", buf); 1516 1517 sprintf(pg_result, "OK: dst6_max=%s", buf); 1518 return count; 1519 } 1520 if (!strcmp(name, "src6")) { 1521 max = min(sizeof(buf) - 1, count - i); 1522 len = strn_len(&user_buffer[i], max); 1523 if (len < 0) 1524 return len; 1525 1526 pkt_dev->flags |= F_IPV6; 1527 1528 if (copy_from_user(buf, &user_buffer[i], len)) 1529 return -EFAULT; 1530 buf[len] = 0; 1531 1532 in6_pton(buf, -1, pkt_dev->in6_saddr.s6_addr, -1, NULL); 1533 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_saddr); 1534 1535 pkt_dev->cur_in6_saddr = pkt_dev->in6_saddr; 1536 1537 if (debug) 1538 pr_debug("src6 set to: %s\n", buf); 1539 1540 sprintf(pg_result, "OK: src6=%s", buf); 1541 return count; 1542 } 1543 if (!strcmp(name, "src_min")) { 1544 max = min(sizeof(pkt_dev->src_min) - 1, count - i); 1545 len = strn_len(&user_buffer[i], max); 1546 if (len < 0) 1547 return len; 1548 1549 if (copy_from_user(buf, &user_buffer[i], len)) 1550 return -EFAULT; 1551 buf[len] = 0; 1552 if (strcmp(buf, pkt_dev->src_min) != 0) { 1553 memset(pkt_dev->src_min, 0, sizeof(pkt_dev->src_min)); 1554 strcpy(pkt_dev->src_min, buf); 1555 pkt_dev->saddr_min = in_aton(pkt_dev->src_min); 1556 pkt_dev->cur_saddr = pkt_dev->saddr_min; 1557 } 1558 if (debug) 1559 pr_debug("src_min set to: %s\n", pkt_dev->src_min); 1560 1561 sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min); 1562 return count; 1563 } 1564 if (!strcmp(name, "src_max")) { 1565 max = min(sizeof(pkt_dev->src_max) - 1, count - i); 1566 len = strn_len(&user_buffer[i], max); 1567 if (len < 0) 1568 return len; 1569 1570 if (copy_from_user(buf, &user_buffer[i], len)) 1571 return -EFAULT; 1572 buf[len] = 0; 1573 if (strcmp(buf, pkt_dev->src_max) != 0) { 1574 memset(pkt_dev->src_max, 0, sizeof(pkt_dev->src_max)); 1575 strcpy(pkt_dev->src_max, buf); 1576 pkt_dev->saddr_max = in_aton(pkt_dev->src_max); 1577 pkt_dev->cur_saddr = pkt_dev->saddr_max; 1578 } 1579 if (debug) 1580 pr_debug("src_max set to: %s\n", pkt_dev->src_max); 1581 1582 sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max); 1583 return count; 1584 } 1585 if (!strcmp(name, "dst_mac")) { 1586 max = min(sizeof(valstr) - 1, count - i); 1587 len = strn_len(&user_buffer[i], max); 1588 if (len < 0) 1589 return len; 1590 1591 memset(valstr, 0, sizeof(valstr)); 1592 if (copy_from_user(valstr, &user_buffer[i], len)) 1593 return -EFAULT; 1594 1595 if (!mac_pton(valstr, pkt_dev->dst_mac)) 1596 return -EINVAL; 1597 /* Set up Dest MAC */ 1598 ether_addr_copy(&pkt_dev->hh[0], pkt_dev->dst_mac); 1599 1600 sprintf(pg_result, "OK: dstmac %pM", pkt_dev->dst_mac); 1601 return count; 1602 } 1603 if (!strcmp(name, "src_mac")) { 1604 max = min(sizeof(valstr) - 1, count - i); 1605 len = strn_len(&user_buffer[i], max); 1606 if (len < 0) 1607 return len; 1608 1609 memset(valstr, 0, sizeof(valstr)); 1610 if (copy_from_user(valstr, &user_buffer[i], len)) 1611 return -EFAULT; 1612 1613 if (!mac_pton(valstr, pkt_dev->src_mac)) 1614 return -EINVAL; 1615 /* Set up Src MAC */ 1616 ether_addr_copy(&pkt_dev->hh[6], pkt_dev->src_mac); 1617 1618 sprintf(pg_result, "OK: srcmac %pM", pkt_dev->src_mac); 1619 return count; 1620 } 1621 1622 if (!strcmp(name, "clear_counters")) { 1623 pktgen_clear_counters(pkt_dev); 1624 sprintf(pg_result, "OK: Clearing counters.\n"); 1625 return count; 1626 } 1627 1628 if (!strcmp(name, "flows")) { 1629 max = min(10, count - i); 1630 len = num_arg(&user_buffer[i], max, &value); 1631 if (len < 0) 1632 return len; 1633 1634 if (value > MAX_CFLOWS) 1635 value = MAX_CFLOWS; 1636 1637 pkt_dev->cflows = value; 1638 sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows); 1639 return count; 1640 } 1641 #ifdef CONFIG_XFRM 1642 if (!strcmp(name, "spi")) { 1643 max = min(10, count - i); 1644 len = num_arg(&user_buffer[i], max, &value); 1645 if (len < 0) 1646 return len; 1647 1648 pkt_dev->spi = value; 1649 sprintf(pg_result, "OK: spi=%u", pkt_dev->spi); 1650 return count; 1651 } 1652 #endif 1653 if (!strcmp(name, "flowlen")) { 1654 max = min(10, count - i); 1655 len = num_arg(&user_buffer[i], max, &value); 1656 if (len < 0) 1657 return len; 1658 1659 pkt_dev->lflow = value; 1660 sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow); 1661 return count; 1662 } 1663 1664 if (!strcmp(name, "queue_map_min")) { 1665 max = min(5, count - i); 1666 len = num_arg(&user_buffer[i], max, &value); 1667 if (len < 0) 1668 return len; 1669 1670 pkt_dev->queue_map_min = value; 1671 sprintf(pg_result, "OK: queue_map_min=%u", pkt_dev->queue_map_min); 1672 return count; 1673 } 1674 1675 if (!strcmp(name, "queue_map_max")) { 1676 max = min(5, count - i); 1677 len = num_arg(&user_buffer[i], max, &value); 1678 if (len < 0) 1679 return len; 1680 1681 pkt_dev->queue_map_max = value; 1682 sprintf(pg_result, "OK: queue_map_max=%u", pkt_dev->queue_map_max); 1683 return count; 1684 } 1685 1686 if (!strcmp(name, "mpls")) { 1687 unsigned int n, cnt; 1688 1689 max = count - i; 1690 len = get_labels(&user_buffer[i], max, pkt_dev); 1691 if (len < 0) 1692 return len; 1693 1694 cnt = sprintf(pg_result, "OK: mpls="); 1695 for (n = 0; n < pkt_dev->nr_labels; n++) 1696 cnt += sprintf(pg_result + cnt, 1697 "%08x%s", ntohl(pkt_dev->labels[n]), 1698 n == pkt_dev->nr_labels-1 ? "" : ","); 1699 1700 if (pkt_dev->nr_labels && pkt_dev->vlan_id != 0xffff) { 1701 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1702 pkt_dev->svlan_id = 0xffff; 1703 1704 if (debug) 1705 pr_debug("VLAN/SVLAN auto turned off\n"); 1706 } 1707 return count; 1708 } 1709 1710 if (!strcmp(name, "vlan_id")) { 1711 max = min(4, count - i); 1712 len = num_arg(&user_buffer[i], max, &value); 1713 if (len < 0) 1714 return len; 1715 1716 if (value <= 4095) { 1717 pkt_dev->vlan_id = value; /* turn on VLAN */ 1718 1719 if (debug) 1720 pr_debug("VLAN turned on\n"); 1721 1722 if (debug && pkt_dev->nr_labels) 1723 pr_debug("MPLS auto turned off\n"); 1724 1725 pkt_dev->nr_labels = 0; /* turn off MPLS */ 1726 sprintf(pg_result, "OK: vlan_id=%u", pkt_dev->vlan_id); 1727 } else { 1728 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1729 pkt_dev->svlan_id = 0xffff; 1730 1731 if (debug) 1732 pr_debug("VLAN/SVLAN turned off\n"); 1733 } 1734 return count; 1735 } 1736 1737 if (!strcmp(name, "vlan_p")) { 1738 max = min(1, count - i); 1739 len = num_arg(&user_buffer[i], max, &value); 1740 if (len < 0) 1741 return len; 1742 1743 if ((value <= 7) && (pkt_dev->vlan_id != 0xffff)) { 1744 pkt_dev->vlan_p = value; 1745 sprintf(pg_result, "OK: vlan_p=%u", pkt_dev->vlan_p); 1746 } else { 1747 sprintf(pg_result, "ERROR: vlan_p must be 0-7"); 1748 } 1749 return count; 1750 } 1751 1752 if (!strcmp(name, "vlan_cfi")) { 1753 max = min(1, count - i); 1754 len = num_arg(&user_buffer[i], max, &value); 1755 if (len < 0) 1756 return len; 1757 1758 if ((value <= 1) && (pkt_dev->vlan_id != 0xffff)) { 1759 pkt_dev->vlan_cfi = value; 1760 sprintf(pg_result, "OK: vlan_cfi=%u", pkt_dev->vlan_cfi); 1761 } else { 1762 sprintf(pg_result, "ERROR: vlan_cfi must be 0-1"); 1763 } 1764 return count; 1765 } 1766 1767 if (!strcmp(name, "svlan_id")) { 1768 max = min(4, count - i); 1769 len = num_arg(&user_buffer[i], max, &value); 1770 if (len < 0) 1771 return len; 1772 1773 if ((value <= 4095) && ((pkt_dev->vlan_id != 0xffff))) { 1774 pkt_dev->svlan_id = value; /* turn on SVLAN */ 1775 1776 if (debug) 1777 pr_debug("SVLAN turned on\n"); 1778 1779 if (debug && pkt_dev->nr_labels) 1780 pr_debug("MPLS auto turned off\n"); 1781 1782 pkt_dev->nr_labels = 0; /* turn off MPLS */ 1783 sprintf(pg_result, "OK: svlan_id=%u", pkt_dev->svlan_id); 1784 } else { 1785 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1786 pkt_dev->svlan_id = 0xffff; 1787 1788 if (debug) 1789 pr_debug("VLAN/SVLAN turned off\n"); 1790 } 1791 return count; 1792 } 1793 1794 if (!strcmp(name, "svlan_p")) { 1795 max = min(1, count - i); 1796 len = num_arg(&user_buffer[i], max, &value); 1797 if (len < 0) 1798 return len; 1799 1800 if ((value <= 7) && (pkt_dev->svlan_id != 0xffff)) { 1801 pkt_dev->svlan_p = value; 1802 sprintf(pg_result, "OK: svlan_p=%u", pkt_dev->svlan_p); 1803 } else { 1804 sprintf(pg_result, "ERROR: svlan_p must be 0-7"); 1805 } 1806 return count; 1807 } 1808 1809 if (!strcmp(name, "svlan_cfi")) { 1810 max = min(1, count - i); 1811 len = num_arg(&user_buffer[i], max, &value); 1812 if (len < 0) 1813 return len; 1814 1815 if ((value <= 1) && (pkt_dev->svlan_id != 0xffff)) { 1816 pkt_dev->svlan_cfi = value; 1817 sprintf(pg_result, "OK: svlan_cfi=%u", pkt_dev->svlan_cfi); 1818 } else { 1819 sprintf(pg_result, "ERROR: svlan_cfi must be 0-1"); 1820 } 1821 return count; 1822 } 1823 1824 if (!strcmp(name, "tos")) { 1825 __u32 tmp_value; 1826 1827 max = min(2, count - i); 1828 len = hex32_arg(&user_buffer[i], max, &tmp_value); 1829 if (len < 0) 1830 return len; 1831 1832 if (len == 2) { 1833 pkt_dev->tos = tmp_value; 1834 sprintf(pg_result, "OK: tos=0x%02x", pkt_dev->tos); 1835 } else { 1836 sprintf(pg_result, "ERROR: tos must be 00-ff"); 1837 } 1838 return count; 1839 } 1840 1841 if (!strcmp(name, "traffic_class")) { 1842 __u32 tmp_value; 1843 1844 max = min(2, count - i); 