1 // SPDX-License-Identifier: GPL-2.0-or-later
4 * Copyright 2001, 2002 by Robert Olsson <robert.olsson@its.uu.se>
5 * Uppsala University and
6 * Swedish University of Agricultural Sciences
8 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
9 * Ben Greear <greearb@candelatech.com>
10 * Jens Låås <jens.laas@data.slu.se>
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.
18 * Additional hacking by:
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>
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
49 * Also moved to /proc/net/pktgen/
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>
56 * Integrated to 2.5.x 021029 --Lucio Maciel (luciomaciel@zipmail.com.br)
58 * 021124 Finished major redesign and rewrite for new functionality.
59 * See Documentation/networking/pktgen.rst for how to use this.
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
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).
70 * By design there should only be *one* "controlling" process. In practice
71 * multiple write accesses gives unpredictable result. Understood by "write"
72 * to /proc gives result code thats should be read be the "writer".
73 * For practical use this should be no problem.
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.
79 * Fix refcount off by one if first packet fails, potential null deref,
82 * First "ranges" functionality for ipv6 030726 --ro
84 * Included flow support. 030802 ANK.
86 * Fixed unaligned access on IA-64 Grant Grundler <grundler@parisc-linux.org>
88 * Remove if fix from added Harald Welte <laforge@netfilter.org> 040419
89 * ia64 compilation fix from Aron Griffis <aron@hp.com> 040604
91 * New xmit() return, do_div and misc clean up by Stephen Hemminger
92 * <shemminger@osdl.org> 040923
94 * Randy Dunlap fixed u64 printk compiler warning
96 * Remove FCS from BW calculation. Lennert Buytenhek <buytenh@wantstofly.org>
97 * New time handling. Lennert Buytenhek <buytenh@wantstofly.org> 041213
99 * Corrections from Nikolai Malykh (nmalykh@bilim.com)
100 * Removed unused flags F_SET_SRCMAC & F_SET_SRCIP 041230
102 * interruptible_sleep_on_timeout() replaced Nishanth Aravamudan <nacc@us.ibm.com>
105 * MPLS support by Steven Whitehouse <steve@chygwyn.com>
107 * 802.1Q/Q-in-Q support by Francesco Fondelli (FF) <francesco.fondelli@gmail.com>
109 * Fixed src_mac command to set source mac of packet to value specified in
110 * command by Adit Ranadive <adit.262@gmail.com>
113 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
159 #include <net/ip6_checksum.h>
160 #include <net/addrconf.h>
162 #include <net/xfrm.h>
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>
172 #include <asm/div64.h> /* do_div */
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 */
182 #define func_enter() pr_debug("entering %s\n", __func__);
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*/ \
204 #define pf(flag) flag##_SHIFT,
210 /* Device flag bits */
211 #define pf(flag) static const __u32 F_##flag = (1<<flag##_SHIFT);
215 #define pf(flag) __stringify(flag),
216 static char *pkt_flag_names[] = {
221 #define NR_PKT_FLAGS ARRAY_SIZE(pkt_flag_names)
223 /* Thread control flag bits */
224 #define T_STOP (1<<0) /* Stop run */
225 #define T_RUN (1<<1) /* Start run */
226 #define T_REMDEVALL (1<<2) /* Remove all devs */
227 #define T_REMDEV (1<<3) /* Remove one dev */
230 #define M_START_XMIT 0 /* Default normal TX */
231 #define M_NETIF_RECEIVE 1 /* Inject packets into stack */
232 #define M_QUEUE_XMIT 2 /* Inject packet into qdisc */
234 /* If lock -- protects updating of if_list */
235 #define if_lock(t) mutex_lock(&(t->if_lock));
236 #define if_unlock(t) mutex_unlock(&(t->if_lock));
238 /* Used to help with determining the pkts on receive */
239 #define PKTGEN_MAGIC 0xbe9be955
240 #define PG_PROC_DIR "pktgen"
241 #define PGCTRL "pgctrl"
243 #define MAX_CFLOWS 65536
245 #define VLAN_TAG_SIZE(x) ((x)->vlan_id == 0xffff ? 0 : 4)
246 #define SVLAN_TAG_SIZE(x) ((x)->svlan_id == 0xffff ? 0 : 4)
258 struct xfrm_state *x;
264 #define F_INIT (1<<0) /* flow has been initialized */
268 * Try to keep frequent/infrequent used vars. separated.
270 struct proc_dir_entry *entry; /* proc file */
271 struct pktgen_thread *pg_thread;/* the owner */
272 struct list_head list; /* chaining in the thread's run-queue */
273 struct rcu_head rcu; /* freed by RCU */
275 int running; /* if false, the test will stop */
277 /* If min != max, then we will either do a linear iteration, or
278 * we will do a random selection from within the range.
284 int pkt_overhead; /* overhead for MPLS, VLANs, IPSEC etc */
286 int removal_mark; /* non-zero => the device is marked for
287 * removal by worker thread */
290 u64 delay; /* nano-seconds */
292 __u64 count; /* Default No packets to send */
293 __u64 sofar; /* How many pkts we've sent so far */
294 __u64 tx_bytes; /* How many bytes we've transmitted */
295 __u64 errors; /* Errors when trying to transmit, */
297 /* runtime counters relating to clone_skb */
300 int last_ok; /* Was last skb sent?
301 * Or a failed transmit of some sort?
302 * This will keep sequence numbers in order
307 u64 idle_acc; /* nano-seconds */
312 * Use multiple SKBs during packet gen.
313 * If this number is greater than 1, then
314 * that many copies of the same packet will be
315 * sent before a new packet is allocated.
316 * If you want to send 1024 identical packets
317 * before creating a new packet,
318 * set clone_skb to 1024.
321 char dst_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
322 char dst_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
323 char src_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
324 char src_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
326 struct in6_addr in6_saddr;
327 struct in6_addr in6_daddr;
328 struct in6_addr cur_in6_daddr;
329 struct in6_addr cur_in6_saddr;
331 struct in6_addr min_in6_daddr;
332 struct in6_addr max_in6_daddr;
333 struct in6_addr min_in6_saddr;
334 struct in6_addr max_in6_saddr;
336 /* If we're doing ranges, random or incremental, then this
337 * defines the min/max for those ranges.
339 __be32 saddr_min; /* inclusive, source IP address */
340 __be32 saddr_max; /* exclusive, source IP address */
341 __be32 daddr_min; /* inclusive, dest IP address */
342 __be32 daddr_max; /* exclusive, dest IP address */
344 __u16 udp_src_min; /* inclusive, source UDP port */
345 __u16 udp_src_max; /* exclusive, source UDP port */
346 __u16 udp_dst_min; /* inclusive, dest UDP port */
347 __u16 udp_dst_max; /* exclusive, dest UDP port */
350 __u8 tos; /* six MSB of (former) IPv4 TOS
351 are for dscp codepoint */
352 __u8 traffic_class; /* ditto for the (former) Traffic Class in IPv6
353 (see RFC 3260, sec. 4) */
356 unsigned int n_imix_entries;
357 struct imix_pkt imix_entries[MAX_IMIX_ENTRIES];
358 /* Maps 0-IMIX_PRECISION range to imix_entry based on probability*/
359 __u8 imix_distribution[IMIX_PRECISION];
362 unsigned int nr_labels; /* Depth of stack, 0 = no MPLS */
363 __be32 labels[MAX_MPLS_LABELS];
365 /* VLAN/SVLAN (802.1Q/Q-in-Q) */
368 __u16 vlan_id; /* 0xffff means no vlan tag */
372 __u16 svlan_id; /* 0xffff means no svlan tag */
374 __u32 src_mac_count; /* How many MACs to iterate through */
375 __u32 dst_mac_count; /* How many MACs to iterate through */
377 unsigned char dst_mac[ETH_ALEN];
378 unsigned char src_mac[ETH_ALEN];
380 __u32 cur_dst_mac_offset;
381 __u32 cur_src_mac_offset;
393 0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB,
395 We fill in SRC address later
396 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
400 __u16 pad; /* pad out the hh struct to an even 16 bytes */
402 struct sk_buff *skb; /* skb we are to transmit next, used for when we
403 * are transmitting the same one multiple times
405 struct net_device *odev; /* The out-going device.
406 * Note that the device should have it's
407 * pg_info pointer pointing back to this
409 * Set when the user specifies the out-going
410 * device name (not when the inject is
411 * started as it used to do.)
413 netdevice_tracker dev_tracker;
415 struct flow_state *flows;
416 unsigned int cflows; /* Concurrent flows (config) */
417 unsigned int lflow; /* Flow length (config) */
418 unsigned int nflows; /* accumulated flows (stats) */
419 unsigned int curfl; /* current sequenced flow (state)*/
423 __u32 skb_priority; /* skb priority field */
424 unsigned int burst; /* number of duplicated packets to burst */
425 int node; /* Memory node */
428 __u8 ipsmode; /* IPSEC mode (config) */
429 __u8 ipsproto; /* IPSEC type (config) */
431 struct xfrm_dst xdst;
432 struct dst_ops dstops;
445 static unsigned int pg_net_id __read_mostly;
449 struct proc_dir_entry *proc_dir;
450 struct list_head pktgen_threads;
454 struct pktgen_thread {
455 struct mutex if_lock; /* for list of devices */
456 struct list_head if_list; /* All device here */
457 struct list_head th_list;
458 struct task_struct *tsk;
461 /* Field for thread to receive "posted" events terminate,
467 wait_queue_head_t queue;
468 struct completion start_done;
469 struct pktgen_net *net;
475 static const char version[] =
476 "Packet Generator for packet performance testing. "
477 "Version: " VERSION "\n";
479 static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i);
480 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname);
481 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
482 const char *ifname, bool exact);
483 static int pktgen_device_event(struct notifier_block *, unsigned long, void *);
484 static void pktgen_run_all_threads(struct pktgen_net *pn);
485 static void pktgen_reset_all_threads(struct pktgen_net *pn);
486 static void pktgen_stop_all_threads(struct pktgen_net *pn);
488 static void pktgen_stop(struct pktgen_thread *t);
489 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev);
490 static void fill_imix_distribution(struct pktgen_dev *pkt_dev);
492 /* Module parameters, defaults. */
493 static int pg_count_d __read_mostly = 1000;
494 static int pg_delay_d __read_mostly;
495 static int pg_clone_skb_d __read_mostly;
496 static int debug __read_mostly;
498 static DEFINE_MUTEX(pktgen_thread_lock);
500 static struct notifier_block pktgen_notifier_block = {
501 .notifier_call = pktgen_device_event,
505 * /proc handling functions
509 static int pgctrl_show(struct seq_file *seq, void *v)
511 seq_puts(seq, version);
515 static ssize_t pgctrl_write(struct file *file, const char __user *buf,
516 size_t count, loff_t *ppos)
519 struct pktgen_net *pn = net_generic(current->nsproxy->net_ns, pg_net_id);
521 if (!capable(CAP_NET_ADMIN))
527 if (count > sizeof(data))
528 count = sizeof(data);
530 if (copy_from_user(data, buf, count))
533 data[count - 1] = 0; /* Strip trailing '\n' and terminate string */
535 if (!strcmp(data, "stop"))
536 pktgen_stop_all_threads(pn);
537 else if (!strcmp(data, "start"))
538 pktgen_run_all_threads(pn);
539 else if (!strcmp(data, "reset"))
540 pktgen_reset_all_threads(pn);
547 static int pgctrl_open(struct inode *inode, struct file *file)
549 return single_open(file, pgctrl_show, pde_data(inode));
552 static const struct proc_ops pktgen_proc_ops = {
553 .proc_open = pgctrl_open,
554 .proc_read = seq_read,
555 .proc_lseek = seq_lseek,
556 .proc_write = pgctrl_write,
557 .proc_release = single_release,
560 static int pktgen_if_show(struct seq_file *seq, void *v)
562 const struct pktgen_dev *pkt_dev = seq->private;
568 "Params: count %llu min_pkt_size: %u max_pkt_size: %u\n",
569 (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size,
570 pkt_dev->max_pkt_size);
572 if (pkt_dev->n_imix_entries > 0) {
573 seq_puts(seq, " imix_weights: ");
574 for (i = 0; i < pkt_dev->n_imix_entries; i++) {
575 seq_printf(seq, "%llu,%llu ",
576 pkt_dev->imix_entries[i].size,
577 pkt_dev->imix_entries[i].weight);
583 " frags: %d delay: %llu clone_skb: %d ifname: %s\n",
584 pkt_dev->nfrags, (unsigned long long) pkt_dev->delay,
585 pkt_dev->clone_skb, pkt_dev->odevname);
587 seq_printf(seq, " flows: %u flowlen: %u\n", pkt_dev->cflows,
591 " queue_map_min: %u queue_map_max: %u\n",
592 pkt_dev->queue_map_min,
593 pkt_dev->queue_map_max);
595 if (pkt_dev->skb_priority)
596 seq_printf(seq, " skb_priority: %u\n",
597 pkt_dev->skb_priority);
599 if (pkt_dev->flags & F_IPV6) {
601 " saddr: %pI6c min_saddr: %pI6c max_saddr: %pI6c\n"
602 " daddr: %pI6c min_daddr: %pI6c max_daddr: %pI6c\n",
604 &pkt_dev->min_in6_saddr, &pkt_dev->max_in6_saddr,
606 &pkt_dev->min_in6_daddr, &pkt_dev->max_in6_daddr);
609 " dst_min: %s dst_max: %s\n",
610 pkt_dev->dst_min, pkt_dev->dst_max);
612 " src_min: %s src_max: %s\n",
613 pkt_dev->src_min, pkt_dev->src_max);
616 seq_puts(seq, " src_mac: ");
618 seq_printf(seq, "%pM ",
619 is_zero_ether_addr(pkt_dev->src_mac) ?
