Merge branches 'pm-cpufreq-fixes' and 'pm-cpuidle-fixes'
[linux/fpc-iii.git] / net / core / pktgen.c
blob96947f5d41e42a9a42f91c40457b87824756510e
1 /*
2 * Authors:
3 * Copyright 2001, 2002 by Robert Olsson <robert.olsson@its.uu.se>
4 * Uppsala University and
5 * Swedish University of Agricultural Sciences
7 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
8 * Ben Greear <greearb@candelatech.com>
9 * Jens Låås <jens.laas@data.slu.se>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
17 * A tool for loading the network with preconfigurated packets.
18 * The tool is implemented as a linux module. Parameters are output
19 * device, delay (to hard_xmit), number of packets, and whether
20 * to use multiple SKBs or just the same one.
21 * pktgen uses the installed interface's output routine.
23 * Additional hacking by:
25 * Jens.Laas@data.slu.se
26 * Improved by ANK. 010120.
27 * Improved by ANK even more. 010212.
28 * MAC address typo fixed. 010417 --ro
29 * Integrated. 020301 --DaveM
30 * Added multiskb option 020301 --DaveM
31 * Scaling of results. 020417--sigurdur@linpro.no
32 * Significant re-work of the module:
33 * * Convert to threaded model to more efficiently be able to transmit
34 * and receive on multiple interfaces at once.
35 * * Converted many counters to __u64 to allow longer runs.
36 * * Allow configuration of ranges, like min/max IP address, MACs,
37 * and UDP-ports, for both source and destination, and can
38 * set to use a random distribution or sequentially walk the range.
39 * * Can now change most values after starting.
40 * * Place 12-byte packet in UDP payload with magic number,
41 * sequence number, and timestamp.
42 * * Add receiver code that detects dropped pkts, re-ordered pkts, and
43 * latencies (with micro-second) precision.
44 * * Add IOCTL interface to easily get counters & configuration.
45 * --Ben Greear <greearb@candelatech.com>
47 * Renamed multiskb to clone_skb and cleaned up sending core for two distinct
48 * skb modes. A clone_skb=0 mode for Ben "ranges" work and a clone_skb != 0
49 * as a "fastpath" with a configurable number of clones after alloc's.
50 * clone_skb=0 means all packets are allocated this also means ranges time
51 * stamps etc can be used. clone_skb=100 means 1 malloc is followed by 100
52 * clones.
54 * Also moved to /proc/net/pktgen/
55 * --ro
57 * Sept 10: Fixed threading/locking. Lots of bone-headed and more clever
58 * mistakes. Also merged in DaveM's patch in the -pre6 patch.
59 * --Ben Greear <greearb@candelatech.com>
61 * Integrated to 2.5.x 021029 --Lucio Maciel (luciomaciel@zipmail.com.br)
64 * 021124 Finished major redesign and rewrite for new functionality.
65 * See Documentation/networking/pktgen.txt for how to use this.
67 * The new operation:
68 * For each CPU one thread/process is created at start. This process checks
69 * for running devices in the if_list and sends packets until count is 0 it
70 * also the thread checks the thread->control which is used for inter-process
71 * communication. controlling process "posts" operations to the threads this
72 * way.
73 * The if_list is RCU protected, and the if_lock remains to protect updating
74 * of if_list, from "add_device" as it invoked from userspace (via proc write).
76 * By design there should only be *one* "controlling" process. In practice
77 * multiple write accesses gives unpredictable result. Understood by "write"
78 * to /proc gives result code thats should be read be the "writer".
79 * For practical use this should be no problem.
81 * Note when adding devices to a specific CPU there good idea to also assign
82 * /proc/irq/XX/smp_affinity so TX-interrupts gets bound to the same CPU.
83 * --ro
85 * Fix refcount off by one if first packet fails, potential null deref,
86 * memleak 030710- KJP
88 * First "ranges" functionality for ipv6 030726 --ro
90 * Included flow support. 030802 ANK.