1845 len = hex32_arg(&user_buffer[i], max, &tmp_value); 1846 if (len < 0) 1847 return len; 1848 1849 if (len == 2) { 1850 pkt_dev->traffic_class = tmp_value; 1851 sprintf(pg_result, "OK: traffic_class=0x%02x", pkt_dev->traffic_class); 1852 } else { 1853 sprintf(pg_result, "ERROR: traffic_class must be 00-ff"); 1854 } 1855 return count; 1856 } 1857 1858 if (!strcmp(name, "skb_priority")) { 1859 max = min(9, count - i); 1860 len = num_arg(&user_buffer[i], max, &value); 1861 if (len < 0) 1862 return len; 1863 1864 pkt_dev->skb_priority = value; 1865 sprintf(pg_result, "OK: skb_priority=%i", 1866 pkt_dev->skb_priority); 1867 return count; 1868 } 1869 1870 sprintf(pkt_dev->result, "No such parameter \"%s\"", name); 1871 return -EINVAL; 1872 } 1873 1874 static int pktgen_if_open(struct inode *inode, struct file *file) 1875 { 1876 return single_open(file, pktgen_if_show, pde_data(inode)); 1877 } 1878 1879 static const struct proc_ops pktgen_if_proc_ops = { 1880 .proc_open = pktgen_if_open, 1881 .proc_read = seq_read, 1882 .proc_lseek = seq_lseek, 1883 .proc_write = pktgen_if_write, 1884 .proc_release = single_release, 1885 }; 1886 1887 static int pktgen_thread_show(struct seq_file *seq, void *v) 1888 { 1889 struct pktgen_thread *t = seq->private; 1890 const struct pktgen_dev *pkt_dev; 1891 1892 BUG_ON(!t); 1893 1894 seq_puts(seq, "Running: "); 1895 1896 rcu_read_lock(); 1897 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 1898 if (pkt_dev->running) 1899 seq_printf(seq, "%s ", pkt_dev->odevname); 1900 1901 seq_puts(seq, "\nStopped: "); 1902 1903 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 1904 if (!pkt_dev->running) 1905 seq_printf(seq, "%s ", pkt_dev->odevname); 1906 1907 if (t->result[0]) 1908 seq_printf(seq, "\nResult: %s\n", t->result); 1909 else 1910 seq_puts(seq, "\nResult: NA\n"); 1911 1912 rcu_read_unlock(); 1913 1914 return 0; 1915 } 1916 1917 static ssize_t pktgen_thread_write(struct file *file, 1918 const char __user *user_buffer, 1919 size_t count, loff_t *offset) 1920 { 1921 struct seq_file *seq = file->private_data; 1922 struct pktgen_thread *t = seq->private; 1923 size_t i, max; 1924 ssize_t len, ret; 1925 char name[40]; 1926 char *pg_result; 1927 1928 if (count < 1) { 1929 // sprintf(pg_result, "Wrong command format"); 1930 return -EINVAL; 1931 } 1932 1933 max = count; 1934 len = count_trail_chars(user_buffer, max); 1935 if (len < 0) 1936 return len; 1937 1938 i = len; 1939 1940 /* Read variable name */ 1941 max = min(sizeof(name) - 1, count - i); 1942 len = strn_len(&user_buffer[i], max); 1943 if (len < 0) 1944 return len; 1945 1946 memset(name, 0, sizeof(name)); 1947 if (copy_from_user(name, &user_buffer[i], len)) 1948 return -EFAULT; 1949 i += len; 1950 1951 max = count - i; 1952 len = count_trail_chars(&user_buffer[i], max); 1953 if (len < 0) 1954 return len; 1955 1956 i += len; 1957 1958 if (debug) 1959 pr_debug("t=%s, count=%lu\n", name, (unsigned long)count); 1960 1961 if (!t) { 1962 pr_err("ERROR: No thread\n"); 1963 ret = -EINVAL; 1964 goto out; 1965 } 1966 1967 pg_result = &(t->result[0]); 1968 1969 if (!strcmp(name, "add_device")) { 1970 char f[32]; 1971 1972 memset(f, 0, 32); 1973 max = min(sizeof(f) - 1, count - i); 1974 len = strn_len(&user_buffer[i], max); 1975 if (len < 0) { 1976 ret = len; 1977 goto out; 1978 } 1979 if (copy_from_user(f, &user_buffer[i], len)) 1980 return -EFAULT; 1981 1982 mutex_lock(&pktgen_thread_lock); 1983 ret = pktgen_add_device(t, f); 1984 mutex_unlock(&pktgen_thread_lock); 1985 if (!ret) { 1986 ret = count; 1987 sprintf(pg_result, "OK: add_device=%s", f); 1988 } else 1989 sprintf(pg_result, "ERROR: can not add device %s", f); 1990 goto out; 1991 } 1992 1993 if (!strcmp(name, "rem_device_all")) { 1994 mutex_lock(&pktgen_thread_lock); 1995 t->control |= T_REMDEVALL; 1996 mutex_unlock(&pktgen_thread_lock); 1997 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */ 1998 ret = count; 1999 sprintf(pg_result, "OK: rem_device_all"); 2000 goto out; 2001 } 2002 2003 if (!strcmp(name, "max_before_softirq")) { 2004 sprintf(pg_result, "OK: Note! max_before_softirq is obsoleted -- Do not use"); 2005 ret = count; 2006 goto out; 2007 } 2008 2009 ret = -EINVAL; 2010 out: 2011 return ret; 2012 } 2013 2014 static int pktgen_thread_open(struct inode *inode, struct file *file) 2015 { 2016 return single_open(file, pktgen_thread_show, pde_data(inode)); 2017 } 2018 2019 static const struct proc_ops pktgen_thread_proc_ops = { 2020 .proc_open = pktgen_thread_open, 2021 .proc_read = seq_read, 2022 .proc_lseek = seq_lseek, 2023 .proc_write = pktgen_thread_write, 2024 .proc_release = single_release, 2025 }; 2026 2027 /* Think find or remove for NN */ 2028 static struct pktgen_dev *__pktgen_NN_threads(const struct pktgen_net *pn, 2029 const char *ifname, int remove) 2030 { 2031 struct pktgen_thread *t; 2032 struct pktgen_dev *pkt_dev = NULL; 2033 bool exact = (remove == FIND); 2034 2035 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 2036 pkt_dev = pktgen_find_dev(t, ifname, exact); 2037 if (pkt_dev) { 2038 if (remove) { 2039 pkt_dev->removal_mark = 1; 2040 t->control |= T_REMDEV; 2041 } 2042 break; 2043 } 2044 } 2045 return pkt_dev; 2046 } 2047 2048 /* 2049 * mark a device for removal 2050 */ 2051 static void pktgen_mark_device(const struct pktgen_net *pn, const char *ifname) 2052 { 2053 struct pktgen_dev *pkt_dev = NULL; 2054 const int max_tries = 10, msec_per_try = 125; 2055 int i = 0; 2056 2057 mutex_lock(&pktgen_thread_lock); 2058 pr_debug("%s: marking %s for removal\n", __func__, ifname); 2059 2060 while (1) { 2061 2062 pkt_dev = __pktgen_NN_threads(pn, ifname, REMOVE); 2063 if (pkt_dev == NULL) 2064 break; /* success */ 2065 2066 mutex_unlock(&pktgen_thread_lock); 2067 pr_debug("%s: waiting for %s to disappear....\n", 2068 __func__, ifname); 2069 schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try)); 2070 mutex_lock(&pktgen_thread_lock); 2071 2072 if (++i >= max_tries) { 2073 pr_err("%s: timed out after waiting %d msec for device %s to be removed\n", 2074 __func__, msec_per_try * i, ifname); 2075 break; 2076 } 2077 2078 } 2079 2080 mutex_unlock(&pktgen_thread_lock); 2081 } 2082 2083 static void pktgen_change_name(const struct pktgen_net *pn, struct net_device *dev) 2084 { 2085 struct pktgen_thread *t; 2086 2087 mutex_lock(&pktgen_thread_lock); 2088 2089 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 2090 struct pktgen_dev *pkt_dev; 2091 2092 if_lock(t); 2093 list_for_each_entry(pkt_dev, &t->if_list, list) { 2094 if (pkt_dev->odev != dev) 2095 continue; 2096 2097 proc_remove(pkt_dev->entry); 2098 2099 pkt_dev->entry = proc_create_data(dev->name, 0600, 2100 pn->proc_dir, 2101 &pktgen_if_proc_ops, 2102 pkt_dev); 2103 if (!pkt_dev->entry) 2104 pr_err("can't move proc entry for '%s'\n", 2105 dev->name); 2106 break; 2107 } 2108 if_unlock(t); 2109 } 2110 mutex_unlock(&pktgen_thread_lock); 2111 } 2112 2113 static int pktgen_device_event(struct notifier_block *unused, 2114 unsigned long event, void *ptr) 2115 { 2116 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 2117 struct pktgen_net *pn = net_generic(dev_net(dev), pg_net_id); 2118 2119 if (pn->pktgen_exiting) 2120 return NOTIFY_DONE; 2121 2122 /* It is OK that we do not hold the group lock right now, 2123 * as we run under the RTNL lock. 2124 */ 2125 2126 switch (event) { 2127 case NETDEV_CHANGENAME: 2128 pktgen_change_name(pn, dev); 2129 break; 2130 2131 case NETDEV_UNREGISTER: 2132 pktgen_mark_device(pn, dev->name); 2133 break; 2134 } 2135 2136 return NOTIFY_DONE; 2137 } 2138 2139 static struct net_device *pktgen_dev_get_by_name(const struct pktgen_net *pn, 2140 struct pktgen_dev *pkt_dev, 2141 const char *ifname) 2142 { 2143 char b[IFNAMSIZ+5]; 2144 int i; 2145 2146 for (i = 0; ifname[i] != '@'; i++) { 2147 if (i == IFNAMSIZ) 2148 break; 2149 2150 b[i] = ifname[i]; 2151 } 2152 b[i] = 0; 2153 2154 return dev_get_by_name(pn->net, b); 2155 } 2156 2157 2158 /* Associate pktgen_dev with a device. */ 2159 2160 static int pktgen_setup_dev(const struct pktgen_net *pn, 2161 struct pktgen_dev *pkt_dev, const char *ifname) 2162 { 2163 struct net_device *odev; 2164 int err; 2165 2166 /* Clean old setups */ 2167 if (pkt_dev->odev) { 2168 netdev_put(pkt_dev->odev, &pkt_dev->dev_tracker); 2169 pkt_dev->odev = NULL; 2170 } 2171 2172 odev = pktgen_dev_get_by_name(pn, pkt_dev, ifname); 2173 if (!odev) { 2174 pr_err("no such netdevice: \"%s\"\n", ifname); 2175 return -ENODEV; 2176 } 2177 2178 if (odev->type != ARPHRD_ETHER && odev->type != ARPHRD_LOOPBACK) { 2179 pr_err("not an ethernet or loopback device: \"%s\"\n", ifname); 2180 err = -EINVAL; 2181 } else if (!netif_running(odev)) { 2182 pr_err("device is down: \"%s\"\n", ifname); 2183 err = -ENETDOWN; 2184 } else { 2185 pkt_dev->odev = odev; 2186 netdev_tracker_alloc(odev, &pkt_dev->dev_tracker, GFP_KERNEL); 2187 return 0; 2188 } 2189 2190 dev_put(odev); 2191 return err; 2192 } 2193 2194 /* Read pkt_dev from the interface and set up internal pktgen_dev 2195 * structure to have the right information to create/send packets 2196 */ 2197 static void pktgen_setup_inject(struct pktgen_dev *pkt_dev) 2198 { 2199 int ntxq; 2200 2201 if (!pkt_dev->odev) { 2202 pr_err("ERROR: pkt_dev->odev == NULL in setup_inject\n"); 2203 sprintf(pkt_dev->result, 2204 "ERROR: pkt_dev->odev == NULL in setup_inject.