620 pkt_dev->odev->dev_addr : pkt_dev->src_mac);
622 seq_puts(seq, "dst_mac: ");
623 seq_printf(seq, "%pM\n", pkt_dev->dst_mac);
626 " udp_src_min: %d udp_src_max: %d"
627 " udp_dst_min: %d udp_dst_max: %d\n",
628 pkt_dev->udp_src_min, pkt_dev->udp_src_max,
629 pkt_dev->udp_dst_min, pkt_dev->udp_dst_max);
632 " src_mac_count: %d dst_mac_count: %d\n",
633 pkt_dev->src_mac_count, pkt_dev->dst_mac_count);
635 if (pkt_dev->nr_labels) {
636 seq_puts(seq, " mpls: ");
637 for (i = 0; i < pkt_dev->nr_labels; i++)
638 seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]),
639 i == pkt_dev->nr_labels-1 ? "\n" : ", ");
642 if (pkt_dev->vlan_id != 0xffff)
643 seq_printf(seq, " vlan_id: %u vlan_p: %u vlan_cfi: %u\n",
644 pkt_dev->vlan_id, pkt_dev->vlan_p,
647 if (pkt_dev->svlan_id != 0xffff)
648 seq_printf(seq, " svlan_id: %u vlan_p: %u vlan_cfi: %u\n",
649 pkt_dev->svlan_id, pkt_dev->svlan_p,
653 seq_printf(seq, " tos: 0x%02x\n", pkt_dev->tos);
655 if (pkt_dev->traffic_class)
656 seq_printf(seq, " traffic_class: 0x%02x\n", pkt_dev->traffic_class);
658 if (pkt_dev->burst > 1)
659 seq_printf(seq, " burst: %d\n", pkt_dev->burst);
661 if (pkt_dev->node >= 0)
662 seq_printf(seq, " node: %d\n", pkt_dev->node);
664 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE)
665 seq_puts(seq, " xmit_mode: netif_receive\n");
666 else if (pkt_dev->xmit_mode == M_QUEUE_XMIT)
667 seq_puts(seq, " xmit_mode: xmit_queue\n");
669 seq_puts(seq, " Flags: ");
671 for (i = 0; i < NR_PKT_FLAGS; i++) {
673 if (!pkt_dev->cflows)
676 if (pkt_dev->flags & (1 << i))
677 seq_printf(seq, "%s ", pkt_flag_names[i]);
678 else if (i == F_FLOW_SEQ)
679 seq_puts(seq, "FLOW_RND ");
682 if (i == F_IPSEC && pkt_dev->spi)
683 seq_printf(seq, "spi:%u", pkt_dev->spi);
689 /* not really stopped, more like last-running-at */
690 stopped = pkt_dev->running ? ktime_get() : pkt_dev->stopped_at;
691 idle = pkt_dev->idle_acc;
692 do_div(idle, NSEC_PER_USEC);
695 "Current:\n pkts-sofar: %llu errors: %llu\n",
696 (unsigned long long)pkt_dev->sofar,
697 (unsigned long long)pkt_dev->errors);
699 if (pkt_dev->n_imix_entries > 0) {
702 seq_puts(seq, " imix_size_counts: ");
703 for (i = 0; i < pkt_dev->n_imix_entries; i++) {
704 seq_printf(seq, "%llu,%llu ",
705 pkt_dev->imix_entries[i].size,
706 pkt_dev->imix_entries[i].count_so_far);
712 " started: %lluus stopped: %lluus idle: %lluus\n",
713 (unsigned long long) ktime_to_us(pkt_dev->started_at),
714 (unsigned long long) ktime_to_us(stopped),
715 (unsigned long long) idle);
718 " seq_num: %d cur_dst_mac_offset: %d cur_src_mac_offset: %d\n",
719 pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset,
720 pkt_dev->cur_src_mac_offset);
722 if (pkt_dev->flags & F_IPV6) {
723 seq_printf(seq, " cur_saddr: %pI6c cur_daddr: %pI6c\n",
724 &pkt_dev->cur_in6_saddr,
725 &pkt_dev->cur_in6_daddr);
727 seq_printf(seq, " cur_saddr: %pI4 cur_daddr: %pI4\n",
728 &pkt_dev->cur_saddr, &pkt_dev->cur_daddr);
730 seq_printf(seq, " cur_udp_dst: %d cur_udp_src: %d\n",
731 pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src);
733 seq_printf(seq, " cur_queue_map: %u\n", pkt_dev->cur_queue_map);
735 seq_printf(seq, " flows: %u\n", pkt_dev->nflows);
737 if (pkt_dev->result[0])
738 seq_printf(seq, "Result: %s\n", pkt_dev->result);
740 seq_puts(seq, "Result: Idle\n");
746 static int hex32_arg(const char __user *user_buffer, unsigned long maxlen,
752 for (; i < maxlen; i++) {
756 if (get_user(c, &user_buffer[i]))
758 value = hex_to_bin(c);
767 static int count_trail_chars(const char __user * user_buffer,
772 for (i = 0; i < maxlen; i++) {
774 if (get_user(c, &user_buffer[i]))
792 static long num_arg(const char __user *user_buffer, unsigned long maxlen,
798 for (i = 0; i < maxlen; i++) {
800 if (get_user(c, &user_buffer[i]))
802 if ((c >= '0') && (c <= '9')) {
811 static int strn_len(const char __user * user_buffer, unsigned int maxlen)
815 for (i = 0; i < maxlen; i++) {
817 if (get_user(c, &user_buffer[i]))
834 /* Parses imix entries from user buffer.
835 * The user buffer should consist of imix entries separated by spaces
836 * where each entry consists of size and weight delimited by commas.
837 * "size1,weight_1 size2,weight_2 ... size_n,weight_n" for example.
839 static ssize_t get_imix_entries(const char __user *buffer,
840 struct pktgen_dev *pkt_dev)
842 const int max_digits = 10;
847 pkt_dev->n_imix_entries = 0;
850 unsigned long weight;
853 len = num_arg(&buffer[i], max_digits, &size);
857 if (get_user(c, &buffer[i]))
859 /* Check for comma between size_i and weight_i */
864 if (size < 14 + 20 + 8)
867 len = num_arg(&buffer[i], max_digits, &weight);
873 pkt_dev->imix_entries[pkt_dev->n_imix_entries].size = size;
874 pkt_dev->imix_entries[pkt_dev->n_imix_entries].weight = weight;
877 if (get_user(c, &buffer[i]))
881 pkt_dev->n_imix_entries++;
883 if (pkt_dev->n_imix_entries > MAX_IMIX_ENTRIES)
890 static ssize_t get_labels(const char __user *buffer, struct pktgen_dev *pkt_dev)
897 pkt_dev->nr_labels = 0;
900 len = hex32_arg(&buffer[i], 8, &tmp);
903 pkt_dev->labels[n] = htonl(tmp);
904 if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM)
905 pkt_dev->flags |= F_MPLS_RND;
907 if (get_user(c, &buffer[i]))
911 if (n >= MAX_MPLS_LABELS)
915 pkt_dev->nr_labels = n;
919 static __u32 pktgen_read_flag(const char *f, bool *disable)
928 for (i = 0; i < NR_PKT_FLAGS; i++) {
929 if (!IS_ENABLED(CONFIG_XFRM) && i == IPSEC_SHIFT)
932 /* allow only disabling ipv6 flag */
933 if (!*disable && i == IPV6_SHIFT)
936 if (strcmp(f, pkt_flag_names[i]) == 0)
940 if (strcmp(f, "FLOW_RND") == 0) {
941 *disable = !*disable;
948 static ssize_t pktgen_if_write(struct file *file,
949 const char __user * user_buffer, size_t count,
952 struct seq_file *seq = file->private_data;
953 struct pktgen_dev *pkt_dev = seq->private;
955 char name[16], valstr[32];
956 unsigned long value = 0;
957 char *pg_result = NULL;
961 pg_result = &(pkt_dev->result[0]);
964 pr_warn("wrong command format\n");
969 tmp = count_trail_chars(user_buffer, max);
971 pr_warn("illegal format\n");
976 /* Read variable name */
978 len = strn_len(&user_buffer[i], sizeof(name) - 1);
982 memset(name, 0, sizeof(name));
983 if (copy_from_user(name, &user_buffer[i], len))
988 len = count_trail_chars(&user_buffer[i], max);
995 size_t copy = min_t(size_t, count + 1, 1024);
996 char *tp = strndup_user(user_buffer, copy);
1001 pr_debug("%s,%zu buffer -:%s:-\n", name, count, tp);
1005 if (!strcmp(name, "min_pkt_size")) {
1006 len = num_arg(&user_buffer[i], 10, &value);
1011 if (value < 14 + 20 + 8)
1012 value = 14 + 20 + 8;
1013 if (value != pkt_dev->min_pkt_size) {
1014 pkt_dev->min_pkt_size = value;
1015 pkt_dev->cur_pkt_size = value;
1017 sprintf(pg_result, "OK: min_pkt_size=%d",
1018 pkt_dev->min_pkt_size);
1022 if (!strcmp(name, "max_pkt_size")) {
1023 len = num_arg(&user_buffer[i], 10, &value);
1028 if (value < 14 + 20 + 8)
1029 value = 14 + 20 + 8;
1030 if (value != pkt_dev->max_pkt_size) {
1031 pkt_dev->max_pkt_size = value;
1032 pkt_dev->cur_pkt_size = value;
1034 sprintf(pg_result, "OK: max_pkt_size=%d",
1035 pkt_dev->max_pkt_size);
1039 /* Shortcut for min = max */
1041 if (!strcmp(name, "pkt_size")) {
1042 len = num_arg(&user_buffer[i], 10, &value);
1047 if (value < 14 + 20 + 8)
1048 value = 14 + 20 + 8;
1049 if (value != pkt_dev->min_pkt_size) {
1050 pkt_dev->min_pkt_size = value;
1051 pkt_dev->max_pkt_size = value;
1052 pkt_dev->cur_pkt_size = value;
1054 sprintf(pg_result, "OK: pkt_size=%d", pkt_dev->min_pkt_size);
1058 if (!strcmp(name, "imix_weights")) {
1059 if (pkt_dev->clone_skb > 0)
1062 len = get_imix_entries(&user_buffer[i], pkt_dev);
1066 fill_imix_distribution(pkt_dev);
1072 if (!strcmp(name, "debug")) {
1073 len = num_arg(&user_buffer[i], 10, &value);
1079 sprintf(pg_result, "OK: debug=%u", debug);
1083 if (!strcmp(name, "frags")) {
1084 len = num_arg(&user_buffer[i], 10, &value);
1089 pkt_dev->nfrags = value;
1090 sprintf(pg_result, "OK: frags=%d", pkt_dev->nfrags);
1093 if (!strcmp(name, "delay")) {
1094 len = num_arg(&user_buffer[i], 10, &value);
1099 if (value == 0x7FFFFFFF)
1100 pkt_dev->delay = ULLONG_MAX;
1102 pkt_dev->delay = (u64)value;
1104 sprintf(pg_result, "OK: delay=%llu",
1105 (unsigned long long) pkt_dev->delay);
1108 if (!strcmp(name, "rate")) {
1109 len = num_arg(&user_buffer[i], 10, &value);
1116 pkt_dev->delay = pkt_dev->min_pkt_size*8*NSEC_PER_USEC/value;
1118 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1120 sprintf(pg_result, "OK: rate=%lu", value);
1123 if (!strcmp(name, "ratep")) {
1124 len = num_arg(&user_buffer[i], 10, &value);
1131 pkt_dev->delay = NSEC_PER_SEC/value;
1133 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1135 sprintf(pg_result, "OK: rate=%lu", value);
1138 if (!strcmp(name, "udp_src_min")) {
1139 len = num_arg(&user_buffer[i], 10, &value);
1144 if (value != pkt_dev->udp_src_min) {
1145 pkt_dev->udp_src_min = value;
1146 pkt_dev->cur_udp_src = value;
1148 sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min);
1151 if (!strcmp(name, "udp_dst_min")) {
1152 len = num_arg(&user_buffer[i], 10, &value);
1157 if (value != pkt_dev->udp_dst_min) {
1158 pkt_dev->udp_dst_min = value;
1159 pkt_dev->cur_udp_dst = value;
1161 sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min);
1164 if (!strcmp(name, "udp_src_max")) {
1165 len = num_arg(&user_buffer[i], 10, &value);
1170 if (value != pkt_dev->udp_src_max) {
1171 pkt_dev->udp_src_max = value;
1172 pkt_dev->cur_udp_src = value;
1174 sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max);
1177 if (!strcmp(name, "udp_dst_max")) {
1178 len = num_arg(&user_buffer[i], 10, &value);
1183 if (value != pkt_dev->udp_dst_max) {
1184 pkt_dev->udp_dst_max = value;
1185 pkt_dev->cur_udp_dst = value;
1187 sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max);
1190 if (!strcmp(name, "clone_skb")) {
1191 len = num_arg(&user_buffer[i], 10, &value);
1194 /* clone_skb is not supported for netif_receive xmit_mode and
1198 ((pkt_dev->xmit_mode == M_NETIF_RECEIVE) ||
1199 !(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))