92 * Fixed unaligned access on IA-64 Grant Grundler <grundler@parisc-linux.org>
94 * Remove if fix from added Harald Welte <laforge@netfilter.org> 040419
95 * ia64 compilation fix from Aron Griffis <aron@hp.com> 040604
97 * New xmit() return, do_div and misc clean up by Stephen Hemminger
98 * <shemminger@osdl.org> 040923
100 * Randy Dunlap fixed u64 printk compiler warning
102 * Remove FCS from BW calculation. Lennert Buytenhek <buytenh@wantstofly.org>
103 * New time handling. Lennert Buytenhek <buytenh@wantstofly.org> 041213
105 * Corrections from Nikolai Malykh (nmalykh@bilim.com)
106 * Removed unused flags F_SET_SRCMAC & F_SET_SRCIP 041230
108 * interruptible_sleep_on_timeout() replaced Nishanth Aravamudan <nacc@us.ibm.com>
109 * 050103
111 * MPLS support by Steven Whitehouse <steve@chygwyn.com>
113 * 802.1Q/Q-in-Q support by Francesco Fondelli (FF) <francesco.fondelli@gmail.com>
115 * Fixed src_mac command to set source mac of packet to value specified in
116 * command by Adit Ranadive <adit.262@gmail.com>
120 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
122 #include <linux/sys.h>
123 #include <linux/types.h>
124 #include <linux/module.h>
125 #include <linux/moduleparam.h>
126 #include <linux/kernel.h>
127 #include <linux/mutex.h>
128 #include <linux/sched.h>
129 #include <linux/slab.h>
130 #include <linux/vmalloc.h>
131 #include <linux/unistd.h>
132 #include <linux/string.h>
133 #include <linux/ptrace.h>
134 #include <linux/errno.h>
135 #include <linux/ioport.h>
136 #include <linux/interrupt.h>
137 #include <linux/capability.h>
138 #include <linux/hrtimer.h>
139 #include <linux/freezer.h>
140 #include <linux/delay.h>
141 #include <linux/timer.h>
142 #include <linux/list.h>
143 #include <linux/init.h>
144 #include <linux/skbuff.h>
145 #include <linux/netdevice.h>
146 #include <linux/inet.h>
147 #include <linux/inetdevice.h>
148 #include <linux/rtnetlink.h>
149 #include <linux/if_arp.h>
150 #include <linux/if_vlan.h>
151 #include <linux/in.h>
152 #include <linux/ip.h>
153 #include <linux/ipv6.h>
154 #include <linux/udp.h>
155 #include <linux/proc_fs.h>
156 #include <linux/seq_file.h>
157 #include <linux/wait.h>
158 #include <linux/etherdevice.h>
159 #include <linux/kthread.h>
160 #include <linux/prefetch.h>
161 #include <net/net_namespace.h>
162 #include <net/checksum.h>
163 #include <net/ipv6.h>
164 #include <net/udp.h>
165 #include <net/ip6_checksum.h>
166 #include <net/addrconf.h>
167 #ifdef CONFIG_XFRM
168 #include <net/xfrm.h>
169 #endif
170 #include <net/netns/generic.h>
171 #include <asm/byteorder.h>
172 #include <linux/rcupdate.h>
173 #include <linux/bitops.h>
174 #include <linux/io.h>
175 #include <linux/timex.h>
176 #include <linux/uaccess.h>
177 #include <asm/dma.h>
178 #include <asm/div64.h> /* do_div */
180 #define VERSION "2.75"
181 #define IP_NAME_SZ 32
182 #define MAX_MPLS_LABELS 16 /* This is the max label stack depth */
183 #define MPLS_STACK_BOTTOM htonl(0x00000100)
185 #define func_enter() pr_debug("entering %s\n", __func__);
187 /* Device flag bits */
188 #define F_IPSRC_RND (1<<0) /* IP-Src Random */
189 #define F_IPDST_RND (1<<1) /* IP-Dst Random */
190 #define F_UDPSRC_RND (1<<2) /* UDP-Src Random */
191 #define F_UDPDST_RND (1<<3) /* UDP-Dst Random */
192 #define F_MACSRC_RND (1<<4) /* MAC-Src Random */
193 #define F_MACDST_RND (1<<5) /* MAC-Dst Random */
194 #define F_TXSIZE_RND (1<<6) /* Transmit size is random */
195 #define F_IPV6 (1<<7) /* Interface in IPV6 Mode */
196 #define F_MPLS_RND (1<<8) /* Random MPLS labels */
197 #define F_VID_RND (1<<9) /* Random VLAN ID */
198 #define F_SVID_RND (1<<10) /* Random SVLAN ID */
199 #define F_FLOW_SEQ (1<<11) /* Sequential flows */
200 #define F_IPSEC_ON (1<<12) /* ipsec on for flows */
201 #define F_QUEUE_MAP_RND (1<<13) /* queue map Random */
202 #define F_QUEUE_MAP_CPU (1<<14) /* queue map mirrors smp_processor_id() */
203 #define F_NODE (1<<15) /* Node memory alloc*/
204 #define F_UDPCSUM (1<<16) /* Include UDP checksum */
205 #define F_NO_TIMESTAMP (1<<17) /* Don't timestamp packets (default TS) */
207 /* Thread control flag bits */
208 #define T_STOP (1<<0) /* Stop run */
209 #define T_RUN (1<<1) /* Start run */
210 #define T_REMDEVALL (1<<2) /* Remove all devs */
211 #define T_REMDEV (1<<3) /* Remove one dev */
213 /* Xmit modes */
214 #define M_START_XMIT 0 /* Default normal TX */
215 #define M_NETIF_RECEIVE 1 /* Inject packets into stack */
216 #define M_QUEUE_XMIT 2 /* Inject packet into qdisc */
218 /* If lock -- protects updating of if_list */
219 #define if_lock(t) mutex_lock(&(t->if_lock));
220 #define if_unlock(t) mutex_unlock(&(t->if_lock));
222 /* Used to help with determining the pkts on receive */
223 #define PKTGEN_MAGIC 0xbe9be955
224 #define PG_PROC_DIR "pktgen"
225 #define PGCTRL "pgctrl"
227 #define MAX_CFLOWS 65536
229 #define VLAN_TAG_SIZE(x) ((x)->vlan_id == 0xffff ? 0 : 4)
230 #define SVLAN_TAG_SIZE(x) ((x)->svlan_id == 0xffff ? 0 : 4)
232 struct flow_state {
233 __be32 cur_daddr;
234 int count;
235 #ifdef CONFIG_XFRM
236 struct xfrm_state *x;
237 #endif
238 __u32 flags;
241 /* flow flag bits */
242 #define F_INIT (1<<0) /* flow has been initialized */
244 struct pktgen_dev {
246 * Try to keep frequent/infrequent used vars. separated.