\n"); 2205 return; 2206 } 2207 2208 /* make sure that we don't pick a non-existing transmit queue */ 2209 ntxq = pkt_dev->odev->real_num_tx_queues; 2210 2211 if (ntxq <= pkt_dev->queue_map_min) { 2212 pr_warn("WARNING: Requested queue_map_min (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n", 2213 pkt_dev->queue_map_min, (ntxq ?: 1) - 1, ntxq, 2214 pkt_dev->odevname); 2215 pkt_dev->queue_map_min = (ntxq ?: 1) - 1; 2216 } 2217 if (pkt_dev->queue_map_max >= ntxq) { 2218 pr_warn("WARNING: Requested queue_map_max (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n", 2219 pkt_dev->queue_map_max, (ntxq ?: 1) - 1, ntxq, 2220 pkt_dev->odevname); 2221 pkt_dev->queue_map_max = (ntxq ?: 1) - 1; 2222 } 2223 2224 /* Default to the interface's mac if not explicitly set. */ 2225 2226 if (is_zero_ether_addr(pkt_dev->src_mac)) 2227 ether_addr_copy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr); 2228 2229 /* Set up Dest MAC */ 2230 ether_addr_copy(&(pkt_dev->hh[0]), pkt_dev->dst_mac); 2231 2232 if (pkt_dev->flags & F_IPV6) { 2233 int i, set = 0, err = 1; 2234 struct inet6_dev *idev; 2235 2236 if (pkt_dev->min_pkt_size == 0) { 2237 pkt_dev->min_pkt_size = 14 + sizeof(struct ipv6hdr) 2238 + sizeof(struct udphdr) 2239 + sizeof(struct pktgen_hdr) 2240 + pkt_dev->pkt_overhead; 2241 } 2242 2243 for (i = 0; i < sizeof(struct in6_addr); i++) 2244 if (pkt_dev->cur_in6_saddr.s6_addr[i]) { 2245 set = 1; 2246 break; 2247 } 2248 2249 if (!set) { 2250 2251 /* 2252 * Use linklevel address if unconfigured. 2253 * 2254 * use ipv6_get_lladdr if/when it's get exported 2255 */ 2256 2257 rcu_read_lock(); 2258 idev = __in6_dev_get(pkt_dev->odev); 2259 if (idev) { 2260 struct inet6_ifaddr *ifp; 2261 2262 read_lock_bh(&idev->lock); 2263 list_for_each_entry(ifp, &idev->addr_list, if_list) { 2264 if ((ifp->scope & IFA_LINK) && 2265 !(ifp->flags & IFA_F_TENTATIVE)) { 2266 pkt_dev->cur_in6_saddr = ifp->addr; 2267 err = 0; 2268 break; 2269 } 2270 } 2271 read_unlock_bh(&idev->lock); 2272 } 2273 rcu_read_unlock(); 2274 if (err) 2275 pr_err("ERROR: IPv6 link address not available\n"); 2276 } 2277 } else { 2278 if (pkt_dev->min_pkt_size == 0) { 2279 pkt_dev->min_pkt_size = 14 + sizeof(struct iphdr) 2280 + sizeof(struct udphdr) 2281 + sizeof(struct pktgen_hdr) 2282 + pkt_dev->pkt_overhead; 2283 } 2284 2285 pkt_dev->saddr_min = 0; 2286 pkt_dev->saddr_max = 0; 2287 if (strlen(pkt_dev->src_min) == 0) { 2288 2289 struct in_device *in_dev; 2290 2291 rcu_read_lock(); 2292 in_dev = __in_dev_get_rcu(pkt_dev->odev); 2293 if (in_dev) { 2294 const struct in_ifaddr *ifa; 2295 2296 ifa = rcu_dereference(in_dev->ifa_list); 2297 if (ifa) { 2298 pkt_dev->saddr_min = ifa->ifa_address; 2299 pkt_dev->saddr_max = pkt_dev->saddr_min; 2300 } 2301 } 2302 rcu_read_unlock(); 2303 } else { 2304 pkt_dev->saddr_min = in_aton(pkt_dev->src_min); 2305 pkt_dev->saddr_max = in_aton(pkt_dev->src_max); 2306 } 2307 2308 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min); 2309 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max); 2310 } 2311 /* Initialize current values. */ 2312 pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size; 2313 if (pkt_dev->min_pkt_size > pkt_dev->max_pkt_size) 2314 pkt_dev->max_pkt_size = pkt_dev->min_pkt_size; 2315 2316 pkt_dev->cur_dst_mac_offset = 0; 2317 pkt_dev->cur_src_mac_offset = 0; 2318 pkt_dev->cur_saddr = pkt_dev->saddr_min; 2319 pkt_dev->cur_daddr = pkt_dev->daddr_min; 2320 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min; 2321 pkt_dev->cur_udp_src = pkt_dev->udp_src_min; 2322 pkt_dev->nflows = 0; 2323 } 2324 2325 2326 static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until) 2327 { 2328 ktime_t start_time, end_time; 2329 s64 remaining; 2330 struct hrtimer_sleeper t; 2331 2332 hrtimer_setup_sleeper_on_stack(&t, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); 2333 hrtimer_set_expires(&t.timer, spin_until); 2334 2335 remaining = ktime_to_ns(hrtimer_expires_remaining(&t.timer)); 2336 if (remaining <= 0) 2337 goto out; 2338 2339 start_time = ktime_get(); 2340 if (remaining < 100000) { 2341 /* for small delays (<100us), just loop until limit is reached */ 2342 do { 2343 end_time = ktime_get(); 2344 } while (ktime_compare(end_time, spin_until) < 0); 2345 } else { 2346 do { 2347 set_current_state(TASK_INTERRUPTIBLE); 2348 hrtimer_sleeper_start_expires(&t, HRTIMER_MODE_ABS); 2349 2350 if (likely(t.task)) 2351 schedule(); 2352 2353 hrtimer_cancel(&t.timer); 2354 } while (t.task && pkt_dev->running && !signal_pending(current)); 2355 __set_current_state(TASK_RUNNING); 2356 end_time = ktime_get(); 2357 } 2358 2359 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(end_time, start_time)); 2360 out: 2361 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay); 2362 destroy_hrtimer_on_stack(&t.timer); 2363 } 2364 2365 static inline void set_pkt_overhead(struct pktgen_dev *pkt_dev) 2366 { 2367 pkt_dev->pkt_overhead = 0; 2368 pkt_dev->pkt_overhead += pkt_dev->nr_labels*sizeof(u32); 2369 pkt_dev->pkt_overhead += VLAN_TAG_SIZE(pkt_dev); 2370 pkt_dev->pkt_overhead += SVLAN_TAG_SIZE(pkt_dev); 2371 } 2372 2373 static inline int f_seen(const struct pktgen_dev *pkt_dev, int flow) 2374 { 2375 return !!(pkt_dev->flows[flow].flags & F_INIT); 2376 } 2377 2378 static inline int f_pick(struct pktgen_dev *pkt_dev) 2379 { 2380 int flow = pkt_dev->curfl; 2381 2382 if (pkt_dev->flags & F_FLOW_SEQ) { 2383 if (pkt_dev->flows[flow].count >= pkt_dev->lflow) { 2384 /* reset time */ 2385 pkt_dev->flows[flow].count = 0; 2386 pkt_dev->flows[flow].flags = 0; 2387 pkt_dev->curfl += 1; 2388 if (pkt_dev->curfl >= pkt_dev->cflows) 2389 pkt_dev->curfl = 0; /*reset */ 2390 } 2391 } else { 2392 flow = get_random_u32_below(pkt_dev->cflows); 2393 pkt_dev->curfl = flow; 2394 2395 if (pkt_dev->flows[flow].count > pkt_dev->lflow) { 2396 pkt_dev->flows[flow].count = 0; 2397 pkt_dev->flows[flow].flags = 0; 2398 } 2399 } 2400 2401 return pkt_dev->curfl; 2402 } 2403 2404 2405 /* If there was already an IPSEC SA, we keep it as is, else 2406 * we go look for it ... 2407 */ 2408 #define DUMMY_MARK 0 2409 static void get_ipsec_sa(struct pktgen_dev *pkt_dev, int flow) 2410 { 2411 #ifdef CONFIG_XFRM 2412 struct xfrm_state *x = pkt_dev->flows[flow].x; 2413 struct pktgen_net *pn = net_generic(dev_net(pkt_dev->odev), pg_net_id); 2414 2415 if (!x) { 2416 2417 if (pkt_dev->spi) { 2418 /* We need as quick as possible to find the right SA 2419 * Searching with minimum criteria to achieve, this. 2420 */ 2421 x = xfrm_state_lookup_byspi(pn->net, htonl(pkt_dev->spi), AF_INET); 2422 } else { 2423 /* slow path: we don't already have xfrm_state */ 2424 x = xfrm_stateonly_find(pn->net, DUMMY_MARK, 0, 2425 (xfrm_address_t *)&pkt_dev->cur_daddr, 2426 (xfrm_address_t *)&pkt_dev->cur_saddr, 2427 AF_INET, 2428 pkt_dev->ipsmode, 2429 pkt_dev->ipsproto, 0); 2430 } 2431 if (x) { 2432 pkt_dev->flows[flow].x = x; 2433 set_pkt_overhead(pkt_dev); 2434 pkt_dev->pkt_overhead += x->props.header_len; 2435 } 2436 2437 } 2438 #endif 2439 } 2440 static void set_cur_queue_map(struct pktgen_dev *pkt_dev) 2441 { 2442 if (pkt_dev->flags & F_QUEUE_MAP_CPU) 2443 pkt_dev->cur_queue_map = smp_processor_id(); 2444 2445 else if (pkt_dev->queue_map_min <= pkt_dev->queue_map_max) { 2446 __u16 t; 2447 2448 if (pkt_dev->flags & F_QUEUE_MAP_RND) { 2449 t = get_random_u32_inclusive(pkt_dev->queue_map_min, 2450 pkt_dev->queue_map_max); 2451 } else { 2452 t = pkt_dev->cur_queue_map + 1; 2453 if (t > pkt_dev->queue_map_max) 2454 t = pkt_dev->queue_map_min; 2455 } 2456 pkt_dev->cur_queue_map = t; 2457 } 2458 pkt_dev->cur_queue_map = pkt_dev->cur_queue_map % pkt_dev->odev->real_num_tx_queues; 2459 } 2460 2461 /* Increment/randomize headers according to flags and current values 2462 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst 2463 */ 2464 static void mod_cur_headers(struct pktgen_dev *pkt_dev) 2465 { 2466 __u32 imn; 2467 __u32 imx; 2468 int flow = 0; 2469 2470 if (pkt_dev->cflows) 2471 flow = f_pick(pkt_dev); 2472 2473 /* Deal with source MAC */ 2474 if (pkt_dev->src_mac_count > 1) { 2475 __u32 mc; 2476 __u32 tmp; 2477 2478 if (pkt_dev->flags & F_MACSRC_RND) 2479 mc = get_random_u32_below(pkt_dev->src_mac_count); 2480 else { 2481 mc = pkt_dev->cur_src_mac_offset++; 2482 if (pkt_dev->cur_src_mac_offset >= 2483 pkt_dev->src_mac_count) 2484 pkt_dev->cur_src_mac_offset = 0; 2485 } 2486 2487 tmp = pkt_dev->src_mac[5] + (mc & 0xFF); 2488 pkt_dev->hh[11] = tmp; 2489 tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8)); 2490 pkt_dev->hh[10] = tmp; 2491 tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8)); 2492 pkt_dev->hh[9] = tmp; 2493 tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8)); 2494 pkt_dev->hh[8] = tmp; 2495 tmp = (pkt_dev->src_mac[1] + (tmp >> 8)); 2496 pkt_dev->hh[7] = tmp; 2497 } 2498 2499 /* Deal with Destination MAC */ 2500 if (pkt_dev->dst_mac_count > 1) { 2501 __u32 mc; 2502 __u32 tmp; 2503 2504 if (pkt_dev->flags & F_MACDST_RND) 2505 mc = get_random_u32_below(pkt_dev->dst_mac_count); 2506 2507 else { 2508 mc = pkt_dev->cur_dst_mac_offset++; 2509 if (pkt_dev->cur_dst_mac_offset >= 2510 pkt_dev->dst_mac_count) { 2511 pkt_dev->cur_dst_mac_offset = 0; 2512 } 2513 } 2514 2515 tmp = pkt_dev->dst_mac[5] + (mc & 0xFF); 2516 pkt_dev->hh[5] = tmp; 2517 tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8)); 2518 pkt_dev->hh[4] = tmp; 2519 tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8)); 2520 pkt_dev->hh[3] = tmp; 2521 tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8)); 2522 pkt_dev->hh[2] = tmp; 2523 tmp = (pkt_dev->dst_mac[1] + (tmp >> 8)); 2524 pkt_dev->hh[1] = tmp; 2525 } 2526 2527 if (pkt_dev->flags & F_MPLS_RND) { 2528 unsigned int i; 2529 2530 for (i = 0; i < pkt_dev->nr_labels; i++) 2531 if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM) 2532 pkt_dev->labels[i] = MPLS_STACK_BOTTOM | 2533 ((__force __be32)get_random_u32() & 2534 htonl(0x000fffff)); 2535 } 2536 2537 if ((pkt_dev->flags & F_VID_RND) && (pkt_dev->vlan_id != 0xffff)) { 2538 pkt_dev->vlan_id = get_random_u32_below(4096); 2539 } 2540 2541 if ((pkt_dev->flags & F_SVID_RND) && (pkt_dev->svlan_id != 0xffff)) { 2542 pkt_dev->svlan_id = get_random_u32_below(4096); 2543 } 2544 2545 if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) { 2546 if (pkt_dev->flags & F_UDPSRC_RND) 2547 pkt_dev->cur_udp_src = get_random_u32_inclusive(pkt_dev->udp_src_min, 2548 pkt_dev->udp_src_max - 1); 2549 2550 else { 2551 pkt_dev->cur_udp_src++; 2552 if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max) 2553 pkt_dev->cur_udp_src = pkt_dev->udp_src_min; 2554 } 2555 } 2556 2557 if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) { 2558 if (pkt_dev->flags & F_UDPDST_RND) { 2559 pkt_dev->cur_udp_dst = get_random_u32_inclusive(pkt_dev->udp_dst_min, 2560 pkt_dev->udp_dst_max - 1); 2561 } else { 2562 pkt_dev->cur_udp_dst++; 2563 if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max) 2564 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min; 2565 } 2566 } 2567 2568 if (!(pkt_dev->flags & F_IPV6)) { 2569 2570 imn = ntohl(pkt_dev->saddr_min); 2571 imx = ntohl(pkt_dev->saddr_max); 2572 if (imn < imx) { 2573 __u32 t; 2574 2575 if (pkt_dev->flags & F_IPSRC_RND) 2576 t = get_random_u32_inclusive(imn, imx - 1); 2577 else { 2578 t = ntohl(pkt_dev->cur_saddr); 2579 t++; 2580 if (t > imx) 2581 t = imn; 2582 2583 } 2584 pkt_dev->cur_saddr = htonl(t); 2585 } 2586 2587 if (pkt_dev->cflows && f_seen(pkt_dev, flow)) { 2588 pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr; 2589 } else { 2590 imn = ntohl(pkt_dev->daddr_min); 2591 imx = ntohl(pkt_dev->daddr_max); 2592 if (imn < imx) { 2593 __u32 t; 2594 __be32 s; 2595 2596 if (pkt_dev->flags & F_IPDST_RND) { 2597 2598 do { 2599 t = get_random_u32_inclusive(imn, imx - 1); 2600 s = htonl(t); 2601 } while (ipv4_is_loopback(s) || 2602 ipv4_is_multicast(s) || 2603 ipv4_is_lbcast(s) || 2604 ipv4_is_zeronet(s) || 2605 ipv4_is_local_multicast(s)); 2606 pkt_dev->cur_daddr = s; 2607 } else { 2608 t = ntohl(pkt_dev->cur_daddr); 2609 t++; 2610 if (t > imx) { 2611 t = imn; 2612 } 2613 pkt_dev->cur_daddr = htonl(t); 2614 } 2615 } 2616 if (pkt_dev->cflows) { 2617 pkt_dev->flows[flow].flags |= F_INIT; 2618 pkt_dev->flows[flow].cur_daddr = 2619 pkt_dev->cur_daddr; 2620 if (pkt_dev->flags & F_IPSEC) 2621 get_ipsec_sa(pkt_dev, flow); 2622 pkt_dev->nflows++; 2623 } 2624 } 2625 } else { /* IPV6 * */ 2626 2627 if (!ipv6_addr_any(&pkt_dev->min_in6_daddr)) { 2628 int i; 2629 2630 /* Only random destinations yet */ 2631 2632 for (i = 0; i < 4; i++) { 2633 pkt_dev->cur_in6_daddr.s6_addr32[i] = 2634 (((__force __be32)get_random_u32() | 2635 pkt_dev->min_in6_daddr.s6_addr32[i]) & 2636 pkt_dev->max_in6_daddr.s6_addr32[i]); 2637 } 2638 } 2639 } 2640 2641 if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) { 2642 __u32 t; 2643 2644 if (pkt_dev->flags & F_TXSIZE_RND) { 2645 t = get_random_u32_inclusive(pkt_dev->min_pkt_size, 2646 pkt_dev->max_pkt_size - 1); 2647 } else { 2648 t = pkt_dev->cur_pkt_size + 1; 2649 if (t > pkt_dev->max_pkt_size) 2650 t = pkt_dev->min_pkt_size; 2651 } 2652 pkt_dev->cur_pkt_size = t; 2653 } else if (pkt_dev->n_imix_entries > 0) { 2654 struct imix_pkt *entry; 2655 __u32 t = get_random_u32_below(IMIX_PRECISION); 2656 __u8 entry_index = pkt_dev->imix_distribution[t]; 2657 2658 entry = &pkt_dev->imix_entries[entry_index]; 2659 entry->count_so_far++; 2660 pkt_dev->cur_pkt_size = entry->size; 2661 } 2662 2663 set_cur_queue_map(pkt_dev); 2664 2665 pkt_dev->flows[flow].count++; 2666 } 2667 2668 static void fill_imix_distribution(struct pktgen_dev *pkt_dev) 2669 { 2670 int cumulative_probabilites[MAX_IMIX_ENTRIES]; 2671 int j = 0; 2672 __u64 cumulative_prob = 0; 2673 __u64 total_weight = 0; 2674 int i = 0; 2675 2676 for (i = 0; i < pkt_dev->n_imix_entries; i++) 2677 total_weight += pkt_dev->imix_entries[i].weight; 2678 2679 /* Fill cumulative_probabilites with sum of normalized probabilities */ 2680 for (i = 0; i < pkt_dev->n_imix_entries - 1; i++) { 2681 cumulative_prob += div64_u64(pkt_dev->imix_entries[i].weight * 2682 IMIX_PRECISION, 2683 total_weight); 2684 cumulative_probabilites[i] = cumulative_prob; 2685 } 2686 cumulative_probabilites[pkt_dev->n_imix_entries - 1] = 100; 2687 2688 for (i = 0; i < IMIX_PRECISION; i++) { 2689 if (i == cumulative_probabilites[j]) 2690 j++; 2691 pkt_dev->imix_distribution[i] = j; 2692 } 2693 } 2694 2695 #ifdef CONFIG_XFRM 2696 static u32 pktgen_dst_metrics[RTAX_MAX + 1] = { 2697 2698 [RTAX_HOPLIMIT] = 0x5, /* Set a static hoplimit */ 2699 }; 2700 2701 static int pktgen_output_ipsec(struct sk_buff *skb, struct pktgen_dev *pkt_dev) 2702 { 2703 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x; 2704 int err = 0; 2705 struct net *net = dev_net(pkt_dev->odev); 2706 2707 if (!x) 2708 return 0; 2709 /* XXX: we dont support tunnel mode for now until 2710 * we resolve the dst issue 2711 */ 2712 if ((x->props.mode != XFRM_MODE_TRANSPORT) && (pkt_dev->spi == 0)) 2713 return 0; 2714 2715 /* But when user specify an valid SPI, transformation 2716 * supports both transport/tunnel mode + ESP/AH type. 2717 */ 2718 if ((x->props.mode == XFRM_MODE_TUNNEL) && (pkt_dev->spi != 0)) 2719 skb->_skb_refdst = (unsigned long)&pkt_dev->xdst.u.dst | SKB_DST_NOREF; 2720 2721 rcu_read_lock_bh(); 2722 err = pktgen_xfrm_outer_mode_output(x, skb); 2723 rcu_read_unlock_bh(); 2724 if (err) { 2725 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR); 2726 goto error; 2727 } 2728 err = x->type->output(x, skb); 2729 if (err) { 2730 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR); 2731 goto error; 2732 } 2733 spin_lock_bh(&x->lock); 2734 x->curlft.bytes += skb->len; 2735 x->curlft.packets++; 2736 spin_unlock_bh(&x->lock); 2737 error: 2738 return err; 2739 } 2740 2741 static void free_SAs(struct pktgen_dev *pkt_dev) 2742 { 2743 if (pkt_dev->cflows) { 2744 /* let go of the SAs if we have them */ 2745 int i; 2746 2747 for (i = 0; i < pkt_dev->cflows; i++) { 2748 struct xfrm_state *x = pkt_dev->flows[i].x; 2749 2750 if (x) { 2751 xfrm_state_put(x); 2752 pkt_dev->flows[i].x = NULL; 2753 } 2754 } 2755 } 2756 } 2757 2758 static int process_ipsec(struct pktgen_dev *pkt_dev, 2759 struct sk_buff *skb, __be16 protocol) 2760 { 2761 if (pkt_dev->flags & F_IPSEC) { 2762 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x; 2763 int nhead = 0; 2764 2765 if (x) { 2766 struct ethhdr *eth; 2767 struct iphdr *iph; 2768 int ret; 2769 2770 nhead = x->props.header_len - skb_headroom(skb); 2771 if (nhead > 0) { 2772 ret = pskb_expand_head(skb, nhead, 0, GFP_ATOMIC); 2773 if (ret < 0) { 2774 pr_err("Error expanding ipsec packet %d\n", 2775 ret); 2776 goto err; 2777 } 2778 } 2779 2780 /* ipsec is not expecting ll header */ 2781 skb_pull(skb, ETH_HLEN); 2782 ret = pktgen_output_ipsec(skb, pkt_dev); 2783 if (ret) { 2784 pr_err("Error creating ipsec packet %d\n", ret); 2785 goto err; 2786 } 2787 /* restore ll */ 2788 eth = skb_push(skb, ETH_HLEN); 2789 memcpy(eth, pkt_dev->hh, 2 * ETH_ALEN); 2790 eth->h_proto = protocol; 2791 2792 /* Update IPv4 header len as well as checksum value */ 2793 iph = ip_hdr(skb); 2794 iph->tot_len = htons(skb->len - ETH_HLEN); 2795 ip_send_check(iph); 2796 } 2797 } 2798 return 1; 2799 err: 2800 kfree_skb(skb); 2801 return 0; 2802 } 2803 #endif 2804 2805 static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev) 2806 { 2807 unsigned int i; 2808 2809 for (i = 0; i < pkt_dev->nr_labels; i++) 2810 *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM; 2811 2812 mpls--; 2813 *mpls |= MPLS_STACK_BOTTOM; 2814 } 2815 2816 static inline __be16 build_tci(unsigned int id, unsigned int cfi, 2817 unsigned int prio) 2818 { 2819 return htons(id | (cfi << 12) | (prio << 13)); 2820 } 2821 2822 static void pktgen_finalize_skb(struct pktgen_dev *pkt_dev, struct sk_buff *skb, 2823 int datalen) 2824 { 2825 struct timespec64 timestamp; 2826 struct pktgen_hdr *pgh; 2827 2828 pgh = skb_put(skb, sizeof(*pgh)); 2829 datalen -= sizeof(*pgh); 2830 2831 if (pkt_dev->nfrags <= 0) { 2832 skb_put_zero(skb, datalen); 2833 } else { 2834 int frags = pkt_dev->nfrags; 2835 int i, len; 2836 int frag_len; 2837 2838 2839 if (frags > MAX_SKB_FRAGS) 2840 frags = MAX_SKB_FRAGS; 2841 len = datalen - frags * PAGE_SIZE; 2842 if (len > 0) { 2843 skb_put_zero(skb, len); 2844 datalen = frags * PAGE_SIZE; 2845 } 2846 2847 i = 0; 2848 frag_len = (datalen/frags) < PAGE_SIZE ? 