1201 if (value > 0 && pkt_dev->n_imix_entries > 0)
1205 pkt_dev->clone_skb = value;
1207 sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb);
1210 if (!strcmp(name, "count")) {
1211 len = num_arg(&user_buffer[i], 10, &value);
1216 pkt_dev->count = value;
1217 sprintf(pg_result, "OK: count=%llu",
1218 (unsigned long long)pkt_dev->count);
1221 if (!strcmp(name, "src_mac_count")) {
1222 len = num_arg(&user_buffer[i], 10, &value);
1227 if (pkt_dev->src_mac_count != value) {
1228 pkt_dev->src_mac_count = value;
1229 pkt_dev->cur_src_mac_offset = 0;
1231 sprintf(pg_result, "OK: src_mac_count=%d",
1232 pkt_dev->src_mac_count);
1235 if (!strcmp(name, "dst_mac_count")) {
1236 len = num_arg(&user_buffer[i], 10, &value);
1241 if (pkt_dev->dst_mac_count != value) {
1242 pkt_dev->dst_mac_count = value;
1243 pkt_dev->cur_dst_mac_offset = 0;
1245 sprintf(pg_result, "OK: dst_mac_count=%d",
1246 pkt_dev->dst_mac_count);
1249 if (!strcmp(name, "burst")) {
1250 len = num_arg(&user_buffer[i], 10, &value);
1256 ((pkt_dev->xmit_mode == M_QUEUE_XMIT) ||
1257 ((pkt_dev->xmit_mode == M_START_XMIT) &&
1258 (!(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))))
1260 pkt_dev->burst = value < 1 ? 1 : value;
1261 sprintf(pg_result, "OK: burst=%u", pkt_dev->burst);
1264 if (!strcmp(name, "node")) {
1265 len = num_arg(&user_buffer[i], 10, &value);
1271 if (node_possible(value)) {
1272 pkt_dev->node = value;
1273 sprintf(pg_result, "OK: node=%d", pkt_dev->node);
1274 if (pkt_dev->page) {
1275 put_page(pkt_dev->page);
1276 pkt_dev->page = NULL;
1280 sprintf(pg_result, "ERROR: node not possible");
1283 if (!strcmp(name, "xmit_mode")) {
1287 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1291 if (copy_from_user(f, &user_buffer[i], len))
1295 if (strcmp(f, "start_xmit") == 0) {
1296 pkt_dev->xmit_mode = M_START_XMIT;
1297 } else if (strcmp(f, "netif_receive") == 0) {
1298 /* clone_skb set earlier, not supported in this mode */
1299 if (pkt_dev->clone_skb > 0)
1302 pkt_dev->xmit_mode = M_NETIF_RECEIVE;
1304 /* make sure new packet is allocated every time
1305 * pktgen_xmit() is called
1307 pkt_dev->last_ok = 1;
1308 } else if (strcmp(f, "queue_xmit") == 0) {
1309 pkt_dev->xmit_mode = M_QUEUE_XMIT;
1310 pkt_dev->last_ok = 1;
1313 "xmit_mode -:%s:- unknown\nAvailable modes: %s",
1314 f, "start_xmit, netif_receive\n");
1317 sprintf(pg_result, "OK: xmit_mode=%s", f);
1320 if (!strcmp(name, "flag")) {
1323 bool disable = false;
1326 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1330 if (copy_from_user(f, &user_buffer[i], len))
1334 flag = pktgen_read_flag(f, &disable);
1338 pkt_dev->flags &= ~flag;
1340 pkt_dev->flags |= flag;
1343 "Flag -:%s:- unknown\nAvailable flags, (prepend ! to un-set flag):\n%s",
1345 "IPSRC_RND, IPDST_RND, UDPSRC_RND, UDPDST_RND, "
1346 "MACSRC_RND, MACDST_RND, TXSIZE_RND, IPV6, "
1347 "MPLS_RND, VID_RND, SVID_RND, FLOW_SEQ, "
1348 "QUEUE_MAP_RND, QUEUE_MAP_CPU, UDPCSUM, "
1356 sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags);
1359 if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) {
1360 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_min) - 1);
1364 if (copy_from_user(buf, &user_buffer[i], len))
1367 if (strcmp(buf, pkt_dev->dst_min) != 0) {
1368 memset(pkt_dev->dst_min, 0, sizeof(pkt_dev->dst_min));
1369 strcpy(pkt_dev->dst_min, buf);
1370 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
1371 pkt_dev->cur_daddr = pkt_dev->daddr_min;
1374 pr_debug("dst_min set to: %s\n", pkt_dev->dst_min);
1376 sprintf(pg_result, "OK: dst_min=%s", pkt_dev->dst_min);
1379 if (!strcmp(name, "dst_max")) {
1380 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_max) - 1);
1384 if (copy_from_user(buf, &user_buffer[i], len))
1387 if (strcmp(buf, pkt_dev->dst_max) != 0) {
1388 memset(pkt_dev->dst_max, 0, sizeof(pkt_dev->dst_max));
1389 strcpy(pkt_dev->dst_max, buf);
1390 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
1391 pkt_dev->cur_daddr = pkt_dev->daddr_max;
1394 pr_debug("dst_max set to: %s\n", pkt_dev->dst_max);
1396 sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max);
1399 if (!strcmp(name, "dst6")) {
1400 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1404 pkt_dev->flags |= F_IPV6;
1406 if (copy_from_user(buf, &user_buffer[i], len))
1410 in6_pton(buf, -1, pkt_dev->in6_daddr.s6_addr, -1, NULL);
1411 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_daddr);
1413 pkt_dev->cur_in6_daddr = pkt_dev->in6_daddr;
1416 pr_debug("dst6 set to: %s\n", buf);
1419 sprintf(pg_result, "OK: dst6=%s", buf);
1422 if (!strcmp(name, "dst6_min")) {
1423 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1427 pkt_dev->flags |= F_IPV6;
1429 if (copy_from_user(buf, &user_buffer[i], len))
1433 in6_pton(buf, -1, pkt_dev->min_in6_daddr.s6_addr, -1, NULL);
1434 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->min_in6_daddr);
1436 pkt_dev->cur_in6_daddr = pkt_dev->min_in6_daddr;
1438 pr_debug("dst6_min set to: %s\n", buf);
1441 sprintf(pg_result, "OK: dst6_min=%s", buf);
1444 if (!strcmp(name, "dst6_max")) {
1445 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1449 pkt_dev->flags |= F_IPV6;
1451 if (copy_from_user(buf, &user_buffer[i], len))
1455 in6_pton(buf, -1, pkt_dev->max_in6_daddr.s6_addr, -1, NULL);
1456 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->max_in6_daddr);
1459 pr_debug("dst6_max set to: %s\n", buf);
1462 sprintf(pg_result, "OK: dst6_max=%s", buf);
1465 if (!strcmp(name, "src6")) {
1466 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1470 pkt_dev->flags |= F_IPV6;
1472 if (copy_from_user(buf, &user_buffer[i], len))
1476 in6_pton(buf, -1, pkt_dev->in6_saddr.s6_addr, -1, NULL);
1477 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_saddr);
1479 pkt_dev->cur_in6_saddr = pkt_dev->in6_saddr;
1482 pr_debug("src6 set to: %s\n", buf);
1485 sprintf(pg_result, "OK: src6=%s", buf);
1488 if (!strcmp(name, "src_min")) {
1489 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_min) - 1);
1493 if (copy_from_user(buf, &user_buffer[i], len))
1496 if (strcmp(buf, pkt_dev->src_min) != 0) {
1497 memset(pkt_dev->src_min, 0, sizeof(pkt_dev->src_min));
1498 strcpy(pkt_dev->src_min, buf);
1499 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
1500 pkt_dev->cur_saddr = pkt_dev->saddr_min;
1503 pr_debug("src_min set to: %s\n", pkt_dev->src_min);
1505 sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min);
1508 if (!strcmp(name, "src_max")) {
1509 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_max) - 1);
1513 if (copy_from_user(buf, &user_buffer[i], len))
1516 if (strcmp(buf, pkt_dev->src_max) != 0) {
1517 memset(pkt_dev->src_max, 0, sizeof(pkt_dev->src_max));
1518 strcpy(pkt_dev->src_max, buf);
1519 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
1520 pkt_dev->cur_saddr = pkt_dev->saddr_max;
1523 pr_debug("src_max set to: %s\n", pkt_dev->src_max);
1525 sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max);
1528 if (!strcmp(name, "dst_mac")) {
1529 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1533 memset(valstr, 0, sizeof(valstr));
1534 if (copy_from_user(valstr, &user_buffer[i], len))
1537 if (!mac_pton(valstr, pkt_dev->dst_mac))
1539 /* Set up Dest MAC */
1540 ether_addr_copy(&pkt_dev->hh[0], pkt_dev->dst_mac);
1542 sprintf(pg_result, "OK: dstmac %pM", pkt_dev->dst_mac);
1545 if (!strcmp(name, "src_mac")) {
1546 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1550 memset(valstr, 0, sizeof(valstr));
1551 if (copy_from_user(valstr, &user_buffer[i], len))
1554 if (!mac_pton(valstr, pkt_dev->src_mac))
1556 /* Set up Src MAC */
1557 ether_addr_copy(&pkt_dev->hh[6], pkt_dev->src_mac);
1559 sprintf(pg_result, "OK: srcmac %pM", pkt_dev->src_mac);
1563 if (!strcmp(name, "clear_counters")) {
1564 pktgen_clear_counters(pkt_dev);
1565 sprintf(pg_result, "OK: Clearing counters.\n");
1569 if (!strcmp(name, "flows")) {
1570 len = num_arg(&user_buffer[i], 10, &value);
1575 if (value > MAX_CFLOWS)
1578 pkt_dev->cflows = value;
1579 sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows);
1583 if (!strcmp(name, "spi")) {
1584 len = num_arg(&user_buffer[i], 10, &value);
1589 pkt_dev->spi = value;
1590 sprintf(pg_result, "OK: spi=%u", pkt_dev->spi);
1594 if (!strcmp(name, "flowlen")) {
1595 len = num_arg(&user_buffer[i], 10, &value);
1600 pkt_dev->lflow = value;
1601 sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow);
1605 if (!strcmp(name, "queue_map_min")) {
1606 len = num_arg(&user_buffer[i], 5, &value);
1611 pkt_dev->queue_map_min = value;
1612 sprintf(pg_result, "OK: queue_map_min=%u", pkt_dev->queue_map_min);
1616 if (!strcmp(name, "queue_map_max")) {
1617 len = num_arg(&user_buffer[i], 5, &value);
1622 pkt_dev->queue_map_max = value;
1623 sprintf(pg_result, "OK: queue_map_max=%u", pkt_dev->queue_map_max);
1627 if (!strcmp(name, "mpls")) {
1628 unsigned int n, cnt;
1630 len = get_labels(&user_buffer[i], pkt_dev);
1634 cnt = sprintf(pg_result, "OK: mpls=");
1635 for (n = 0; n < pkt_dev->nr_labels; n++)
1636 cnt += sprintf(pg_result + cnt,
1637 "%08x%s", ntohl(pkt_dev->labels[n]),
1638 n == pkt_dev->nr_labels-1 ? "" : ",");
1640 if (pkt_dev->nr_labels && pkt_dev->vlan_id != 0xffff) {
1641 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1642 pkt_dev->svlan_id = 0xffff;
1645 pr_debug("VLAN/SVLAN auto turned off\n");
1650 if (!strcmp(name, "vlan_id")) {
1651 len = num_arg(&user_buffer[i], 4, &value);
1656 if (value <= 4095) {
1657 pkt_dev->vlan_id = value; /* turn on VLAN */
1660 pr_debug("VLAN turned on\n");
1662 if (debug && pkt_dev->nr_labels)
1663 pr_debug("MPLS auto turned off\n");
1665 pkt_dev->nr_labels = 0; /* turn off MPLS */
1666 sprintf(pg_result, "OK: vlan_id=%u", pkt_dev->vlan_id);
1668 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1669 pkt_dev->svlan_id = 0xffff;
1672 pr_debug("VLAN/SVLAN turned off\n");
1677 if (!strcmp(name, "vlan_p")) {
1678 len = num_arg(&user_buffer[i], 1, &value);
1683 if ((value <= 7) && (pkt_dev->vlan_id != 0xffff)) {
1684 pkt_dev->vlan_p = value;
1685 sprintf(pg_result, "OK: vlan_p=%u", pkt_dev->vlan_p);
1687 sprintf(pg_result, "ERROR: vlan_p must be 0-7");