248 struct proc_dir_entry *entry; /* proc file */
249 struct pktgen_thread *pg_thread;/* the owner */
250 struct list_head list; /* chaining in the thread's run-queue */
251 struct rcu_head rcu; /* freed by RCU */
253 int running; /* if false, the test will stop */
255 /* If min != max, then we will either do a linear iteration, or
256 * we will do a random selection from within the range.
258 __u32 flags;
259 int xmit_mode;
260 int min_pkt_size;
261 int max_pkt_size;
262 int pkt_overhead; /* overhead for MPLS, VLANs, IPSEC etc */
263 int nfrags;
264 int removal_mark; /* non-zero => the device is marked for
265 * removal by worker thread */
267 struct page *page;
268 u64 delay; /* nano-seconds */
270 __u64 count; /* Default No packets to send */
271 __u64 sofar; /* How many pkts we've sent so far */
272 __u64 tx_bytes; /* How many bytes we've transmitted */
273 __u64 errors; /* Errors when trying to transmit, */
275 /* runtime counters relating to clone_skb */
277 __u32 clone_count;
278 int last_ok; /* Was last skb sent?
279 * Or a failed transmit of some sort?
280 * This will keep sequence numbers in order
282 ktime_t next_tx;
283 ktime_t started_at;
284 ktime_t stopped_at;
285 u64 idle_acc; /* nano-seconds */
287 __u32 seq_num;
289 int clone_skb; /*
290 * Use multiple SKBs during packet gen.
291 * If this number is greater than 1, then
292 * that many copies of the same packet will be
293 * sent before a new packet is allocated.
294 * If you want to send 1024 identical packets
295 * before creating a new packet,
296 * set clone_skb to 1024.
299 char dst_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
300 char dst_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
301 char src_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
302 char src_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
304 struct in6_addr in6_saddr;
305 struct in6_addr in6_daddr;
306 struct in6_addr cur_in6_daddr;
307 struct in6_addr cur_in6_saddr;
308 /* For ranges */
309 struct in6_addr min_in6_daddr;
310 struct in6_addr max_in6_daddr;
311 struct in6_addr min_in6_saddr;
312 struct in6_addr max_in6_saddr;
314 /* If we're doing ranges, random or incremental, then this
315 * defines the min/max for those ranges.
317 __be32 saddr_min; /* inclusive, source IP address */
318 __be32 saddr_max; /* exclusive, source IP address */
319 __be32 daddr_min; /* inclusive, dest IP address */
320 __be32 daddr_max; /* exclusive, dest IP address */
322 __u16 udp_src_min; /* inclusive, source UDP port */
323 __u16 udp_src_max; /* exclusive, source UDP port */
324 __u16 udp_dst_min; /* inclusive, dest UDP port */
325 __u16 udp_dst_max; /* exclusive, dest UDP port */
327 /* DSCP + ECN */
328 __u8 tos; /* six MSB of (former) IPv4 TOS
329 are for dscp codepoint */
330 __u8 traffic_class; /* ditto for the (former) Traffic Class in IPv6
331 (see RFC 3260, sec. 4) */
333 /* MPLS */
334 unsigned int nr_labels; /* Depth of stack, 0 = no MPLS */
335 __be32 labels[MAX_MPLS_LABELS];
337 /* VLAN/SVLAN (802.1Q/Q-in-Q) */
338 __u8 vlan_p;
339 __u8 vlan_cfi;
340 __u16 vlan_id; /* 0xffff means no vlan tag */
342 __u8 svlan_p;
343 __u8 svlan_cfi;
344 __u16 svlan_id; /* 0xffff means no svlan tag */
346 __u32 src_mac_count; /* How many MACs to iterate through */
347 __u32 dst_mac_count; /* How many MACs to iterate through */
349 unsigned char dst_mac[ETH_ALEN];
350 unsigned char src_mac[ETH_ALEN];
352 __u32 cur_dst_mac_offset;
353 __u32 cur_src_mac_offset;
354 __be32 cur_saddr;
355 __be32 cur_daddr;
356 __u16 ip_id;
357 __u16 cur_udp_dst;
358 __u16 cur_udp_src;
359 __u16 cur_queue_map;
360 __u32 cur_pkt_size;
361 __u32 last_pkt_size;
363 __u8 hh[14];
364 /* = {
365 0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB,
367 We fill in SRC address later
368 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
369 0x08, 0x00
372 __u16 pad; /* pad out the hh struct to an even 16 bytes */
374 struct sk_buff *skb; /* skb we are to transmit next, used for when we
375 * are transmitting the same one multiple times
377 struct net_device *odev; /* The out-going device.
378 * Note that the device should have it's
379 * pg_info pointer pointing back to this
380 * device.
381 * Set when the user specifies the out-going
382 * device name (not when the inject is
383 * started as it used to do.)