2849 (datalen/frags) : PAGE_SIZE; 2850 while (datalen > 0) { 2851 if (unlikely(!pkt_dev->page)) { 2852 int node = numa_node_id(); 2853 2854 if (pkt_dev->node >= 0 && (pkt_dev->flags & F_NODE)) 2855 node = pkt_dev->node; 2856 pkt_dev->page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0); 2857 if (!pkt_dev->page) 2858 break; 2859 } 2860 get_page(pkt_dev->page); 2861 2862 /*last fragment, fill rest of data*/ 2863 if (i == (frags - 1)) 2864 skb_frag_fill_page_desc(&skb_shinfo(skb)->frags[i], 2865 pkt_dev->page, 0, 2866 (datalen < PAGE_SIZE ? 2867 datalen : PAGE_SIZE)); 2868 else 2869 skb_frag_fill_page_desc(&skb_shinfo(skb)->frags[i], 2870 pkt_dev->page, 0, frag_len); 2871 2872 datalen -= skb_frag_size(&skb_shinfo(skb)->frags[i]); 2873 skb->len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2874 skb->data_len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2875 i++; 2876 skb_shinfo(skb)->nr_frags = i; 2877 } 2878 } 2879 2880 /* Stamp the time, and sequence number, 2881 * convert them to network byte order 2882 */ 2883 pgh->pgh_magic = htonl(PKTGEN_MAGIC); 2884 pgh->seq_num = htonl(pkt_dev->seq_num); 2885 2886 if (pkt_dev->flags & F_NO_TIMESTAMP) { 2887 pgh->tv_sec = 0; 2888 pgh->tv_usec = 0; 2889 } else { 2890 /* 2891 * pgh->tv_sec wraps in y2106 when interpreted as unsigned 2892 * as done by wireshark, or y2038 when interpreted as signed. 2893 * This is probably harmless, but if anyone wants to improve 2894 * it, we could introduce a variant that puts 64-bit nanoseconds 2895 * into the respective header bytes. 2896 * This would also be slightly faster to read. 2897 */ 2898 ktime_get_real_ts64(×tamp); 2899 pgh->tv_sec = htonl(timestamp.tv_sec); 2900 pgh->tv_usec = htonl(timestamp.tv_nsec / NSEC_PER_USEC); 2901 } 2902 } 2903 2904 static struct sk_buff *pktgen_alloc_skb(struct net_device *dev, 2905 struct pktgen_dev *pkt_dev) 2906 { 2907 unsigned int extralen = LL_RESERVED_SPACE(dev); 2908 struct sk_buff *skb = NULL; 2909 unsigned int size; 2910 2911 size = pkt_dev->cur_pkt_size + 64 + extralen + pkt_dev->pkt_overhead; 2912 if (pkt_dev->flags & F_NODE) { 2913 int node = pkt_dev->node >= 0 ? pkt_dev->node : numa_node_id(); 2914 2915 skb = __alloc_skb(NET_SKB_PAD + size, GFP_NOWAIT, 0, node); 2916 if (likely(skb)) { 2917 skb_reserve(skb, NET_SKB_PAD); 2918 skb->dev = dev; 2919 } 2920 } else { 2921 skb = __netdev_alloc_skb(dev, size, GFP_NOWAIT); 2922 } 2923 2924 /* the caller pre-fetches from skb->data and reserves for the mac hdr */ 2925 if (likely(skb)) 2926 skb_reserve(skb, extralen - 16); 2927 2928 return skb; 2929 } 2930 2931 static struct sk_buff *fill_packet_ipv4(struct net_device *odev, 2932 struct pktgen_dev *pkt_dev) 2933 { 2934 struct sk_buff *skb = NULL; 2935 __u8 *eth; 2936 struct udphdr *udph; 2937 int datalen, iplen; 2938 struct iphdr *iph; 2939 __be16 protocol = htons(ETH_P_IP); 2940 __be32 *mpls; 2941 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */ 2942 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */ 2943 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */ 2944 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */ 2945 u16 queue_map; 2946 2947 if (pkt_dev->nr_labels) 2948 protocol = htons(ETH_P_MPLS_UC); 2949 2950 if (pkt_dev->vlan_id != 0xffff) 2951 protocol = htons(ETH_P_8021Q); 2952 2953 /* Update any of the values, used when we're incrementing various 2954 * fields. 2955 */ 2956 mod_cur_headers(pkt_dev); 2957 queue_map = pkt_dev->cur_queue_map; 2958 2959 skb = pktgen_alloc_skb(odev, pkt_dev); 2960 if (!skb) { 2961 sprintf(pkt_dev->result, "No memory"); 2962 return NULL; 2963 } 2964 2965 prefetchw(skb->data); 2966 skb_reserve(skb, 16); 2967 2968 /* Reserve for ethernet and IP header */ 2969 eth = skb_push(skb, 14); 2970 mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32)); 2971 if (pkt_dev->nr_labels) 2972 mpls_push(mpls, pkt_dev); 2973 2974 if (pkt_dev->vlan_id != 0xffff) { 2975 if (pkt_dev->svlan_id != 0xffff) { 2976 svlan_tci = skb_put(skb, sizeof(__be16)); 2977 *svlan_tci = build_tci(pkt_dev->svlan_id, 2978 pkt_dev->svlan_cfi, 2979 pkt_dev->svlan_p); 2980 svlan_encapsulated_proto = skb_put(skb, 2981 sizeof(__be16)); 2982 *svlan_encapsulated_proto = htons(ETH_P_8021Q); 2983 } 2984 vlan_tci = skb_put(skb, sizeof(__be16)); 2985 *vlan_tci = build_tci(pkt_dev->vlan_id, 2986 pkt_dev->vlan_cfi, 2987 pkt_dev->vlan_p); 2988 vlan_encapsulated_proto = skb_put(skb, sizeof(__be16)); 2989 *vlan_encapsulated_proto = htons(ETH_P_IP); 2990 } 2991 2992 skb_reset_mac_header(skb); 2993 skb_set_network_header(skb, skb->len); 2994 iph = skb_put(skb, sizeof(struct iphdr)); 2995 2996 skb_set_transport_header(skb, skb->len); 2997 udph = skb_put(skb, sizeof(struct udphdr)); 2998 skb_set_queue_mapping(skb, queue_map); 2999 skb->priority = pkt_dev->skb_priority; 3000 3001 memcpy(eth, pkt_dev->hh, 12); 3002 *(__be16 *)ð[12] = protocol; 3003 3004 /* Eth + IPh + UDPh + mpls */ 3005 datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 - 3006 pkt_dev->pkt_overhead; 3007 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) 3008 datalen = sizeof(struct pktgen_hdr); 3009 3010 udph->source = htons(pkt_dev->cur_udp_src); 3011 udph->dest = htons(pkt_dev->cur_udp_dst); 3012 udph->len = htons(datalen + 8); /* DATA + udphdr */ 3013 udph->check = 0; 3014 3015 iph->ihl = 5; 3016 iph->version = 4; 3017 iph->ttl = 32; 3018 iph->tos = pkt_dev->tos; 3019 iph->protocol = IPPROTO_UDP; /* UDP */ 3020 iph->saddr = pkt_dev->cur_saddr; 3021 iph->daddr = pkt_dev->cur_daddr; 3022 iph->id = htons(pkt_dev->ip_id); 3023 pkt_dev->ip_id++; 3024 iph->frag_off = 0; 3025 iplen = 20 + 8 + datalen; 3026 iph->tot_len = htons(iplen); 3027 ip_send_check(iph); 3028 skb->protocol = protocol; 3029 skb->dev = odev; 3030 skb->pkt_type = PACKET_HOST; 3031 3032 pktgen_finalize_skb(pkt_dev, skb, datalen); 3033 3034 if (!(pkt_dev->flags & F_UDPCSUM)) { 3035 skb->ip_summed = CHECKSUM_NONE; 3036 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM)) { 3037 skb->ip_summed = CHECKSUM_PARTIAL; 3038 skb->csum = 0; 3039 udp4_hwcsum(skb, iph->saddr, iph->daddr); 3040 } else { 3041 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), datalen + 8, 0); 3042 3043 /* add protocol-dependent pseudo-header */ 3044 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr, 3045 datalen + 8, IPPROTO_UDP, csum); 3046 3047 if (udph->check == 0) 3048 udph->check = CSUM_MANGLED_0; 3049 } 3050 3051 #ifdef CONFIG_XFRM 3052 if (!process_ipsec(pkt_dev, skb, protocol)) 3053 return NULL; 3054 #endif 3055 3056 return skb; 3057 } 3058 3059 static struct sk_buff *fill_packet_ipv6(struct net_device *odev, 3060 struct pktgen_dev *pkt_dev) 3061 { 3062 struct sk_buff *skb = NULL; 3063 __u8 *eth; 3064 struct udphdr *udph; 3065 int datalen, udplen; 3066 struct ipv6hdr *iph; 3067 __be16 protocol = htons(ETH_P_IPV6); 3068 __be32 *mpls; 3069 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */ 3070 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */ 3071 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */ 3072 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */ 3073 u16 queue_map; 3074 3075 if (pkt_dev->nr_labels) 3076 protocol = htons(ETH_P_MPLS_UC); 3077 3078 if (pkt_dev->vlan_id != 0xffff) 3079 protocol = htons(ETH_P_8021Q); 3080 3081 /* Update any of the values, used when we're incrementing various 3082 * fields. 3083 */ 3084 mod_cur_headers(pkt_dev); 3085 queue_map = pkt_dev->cur_queue_map; 3086 3087 skb = pktgen_alloc_skb(odev, pkt_dev); 3088 if (!skb) { 3089 sprintf(pkt_dev->result, "No memory"); 3090 return NULL; 3091 } 3092 3093 prefetchw(skb->data); 3094 skb_reserve(skb, 16); 3095 3096 /* Reserve for ethernet and IP header */ 3097 eth = skb_push(skb, 14); 3098 mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32)); 3099 if (pkt_dev->nr_labels) 3100 mpls_push(mpls, pkt_dev); 3101 3102 if (pkt_dev->vlan_id != 0xffff) { 3103 if (pkt_dev->svlan_id != 0xffff) { 3104 svlan_tci = skb_put(skb, sizeof(__be16)); 3105 *svlan_tci = build_tci(pkt_dev->svlan_id, 3106 pkt_dev->svlan_cfi, 3107 pkt_dev->svlan_p); 3108 svlan_encapsulated_proto = skb_put(skb, 3109 sizeof(__be16)); 3110 *svlan_encapsulated_proto = htons(ETH_P_8021Q); 3111 } 3112 vlan_tci = skb_put(skb, sizeof(__be16)); 3113 *vlan_tci = build_tci(pkt_dev->vlan_id, 3114 pkt_dev->vlan_cfi, 3115 pkt_dev->vlan_p); 3116 vlan_encapsulated_proto = skb_put(skb, sizeof(__be16)); 3117 *vlan_encapsulated_proto = htons(ETH_P_IPV6); 3118 } 3119 3120 skb_reset_mac_header(skb); 3121 skb_set_network_header(skb, skb->len); 3122 iph = skb_put(skb, sizeof(struct ipv6hdr)); 3123 3124 skb_set_transport_header(skb, skb->len); 3125 udph = skb_put(skb, sizeof(struct udphdr)); 3126 skb_set_queue_mapping(skb, queue_map); 3127 skb->priority = pkt_dev->skb_priority; 3128 3129 memcpy(eth, pkt_dev->hh, 12); 3130 *(__be16 *) ð[12] = protocol; 3131 3132 /* Eth + IPh + UDPh + mpls */ 3133 datalen = pkt_dev->cur_pkt_size - 14 - 3134 sizeof(struct ipv6hdr) - sizeof(struct udphdr) - 3135 pkt_dev->pkt_overhead; 3136 3137 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) { 3138 datalen = sizeof(struct pktgen_hdr); 3139 net_info_ratelimited("increased datalen to %d\n", datalen); 3140 } 3141 3142 udplen = datalen + sizeof(struct udphdr); 3143 udph->source = htons(pkt_dev->cur_udp_src); 3144 udph->dest = htons(pkt_dev->cur_udp_dst); 3145 udph->len = htons(udplen); 3146 udph->check = 0; 3147 3148 *(__be32 *) iph = htonl(0x60000000); /* Version + flow */ 3149 3150 if (pkt_dev->traffic_class) { 3151 /* Version + traffic class + flow (0) */ 3152 *(__be32 *)iph |= htonl(0x60000000 | (pkt_dev->traffic_class << 20)); 3153 } 3154 3155 iph->hop_limit = 32; 3156 3157 iph->payload_len = htons(udplen); 3158 iph->nexthdr = IPPROTO_UDP; 3159 3160 iph->daddr = pkt_dev->cur_in6_daddr; 3161 iph->saddr = pkt_dev->cur_in6_saddr; 3162 3163 skb->protocol = protocol; 3164 skb->dev = odev; 3165 skb->pkt_type = PACKET_HOST; 3166 3167 pktgen_finalize_skb(pkt_dev, skb, datalen); 3168 3169 if (!