1692 if (!strcmp(name, "vlan_cfi")) {
1693 len = num_arg(&user_buffer[i], 1, &value);
1698 if ((value <= 1) && (pkt_dev->vlan_id != 0xffff)) {
1699 pkt_dev->vlan_cfi = value;
1700 sprintf(pg_result, "OK: vlan_cfi=%u", pkt_dev->vlan_cfi);
1702 sprintf(pg_result, "ERROR: vlan_cfi must be 0-1");
1707 if (!strcmp(name, "svlan_id")) {
1708 len = num_arg(&user_buffer[i], 4, &value);
1713 if ((value <= 4095) && ((pkt_dev->vlan_id != 0xffff))) {
1714 pkt_dev->svlan_id = value; /* turn on SVLAN */
1717 pr_debug("SVLAN turned on\n");
1719 if (debug && pkt_dev->nr_labels)
1720 pr_debug("MPLS auto turned off\n");
1722 pkt_dev->nr_labels = 0; /* turn off MPLS */
1723 sprintf(pg_result, "OK: svlan_id=%u", pkt_dev->svlan_id);
1725 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1726 pkt_dev->svlan_id = 0xffff;
1729 pr_debug("VLAN/SVLAN turned off\n");
1734 if (!strcmp(name, "svlan_p")) {
1735 len = num_arg(&user_buffer[i], 1, &value);
1740 if ((value <= 7) && (pkt_dev->svlan_id != 0xffff)) {
1741 pkt_dev->svlan_p = value;
1742 sprintf(pg_result, "OK: svlan_p=%u", pkt_dev->svlan_p);
1744 sprintf(pg_result, "ERROR: svlan_p must be 0-7");
1749 if (!strcmp(name, "svlan_cfi")) {
1750 len = num_arg(&user_buffer[i], 1, &value);
1755 if ((value <= 1) && (pkt_dev->svlan_id != 0xffff)) {
1756 pkt_dev->svlan_cfi = value;
1757 sprintf(pg_result, "OK: svlan_cfi=%u", pkt_dev->svlan_cfi);
1759 sprintf(pg_result, "ERROR: svlan_cfi must be 0-1");
1764 if (!strcmp(name, "tos")) {
1765 __u32 tmp_value = 0;
1766 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1772 pkt_dev->tos = tmp_value;
1773 sprintf(pg_result, "OK: tos=0x%02x", pkt_dev->tos);
1775 sprintf(pg_result, "ERROR: tos must be 00-ff");
1780 if (!strcmp(name, "traffic_class")) {
1781 __u32 tmp_value = 0;
1782 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1788 pkt_dev->traffic_class = tmp_value;
1789 sprintf(pg_result, "OK: traffic_class=0x%02x", pkt_dev->traffic_class);
1791 sprintf(pg_result, "ERROR: traffic_class must be 00-ff");
1796 if (!strcmp(name, "skb_priority")) {
1797 len = num_arg(&user_buffer[i], 9, &value);
1802 pkt_dev->skb_priority = value;
1803 sprintf(pg_result, "OK: skb_priority=%i",
1804 pkt_dev->skb_priority);
1808 sprintf(pkt_dev->result, "No such parameter \"%s\"", name);
1812 static int pktgen_if_open(struct inode *inode, struct file *file)
1814 return single_open(file, pktgen_if_show, pde_data(inode));
1817 static const struct proc_ops pktgen_if_proc_ops = {
1818 .proc_open = pktgen_if_open,
1819 .proc_read = seq_read,
1820 .proc_lseek = seq_lseek,
1821 .proc_write = pktgen_if_write,
1822 .proc_release = single_release,
1825 static int pktgen_thread_show(struct seq_file *seq, void *v)
1827 struct pktgen_thread *t = seq->private;
1828 const struct pktgen_dev *pkt_dev;
1832 seq_puts(seq, "Running: ");
1835 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1836 if (pkt_dev->running)
1837 seq_printf(seq, "%s ", pkt_dev->odevname);
1839 seq_puts(seq, "\nStopped: ");
1841 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1842 if (!pkt_dev->running)
1843 seq_printf(seq, "%s ", pkt_dev->odevname);
1846 seq_printf(seq, "\nResult: %s\n", t->result);
1848 seq_puts(seq, "\nResult: NA\n");
1855 static ssize_t pktgen_thread_write(struct file *file,
1856 const char __user * user_buffer,
1857 size_t count, loff_t * offset)
1859 struct seq_file *seq = file->private_data;
1860 struct pktgen_thread *t = seq->private;
1861 int i, max, len, ret;
1866 // sprintf(pg_result, "Wrong command format");
1871 len = count_trail_chars(user_buffer, max);
1877 /* Read variable name */
1879 len = strn_len(&user_buffer[i], sizeof(name) - 1);
1883 memset(name, 0, sizeof(name));
1884 if (copy_from_user(name, &user_buffer[i], len))
1889 len = count_trail_chars(&user_buffer[i], max);
1896 pr_debug("t=%s, count=%lu\n", name, (unsigned long)count);
1899 pr_err("ERROR: No thread\n");
1904 pg_result = &(t->result[0]);
1906 if (!strcmp(name, "add_device")) {
1909 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1914 if (copy_from_user(f, &user_buffer[i], len))
1917 mutex_lock(&pktgen_thread_lock);
1918 ret = pktgen_add_device(t, f);
1919 mutex_unlock(&pktgen_thread_lock);
1922 sprintf(pg_result, "OK: add_device=%s", f);
1924 sprintf(pg_result, "ERROR: can not add device %s", f);
1928 if (!strcmp(name, "rem_device_all")) {
1929 mutex_lock(&pktgen_thread_lock);
1930 t->control |= T_REMDEVALL;
1931 mutex_unlock(&pktgen_thread_lock);
1932 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */
1934 sprintf(pg_result, "OK: rem_device_all");
1938 if (!strcmp(name, "max_before_softirq")) {
1939 sprintf(pg_result, "OK: Note! max_before_softirq is obsoleted -- Do not use");
1949 static int pktgen_thread_open(struct inode *inode, struct file *file)
1951 return single_open(file, pktgen_thread_show, pde_data(inode));
1954 static const struct proc_ops pktgen_thread_proc_ops = {
1955 .proc_open = pktgen_thread_open,
1956 .proc_read = seq_read,
1957 .proc_lseek = seq_lseek,
1958 .proc_write = pktgen_thread_write,
1959 .proc_release = single_release,
1962 /* Think find or remove for NN */
1963 static struct pktgen_dev *__pktgen_NN_threads(const struct pktgen_net *pn,
1964 const char *ifname, int remove)
1966 struct pktgen_thread *t;
1967 struct pktgen_dev *pkt_dev = NULL;
1968 bool exact = (remove == FIND);
1970 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
1971 pkt_dev = pktgen_find_dev(t, ifname, exact);
1974 pkt_dev->removal_mark = 1;
1975 t->control |= T_REMDEV;
1984 * mark a device for removal
1986 static void pktgen_mark_device(const struct pktgen_net *pn, const char *ifname)
1988 struct pktgen_dev *pkt_dev = NULL;
1989 const int max_tries = 10, msec_per_try = 125;
1992 mutex_lock(&pktgen_thread_lock);
1993 pr_debug("%s: marking %s for removal\n", __func__, ifname);
1997 pkt_dev = __pktgen_NN_threads(pn, ifname, REMOVE);
1998 if (pkt_dev == NULL)
1999 break; /* success */
2001 mutex_unlock(&pktgen_thread_lock);
2002 pr_debug("%s: waiting for %s to disappear....\n",
2004 schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try));
2005 mutex_lock(&pktgen_thread_lock);
2007 if (++i >= max_tries) {
2008 pr_err("%s: timed out after waiting %d msec for device %s to be removed\n",
2009 __func__, msec_per_try * i, ifname);
2015 mutex_unlock(&pktgen_thread_lock);
2018 static void pktgen_change_name(const struct pktgen_net *pn, struct net_device *dev)
2020 struct pktgen_thread *t;
2022 mutex_lock(&pktgen_thread_lock);
2024 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
2025 struct pktgen_dev *pkt_dev;
2028 list_for_each_entry(pkt_dev, &t->if_list, list) {
2029 if (pkt_dev->odev != dev)
2032 proc_remove(pkt_dev->entry);
2034 pkt_dev->entry = proc_create_data(dev->name, 0600,
2036 &pktgen_if_proc_ops,
2038 if (!pkt_dev->entry)
2039 pr_err("can't move proc entry for '%s'\n",
2045 mutex_unlock(&pktgen_thread_lock);
2048 static int pktgen_device_event(struct notifier_block *unused,
2049 unsigned long event, void *ptr)
2051 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2052 struct pktgen_net *pn = net_generic(dev_net(dev), pg_net_id);
2054 if (pn->pktgen_exiting)
2057 /* It is OK that we do not hold the group lock right now,
2058 * as we run under the RTNL lock.
2062 case NETDEV_CHANGENAME:
2063 pktgen_change_name(pn, dev);
2066 case NETDEV_UNREGISTER:
2067 pktgen_mark_device(pn, dev->name);
2074 static struct net_device *pktgen_dev_get_by_name(const struct pktgen_net *pn,
2075 struct pktgen_dev *pkt_dev,
2081 for (i = 0; ifname[i] != '@'; i++) {
2089 return dev_get_by_name(pn->net, b);
2093 /* Associate pktgen_dev with a device. */
2095 static int pktgen_setup_dev(const struct pktgen_net *pn,
2096 struct pktgen_dev *pkt_dev, const char *ifname)
2098 struct net_device *odev;
2101 /* Clean old setups */
2102 if (pkt_dev->odev) {
2103 netdev_put(pkt_dev->odev, &pkt_dev->dev_tracker);
2104 pkt_dev->odev = NULL;
2107 odev = pktgen_dev_get_by_name(pn, pkt_dev, ifname);
2109 pr_err("no such netdevice: \"%s\"\n", ifname);
2113 if (odev->type != ARPHRD_ETHER && odev->type != ARPHRD_LOOPBACK) {
2114 pr_err("not an ethernet or loopback device: \"%s\"\n", ifname);
2116 } else if (!netif_running(odev)) {
2117 pr_err("device is down: \"%s\"\n", ifname);
2120 pkt_dev->odev = odev;
2121 netdev_tracker_alloc(odev, &pkt_dev->dev_tracker, GFP_KERNEL);
2129 /* Read pkt_dev from the interface and set up internal pktgen_dev
2130 * structure to have the right information to create/send packets
2132 static void pktgen_setup_inject(struct pktgen_dev *pkt_dev)
2136 if (!pkt_dev->odev) {
2137 pr_err("ERROR: pkt_dev->odev == NULL in setup_inject\n");
2138 sprintf(pkt_dev->result,
2139 "ERROR: pkt_dev->odev == NULL in setup_inject.\n");
2143 /* make sure that we don't pick a non-existing transmit queue */
2144 ntxq = pkt_dev->odev->real_num_tx_queues;
2146 if (ntxq <= pkt_dev->queue_map_min) {
2147 pr_warn("WARNING: Requested queue_map_min (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2148 pkt_dev->queue_map_min, (ntxq ?: 1) - 1, ntxq,
2150 pkt_dev->queue_map_min = (ntxq ?: 1) - 1;
2152 if (pkt_dev->queue_map_max >= ntxq) {
2153 pr_warn("WARNING: Requested queue_map_max (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2154 pkt_dev->queue_map_max, (ntxq ?: 1) - 1, ntxq,
2156 pkt_dev->queue_map_max = (ntxq ?: 1) - 1;
2159 /* Default to the interface's mac if not explicitly set. */
2161 if (is_zero_ether_addr(pkt_dev->src_mac))
2162 ether_addr_copy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr);
2164 /* Set up Dest MAC */
2165 ether_addr_copy(&(pkt_dev->hh[0]), pkt_dev->dst_mac);
2167 if (pkt_dev->flags & F_IPV6) {
2168 int i, set = 0, err = 1;
2169 struct inet6_dev *idev;
2171 if (pkt_dev->min_pkt_size == 0) {
2172 pkt_dev->min_pkt_size = 14 + sizeof(struct ipv6hdr)
2173 + sizeof(struct udphdr)
2174 + sizeof(struct pktgen_hdr)
2175 + pkt_dev->pkt_overhead;
2178 for (i = 0; i < sizeof(struct in6_addr); i++)
2179 if (pkt_dev->cur_in6_saddr.s6_addr[i]) {
2187 * Use linklevel address if unconfigured.