385 char odevname[32];
386 struct flow_state *flows;
387 unsigned int cflows; /* Concurrent flows (config) */
388 unsigned int lflow; /* Flow length (config) */
389 unsigned int nflows; /* accumulated flows (stats) */
390 unsigned int curfl; /* current sequenced flow (state)*/
392 u16 queue_map_min;
393 u16 queue_map_max;
394 __u32 skb_priority; /* skb priority field */
395 unsigned int burst; /* number of duplicated packets to burst */
396 int node; /* Memory node */
398 #ifdef CONFIG_XFRM
399 __u8 ipsmode; /* IPSEC mode (config) */
400 __u8 ipsproto; /* IPSEC type (config) */
401 __u32 spi;
402 struct dst_entry dst;
403 struct dst_ops dstops;
404 #endif
405 char result[512];
408 struct pktgen_hdr {
409 __be32 pgh_magic;
410 __be32 seq_num;
411 __be32 tv_sec;
412 __be32 tv_usec;
416 static unsigned int pg_net_id __read_mostly;
418 struct pktgen_net {
419 struct net *net;
420 struct proc_dir_entry *proc_dir;
421 struct list_head pktgen_threads;
422 bool pktgen_exiting;
425 struct pktgen_thread {
426 struct mutex if_lock; /* for list of devices */
427 struct list_head if_list; /* All device here */
428 struct list_head th_list;
429 struct task_struct *tsk;
430 char result[512];
432 /* Field for thread to receive "posted" events terminate,
433 stop ifs etc. */
435 u32 control;
436 int cpu;
438 wait_queue_head_t queue;
439 struct completion start_done;
440 struct pktgen_net *net;
443 #define REMOVE 1
444 #define FIND 0
446 static const char version[] =
447 "Packet Generator for packet performance testing. "
448 "Version: " VERSION "\n";
450 static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i);
451 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname);
452 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
453 const char *ifname, bool exact);
454 static int pktgen_device_event(struct notifier_block *, unsigned long, void *);
455 static void pktgen_run_all_threads(struct pktgen_net *pn);
456 static void pktgen_reset_all_threads(struct pktgen_net *pn);
457 static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn);
459 static void pktgen_stop(struct pktgen_thread *t);
460 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev);
462 /* Module parameters, defaults. */
463 static int pg_count_d __read_mostly = 1000;
464 static int pg_delay_d __read_mostly;
465 static int pg_clone_skb_d __read_mostly;
466 static int debug __read_mostly;
468 static DEFINE_MUTEX(pktgen_thread_lock);
470 static struct notifier_block pktgen_notifier_block = {
471 .notifier_call = pktgen_device_event,
475 * /proc handling functions
479 static int pgctrl_show(struct seq_file *seq, void *v)
481 seq_puts(seq, version);
482 return 0;
485 static ssize_t pgctrl_write(struct file *file, const char __user *buf,
486 size_t count, loff_t *ppos)
488 char data[128];
489 struct pktgen_net *pn = net_generic(current->nsproxy->net_ns, pg_net_id);
491 if (!capable(CAP_NET_ADMIN))
492 return -EPERM;
494 if (count == 0)
495 return -EINVAL;
497 if (count > sizeof(data))
498 count = sizeof(data);
500 if (copy_from_user(data, buf, count))
501 return -EFAULT;
503 data[count - 1] = 0; /* Strip trailing '\n' and terminate string */
505 if (!strcmp(data, "stop"))
506 pktgen_stop_all_threads_ifs(pn);
508 else if (!strcmp(data, "start"))
509 pktgen_run_all_threads(pn);
511 else if (!strcmp(data, "reset"))
512 pktgen_reset_all_threads(pn);
514 else
515 return -EINVAL;
517 return count;
520 static int pgctrl_open(struct inode *inode, struct file *file)
522 return single_open(file, pgctrl_show, PDE_DATA(inode));
525 static const struct file_operations pktgen_fops = {
526 .owner = THIS_MODULE,
527 .open = pgctrl_open,
528 .read = seq_read,
529 .llseek = seq_lseek,
530 .write = pgctrl_write,
531 .release = single_release,
534 static int pktgen_if_show(struct seq_file *seq, void *v)
536 const struct pktgen_dev *pkt_dev = seq->private;
537 ktime_t stopped;
538 u64 idle;
540 seq_printf(seq,
541 "Params: count %llu min_pkt_size: %u max_pkt_size: %u\n",
542 (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size,
543 pkt_dev->max_pkt_size);
545 seq_printf(seq,
546 " frags: %d delay: %llu clone_skb: %d ifname: %s\n",
547 pkt_dev->nfrags, (unsigned long long) pkt_dev->delay,
548 pkt_dev->clone_skb, pkt_dev->odevname);
550 seq_printf(seq, " flows: %u flowlen: %u\n", pkt_dev->cflows,
551 pkt_dev->lflow);
553 seq_printf(seq,
554 " queue_map_min: %u queue_map_max: %u\n",
555 pkt_dev->queue_map_min,
556 pkt_dev->queue_map_max);
558 if (pkt_dev->skb_priority)
559 seq_printf(seq, " skb_priority: %u\n",
560 pkt_dev->skb_priority);
562 if (pkt_dev->flags & F_IPV6) {
563 seq_printf(seq,
564 " saddr: %pI6c min_saddr: %pI6c max_saddr: %pI6c\n"
565 " daddr: %pI6c min_daddr: %pI6c max_daddr: %pI6c\n",
566 &pkt_dev->in6_saddr,
567 &pkt_dev->min_in6_saddr, &pkt_dev->max_in6_saddr,
568 &pkt_dev->in6_daddr,
569 &pkt_dev->min_in6_daddr, &pkt_dev->max_in6_daddr);
570 } else {
571 seq_printf(seq,
572 " dst_min: %s dst_max: %s\n",
573 pkt_dev->dst_min, pkt_dev->dst_max);
574 seq_printf(seq,
575 " src_min: %s src_max: %s\n",
576 pkt_dev->src_min, pkt_dev->src_max);
579 seq_puts(seq, " src_mac: ");
581 seq_printf(seq, "%pM ",
582 is_zero_ether_addr(pkt_dev->src_mac) ?