(pkt_dev->flags & F_UDPCSUM)) { 3170 skb->ip_summed = CHECKSUM_NONE; 3171 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM)) { 3172 skb->ip_summed = CHECKSUM_PARTIAL; 3173 skb->csum_start = skb_transport_header(skb) - skb->head; 3174 skb->csum_offset = offsetof(struct udphdr, check); 3175 udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0); 3176 } else { 3177 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), udplen, 0); 3178 3179 /* add protocol-dependent pseudo-header */ 3180 udph->check = csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, csum); 3181 3182 if (udph->check == 0) 3183 udph->check = CSUM_MANGLED_0; 3184 } 3185 3186 return skb; 3187 } 3188 3189 static struct sk_buff *fill_packet(struct net_device *odev, 3190 struct pktgen_dev *pkt_dev) 3191 { 3192 if (pkt_dev->flags & F_IPV6) 3193 return fill_packet_ipv6(odev, pkt_dev); 3194 else 3195 return fill_packet_ipv4(odev, pkt_dev); 3196 } 3197 3198 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev) 3199 { 3200 pkt_dev->seq_num = 1; 3201 pkt_dev->idle_acc = 0; 3202 pkt_dev->sofar = 0; 3203 pkt_dev->tx_bytes = 0; 3204 pkt_dev->errors = 0; 3205 } 3206 3207 /* Set up structure for sending pkts, clear counters */ 3208 3209 static void pktgen_run(struct pktgen_thread *t) 3210 { 3211 struct pktgen_dev *pkt_dev; 3212 int started = 0; 3213 3214 func_enter(); 3215 3216 rcu_read_lock(); 3217 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3218 3219 /* 3220 * setup odev and create initial packet. 3221 */ 3222 pktgen_setup_inject(pkt_dev); 3223 3224 if (pkt_dev->odev) { 3225 pktgen_clear_counters(pkt_dev); 3226 pkt_dev->skb = NULL; 3227 pkt_dev->started_at = pkt_dev->next_tx = ktime_get(); 3228 3229 set_pkt_overhead(pkt_dev); 3230 3231 strcpy(pkt_dev->result, "Starting"); 3232 pkt_dev->running = 1; /* Cranke yeself! */ 3233 started++; 3234 } else 3235 strcpy(pkt_dev->result, "Error starting"); 3236 } 3237 rcu_read_unlock(); 3238 if (started) 3239 t->control &= ~(T_STOP); 3240 } 3241 3242 static void pktgen_handle_all_threads(struct pktgen_net *pn, u32 flags) 3243 { 3244 struct pktgen_thread *t; 3245 3246 mutex_lock(&pktgen_thread_lock); 3247 3248 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3249 t->control |= (flags); 3250 3251 mutex_unlock(&pktgen_thread_lock); 3252 } 3253 3254 static void pktgen_stop_all_threads(struct pktgen_net *pn) 3255 { 3256 func_enter(); 3257 3258 pktgen_handle_all_threads(pn, T_STOP); 3259 } 3260 3261 static int thread_is_running(const struct pktgen_thread *t) 3262 { 3263 const struct pktgen_dev *pkt_dev; 3264 3265 rcu_read_lock(); 3266 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 3267 if (pkt_dev->running) { 3268 rcu_read_unlock(); 3269 return 1; 3270 } 3271 rcu_read_unlock(); 3272 return 0; 3273 } 3274 3275 static int pktgen_wait_thread_run(struct pktgen_thread *t) 3276 { 3277 while (thread_is_running(t)) { 3278 3279 /* note: 't' will still be around even after the unlock/lock 3280 * cycle because pktgen_thread threads are only cleared at 3281 * net exit 3282 */ 3283 mutex_unlock(&pktgen_thread_lock); 3284 msleep_interruptible(100); 3285 mutex_lock(&pktgen_thread_lock); 3286 3287 if (signal_pending(current)) 3288 goto signal; 3289 } 3290 return 1; 3291 signal: 3292 return 0; 3293 } 3294 3295 static int pktgen_wait_all_threads_run(struct pktgen_net *pn) 3296 { 3297 struct pktgen_thread *t; 3298 int sig = 1; 3299 3300 /* prevent from racing with rmmod */ 3301 if (!try_module_get(THIS_MODULE)) 3302 return sig; 3303 3304 mutex_lock(&pktgen_thread_lock); 3305 3306 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 3307 sig = pktgen_wait_thread_run(t); 3308 if (sig == 0) 3309 break; 3310 } 3311 3312 if (sig == 0) 3313 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3314 t->control |= (T_STOP); 3315 3316 mutex_unlock(&pktgen_thread_lock); 3317 module_put(THIS_MODULE); 3318 return sig; 3319 } 3320 3321 static void pktgen_run_all_threads(struct pktgen_net *pn) 3322 { 3323 func_enter(); 3324 3325 pktgen_handle_all_threads(pn, T_RUN); 3326 3327 /* Propagate thread->control */ 3328 schedule_timeout_interruptible(msecs_to_jiffies(125)); 3329 3330 pktgen_wait_all_threads_run(pn); 3331 } 3332 3333 static void pktgen_reset_all_threads(struct pktgen_net *pn) 3334 { 3335 func_enter(); 3336 3337 pktgen_handle_all_threads(pn, T_REMDEVALL); 3338 3339 /* Propagate thread->control */ 3340 schedule_timeout_interruptible(msecs_to_jiffies(125)); 3341 3342 pktgen_wait_all_threads_run(pn); 3343 } 3344 3345 static void show_results(struct pktgen_dev *pkt_dev, int nr_frags) 3346 { 3347 __u64 bps, mbps, pps; 3348 char *p = pkt_dev->result; 3349 ktime_t elapsed = ktime_sub(pkt_dev->stopped_at, 3350 pkt_dev->started_at); 3351 ktime_t idle = ns_to_ktime(pkt_dev->idle_acc); 3352 3353 p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n", 3354 (unsigned long long)ktime_to_us(elapsed), 3355 (unsigned long long)ktime_to_us(ktime_sub(elapsed, idle)), 3356 (unsigned long long)ktime_to_us(idle), 3357 (unsigned long long)pkt_dev->sofar, 3358 pkt_dev->cur_pkt_size, nr_frags); 3359 3360 pps = div64_u64(pkt_dev->sofar * NSEC_PER_SEC, 3361 ktime_to_ns(elapsed)); 3362 3363 if (pkt_dev->n_imix_entries > 0) { 3364 int i; 3365 struct imix_pkt *entry; 3366 3367 bps = 0; 3368 for (i = 0; i < pkt_dev->n_imix_entries; i++) { 3369 entry = &pkt_dev->imix_entries[i]; 3370 bps += entry->size * entry->count_so_far; 3371 } 3372 bps = div64_u64(bps * 8 * NSEC_PER_SEC, ktime_to_ns(elapsed)); 3373 } else { 3374 bps = pps * 8 * pkt_dev->cur_pkt_size; 3375 } 3376 3377 mbps = bps; 3378 do_div(mbps, 1000000); 3379 p += sprintf(p, " %llupps %lluMb/sec (%llubps) errors: %llu", 3380 (unsigned long long)pps, 3381 (unsigned long long)mbps, 3382 (unsigned long long)bps, 3383 (unsigned long long)pkt_dev->errors); 3384 } 3385 3386 /* Set stopped-at timer, remove from running list, do counters & statistics */ 3387 static int pktgen_stop_device(struct pktgen_dev *pkt_dev) 3388 { 3389 int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1; 3390 3391 if (!pkt_dev->running) { 3392 pr_warn("interface: %s is already stopped\n", 3393 pkt_dev->odevname); 3394 return -EINVAL; 3395 } 3396 3397 pkt_dev->running = 0; 3398 kfree_skb(pkt_dev->skb); 3399 pkt_dev->skb = NULL; 3400 pkt_dev->stopped_at = ktime_get(); 3401 3402 show_results(pkt_dev, nr_frags); 3403 3404 return 0; 3405 } 3406 3407 static struct pktgen_dev *next_to_run(struct pktgen_thread *t) 3408 { 3409 struct pktgen_dev *pkt_dev, *best = NULL; 3410 3411 rcu_read_lock(); 3412 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3413 if (!pkt_dev->running) 3414 continue; 3415 if (best == NULL) 3416 best = pkt_dev; 3417 else if (ktime_compare(pkt_dev->next_tx, best->next_tx) < 0) 3418 best = pkt_dev; 3419 } 3420 rcu_read_unlock(); 3421 3422 return best; 3423 } 3424 3425 static void pktgen_stop(struct pktgen_thread *t) 3426 { 3427 struct pktgen_dev *pkt_dev; 3428 3429 func_enter(); 3430 3431 rcu_read_lock(); 3432 3433 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3434 pktgen_stop_device(pkt_dev); 3435 } 3436 3437 rcu_read_unlock(); 3438 } 3439 3440 /* 3441 * one of our devices needs to be removed - find it 3442 * and remove it 3443 */ 3444 static void pktgen_rem_one_if(struct pktgen_thread *t) 3445 { 3446 struct list_head *q, *n; 3447 struct pktgen_dev *cur; 3448 3449 func_enter(); 3450 3451 list_for_each_safe(q, n, &t->if_list) { 3452 cur = list_entry(q, struct pktgen_dev, list); 3453 3454 if (!cur->removal_mark) 3455 continue; 3456 3457 kfree_skb(cur->skb); 3458 cur->skb = NULL; 3459 3460 pktgen_remove_device(t, cur); 3461 3462 break; 3463 } 3464 } 3465 3466 static void pktgen_rem_all_ifs(struct pktgen_thread *t) 3467 { 3468 struct list_head *q, *n; 3469 struct pktgen_dev *cur; 3470 3471 func_enter(); 3472 3473 /* Remove all devices, free mem */ 3474 3475 list_for_each_safe(q, n, &t->if_list) { 3476 cur = list_entry(q, struct pktgen_dev, list); 3477 3478 kfree_skb(cur->skb); 3479 cur->skb = NULL; 3480 3481 pktgen_remove_device(t, cur); 3482 } 3483 } 3484 3485 static void pktgen_rem_thread(struct pktgen_thread *t) 3486 { 3487 /* Remove from the thread list */ 3488 remove_proc_entry(t->tsk->comm, t->net->proc_dir); 3489 } 3490 3491 static void pktgen_resched(struct pktgen_dev *pkt_dev) 3492 { 3493 ktime_t idle_start = ktime_get(); 3494 3495 schedule(); 3496 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start)); 3497 } 3498 3499 static void pktgen_wait_for_skb(struct pktgen_dev *pkt_dev) 3500 { 3501 ktime_t idle_start = ktime_get(); 3502 3503 while (refcount_read(&(pkt_dev->skb->users)) != 1) { 3504 if (signal_pending(current)) 3505 break; 3506 3507 if (need_resched()) 3508 pktgen_resched(pkt_dev); 3509 else 3510 cpu_relax(); 3511 } 3512 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start)); 3513 } 3514 3515 static void pktgen_xmit(struct pktgen_dev *pkt_dev) 3516 { 3517 bool skb_shared = !!