2189 * use ipv6_get_lladdr if/when it's get exported
2193 idev = __in6_dev_get(pkt_dev->odev);
2195 struct inet6_ifaddr *ifp;
2197 read_lock_bh(&idev->lock);
2198 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2199 if ((ifp->scope & IFA_LINK) &&
2200 !(ifp->flags & IFA_F_TENTATIVE)) {
2201 pkt_dev->cur_in6_saddr = ifp->addr;
2206 read_unlock_bh(&idev->lock);
2210 pr_err("ERROR: IPv6 link address not available\n");
2213 if (pkt_dev->min_pkt_size == 0) {
2214 pkt_dev->min_pkt_size = 14 + sizeof(struct iphdr)
2215 + sizeof(struct udphdr)
2216 + sizeof(struct pktgen_hdr)
2217 + pkt_dev->pkt_overhead;
2220 pkt_dev->saddr_min = 0;
2221 pkt_dev->saddr_max = 0;
2222 if (strlen(pkt_dev->src_min) == 0) {
2224 struct in_device *in_dev;
2227 in_dev = __in_dev_get_rcu(pkt_dev->odev);
2229 const struct in_ifaddr *ifa;
2231 ifa = rcu_dereference(in_dev->ifa_list);
2233 pkt_dev->saddr_min = ifa->ifa_address;
2234 pkt_dev->saddr_max = pkt_dev->saddr_min;
2239 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
2240 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
2243 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
2244 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
2246 /* Initialize current values. */
2247 pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size;
2248 if (pkt_dev->min_pkt_size > pkt_dev->max_pkt_size)
2249 pkt_dev->max_pkt_size = pkt_dev->min_pkt_size;
2251 pkt_dev->cur_dst_mac_offset = 0;
2252 pkt_dev->cur_src_mac_offset = 0;
2253 pkt_dev->cur_saddr = pkt_dev->saddr_min;
2254 pkt_dev->cur_daddr = pkt_dev->daddr_min;
2255 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2256 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2257 pkt_dev->nflows = 0;
2261 static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until)
2263 ktime_t start_time, end_time;
2265 struct hrtimer_sleeper t;
2267 hrtimer_init_sleeper_on_stack(&t, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
2268 hrtimer_set_expires(&t.timer, spin_until);
2270 remaining = ktime_to_ns(hrtimer_expires_remaining(&t.timer));
2274 start_time = ktime_get();
2275 if (remaining < 100000) {
2276 /* for small delays (<100us), just loop until limit is reached */
2278 end_time = ktime_get();
2279 } while (ktime_compare(end_time, spin_until) < 0);
2282 set_current_state(TASK_INTERRUPTIBLE);
2283 hrtimer_sleeper_start_expires(&t, HRTIMER_MODE_ABS);
2288 hrtimer_cancel(&t.timer);
2289 } while (t.task && pkt_dev->running && !signal_pending(current));
2290 __set_current_state(TASK_RUNNING);
2291 end_time = ktime_get();
2294 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(end_time, start_time));
2296 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay);
2297 destroy_hrtimer_on_stack(&t.timer);
2300 static inline void set_pkt_overhead(struct pktgen_dev *pkt_dev)
2302 pkt_dev->pkt_overhead = 0;
2303 pkt_dev->pkt_overhead += pkt_dev->nr_labels*sizeof(u32);
2304 pkt_dev->pkt_overhead += VLAN_TAG_SIZE(pkt_dev);
2305 pkt_dev->pkt_overhead += SVLAN_TAG_SIZE(pkt_dev);
2308 static inline int f_seen(const struct pktgen_dev *pkt_dev, int flow)
2310 return !!(pkt_dev->flows[flow].flags & F_INIT);
2313 static inline int f_pick(struct pktgen_dev *pkt_dev)
2315 int flow = pkt_dev->curfl;
2317 if (pkt_dev->flags & F_FLOW_SEQ) {
2318 if (pkt_dev->flows[flow].count >= pkt_dev->lflow) {
2320 pkt_dev->flows[flow].count = 0;
2321 pkt_dev->flows[flow].flags = 0;
2322 pkt_dev->curfl += 1;
2323 if (pkt_dev->curfl >= pkt_dev->cflows)
2324 pkt_dev->curfl = 0; /*reset */
2327 flow = get_random_u32_below(pkt_dev->cflows);
2328 pkt_dev->curfl = flow;
2330 if (pkt_dev->flows[flow].count > pkt_dev->lflow) {
2331 pkt_dev->flows[flow].count = 0;
2332 pkt_dev->flows[flow].flags = 0;
2336 return pkt_dev->curfl;
2341 /* If there was already an IPSEC SA, we keep it as is, else
2342 * we go look for it ...
2344 #define DUMMY_MARK 0
2345 static void get_ipsec_sa(struct pktgen_dev *pkt_dev, int flow)
2347 struct xfrm_state *x = pkt_dev->flows[flow].x;
2348 struct pktgen_net *pn = net_generic(dev_net(pkt_dev->odev), pg_net_id);
2352 /* We need as quick as possible to find the right SA
2353 * Searching with minimum criteria to archieve this.
2355 x = xfrm_state_lookup_byspi(pn->net, htonl(pkt_dev->spi), AF_INET);
2357 /* slow path: we dont already have xfrm_state */
2358 x = xfrm_stateonly_find(pn->net, DUMMY_MARK, 0,
2359 (xfrm_address_t *)&pkt_dev->cur_daddr,
2360 (xfrm_address_t *)&pkt_dev->cur_saddr,
2363 pkt_dev->ipsproto, 0);
2366 pkt_dev->flows[flow].x = x;
2367 set_pkt_overhead(pkt_dev);
2368 pkt_dev->pkt_overhead += x->props.header_len;
2374 static void set_cur_queue_map(struct pktgen_dev *pkt_dev)
2377 if (pkt_dev->flags & F_QUEUE_MAP_CPU)
2378 pkt_dev->cur_queue_map = smp_processor_id();
2380 else if (pkt_dev->queue_map_min <= pkt_dev->queue_map_max) {
2382 if (pkt_dev->flags & F_QUEUE_MAP_RND) {
2383 t = get_random_u32_inclusive(pkt_dev->queue_map_min,
2384 pkt_dev->queue_map_max);
2386 t = pkt_dev->cur_queue_map + 1;
2387 if (t > pkt_dev->queue_map_max)
2388 t = pkt_dev->queue_map_min;
2390 pkt_dev->cur_queue_map = t;
2392 pkt_dev->cur_queue_map = pkt_dev->cur_queue_map % pkt_dev->odev->real_num_tx_queues;
2395 /* Increment/randomize headers according to flags and current values
2396 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst
2398 static void mod_cur_headers(struct pktgen_dev *pkt_dev)
2404 if (pkt_dev->cflows)
2405 flow = f_pick(pkt_dev);
2407 /* Deal with source MAC */
2408 if (pkt_dev->src_mac_count > 1) {
2412 if (pkt_dev->flags & F_MACSRC_RND)
2413 mc = get_random_u32_below(pkt_dev->src_mac_count);
2415 mc = pkt_dev->cur_src_mac_offset++;
2416 if (pkt_dev->cur_src_mac_offset >=
2417 pkt_dev->src_mac_count)
2418 pkt_dev->cur_src_mac_offset = 0;
2421 tmp = pkt_dev->src_mac[5] + (mc & 0xFF);
2422 pkt_dev->hh[11] = tmp;
2423 tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2424 pkt_dev->hh[10] = tmp;
2425 tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2426 pkt_dev->hh[9] = tmp;
2427 tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2428 pkt_dev->hh[8] = tmp;
2429 tmp = (pkt_dev->src_mac[1] + (tmp >> 8));
2430 pkt_dev->hh[7] = tmp;
2433 /* Deal with Destination MAC */
2434 if (pkt_dev->dst_mac_count > 1) {
2438 if (pkt_dev->flags & F_MACDST_RND)
2439 mc = get_random_u32_below(pkt_dev->dst_mac_count);
2442 mc = pkt_dev->cur_dst_mac_offset++;
2443 if (pkt_dev->cur_dst_mac_offset >=
2444 pkt_dev->dst_mac_count) {
2445 pkt_dev->cur_dst_mac_offset = 0;
2449 tmp = pkt_dev->dst_mac[5] + (mc & 0xFF);
2450 pkt_dev->hh[5] = tmp;
2451 tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2452 pkt_dev->hh[4] = tmp;
2453 tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2454 pkt_dev->hh[3] = tmp;
2455 tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2456 pkt_dev->hh[2] = tmp;
2457 tmp = (pkt_dev->dst_mac[1] + (tmp >> 8));
2458 pkt_dev->hh[1] = tmp;
2461 if (pkt_dev->flags & F_MPLS_RND) {
2463 for (i = 0; i < pkt_dev->nr_labels; i++)
2464 if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM)
2465 pkt_dev->labels[i] = MPLS_STACK_BOTTOM |
2466 ((__force __be32)get_random_u32() &
2470 if ((pkt_dev->flags & F_VID_RND) && (pkt_dev->vlan_id != 0xffff)) {
2471 pkt_dev->vlan_id = get_random_u32_below(4096);
2474 if ((pkt_dev->flags & F_SVID_RND) && (pkt_dev->svlan_id != 0xffff)) {
2475 pkt_dev->svlan_id = get_random_u32_below(4096);
2478 if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) {
2479 if (pkt_dev->flags & F_UDPSRC_RND)
2480 pkt_dev->cur_udp_src = get_random_u32_inclusive(pkt_dev->udp_src_min,
2481 pkt_dev->udp_src_max - 1);
2484 pkt_dev->cur_udp_src++;
2485 if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max)
2486 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2490 if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) {
2491 if (pkt_dev->flags & F_UDPDST_RND) {
2492 pkt_dev->cur_udp_dst = get_random_u32_inclusive(pkt_dev->udp_dst_min,
2493 pkt_dev->udp_dst_max - 1);
2495 pkt_dev->cur_udp_dst++;
2496 if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max)
2497 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2501 if (!(pkt_dev->flags & F_IPV6)) {
2503 imn = ntohl(pkt_dev->saddr_min);
2504 imx = ntohl(pkt_dev->saddr_max);
2507 if (pkt_dev->flags & F_IPSRC_RND)
2508 t = get_random_u32_inclusive(imn, imx - 1);
2510 t = ntohl(pkt_dev->cur_saddr);
2516 pkt_dev->cur_saddr = htonl(t);
2519 if (pkt_dev->cflows && f_seen(pkt_dev, flow)) {
2520 pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr;
2522 imn = ntohl(pkt_dev->daddr_min);
2523 imx = ntohl(pkt_dev->daddr_max);
2527 if (pkt_dev->flags & F_IPDST_RND) {
2530 t = get_random_u32_inclusive(imn, imx - 1);
2532 } while (ipv4_is_loopback(s) ||
2533 ipv4_is_multicast(s) ||
2534 ipv4_is_lbcast(s) ||
2535 ipv4_is_zeronet(s) ||
2536 ipv4_is_local_multicast(s));
2537 pkt_dev->cur_daddr = s;
2539 t = ntohl(pkt_dev->cur_daddr);
2544 pkt_dev->cur_daddr = htonl(t);
2547 if (pkt_dev->cflows) {
2548 pkt_dev->flows[flow].flags |= F_INIT;
2549 pkt_dev->flows[flow].cur_daddr =
2552 if (pkt_dev->flags & F_IPSEC)
2553 get_ipsec_sa(pkt_dev, flow);
2558 } else { /* IPV6 * */
2560 if (!ipv6_addr_any(&pkt_dev->min_in6_daddr)) {
2563 /* Only random destinations yet */
2565 for (i = 0; i < 4; i++) {
2566 pkt_dev->cur_in6_daddr.s6_addr32[i] =
2567 (((__force __be32)get_random_u32() |
2568 pkt_dev->min_in6_daddr.s6_addr32[i]) &
2569 pkt_dev->max_in6_daddr.s6_addr32[i]);
2574 if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) {
2576 if (pkt_dev->flags & F_TXSIZE_RND) {
2577 t = get_random_u32_inclusive(pkt_dev->min_pkt_size,
2578 pkt_dev->max_pkt_size - 1);
2580 t = pkt_dev->cur_pkt_size + 1;
2581 if (t > pkt_dev->max_pkt_size)
2582 t = pkt_dev->min_pkt_size;
2584 pkt_dev->cur_pkt_size = t;
2585 } else if (pkt_dev->n_imix_entries > 0) {
2586 struct imix_pkt *entry;
2587 __u32 t = get_random_u32_below(IMIX_PRECISION);
2588 __u8 entry_index = pkt_dev->imix_distribution[t];
2590 entry = &pkt_dev->imix_entries[entry_index];
2591 entry->count_so_far++;
2592 pkt_dev->cur_pkt_size = entry->size;
2595 set_cur_queue_map(pkt_dev);
2597 pkt_dev->flows[flow].count++;
2600 static void fill_imix_distribution(struct pktgen_dev *pkt_dev)
2602 int cumulative_probabilites[MAX_IMIX_ENTRIES];
2604 __u64 cumulative_prob = 0;
2605 __u64 total_weight = 0;
2608 for (i = 0; i < pkt_dev->n_imix_entries; i++)
2609 total_weight += pkt_dev->imix_entries[i].weight;
2611 /* Fill cumulative_probabilites with sum of normalized probabilities */
2612 for (i = 0; i < pkt_dev->n_imix_entries - 1; i++) {
2613 cumulative_prob += div64_u64(pkt_dev->imix_entries[i].weight *
2616 cumulative_probabilites[i] = cumulative_prob;
2618 cumulative_probabilites[pkt_dev->n_imix_entries - 1] = 100;
2620 for (i = 0; i < IMIX_PRECISION; i++) {
2621 if (i == cumulative_probabilites[j])
2623 pkt_dev->imix_distribution[i] = j;
2628 static u32 pktgen_dst_metrics[RTAX_MAX + 1] = {
2630 [RTAX_HOPLIMIT] = 0x5, /* Set a static hoplimit */
2633 static int pktgen_output_ipsec(struct sk_buff *skb, struct pktgen_dev *pkt_dev)
2635 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2637 struct net *net = dev_net(pkt_dev->odev);
2641 /* XXX: we dont support tunnel mode for now until
2642 * we resolve the dst issue */
2643 if ((x->props.mode != XFRM_MODE_TRANSPORT) && (pkt_dev->spi == 0))
2646 /* But when user specify an valid SPI, transformation
2647 * supports both transport/tunnel mode + ESP/AH type.