583 pkt_dev->odev->dev_addr : pkt_dev->src_mac);
585 seq_puts(seq, "dst_mac: ");
586 seq_printf(seq, "%pM\n", pkt_dev->dst_mac);
588 seq_printf(seq,
589 " udp_src_min: %d udp_src_max: %d"
590 " udp_dst_min: %d udp_dst_max: %d\n",
591 pkt_dev->udp_src_min, pkt_dev->udp_src_max,
592 pkt_dev->udp_dst_min, pkt_dev->udp_dst_max);
594 seq_printf(seq,
595 " src_mac_count: %d dst_mac_count: %d\n",
596 pkt_dev->src_mac_count, pkt_dev->dst_mac_count);
598 if (pkt_dev->nr_labels) {
599 unsigned int i;
600 seq_puts(seq, " mpls: ");
601 for (i = 0; i < pkt_dev->nr_labels; i++)
602 seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]),
603 i == pkt_dev->nr_labels-1 ? "\n" : ", ");
606 if (pkt_dev->vlan_id != 0xffff)
607 seq_printf(seq, " vlan_id: %u vlan_p: %u vlan_cfi: %u\n",
608 pkt_dev->vlan_id, pkt_dev->vlan_p,
609 pkt_dev->vlan_cfi);
611 if (pkt_dev->svlan_id != 0xffff)
612 seq_printf(seq, " svlan_id: %u vlan_p: %u vlan_cfi: %u\n",
613 pkt_dev->svlan_id, pkt_dev->svlan_p,
614 pkt_dev->svlan_cfi);
616 if (pkt_dev->tos)
617 seq_printf(seq, " tos: 0x%02x\n", pkt_dev->tos);
619 if (pkt_dev->traffic_class)
620 seq_printf(seq, " traffic_class: 0x%02x\n", pkt_dev->traffic_class);
622 if (pkt_dev->burst > 1)
623 seq_printf(seq, " burst: %d\n", pkt_dev->burst);
625 if (pkt_dev->node >= 0)
626 seq_printf(seq, " node: %d\n", pkt_dev->node);
628 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE)
629 seq_puts(seq, " xmit_mode: netif_receive\n");
630 else if (pkt_dev->xmit_mode == M_QUEUE_XMIT)
631 seq_puts(seq, " xmit_mode: xmit_queue\n");
633 seq_puts(seq, " Flags: ");
635 if (pkt_dev->flags & F_IPV6)
636 seq_puts(seq, "IPV6 ");
638 if (pkt_dev->flags & F_IPSRC_RND)
639 seq_puts(seq, "IPSRC_RND ");
641 if (pkt_dev->flags & F_IPDST_RND)
642 seq_puts(seq, "IPDST_RND ");
644 if (pkt_dev->flags & F_TXSIZE_RND)
645 seq_puts(seq, "TXSIZE_RND ");
647 if (pkt_dev->flags & F_UDPSRC_RND)
648 seq_puts(seq, "UDPSRC_RND ");
650 if (pkt_dev->flags & F_UDPDST_RND)
651 seq_puts(seq, "UDPDST_RND ");
653 if (pkt_dev->flags & F_UDPCSUM)
654 seq_puts(seq, "UDPCSUM ");
656 if (pkt_dev->flags & F_NO_TIMESTAMP)
657 seq_puts(seq, "NO_TIMESTAMP ");
659 if (pkt_dev->flags & F_MPLS_RND)
660 seq_puts(seq, "MPLS_RND ");
662 if (pkt_dev->flags & F_QUEUE_MAP_RND)
663 seq_puts(seq, "QUEUE_MAP_RND ");
665 if (pkt_dev->flags & F_QUEUE_MAP_CPU)
666 seq_puts(seq, "QUEUE_MAP_CPU ");
668 if (pkt_dev->cflows) {
669 if (pkt_dev->flags & F_FLOW_SEQ)
670 seq_puts(seq, "FLOW_SEQ "); /*in sequence flows*/
671 else
672 seq_puts(seq, "FLOW_RND ");
675 #ifdef CONFIG_XFRM
676 if (pkt_dev->flags & F_IPSEC_ON) {
677 seq_puts(seq, "IPSEC ");
678 if (pkt_dev->spi)
679 seq_printf(seq, "spi:%u", pkt_dev->spi);
681 #endif
683 if (pkt_dev->flags & F_MACSRC_RND)
684 seq_puts(seq, "MACSRC_RND ");
686 if (pkt_dev->flags & F_MACDST_RND)
687 seq_puts(seq, "MACDST_RND ");
689 if (pkt_dev->flags & F_VID_RND)
690 seq_puts(seq, "VID_RND ");
692 if (pkt_dev->flags & F_SVID_RND)
693 seq_puts(seq, "SVID_RND ");
695 if (pkt_dev->flags & F_NODE)
696 seq_puts(seq, "NODE_ALLOC ");
698 seq_puts(seq, "\n");
700 /* not really stopped, more like last-running-at */
701 stopped = pkt_dev->running ? ktime_get() : pkt_dev->stopped_at;
702 idle = pkt_dev->idle_acc;
703 do_div(idle, NSEC_PER_USEC);
705 seq_printf(seq,
706 "Current:\n pkts-sofar: %llu errors: %llu\n",
707 (unsigned long long)pkt_dev->sofar,
708 (unsigned long long)pkt_dev->errors);
710 seq_printf(seq,
711 " started: %lluus stopped: %lluus idle: %lluus\n",
712 (unsigned long long) ktime_to_us(pkt_dev->started_at),
713 (unsigned long long) ktime_to_us(stopped),
714 (unsigned long long) idle);
716 seq_printf(seq,
717 " seq_num: %d cur_dst_mac_offset: %d cur_src_mac_offset: %d\n",
718 pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset,
719 pkt_dev->cur_src_mac_offset);