(READ_ONCE(pkt_dev->flags) & F_SHARED); 3518 struct net_device *odev = pkt_dev->odev; 3519 struct netdev_queue *txq; 3520 unsigned int burst = 1; 3521 struct sk_buff *skb; 3522 int clone_skb = 0; 3523 int ret; 3524 3525 /* If 'skb_shared' is false, the read of possible 3526 * new values (if any) for 'burst' and 'clone_skb' will be skipped to 3527 * prevent some concurrent changes from slipping in. And the stabilized 3528 * config will be read in during the next run of pktgen_xmit. 3529 */ 3530 if (skb_shared) { 3531 burst = READ_ONCE(pkt_dev->burst); 3532 clone_skb = READ_ONCE(pkt_dev->clone_skb); 3533 } 3534 3535 /* If device is offline, then don't send */ 3536 if (unlikely(!netif_running(odev) || !netif_carrier_ok(odev))) { 3537 pktgen_stop_device(pkt_dev); 3538 return; 3539 } 3540 3541 /* This is max DELAY, this has special meaning of 3542 * "never transmit" 3543 */ 3544 if (unlikely(pkt_dev->delay == ULLONG_MAX)) { 3545 pkt_dev->next_tx = ktime_add_ns(ktime_get(), ULONG_MAX); 3546 return; 3547 } 3548 3549 /* If no skb or clone count exhausted then get new one */ 3550 if (!pkt_dev->skb || (pkt_dev->last_ok && 3551 ++pkt_dev->clone_count >= clone_skb)) { 3552 /* build a new pkt */ 3553 kfree_skb(pkt_dev->skb); 3554 3555 pkt_dev->skb = fill_packet(odev, pkt_dev); 3556 if (pkt_dev->skb == NULL) { 3557 pr_err("ERROR: couldn't allocate skb in fill_packet\n"); 3558 schedule(); 3559 pkt_dev->clone_count--; /* back out increment, OOM */ 3560 return; 3561 } 3562 pkt_dev->last_pkt_size = pkt_dev->skb->len; 3563 pkt_dev->clone_count = 0; /* reset counter */ 3564 } 3565 3566 if (pkt_dev->delay && pkt_dev->last_ok) 3567 spin(pkt_dev, pkt_dev->next_tx); 3568 3569 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) { 3570 skb = pkt_dev->skb; 3571 skb->protocol = eth_type_trans(skb, skb->dev); 3572 if (skb_shared) 3573 refcount_add(burst, &skb->users); 3574 local_bh_disable(); 3575 do { 3576 ret = netif_receive_skb(skb); 3577 if (ret == NET_RX_DROP) 3578 pkt_dev->errors++; 3579 pkt_dev->sofar++; 3580 pkt_dev->seq_num++; 3581 if (unlikely(!skb_shared)) { 3582 pkt_dev->skb = NULL; 3583 break; 3584 } 3585 if (refcount_read(&skb->users) != burst) { 3586 /* skb was queued by rps/rfs or taps, 3587 * so cannot reuse this skb 3588 */ 3589 WARN_ON(refcount_sub_and_test(burst - 1, &skb->users)); 3590 /* get out of the loop and wait 3591 * until skb is consumed 3592 */ 3593 break; 3594 } 3595 /* skb was 'freed' by stack, so clean few 3596 * bits and reuse it 3597 */ 3598 skb_reset_redirect(skb); 3599 } while (--burst > 0); 3600 goto out; /* Skips xmit_mode M_START_XMIT */ 3601 } else if (pkt_dev->xmit_mode == M_QUEUE_XMIT) { 3602 local_bh_disable(); 3603 if (skb_shared) 3604 refcount_inc(&pkt_dev->skb->users); 3605 3606 ret = dev_queue_xmit(pkt_dev->skb); 3607 3608 if (!skb_shared && dev_xmit_complete(ret)) 3609 pkt_dev->skb = NULL; 3610 3611 switch (ret) { 3612 case NET_XMIT_SUCCESS: 3613 pkt_dev->sofar++; 3614 pkt_dev->seq_num++; 3615 pkt_dev->tx_bytes += pkt_dev->last_pkt_size; 3616 break; 3617 case NET_XMIT_DROP: 3618 case NET_XMIT_CN: 3619 /* These are all valid return codes for a qdisc but 3620 * indicate packets are being dropped or will likely 3621 * be dropped soon. 3622 */ 3623 case NETDEV_TX_BUSY: 3624 /* qdisc may call dev_hard_start_xmit directly in cases 3625 * where no queues exist e.g. loopback device, virtual 3626 * devices, etc. In this case we need to handle 3627 * NETDEV_TX_ codes. 3628 */ 3629 default: 3630 pkt_dev->errors++; 3631 net_info_ratelimited("%s xmit error: %d\n", 3632 pkt_dev->odevname, ret); 3633 break; 3634 } 3635 goto out; 3636 } 3637 3638 txq = skb_get_tx_queue(odev, pkt_dev->skb); 3639 3640 local_bh_disable(); 3641 3642 HARD_TX_LOCK(odev, txq, smp_processor_id()); 3643 3644 if (unlikely(netif_xmit_frozen_or_drv_stopped(txq))) { 3645 pkt_dev->last_ok = 0; 3646 goto unlock; 3647 } 3648 if (skb_shared) 3649 refcount_add(burst, &pkt_dev->skb->users); 3650 3651 xmit_more: 3652 ret = netdev_start_xmit(pkt_dev->skb, odev, txq, --burst > 0); 3653 3654 if (!skb_shared && dev_xmit_complete(ret)) 3655 pkt_dev->skb = NULL; 3656 3657 switch (ret) { 3658 case NETDEV_TX_OK: 3659 pkt_dev->last_ok = 1; 3660 pkt_dev->sofar++; 3661 pkt_dev->seq_num++; 3662 pkt_dev->tx_bytes += pkt_dev->last_pkt_size; 3663 if (burst > 0 && !netif_xmit_frozen_or_drv_stopped(txq)) 3664 goto xmit_more; 3665 break; 3666 case NET_XMIT_DROP: 3667 case NET_XMIT_CN: 3668 /* skb has been consumed */ 3669 pkt_dev->errors++; 3670 break; 3671 default: /* Drivers are not supposed to return other values! */ 3672 net_info_ratelimited("%s xmit error: %d\n", 3673 pkt_dev->odevname, ret); 3674 pkt_dev->errors++; 3675 fallthrough; 3676 case NETDEV_TX_BUSY: 3677 /* Retry it next time */ 3678 if (skb_shared) 3679 refcount_dec(&pkt_dev->skb->users); 3680 pkt_dev->last_ok = 0; 3681 } 3682 if (unlikely(burst)) 3683 WARN_ON(refcount_sub_and_test(burst, &pkt_dev->skb->users)); 3684 unlock: 3685 HARD_TX_UNLOCK(odev, txq); 3686 3687 out: 3688 local_bh_enable(); 3689 3690 /* If pkt_dev->count is zero, then run forever */ 3691 if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) { 3692 if (pkt_dev->skb) 3693 pktgen_wait_for_skb(pkt_dev); 3694 3695 /* Done with this */ 3696 pktgen_stop_device(pkt_dev); 3697 } 3698 } 3699 3700 /* 3701 * Main loop of the thread goes here 3702 */ 3703 3704 static int pktgen_thread_worker(void *arg) 3705 { 3706 struct pktgen_thread *t = arg; 3707 struct pktgen_dev *pkt_dev = NULL; 3708 int cpu = t->cpu; 3709 3710 WARN_ON_ONCE(smp_processor_id() != cpu); 3711 3712 init_waitqueue_head(&t->queue); 3713 complete(&t->start_done); 3714 3715 pr_debug("starting pktgen/%d: pid=%d\n", cpu, task_pid_nr(current)); 3716 3717 set_freezable(); 3718 3719 while (!kthread_should_stop()) { 3720 pkt_dev = next_to_run(t); 3721 3722 if (unlikely(!pkt_dev && t->control == 0)) { 3723 if (t->net->pktgen_exiting) 3724 break; 3725 wait_event_freezable_timeout(t->queue, 3726 t->control != 0, HZ / 10); 3727 continue; 3728 } 3729 3730 if (likely(pkt_dev)) { 3731 pktgen_xmit(pkt_dev); 3732 3733 if (need_resched()) 3734 pktgen_resched(pkt_dev); 3735 else 3736 cpu_relax(); 3737 } 3738 3739 if (t->control & T_STOP) { 3740 pktgen_stop(t); 3741 t->control &= ~(T_STOP); 3742 } 3743 3744 if (t->control & T_RUN) { 3745 pktgen_run(t); 3746 t->control &= ~(T_RUN); 3747 } 3748 3749 if (t->control & T_REMDEVALL) { 3750 pktgen_rem_all_ifs(t); 3751 t->control &= ~(T_REMDEVALL); 3752 } 3753 3754 if (t->control & T_REMDEV) { 3755 pktgen_rem_one_if(t); 3756 t->control &= ~(T_REMDEV); 3757 } 3758 3759 try_to_freeze(); 3760 } 3761 3762 pr_debug("%s stopping all device\n", t->tsk->comm); 3763 pktgen_stop(t); 3764 3765 pr_debug("%s removing all device\n", t->tsk->comm); 3766 pktgen_rem_all_ifs(t); 3767 3768 pr_debug("%s removing thread\n", t->tsk->comm); 3769 pktgen_rem_thread(t); 3770 3771 return 0; 3772 } 3773 3774 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t, 3775 const char *ifname, bool exact) 3776 { 3777 struct pktgen_dev *p, *pkt_dev = NULL; 3778 size_t len = strlen(ifname); 3779 3780 rcu_read_lock(); 3781 list_for_each_entry_rcu(p, &t->if_list, list) 3782 if (strncmp(p->odevname, ifname, len) == 0) { 3783 if (p->odevname[len]) { 3784 if (exact || p->odevname[len] != '@') 3785 continue; 3786 } 3787 pkt_dev = p; 3788 break; 3789 } 3790 3791 rcu_read_unlock(); 3792 pr_debug("find_dev(%s) returning %p\n", ifname, pkt_dev); 3793 return pkt_dev; 3794 } 3795 3796 /* 3797 * Adds a dev at front of if_list. 3798 */ 3799 3800 static int add_dev_to_thread(struct pktgen_thread *t, 3801 struct pktgen_dev *pkt_dev) 3802 { 3803 int rv = 0; 3804 3805 /* This function cannot be called concurrently, as its called 3806 * under pktgen_thread_lock mutex, but it can run from 3807 * userspace on another CPU than the kthread. The if_lock() 3808 * is used here to sync with concurrent instances of 3809 * _rem_dev_from_if_list() invoked via kthread, which is also 3810 * updating the if_list 3811 */ 3812 if_lock(t); 3813 3814 