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