721 if (pkt_dev->flags & F_IPV6) {
722 seq_printf(seq, " cur_saddr: %pI6c cur_daddr: %pI6c\n",
723 &pkt_dev->cur_in6_saddr,
724 &pkt_dev->cur_in6_daddr);
725 } else
726 seq_printf(seq, " cur_saddr: %pI4 cur_daddr: %pI4\n",
727 &pkt_dev->cur_saddr, &pkt_dev->cur_daddr);
729 seq_printf(seq, " cur_udp_dst: %d cur_udp_src: %d\n",
730 pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src);
732 seq_printf(seq, " cur_queue_map: %u\n", pkt_dev->cur_queue_map);
734 seq_printf(seq, " flows: %u\n", pkt_dev->nflows);
736 if (pkt_dev->result[0])
737 seq_printf(seq, "Result: %s\n", pkt_dev->result);
738 else
739 seq_puts(seq, "Result: Idle\n");
741 return 0;
745 static int hex32_arg(const char __user *user_buffer, unsigned long maxlen,
746 __u32 *num)
748 int i = 0;
749 *num = 0;
751 for (; i < maxlen; i++) {
752 int value;
753 char c;
754 *num <<= 4;
755 if (get_user(c, &user_buffer[i]))
756 return -EFAULT;
757 value = hex_to_bin(c);
758 if (value >= 0)
759 *num |= value;
760 else
761 break;
763 return i;
766 static int count_trail_chars(const char __user * user_buffer,
767 unsigned int maxlen)
769 int i;
771 for (i = 0; i < maxlen; i++) {
772 char c;
773 if (get_user(c, &user_buffer[i]))
774 return -EFAULT;
775 switch (c) {
776 case '\"':
777 case '\n':
778 case '\r':
779 case '\t':
780 case ' ':
781 case '=':
782 break;
783 default:
784 goto done;
787 done:
788 return i;
791 static long num_arg(const char __user *user_buffer, unsigned long maxlen,
792 unsigned long *num)
794 int i;
795 *num = 0;
797 for (i = 0; i < maxlen; i++) {
798 char c;
799 if (get_user(c, &user_buffer[i]))
800 return -EFAULT;
801 if ((c >= '0') && (c <= '9')) {
802 *num *= 10;
803 *num += c - '0';
804 } else
805 break;
807 return i;
810 static int strn_len(const char __user * user_buffer, unsigned int maxlen)
812 int i;
814 for (i = 0; i < maxlen; i++) {
815 char c;
816 if (get_user(c, &user_buffer[i]))
817 return -EFAULT;
818 switch (c) {
819 case '\"':
820 case '\n':
821 case '\r':
822 case '\t':
823 case ' ':
824 goto done_str;
825 default:
826 break;
829 done_str:
830 return i;
833 static ssize_t get_labels(const char __user *buffer, struct pktgen_dev *pkt_dev)
835 unsigned int n = 0;
836 char c;
837 ssize_t i = 0;
838 int len;
840 pkt_dev->nr_labels = 0;
841 do {
842 __u32 tmp;
843 len = hex32_arg(&buffer[i], 8, &tmp);
844 if (len <= 0)
845 return len;
846 pkt_dev->labels[n] = htonl(tmp);
847 if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM)
848 pkt_dev->flags |= F_MPLS_RND;
849 i += len;
850 if (get_user(c, &buffer[i]))
851 return -EFAULT;
852 i++;
853 n++;
854 if (n >= MAX_MPLS_LABELS)
855 return -E2BIG;
856 } while (c == ',');
858 pkt_dev->nr_labels = n;
859 return i;
862 static ssize_t pktgen_if_write(struct file *file,
863 const char __user * user_buffer, size_t count,
864 loff_t * offset)
866 struct seq_file *seq = file->private_data;
867 struct pktgen_dev *pkt_dev = seq->private;
868 int i, max, len;
869 char name[16], valstr[32];
870 unsigned long value = 0;
871 char *pg_result = NULL;
872 int tmp = 0;
873 char buf[128];
875 pg_result = &(pkt_dev->result[0]);
877 if (count < 1) {
878 pr_warn("wrong command format\n");
879 return -EINVAL;
882 max = count;
883 tmp = count_trail_chars(user_buffer, max);
884 if (tmp < 0) {
885 pr_warn("illegal format\n");
886 return tmp;
888 i = tmp;
890 /* Read variable name */
892 len = strn_len(&user_buffer[i], sizeof(name) - 1);
893 if (len < 0)
894 return len;
896 memset(name, 0, sizeof(name));
897 if (copy_from_user(name, &user_buffer[i], len))
898 return -EFAULT;
899 i += len;
901 max = count - i;
902 len = count_trail_chars(&user_buffer[i], max);
903 if (len < 0)
904 return len;
906 i += len;
908 if (debug) {
909 size_t copy = min_t(size_t, count, 1023);
910 char tb[copy + 1];