if (pkt_dev->pg_thread) { 3815 pr_err("ERROR: already assigned to a thread\n"); 3816 rv = -EBUSY; 3817 goto out; 3818 } 3819 3820 pkt_dev->running = 0; 3821 pkt_dev->pg_thread = t; 3822 list_add_rcu(&pkt_dev->list, &t->if_list); 3823 3824 out: 3825 if_unlock(t); 3826 return rv; 3827 } 3828 3829 /* Called under thread lock */ 3830 3831 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname) 3832 { 3833 struct pktgen_dev *pkt_dev; 3834 int err; 3835 int node = cpu_to_node(t->cpu); 3836 3837 /* We don't allow a device to be on several threads */ 3838 3839 pkt_dev = __pktgen_NN_threads(t->net, ifname, FIND); 3840 if (pkt_dev) { 3841 pr_err("ERROR: interface already used\n"); 3842 return -EBUSY; 3843 } 3844 3845 pkt_dev = kzalloc_node(sizeof(struct pktgen_dev), GFP_KERNEL, node); 3846 if (!pkt_dev) 3847 return -ENOMEM; 3848 3849 strcpy(pkt_dev->odevname, ifname); 3850 pkt_dev->flows = vzalloc_node(array_size(MAX_CFLOWS, 3851 sizeof(struct flow_state)), 3852 node); 3853 if (pkt_dev->flows == NULL) { 3854 kfree(pkt_dev); 3855 return -ENOMEM; 3856 } 3857 3858 pkt_dev->removal_mark = 0; 3859 pkt_dev->nfrags = 0; 3860 pkt_dev->delay = pg_delay_d; 3861 pkt_dev->count = pg_count_d; 3862 pkt_dev->sofar = 0; 3863 pkt_dev->udp_src_min = 9; /* sink port */ 3864 pkt_dev->udp_src_max = 9; 3865 pkt_dev->udp_dst_min = 9; 3866 pkt_dev->udp_dst_max = 9; 3867 pkt_dev->vlan_p = 0; 3868 pkt_dev->vlan_cfi = 0; 3869 pkt_dev->vlan_id = 0xffff; 3870 pkt_dev->svlan_p = 0; 3871 pkt_dev->svlan_cfi = 0; 3872 pkt_dev->svlan_id = 0xffff; 3873 pkt_dev->burst = 1; 3874 pkt_dev->node = NUMA_NO_NODE; 3875 pkt_dev->flags = F_SHARED; /* SKB shared by default */ 3876 3877 err = pktgen_setup_dev(t->net, pkt_dev, ifname); 3878 if (err) 3879 goto out1; 3880 if (pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING) 3881 pkt_dev->clone_skb = pg_clone_skb_d; 3882 3883 pkt_dev->entry = proc_create_data(ifname, 0600, t->net->proc_dir, 3884 &pktgen_if_proc_ops, pkt_dev); 3885 if (!pkt_dev->entry) { 3886 pr_err("cannot create %s/%s procfs entry\n", 3887 PG_PROC_DIR, ifname); 3888 err = -EINVAL; 3889 goto out2; 3890 } 3891 #ifdef CONFIG_XFRM 3892 pkt_dev->ipsmode = XFRM_MODE_TRANSPORT; 3893 pkt_dev->ipsproto = IPPROTO_ESP; 3894 3895 /* xfrm tunnel mode needs additional dst to extract outer 3896 * ip header protocol/ttl/id field, here create a phony one. 3897 * instead of looking for a valid rt, which definitely hurting 3898 * performance under such circumstance. 3899 */ 3900 pkt_dev->dstops.family = AF_INET; 3901 pkt_dev->xdst.u.dst.dev = pkt_dev->odev; 3902 dst_init_metrics(&pkt_dev->xdst.u.dst, pktgen_dst_metrics, false); 3903 pkt_dev->xdst.child = &pkt_dev->xdst.u.dst; 3904 pkt_dev->xdst.u.dst.ops = &pkt_dev->dstops; 3905 #endif 3906 3907 return add_dev_to_thread(t, pkt_dev); 3908 out2: 3909 netdev_put(pkt_dev->odev, &pkt_dev->dev_tracker); 3910 out1: 3911 #ifdef CONFIG_XFRM 3912 free_SAs(pkt_dev); 3913 #endif 3914 vfree(pkt_dev->flows); 3915 kfree(pkt_dev); 3916 return err; 3917 } 3918 3919 static int __net_init pktgen_create_thread(int cpu, struct pktgen_net *pn) 3920 { 3921 struct pktgen_thread *t; 3922 struct proc_dir_entry *pe; 3923 struct task_struct *p; 3924 3925 t = kzalloc_node(sizeof(struct pktgen_thread), GFP_KERNEL, 3926 cpu_to_node(cpu)); 3927 if (!t) { 3928 pr_err("ERROR: out of memory, can't create new thread\n"); 3929 return -ENOMEM; 3930 } 3931 3932 mutex_init(&t->if_lock); 3933 t->cpu = cpu; 3934 3935 INIT_LIST_HEAD(&t->if_list); 3936 3937 list_add_tail(&t->th_list, &pn->pktgen_threads); 3938 init_completion(&t->start_done); 3939 3940 p = kthread_create_on_cpu(pktgen_thread_worker, t, cpu, "kpktgend_%d"); 3941 if (IS_ERR(p)) { 3942 pr_err("kthread_create_on_node() failed for cpu %d\n", t->cpu); 3943 list_del(&t->th_list); 3944 kfree(t); 3945 return PTR_ERR(p); 3946 } 3947 3948 t->tsk = p; 3949 3950 pe = proc_create_data(t->tsk->comm, 0600, pn->proc_dir, 3951 &pktgen_thread_proc_ops, t); 3952 if (!pe) { 3953 pr_err("cannot create %s/%s procfs entry\n", 3954 PG_PROC_DIR, t->tsk->comm); 3955 kthread_stop(p); 3956 list_del(&t->th_list); 3957 kfree(t); 3958 return -EINVAL; 3959 } 3960 3961 t->net = pn; 3962 get_task_struct(p); 3963 wake_up_process(p); 3964 wait_for_completion(&t->start_done); 3965 3966 return 0; 3967 } 3968 3969 /* 3970 * Removes a device from the thread if_list. 3971 */ 3972 static void _rem_dev_from_if_list(struct pktgen_thread *t, 3973 struct pktgen_dev *pkt_dev) 3974 { 3975 struct list_head *q, *n; 3976 struct pktgen_dev *p; 3977 3978 if_lock(t); 3979 list_for_each_safe(q, n, &t->if_list) { 3980 p = list_entry(q, struct pktgen_dev, list); 3981 if (p == pkt_dev) 3982 list_del_rcu(&p->list); 3983 } 3984 if_unlock(t); 3985 } 3986 3987 static int pktgen_remove_device(struct pktgen_thread *t, 3988 struct pktgen_dev *pkt_dev) 3989 { 3990 pr_debug("remove_device pkt_dev=%p\n", pkt_dev); 3991 3992 if (pkt_dev->running) { 3993 pr_warn("WARNING: trying to remove a running interface, stopping it now\n"); 3994 pktgen_stop_device(pkt_dev); 3995 } 3996 3997 /* Dis-associate from the interface */ 3998 3999 if (pkt_dev->odev) { 4000 netdev_put(pkt_dev->odev, &pkt_dev->dev_tracker); 4001 pkt_dev->odev = NULL; 4002 } 4003 4004 /* Remove proc before if_list entry, because add_device uses 4005 * list to determine if interface already exist, avoid race 4006 * with proc_create_data() 4007 */ 4008 proc_remove(pkt_dev->entry); 4009 4010 /* And update the thread if_list */ 4011 _rem_dev_from_if_list(t, pkt_dev); 4012 4013 #ifdef CONFIG_XFRM 4014 free_SAs(pkt_dev); 4015 #endif 4016 vfree(pkt_dev->flows); 4017 if (pkt_dev->page) 4018 put_page(pkt_dev->page); 4019 kfree_rcu(pkt_dev, rcu); 4020 return 0; 4021 } 4022 4023 static int __net_init pg_net_init(struct net *net) 4024 { 4025 struct pktgen_net *pn = net_generic(net, pg_net_id); 4026 struct proc_dir_entry *pe; 4027 int cpu, ret = 0; 4028 4029 pn->net = net; 4030 INIT_LIST_HEAD(&pn->pktgen_threads); 4031 pn->pktgen_exiting = false; 4032 pn->proc_dir = proc_mkdir(PG_PROC_DIR, pn->net->proc_net); 4033 if (!pn->proc_dir) { 4034 pr_warn("cannot create /proc/net/%s\n", PG_PROC_DIR); 4035 return -ENODEV; 4036 } 4037 pe = proc_create(PGCTRL, 0600, pn->proc_dir, &pktgen_proc_ops); 4038 if (pe == NULL) { 4039 pr_err("cannot create %s procfs entry\n", PGCTRL); 4040 ret = -EINVAL; 4041 goto remove; 4042 } 4043 4044 cpus_read_lock(); 4045 for_each_online_cpu(cpu) { 4046 int err; 4047 4048 err = pktgen_create_thread(cpu, pn); 4049 if (err) 4050 pr_warn("Cannot create thread for cpu %d (%d)\n", 4051 cpu, err); 4052 } 4053 cpus_read_unlock(); 4054 4055 if (list_empty(&pn->pktgen_threads)) { 4056 pr_err("Initialization failed for all threads\n"); 4057 ret = -ENODEV; 4058 goto remove_entry; 4059 } 4060 4061 return 0; 4062 4063 remove_entry: 4064 remove_proc_entry(PGCTRL, pn->proc_dir); 4065 remove: 4066 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net); 4067 return ret; 4068 } 4069 4070 static void __net_exit pg_net_exit(struct net *net) 4071 { 4072 struct pktgen_net *pn = net_generic(net, pg_net_id); 4073 struct pktgen_thread *t; 4074 struct list_head *q, *n; 4075 LIST_HEAD(list); 4076 4077 /* Stop all interfaces & threads */ 4078 pn->pktgen_exiting = true; 4079 4080 mutex_lock(&pktgen_thread_lock); 4081 list_splice_init(&pn->pktgen_threads, &list); 4082 mutex_unlock(&pktgen_thread_lock); 4083 4084 list_for_each_safe(q, n, &list) { 4085 t = list_entry(q, struct pktgen_thread, th_list); 4086 list_del(&t->th_list); 4087 kthread_stop_put(t->tsk); 4088 kfree(t); 4089 } 4090 4091 remove_proc_entry(PGCTRL, pn->proc_dir); 4092 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net); 4093 } 4094 4095 static struct pernet_operations pg_net_ops = { 4096 .init = pg_net_init, 4097 .exit = pg_net_exit, 4098 .id = &pg_net_id, 4099 .size = sizeof(struct pktgen_net), 4100 }; 4101 4102 static int __init pg_init(void) 4103 { 4104 int ret = 0; 4105 4106 pr_info("%s", version); 4107 ret = register_pernet_subsys(&pg_net_ops); 4108 if (ret) 4109 return ret; 4110 ret = register_netdevice_notifier(&pktgen_notifier_block); 4111 if (ret) 4112 unregister_pernet_subsys(&pg_net_ops); 4113 4114 return ret; 4115 } 4116 4117 static void __exit pg_cleanup(void) 4118 { 4119 unregister_netdevice_notifier(&pktgen_notifier_block); 4120 unregister_pernet_subsys(&pg_net_ops); 4121 /* Don't need rcu_barrier() due to use of kfree_rcu() */ 4122 } 4123 4124 module_init(pg_init); 4125 module_exit(pg_cleanup); 4126 4127 MODULE_AUTHOR("Robert Olsson <robert.olsson@its.uu.se>"); 4128 MODULE_DESCRIPTION("Packet Generator tool"); 4129 MODULE_LICENSE("GPL"); 4130 MODULE_VERSION(VERSION); 4131 module_param(pg_count_d, int, 0); 4132 MODULE_PARM_DESC(pg_count_d, "Default number of packets to inject"); 4133 module_param(pg_delay_d, int, 0); 4134 MODULE_PARM_DESC(pg_delay_d, "Default delay between packets (nanoseconds)"); 4135 module_param(pg_clone_skb_d, int, 0); 4136 MODULE_PARM_DESC(pg_clone_skb_d, "Default number of copies of the same packet"); 4137 module_param(debug, int, 0); 4138 MODULE_PARM_DESC(debug, "Enable debugging of pktgen module"); 4139