2649 if ((x->props.mode == XFRM_MODE_TUNNEL) && (pkt_dev->spi != 0))
2650 skb->_skb_refdst = (unsigned long)&pkt_dev->xdst.u.dst | SKB_DST_NOREF;
2653 err = pktgen_xfrm_outer_mode_output(x, skb);
2654 rcu_read_unlock_bh();
2656 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
2659 err = x->type->output(x, skb);
2661 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
2664 spin_lock_bh(&x->lock);
2665 x->curlft.bytes += skb->len;
2666 x->curlft.packets++;
2667 spin_unlock_bh(&x->lock);
2672 static void free_SAs(struct pktgen_dev *pkt_dev)
2674 if (pkt_dev->cflows) {
2675 /* let go of the SAs if we have them */
2677 for (i = 0; i < pkt_dev->cflows; i++) {
2678 struct xfrm_state *x = pkt_dev->flows[i].x;
2681 pkt_dev->flows[i].x = NULL;
2687 static int process_ipsec(struct pktgen_dev *pkt_dev,
2688 struct sk_buff *skb, __be16 protocol)
2690 if (pkt_dev->flags & F_IPSEC) {
2691 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2698 nhead = x->props.header_len - skb_headroom(skb);
2700 ret = pskb_expand_head(skb, nhead, 0, GFP_ATOMIC);
2702 pr_err("Error expanding ipsec packet %d\n",
2708 /* ipsec is not expecting ll header */
2709 skb_pull(skb, ETH_HLEN);
2710 ret = pktgen_output_ipsec(skb, pkt_dev);
2712 pr_err("Error creating ipsec packet %d\n", ret);
2716 eth = skb_push(skb, ETH_HLEN);
2717 memcpy(eth, pkt_dev->hh, 2 * ETH_ALEN);
2718 eth->h_proto = protocol;
2720 /* Update IPv4 header len as well as checksum value */
2722 iph->tot_len = htons(skb->len - ETH_HLEN);
2733 static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev)
2736 for (i = 0; i < pkt_dev->nr_labels; i++)
2737 *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM;
2740 *mpls |= MPLS_STACK_BOTTOM;
2743 static inline __be16 build_tci(unsigned int id, unsigned int cfi,
2746 return htons(id | (cfi << 12) | (prio << 13));
2749 static void pktgen_finalize_skb(struct pktgen_dev *pkt_dev, struct sk_buff *skb,
2752 struct timespec64 timestamp;
2753 struct pktgen_hdr *pgh;
2755 pgh = skb_put(skb, sizeof(*pgh));
2756 datalen -= sizeof(*pgh);
2758 if (pkt_dev->nfrags <= 0) {
2759 skb_put_zero(skb, datalen);
2761 int frags = pkt_dev->nfrags;
2766 if (frags > MAX_SKB_FRAGS)
2767 frags = MAX_SKB_FRAGS;
2768 len = datalen - frags * PAGE_SIZE;
2770 skb_put_zero(skb, len);
2771 datalen = frags * PAGE_SIZE;
2775 frag_len = (datalen/frags) < PAGE_SIZE ?
2776 (datalen/frags) : PAGE_SIZE;
2777 while (datalen > 0) {
2778 if (unlikely(!pkt_dev->page)) {
2779 int node = numa_node_id();
2781 if (pkt_dev->node >= 0 && (pkt_dev->flags & F_NODE))
2782 node = pkt_dev->node;
2783 pkt_dev->page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0);
2787 get_page(pkt_dev->page);
2789 /*last fragment, fill rest of data*/
2790 if (i == (frags - 1))
2791 skb_frag_fill_page_desc(&skb_shinfo(skb)->frags[i],
2793 (datalen < PAGE_SIZE ?
2794 datalen : PAGE_SIZE));
2796 skb_frag_fill_page_desc(&skb_shinfo(skb)->frags[i],
2797 pkt_dev->page, 0, frag_len);
2799 datalen -= skb_frag_size(&skb_shinfo(skb)->frags[i]);
2800 skb->len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2801 skb->data_len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2803 skb_shinfo(skb)->nr_frags = i;
2807 /* Stamp the time, and sequence number,
2808 * convert them to network byte order
2810 pgh->pgh_magic = htonl(PKTGEN_MAGIC);
2811 pgh->seq_num = htonl(pkt_dev->seq_num);
2813 if (pkt_dev->flags & F_NO_TIMESTAMP) {
2818 * pgh->tv_sec wraps in y2106 when interpreted as unsigned
2819 * as done by wireshark, or y2038 when interpreted as signed.
2820 * This is probably harmless, but if anyone wants to improve
2821 * it, we could introduce a variant that puts 64-bit nanoseconds
2822 * into the respective header bytes.
2823 * This would also be slightly faster to read.
2825 ktime_get_real_ts64(×tamp);
2826 pgh->tv_sec = htonl(timestamp.tv_sec);
2827 pgh->tv_usec = htonl(timestamp.tv_nsec / NSEC_PER_USEC);
2831 static struct sk_buff *pktgen_alloc_skb(struct net_device *dev,
2832 struct pktgen_dev *pkt_dev)
2834 unsigned int extralen = LL_RESERVED_SPACE(dev);
2835 struct sk_buff *skb = NULL;
2838 size = pkt_dev->cur_pkt_size + 64 + extralen + pkt_dev->pkt_overhead;
2839 if (pkt_dev->flags & F_NODE) {
2840 int node = pkt_dev->node >= 0 ? pkt_dev->node : numa_node_id();
2842 skb = __alloc_skb(NET_SKB_PAD + size, GFP_NOWAIT, 0, node);
2844 skb_reserve(skb, NET_SKB_PAD);
2848 skb = __netdev_alloc_skb(dev, size, GFP_NOWAIT);
2851 /* the caller pre-fetches from skb->data and reserves for the mac hdr */
2853 skb_reserve(skb, extralen - 16);
2858 static struct sk_buff *fill_packet_ipv4(struct net_device *odev,
2859 struct pktgen_dev *pkt_dev)
2861 struct sk_buff *skb = NULL;
2863 struct udphdr *udph;
2866 __be16 protocol = htons(ETH_P_IP);
2868 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2869 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2870 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2871 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2874 if (pkt_dev->nr_labels)
2875 protocol = htons(ETH_P_MPLS_UC);
2877 if (pkt_dev->vlan_id != 0xffff)
2878 protocol = htons(ETH_P_8021Q);
2880 /* Update any of the values, used when we're incrementing various
2883 mod_cur_headers(pkt_dev);
2884 queue_map = pkt_dev->cur_queue_map;
2886 skb = pktgen_alloc_skb(odev, pkt_dev);
2888 sprintf(pkt_dev->result, "No memory");
2892 prefetchw(skb->data);
2893 skb_reserve(skb, 16);
2895 /* Reserve for ethernet and IP header */
2896 eth = skb_push(skb, 14);
2897 mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32));
2898 if (pkt_dev->nr_labels)
2899 mpls_push(mpls, pkt_dev);
2901 if (pkt_dev->vlan_id != 0xffff) {
2902 if (pkt_dev->svlan_id != 0xffff) {
2903 svlan_tci = skb_put(skb, sizeof(__be16));
2904 *svlan_tci = build_tci(pkt_dev->svlan_id,
2907 svlan_encapsulated_proto = skb_put(skb,
2909 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2911 vlan_tci = skb_put(skb, sizeof(__be16));
2912 *vlan_tci = build_tci(pkt_dev->vlan_id,
2915 vlan_encapsulated_proto = skb_put(skb, sizeof(__be16));
2916 *vlan_encapsulated_proto = htons(ETH_P_IP);
2919 skb_reset_mac_header(skb);
2920 skb_set_network_header(skb, skb->len);
2921 iph = skb_put(skb, sizeof(struct iphdr));
2923 skb_set_transport_header(skb, skb->len);
2924 udph = skb_put(skb, sizeof(struct udphdr));
2925 skb_set_queue_mapping(skb, queue_map);
2926 skb->priority = pkt_dev->skb_priority;
2928 memcpy(eth, pkt_dev->hh, 12);
2929 *(__be16 *) & eth[12] = protocol;
2931 /* Eth + IPh + UDPh + mpls */
2932 datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 -
2933 pkt_dev->pkt_overhead;
2934 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr))
2935 datalen = sizeof(struct pktgen_hdr);
2937 udph->source = htons(pkt_dev->cur_udp_src);
2938 udph->dest = htons(pkt_dev->cur_udp_dst);
2939 udph->len = htons(datalen + 8); /* DATA + udphdr */
2945 iph->tos = pkt_dev->tos;
2946 iph->protocol = IPPROTO_UDP; /* UDP */
2947 iph->saddr = pkt_dev->cur_saddr;
2948 iph->daddr = pkt_dev->cur_daddr;
2949 iph->id = htons(pkt_dev->ip_id);
2952 iplen = 20 + 8 + datalen;
2953 iph->tot_len = htons(iplen);
2955 skb->protocol = protocol;
2957 skb->pkt_type = PACKET_HOST;
2959 pktgen_finalize_skb(pkt_dev, skb, datalen);
2961 if (!(pkt_dev->flags & F_UDPCSUM)) {
2962 skb->ip_summed = CHECKSUM_NONE;
2963 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM)) {
2964 skb->ip_summed = CHECKSUM_PARTIAL;
2966 udp4_hwcsum(skb, iph->saddr, iph->daddr);
2968 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), datalen + 8, 0);
2970 /* add protocol-dependent pseudo-header */
2971 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
2972 datalen + 8, IPPROTO_UDP, csum);
2974 if (udph->check == 0)
2975 udph->check = CSUM_MANGLED_0;
2979 if (!process_ipsec(pkt_dev, skb, protocol))
2986 static struct sk_buff *fill_packet_ipv6(struct net_device *odev,
2987 struct pktgen_dev *pkt_dev)
2989 struct sk_buff *skb = NULL;
2991 struct udphdr *udph;
2992 int datalen, udplen;
2993 struct ipv6hdr *iph;
2994 __be16 protocol = htons(ETH_P_IPV6);
2996 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2997 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2998 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2999 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
3002 if (pkt_dev->nr_labels)
3003 protocol = htons(ETH_P_MPLS_UC);
3005 if (pkt_dev->vlan_id != 0xffff)
3006 protocol = htons(ETH_P_8021Q);
3008 /* Update any of the values, used when we're incrementing various
3011 mod_cur_headers(pkt_dev);
3012 queue_map = pkt_dev->cur_queue_map;
3014 skb = pktgen_alloc_skb(odev, pkt_dev);
3016 sprintf(pkt_dev->result, "No memory");
3020 prefetchw(skb->data);
3021 skb_reserve(skb, 16);
3023 /* Reserve for ethernet and IP header */
3024 eth = skb_push(skb, 14);
3025 mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32));
3026 if (pkt_dev->nr_labels)
3027 mpls_push(mpls, pkt_dev);
3029 if (pkt_dev->vlan_id != 0xffff) {
3030 if (pkt_dev->svlan_id != 0xffff) {
3031 svlan_tci = skb_put(skb, sizeof(__be16));
3032 *svlan_tci = build_tci(pkt_dev->svlan_id,
3035 svlan_encapsulated_proto = skb_put(skb,
3037 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
3039 vlan_tci = skb_put(skb, sizeof(__be16));
3040 *vlan_tci = build_tci(pkt_dev->vlan_id,
3043 vlan_encapsulated_proto = skb_put(skb, sizeof(__be16));
3044 *vlan_encapsulated_proto = htons(ETH_P_IPV6);
3047 skb_reset_mac_header(skb);
3048 skb_set_network_header(skb, skb->len);
3049 iph = skb_put(skb, sizeof(struct ipv6hdr));
3051 skb_set_transport_header(skb, skb->len);
3052 udph = skb_put(skb, sizeof(struct udphdr));
3053 skb_set_queue_mapping(skb, queue_map);
3054 skb->priority = pkt_dev->skb_priority;
3056 memcpy(eth, pkt_dev->hh, 12);
3057 *(__be16 *) ð[12] = protocol;
3059 /* Eth + IPh + UDPh + mpls */
3060 datalen = pkt_dev->cur_pkt_size - 14 -
3061 sizeof(struct ipv6hdr) - sizeof(struct udphdr) -
3062 pkt_dev->pkt_overhead;
3064 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) {
3065 datalen = sizeof(struct pktgen_hdr);
3066 net_info_ratelimited("increased datalen to %d\n", datalen);
3069 udplen = datalen + sizeof(struct udphdr);
3070 udph->source = htons(pkt_dev->cur_udp_src);
3071 udph->dest = htons(pkt_dev->cur_udp_dst);
3072 udph->len = htons(udplen);
3075 *(__be32 *) iph = htonl(0x60000000); /* Version + flow */
3077 if (pkt_dev->traffic_class) {
3078 /* Version + traffic class + flow (0) */
3079 *(__be32 *)iph |= htonl(0x60000000 | (pkt_dev->traffic_class << 20));
3082 iph->hop_limit = 32;
3084 iph->payload_len = htons(udplen);
3085 iph->nexthdr = IPPROTO_UDP;
3087 iph->daddr = pkt_dev->cur_in6_daddr;
3088 iph->saddr = pkt_dev->cur_in6_saddr;
3090 skb->protocol = protocol;
3092 skb->pkt_type = PACKET_HOST;
3094 pktgen_finalize_skb(pkt_dev, skb, datalen);
3096 if (!(pkt_dev->flags & F_UDPCSUM)) {
3097 skb->ip_summed = CHECKSUM_NONE;
3098 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM)) {
3099 skb->ip_summed = CHECKSUM_PARTIAL;
3100 skb->csum_start = skb_transport_header(skb) - skb->head;
3101 skb->csum_offset = offsetof(struct udphdr, check);
3102 udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0);
3104 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), udplen, 0);
3106 /* add protocol-dependent pseudo-header */
3107 udph->check = csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, csum);
3109 if (udph->check == 0)
3110 udph->check = CSUM_MANGLED_0;
3116 static struct sk_buff *fill_packet(struct net_device *odev,
3117 struct pktgen_dev *pkt_dev)
3119 if (pkt_dev->flags & F_IPV6)
3120 return fill_packet_ipv6(odev, pkt_dev);
3122 return fill_packet_ipv4(odev, pkt_dev);
3125 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev)
3127 pkt_dev->seq_num = 1;
3128 pkt_dev->idle_acc = 0;
3130 pkt_dev->tx_bytes = 0;
3131 pkt_dev->errors = 0;
3134 /* Set up structure for sending pkts, clear counters */
3136 static void pktgen_run(struct pktgen_thread *t)
3138 struct pktgen_dev *pkt_dev;
3144 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3147 * setup odev and create initial packet.