911 if (copy_from_user(tb, user_buffer, copy))
912 return -EFAULT;
913 tb[copy] = 0;
914 pr_debug("%s,%lu buffer -:%s:-\n",
915 name, (unsigned long)count, tb);
918 if (!strcmp(name, "min_pkt_size")) {
919 len = num_arg(&user_buffer[i], 10, &value);
920 if (len < 0)
921 return len;
923 i += len;
924 if (value < 14 + 20 + 8)
925 value = 14 + 20 + 8;
926 if (value != pkt_dev->min_pkt_size) {
927 pkt_dev->min_pkt_size = value;
928 pkt_dev->cur_pkt_size = value;
930 sprintf(pg_result, "OK: min_pkt_size=%u",
931 pkt_dev->min_pkt_size);
932 return count;
935 if (!strcmp(name, "max_pkt_size")) {
936 len = num_arg(&user_buffer[i], 10, &value);
937 if (len < 0)
938 return len;
940 i += len;
941 if (value < 14 + 20 + 8)
942 value = 14 + 20 + 8;
943 if (value != pkt_dev->max_pkt_size) {
944 pkt_dev->max_pkt_size = value;
945 pkt_dev->cur_pkt_size = value;
947 sprintf(pg_result, "OK: max_pkt_size=%u",
948 pkt_dev->max_pkt_size);
949 return count;
952 /* Shortcut for min = max */
954 if (!strcmp(name, "pkt_size")) {
955 len = num_arg(&user_buffer[i], 10, &value);
956 if (len < 0)
957 return len;
959 i += len;
960 if (value < 14 + 20 + 8)
961 value = 14 + 20 + 8;
962 if (value != pkt_dev->min_pkt_size) {
963 pkt_dev->min_pkt_size = value;
964 pkt_dev->max_pkt_size = value;
965 pkt_dev->cur_pkt_size = value;
967 sprintf(pg_result, "OK: pkt_size=%u", pkt_dev->min_pkt_size);
968 return count;
971 if (!strcmp(name, "debug")) {
972 len = num_arg(&user_buffer[i], 10, &value);
973 if (len < 0)
974 return len;
976 i += len;
977 debug = value;
978 sprintf(pg_result, "OK: debug=%u", debug);
979 return count;
982 if (!strcmp(name, "frags")) {
983 len = num_arg(&user_buffer[i], 10, &value);
984 if (len < 0)
985 return len;
987 i += len;
988 pkt_dev->nfrags = value;
989 sprintf(pg_result, "OK: frags=%u", pkt_dev->nfrags);
990 return count;
992 if (!strcmp(name, "delay")) {
993 len = num_arg(&user_buffer[i], 10, &value);
994 if (len < 0)
995 return len;
997 i += len;
998 if (value == 0x7FFFFFFF)
999 pkt_dev->delay = ULLONG_MAX;
1000 else
1001 pkt_dev->delay = (u64)value;
1003 sprintf(pg_result, "OK: delay=%llu",
1004 (unsigned long long) pkt_dev->delay);
1005 return count;
1007 if (!strcmp(name, "rate")) {
1008 len = num_arg(&user_buffer[i], 10, &value);
1009 if (len < 0)
1010 return len;
1012 i += len;
1013 if (!value)
1014 return len;
1015 pkt_dev->delay = pkt_dev->min_pkt_size*8*NSEC_PER_USEC/value;
1016 if (debug)
1017 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1019 sprintf(pg_result, "OK: rate=%lu", value);
1020 return count;
1022 if (!strcmp(name, "ratep")) {
1023 len = num_arg(&user_buffer[i], 10, &value);
1024 if (len < 0)
1025 return len;
1027 i += len;
1028 if (!value)
1029 return len;
1030 pkt_dev->delay = NSEC_PER_SEC/value;
1031 if (debug)
1032 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1034 sprintf(pg_result, "OK: rate=%lu", value);
1035 return count;
1037 if (!strcmp(name, "udp_src_min")) {
1038 len = num_arg(&user_buffer[i], 10, &value);
1039 if (len < 0)
1040 return len;
1042 i += len;
1043 if (value != pkt_dev->udp_src_min) {
1044 pkt_dev->udp_src_min = value;
1045 pkt_dev->cur_udp_src = value;
1047 sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min);
1048 return count;
1050 if (!strcmp(name, "udp_dst_min")) {
1051 len = num_arg(&user_buffer[i], 10, &value);
1052 if (len < 0)
1053 return len;
1055 i += len;
1056 if (value != pkt_dev->udp_dst_min) {
1057 pkt_dev->udp_dst_min = value;
1058 pkt_dev->cur_udp_dst = value;
1060 sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min);
1061 return count;
1063 if (!strcmp(name, "udp_src_max")) {
1064 len = num_arg(&user_buffer[i], 10, &value);
1065 if (len < 0)
1066 return len;
1068 i += len;
1069 if (value != pkt_dev->udp_src_max) {
1070 pkt_dev->udp_src_max = value;
1071 pkt_dev->cur_udp_src = value;