3149 pktgen_setup_inject(pkt_dev);
3151 if (pkt_dev->odev) {
3152 pktgen_clear_counters(pkt_dev);
3153 pkt_dev->skb = NULL;
3154 pkt_dev->started_at = pkt_dev->next_tx = ktime_get();
3156 set_pkt_overhead(pkt_dev);
3158 strcpy(pkt_dev->result, "Starting");
3159 pkt_dev->running = 1; /* Cranke yeself! */
3162 strcpy(pkt_dev->result, "Error starting");
3166 t->control &= ~(T_STOP);
3169 static void pktgen_handle_all_threads(struct pktgen_net *pn, u32 flags)
3171 struct pktgen_thread *t;
3173 mutex_lock(&pktgen_thread_lock);
3175 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3176 t->control |= (flags);
3178 mutex_unlock(&pktgen_thread_lock);
3181 static void pktgen_stop_all_threads(struct pktgen_net *pn)
3185 pktgen_handle_all_threads(pn, T_STOP);
3188 static int thread_is_running(const struct pktgen_thread *t)
3190 const struct pktgen_dev *pkt_dev;
3193 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
3194 if (pkt_dev->running) {
3202 static int pktgen_wait_thread_run(struct pktgen_thread *t)
3204 while (thread_is_running(t)) {
3206 /* note: 't' will still be around even after the unlock/lock
3207 * cycle because pktgen_thread threads are only cleared at
3210 mutex_unlock(&pktgen_thread_lock);
3211 msleep_interruptible(100);
3212 mutex_lock(&pktgen_thread_lock);
3214 if (signal_pending(current))
3222 static int pktgen_wait_all_threads_run(struct pktgen_net *pn)
3224 struct pktgen_thread *t;
3227 /* prevent from racing with rmmod */
3228 if (!try_module_get(THIS_MODULE))
3231 mutex_lock(&pktgen_thread_lock);
3233 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
3234 sig = pktgen_wait_thread_run(t);
3240 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3241 t->control |= (T_STOP);
3243 mutex_unlock(&pktgen_thread_lock);
3244 module_put(THIS_MODULE);
3248 static void pktgen_run_all_threads(struct pktgen_net *pn)
3252 pktgen_handle_all_threads(pn, T_RUN);
3254 /* Propagate thread->control */
3255 schedule_timeout_interruptible(msecs_to_jiffies(125));
3257 pktgen_wait_all_threads_run(pn);
3260 static void pktgen_reset_all_threads(struct pktgen_net *pn)
3264 pktgen_handle_all_threads(pn, T_REMDEVALL);
3266 /* Propagate thread->control */
3267 schedule_timeout_interruptible(msecs_to_jiffies(125));
3269 pktgen_wait_all_threads_run(pn);
3272 static void show_results(struct pktgen_dev *pkt_dev, int nr_frags)
3274 __u64 bps, mbps, pps;
3275 char *p = pkt_dev->result;
3276 ktime_t elapsed = ktime_sub(pkt_dev->stopped_at,
3277 pkt_dev->started_at);
3278 ktime_t idle = ns_to_ktime(pkt_dev->idle_acc);
3280 p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n",
3281 (unsigned long long)ktime_to_us(elapsed),
3282 (unsigned long long)ktime_to_us(ktime_sub(elapsed, idle)),
3283 (unsigned long long)ktime_to_us(idle),
3284 (unsigned long long)pkt_dev->sofar,
3285 pkt_dev->cur_pkt_size, nr_frags);
3287 pps = div64_u64(pkt_dev->sofar * NSEC_PER_SEC,
3288 ktime_to_ns(elapsed));
3290 if (pkt_dev->n_imix_entries > 0) {
3292 struct imix_pkt *entry;
3295 for (i = 0; i < pkt_dev->n_imix_entries; i++) {
3296 entry = &pkt_dev->imix_entries[i];
3297 bps += entry->size * entry->count_so_far;
3299 bps = div64_u64(bps * 8 * NSEC_PER_SEC, ktime_to_ns(elapsed));
3301 bps = pps * 8 * pkt_dev->cur_pkt_size;
3305 do_div(mbps, 1000000);
3306 p += sprintf(p, " %llupps %lluMb/sec (%llubps) errors: %llu",
3307 (unsigned long long)pps,
3308 (unsigned long long)mbps,
3309 (unsigned long long)bps,
3310 (unsigned long long)pkt_dev->errors);
3313 /* Set stopped-at timer, remove from running list, do counters & statistics */
3314 static int pktgen_stop_device(struct pktgen_dev *pkt_dev)
3316 int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1;
3318 if (!pkt_dev->running) {
3319 pr_warn("interface: %s is already stopped\n",
3324 pkt_dev->running = 0;
3325 kfree_skb(pkt_dev->skb);
3326 pkt_dev->skb = NULL;
3327 pkt_dev->stopped_at = ktime_get();
3329 show_results(pkt_dev, nr_frags);
3334 static struct pktgen_dev *next_to_run(struct pktgen_thread *t)
3336 struct pktgen_dev *pkt_dev, *best = NULL;
3339 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3340 if (!pkt_dev->running)
3344 else if (ktime_compare(pkt_dev->next_tx, best->next_tx) < 0)
3352 static void pktgen_stop(struct pktgen_thread *t)
3354 struct pktgen_dev *pkt_dev;
3360 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3361 pktgen_stop_device(pkt_dev);
3368 * one of our devices needs to be removed - find it
3371 static void pktgen_rem_one_if(struct pktgen_thread *t)
3373 struct list_head *q, *n;
3374 struct pktgen_dev *cur;
3378 list_for_each_safe(q, n, &t->if_list) {
3379 cur = list_entry(q, struct pktgen_dev, list);
3381 if (!cur->removal_mark)
3384 kfree_skb(cur->skb);
3387 pktgen_remove_device(t, cur);
3393 static void pktgen_rem_all_ifs(struct pktgen_thread *t)
3395 struct list_head *q, *n;
3396 struct pktgen_dev *cur;
3400 /* Remove all devices, free mem */
3402 list_for_each_safe(q, n, &t->if_list) {
3403 cur = list_entry(q, struct pktgen_dev, list);
3405 kfree_skb(cur->skb);
3408 pktgen_remove_device(t, cur);
3412 static void pktgen_rem_thread(struct pktgen_thread *t)
3414 /* Remove from the thread list */
3415 remove_proc_entry(t->tsk->comm, t->net->proc_dir);
3418 static void pktgen_resched(struct pktgen_dev *pkt_dev)
3420 ktime_t idle_start = ktime_get();
3422 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3425 static void pktgen_wait_for_skb(struct pktgen_dev *pkt_dev)
3427 ktime_t idle_start = ktime_get();
3429 while (refcount_read(&(pkt_dev->skb->users)) != 1) {
3430 if (signal_pending(current))
3434 pktgen_resched(pkt_dev);
3438 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3441 static void pktgen_xmit(struct pktgen_dev *pkt_dev)
3443 unsigned int burst = READ_ONCE(pkt_dev->burst);
3444 struct net_device *odev = pkt_dev->odev;
3445 struct netdev_queue *txq;
3446 struct sk_buff *skb;
3449 /* If device is offline, then don't send */
3450 if (unlikely(!netif_running(odev) || !netif_carrier_ok(odev))) {
3451 pktgen_stop_device(pkt_dev);
3455 /* This is max DELAY, this has special meaning of
3458 if (unlikely(pkt_dev->delay == ULLONG_MAX)) {
3459 pkt_dev->next_tx = ktime_add_ns(ktime_get(), ULONG_MAX);
3463 /* If no skb or clone count exhausted then get new one */
3464 if (!pkt_dev->skb || (pkt_dev->last_ok &&
3465 ++pkt_dev->clone_count >= pkt_dev->clone_skb)) {
3466 /* build a new pkt */
3467 kfree_skb(pkt_dev->skb);
3469 pkt_dev->skb = fill_packet(odev, pkt_dev);
3470 if (pkt_dev->skb == NULL) {
3471 pr_err("ERROR: couldn't allocate skb in fill_packet\n");
3473 pkt_dev->clone_count--; /* back out increment, OOM */
3476 pkt_dev->last_pkt_size = pkt_dev->skb->len;
3477 pkt_dev->clone_count = 0; /* reset counter */
3480 if (pkt_dev->delay && pkt_dev->last_ok)
3481 spin(pkt_dev, pkt_dev->next_tx);
3483 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) {
3485 skb->protocol = eth_type_trans(skb, skb->dev);
3486 refcount_add(burst, &skb->users);
3489 ret = netif_receive_skb(skb);
3490 if (ret == NET_RX_DROP)
3494 if (refcount_read(&skb->users) != burst) {
3495 /* skb was queued by rps/rfs or taps,
3496 * so cannot reuse this skb
3498 WARN_ON(refcount_sub_and_test(burst - 1, &skb->users));
3499 /* get out of the loop and wait
3500 * until skb is consumed
3504 /* skb was 'freed' by stack, so clean few
3507 skb_reset_redirect(skb);
3508 } while (--burst > 0);
3509 goto out; /* Skips xmit_mode M_START_XMIT */
3510 } else if (pkt_dev->xmit_mode == M_QUEUE_XMIT) {
3512 refcount_inc(&pkt_dev->skb->users);
3514 ret = dev_queue_xmit(pkt_dev->skb);
3516 case NET_XMIT_SUCCESS:
3519 pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3523 /* These are all valid return codes for a qdisc but
3524 * indicate packets are being dropped or will likely
3527 case NETDEV_TX_BUSY:
3528 /* qdisc may call dev_hard_start_xmit directly in cases
3529 * where no queues exist e.g. loopback device, virtual
3530 * devices, etc. In this case we need to handle
3535 net_info_ratelimited("%s xmit error: %d\n",
3536 pkt_dev->odevname, ret);
3542 txq = skb_get_tx_queue(odev, pkt_dev->skb);
3546 HARD_TX_LOCK(odev, txq, smp_processor_id());
3548 if (unlikely(netif_xmit_frozen_or_drv_stopped(txq))) {
3549 pkt_dev->last_ok = 0;
3552 refcount_add(burst, &pkt_dev->skb->users);
3555 ret = netdev_start_xmit(pkt_dev->skb, odev, txq, --burst > 0);
3559 pkt_dev->last_ok = 1;
3562 pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3563 if (burst > 0 && !netif_xmit_frozen_or_drv_stopped(txq))
3568 /* skb has been consumed */
3571 default: /* Drivers are not supposed to return other values! */
3572 net_info_ratelimited("%s xmit error: %d\n",
3573 pkt_dev->odevname, ret);
3576 case NETDEV_TX_BUSY:
3577 /* Retry it next time */
3578 refcount_dec(&(pkt_dev->skb->users));
3579 pkt_dev->last_ok = 0;
3581 if (unlikely(burst))
3582 WARN_ON(refcount_sub_and_test(burst, &pkt_dev->skb->users));
3584 HARD_TX_UNLOCK(odev, txq);
3589 /* If pkt_dev->count is zero, then run forever */
3590 if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) {
3591 pktgen_wait_for_skb(pkt_dev);
3593 /* Done with this */
3594 pktgen_stop_device(pkt_dev);
3599 * Main loop of the thread goes here
3602 static int pktgen_thread_worker(void *arg)
3604 struct pktgen_thread *t = arg;
3605 struct pktgen_dev *pkt_dev = NULL;
3608 WARN_ON(smp_processor_id() != cpu);
3610 init_waitqueue_head(&t->queue);
3611 complete(&t->start_done);
3613 pr_debug("starting pktgen/%d: pid=%d\n", cpu, task_pid_nr(current));
3617 while (!kthread_should_stop()) {
3618 pkt_dev = next_to_run(t);
3620 if (unlikely(!pkt_dev && t->control == 0)) {
3621 if (t->net->pktgen_exiting)
3623 wait_event_interruptible_timeout(t->queue,
3630 if (likely(pkt_dev)) {
3631 pktgen_xmit(pkt_dev);
3634 pktgen_resched(pkt_dev);
3639 if (t->control & T_STOP) {
3641 t->control &= ~(T_STOP);
3644 if (t->control & T_RUN) {
3646 t->control &= ~(T_RUN);
3649 if (t->control & T_REMDEVALL) {
3650 pktgen_rem_all_ifs(t);
3651 t->control &= ~(T_REMDEVALL);
3654 if (t->control & T_REMDEV) {
3655 pktgen_rem_one_if(t);
3656 t->control &= ~(T_REMDEV);
3662 pr_debug("%s stopping all device\n", t->tsk->comm);
3665 pr_debug("%s removing all device\n", t->tsk->comm);
3666 pktgen_rem_all_ifs(t);
3668 pr_debug("%s removing thread\n", t->tsk->comm);
3669 pktgen_rem_thread(t);
3674 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
3675 const char *ifname, bool exact)
3677 struct pktgen_dev *p, *pkt_dev = NULL;
3678 size_t len = strlen(ifname);
3681 list_for_each_entry_rcu(p, &t->if_list, list)
3682 if (strncmp(p->odevname, ifname, len) == 0) {
3683 if (p->odevname[len]) {
3684 if (exact || p->odevname[len] != '@')
3692 pr_debug("find_dev(%s) returning %p\n", ifname, pkt_dev);
3697 * Adds a dev at front of if_list.
3700 static int add_dev_to_thread(struct pktgen_thread *t,
3701 struct pktgen_dev *pkt_dev)
3705 /* This function cannot be called concurrently, as its called
3706 * under pktgen_thread_lock mutex, but it can run from
3707 * userspace on another CPU than the kthread. The if_lock()
3708 * is used here to sync with concurrent instances of
3709 * _rem_dev_from_if_list() invoked via kthread, which is also
3710 * updating the if_list */
3713 if (pkt_dev->pg_thread) {
3714 pr_err("ERROR: already assigned to a thread\n");
3719 pkt_dev->running = 0;
3720 pkt_dev->pg_thread = t;
3721 list_add_rcu(&pkt_dev->list, &t->if_list);
3728 /* Called under thread lock */
3730 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname)
3732 struct pktgen_dev *pkt_dev;
3734 int node = cpu_to_node(t->cpu);
3736 /* We don't allow a device to be on several threads */
3738 pkt_dev = __pktgen_NN_threads(t->net, ifname, FIND);
3740 pr_err("ERROR: interface already used\n");
3744 pkt_dev = kzalloc_node(sizeof(struct pktgen_dev), GFP_KERNEL, node);
3748 strcpy(pkt_dev->odevname, ifname);
3749 pkt_dev->flows = vzalloc_node(array_size(MAX_CFLOWS,
3750 sizeof(struct flow_state)),
3752 if (pkt_dev->flows == NULL) {
3757 pkt_dev->removal_mark = 0;
3758 pkt_dev->nfrags = 0;
3759 pkt_dev->delay = pg_delay_d;
3760 pkt_dev->count = pg_count_d;
3762 pkt_dev->udp_src_min = 9; /* sink port */
3763 pkt_dev->udp_src_max = 9;
3764 pkt_dev->udp_dst_min = 9;
3765 pkt_dev->udp_dst_max = 9;
3766 pkt_dev->vlan_p = 0;
3767 pkt_dev->vlan_cfi = 0;
3768 pkt_dev->vlan_id = 0xffff;
3769 pkt_dev->svlan_p = 0;
3770 pkt_dev->svlan_cfi = 0;
3771 pkt_dev->svlan_id = 0xffff;
3773 pkt_dev->node = NUMA_NO_NODE;
3775 err = pktgen_setup_dev(t->net, pkt_dev, ifname);
3778 if (pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)
3779 pkt_dev->clone_skb = pg_clone_skb_d;
3781 pkt_dev->entry = proc_create_data(ifname, 0600, t->net->proc_dir,
3782 &pktgen_if_proc_ops, pkt_dev);
3783 if (!pkt_dev->entry) {
3784 pr_err("cannot create %s/%s procfs entry\n",
3785 PG_PROC_DIR, ifname);
3790 pkt_dev->ipsmode = XFRM_MODE_TRANSPORT;
3791 pkt_dev->ipsproto = IPPROTO_ESP;
3793 /* xfrm tunnel mode needs additional dst to extract outter
3794 * ip header protocol/ttl/id field, here creat a phony one.
3795 * instead of looking for a valid rt, which definitely hurting
3796 * performance under such circumstance.
3798 pkt_dev->dstops.family = AF_INET;
3799 pkt_dev->xdst.u.dst.dev = pkt_dev->odev;
3800 dst_init_metrics(&pkt_dev->xdst.u.dst, pktgen_dst_metrics, false);
3801 pkt_dev->xdst.child = &pkt_dev->xdst.u.dst;
3802 pkt_dev->xdst.u.dst.ops = &pkt_dev->dstops;
3805 return add_dev_to_thread(t, pkt_dev);
3807 netdev_put(pkt_dev->odev, &pkt_dev->dev_tracker);
3812 vfree(pkt_dev->flows);
3817 static int __net_init pktgen_create_thread(int cpu, struct pktgen_net *pn)
3819 struct pktgen_thread *t;
3820 struct proc_dir_entry *pe;
3821 struct task_struct *p;
3823 t = kzalloc_node(sizeof(struct pktgen_thread), GFP_KERNEL,
3826 pr_err("ERROR: out of memory, can't create new thread\n");
3830 mutex_init(&t->if_lock);
3833 INIT_LIST_HEAD(&t->if_list);
3835 list_add_tail(&t->th_list, &pn->pktgen_threads);
3836 init_completion(&t->start_done);
3838 p = kthread_create_on_node(pktgen_thread_worker,
3841 "kpktgend_%d", cpu);
3843 pr_err("kthread_create_on_node() failed for cpu %d\n", t->cpu);
3844 list_del(&t->th_list);
3848 kthread_bind(p, cpu);
3851 pe = proc_create_data(t->tsk->comm, 0600, pn->proc_dir,
3852 &pktgen_thread_proc_ops, t);
3854 pr_err("cannot create %s/%s procfs entry\n",
3855 PG_PROC_DIR, t->tsk->comm);
3857 list_del(&t->th_list);
3865 wait_for_completion(&t->start_done);
3871 * Removes a device from the thread if_list.
3873 static void _rem_dev_from_if_list(struct pktgen_thread *t,
3874 struct pktgen_dev *pkt_dev)
3876 struct list_head *q, *n;
3877 struct pktgen_dev *p;
3880 list_for_each_safe(q, n, &t->if_list) {
3881 p = list_entry(q, struct pktgen_dev, list);
3883 list_del_rcu(&p->list);
3888 static int pktgen_remove_device(struct pktgen_thread *t,
3889 struct pktgen_dev *pkt_dev)
3891 pr_debug("remove_device pkt_dev=%p\n", pkt_dev);
3893 if (pkt_dev->running) {
3894 pr_warn("WARNING: trying to remove a running interface, stopping it now\n");
3895 pktgen_stop_device(pkt_dev);
3898 /* Dis-associate from the interface */
3900 if (pkt_dev->odev) {
3901 netdev_put(pkt_dev->odev, &pkt_dev->dev_tracker);
3902 pkt_dev->odev = NULL;
3905 /* Remove proc before if_list entry, because add_device uses
3906 * list to determine if interface already exist, avoid race
3907 * with proc_create_data() */
3908 proc_remove(pkt_dev->entry);
3910 /* And update the thread if_list */
3911 _rem_dev_from_if_list(t, pkt_dev);
3916 vfree(pkt_dev->flows);
3918 put_page(pkt_dev->page);
3919 kfree_rcu(pkt_dev, rcu);
3923 static int __net_init pg_net_init(struct net *net)
3925 struct pktgen_net *pn = net_generic(net, pg_net_id);
3926 struct proc_dir_entry *pe;
3930 INIT_LIST_HEAD(&pn->pktgen_threads);
3931 pn->pktgen_exiting = false;
3932 pn->proc_dir = proc_mkdir(PG_PROC_DIR, pn->net->proc_net);
3933 if (!pn->proc_dir) {
3934 pr_warn("cannot create /proc/net/%s\n", PG_PROC_DIR);
3937 pe = proc_create(PGCTRL, 0600, pn->proc_dir, &pktgen_proc_ops);
3939 pr_err("cannot create %s procfs entry\n", PGCTRL);
3944 for_each_online_cpu(cpu) {
3947 err = pktgen_create_thread(cpu, pn);
3949 pr_warn("Cannot create thread for cpu %d (%d)\n",
3953 if (list_empty(&pn->pktgen_threads)) {
3954 pr_err("Initialization failed for all threads\n");
3962 remove_proc_entry(PGCTRL, pn->proc_dir);
3964 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3968 static void __net_exit pg_net_exit(struct net *net)
3970 struct pktgen_net *pn = net_generic(net, pg_net_id);
3971 struct pktgen_thread *t;
3972 struct list_head *q, *n;
3975 /* Stop all interfaces & threads */
3976 pn->pktgen_exiting = true;
3978 mutex_lock(&pktgen_thread_lock);
3979 list_splice_init(&pn->pktgen_threads, &list);
3980 mutex_unlock(&pktgen_thread_lock);
3982 list_for_each_safe(q, n, &list) {
3983 t = list_entry(q, struct pktgen_thread, th_list);
3984 list_del(&t->th_list);
3985 kthread_stop(t->tsk);
3986 put_task_struct(t->tsk);
3990 remove_proc_entry(PGCTRL, pn->proc_dir);
3991 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3994 static struct pernet_operations pg_net_ops = {
3995 .init = pg_net_init,
3996 .exit = pg_net_exit,
3998 .size = sizeof(struct pktgen_net),
4001 static int __init pg_init(void)
4005 pr_info("%s", version);
4006 ret = register_pernet_subsys(&pg_net_ops);
4009 ret = register_netdevice_notifier(&pktgen_notifier_block);
4011 unregister_pernet_subsys(&pg_net_ops);
4016 static void __exit pg_cleanup(void)
4018 unregister_netdevice_notifier(&pktgen_notifier_block);
4019 unregister_pernet_subsys(&pg_net_ops);
4020 /* Don't need rcu_barrier() due to use of kfree_rcu() */
4023 module_init(pg_init);
4024 module_exit(pg_cleanup);
4026 MODULE_AUTHOR("Robert Olsson <robert.olsson@its.uu.se>");
4027 MODULE_DESCRIPTION("Packet Generator tool");
4028 MODULE_LICENSE("GPL");
4029 MODULE_VERSION(VERSION);
4030 module_param(pg_count_d, int, 0);
4031 MODULE_PARM_DESC(pg_count_d, "Default number of packets to inject");
4032 module_param(pg_delay_d, int, 0);
4033 MODULE_PARM_DESC(pg_delay_d, "Default delay between packets (nanoseconds)");
4034 module_param(pg_clone_skb_d, int, 0);
4035 MODULE_PARM_DESC(pg_clone_skb_d, "Default number of copies of the same packet");
4036 module_param(debug, int, 0);
4037 MODULE_PARM_DESC(debug, "Enable debugging of pktgen module");