1073 sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max);
1074 return count;
1076 if (!strcmp(name, "udp_dst_max")) {
1077 len = num_arg(&user_buffer[i], 10, &value);
1078 if (len < 0)
1079 return len;
1081 i += len;
1082 if (value != pkt_dev->udp_dst_max) {
1083 pkt_dev->udp_dst_max = value;
1084 pkt_dev->cur_udp_dst = value;
1086 sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max);
1087 return count;
1089 if (!strcmp(name, "clone_skb")) {
1090 len = num_arg(&user_buffer[i], 10, &value);
1091 if (len < 0)
1092 return len;
1093 if ((value > 0) &&
1094 ((pkt_dev->xmit_mode == M_NETIF_RECEIVE) ||
1095 !(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))
1096 return -ENOTSUPP;
1097 i += len;
1098 pkt_dev->clone_skb = value;
1100 sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb);
1101 return count;
1103 if (!strcmp(name, "count")) {
1104 len = num_arg(&user_buffer[i], 10, &value);
1105 if (len < 0)
1106 return len;
1108 i += len;
1109 pkt_dev->count = value;
1110 sprintf(pg_result, "OK: count=%llu",
1111 (unsigned long long)pkt_dev->count);
1112 return count;
1114 if (!strcmp(name, "src_mac_count")) {
1115 len = num_arg(&user_buffer[i], 10, &value);
1116 if (len < 0)
1117 return len;
1119 i += len;
1120 if (pkt_dev->src_mac_count != value) {
1121 pkt_dev->src_mac_count = value;
1122 pkt_dev->cur_src_mac_offset = 0;
1124 sprintf(pg_result, "OK: src_mac_count=%d",
1125 pkt_dev->src_mac_count);
1126 return count;
1128 if (!strcmp(name, "dst_mac_count")) {
1129 len = num_arg(&user_buffer[i], 10, &value);
1130 if (len < 0)
1131 return len;
1133 i += len;
1134 if (pkt_dev->dst_mac_count != value) {
1135 pkt_dev->dst_mac_count = value;
1136 pkt_dev->cur_dst_mac_offset = 0;
1138 sprintf(pg_result, "OK: dst_mac_count=%d",
1139 pkt_dev->dst_mac_count);
1140 return count;
1142 if (!strcmp(name, "burst")) {
1143 len = num_arg(&user_buffer[i], 10, &value);
1144 if (len < 0)
1145 return len;
1147 i += len;
1148 if ((value > 1) &&
1149 ((pkt_dev->xmit_mode == M_QUEUE_XMIT) ||
1150 ((pkt_dev->xmit_mode == M_START_XMIT) &&
1151 (!(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))))
1152 return -ENOTSUPP;
1153 pkt_dev->burst = value < 1 ? 1 : value;
1154 sprintf(pg_result, "OK: burst=%d", pkt_dev->burst);
1155 return count;
1157 if (!strcmp(name, "node")) {
1158 len = num_arg(&user_buffer[i], 10, &value);
1159 if (len < 0)
1160 return len;
1162 i += len;
1164 if (node_possible(value)) {
1165 pkt_dev->node = value;
1166 sprintf(pg_result, "OK: node=%d", pkt_dev->node);
1167 if (pkt_dev->page) {
1168 put_page(pkt_dev->page);
1169 pkt_dev->page = NULL;
1172 else
1173 sprintf(pg_result, "ERROR: node not possible");
1174 return count;
1176 if (!strcmp(name, "xmit_mode")) {
1177 char f[32];
1179 memset(f, 0, 32);
1180 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1181 if (len < 0)
1182 return len;
1184 if (copy_from_user(f, &user_buffer[i], len))
1185 return -EFAULT;
1186 i += len;
1188 if (strcmp(f, "start_xmit") == 0) {
1189 pkt_dev->xmit_mode = M_START_XMIT;
1190 } else if (strcmp(f, "netif_receive") == 0) {
1191 /* clone_skb set earlier, not supported in this mode */
1192 if (pkt_dev->clone_skb > 0)
1193 return -ENOTSUPP;
1195 pkt_dev->xmit_mode = M_NETIF_RECEIVE;
1197 /* make sure new packet is allocated every time
1198 * pktgen_xmit() is called
1200 pkt_dev->last_ok = 1;
1202 /* override clone_skb if user passed default value
1203 * at module loading time
1205 pkt_dev->clone_skb = 0;
1206 } else if (strcmp(f, "queue_xmit") == 0) {
1207 pkt_dev->xmit_mode = M_QUEUE_XMIT;
1208 pkt_dev->last_ok = 1;
1209 } else {
1210 sprintf(pg_result,
1211 "xmit_mode -:%s:- unknown\nAvailable modes: %s",
1212 f, "start_xmit, netif_receive\n");
1213 return count;
1215 sprintf(pg_result, "OK: xmit_mode=%s", f);
1216 return count;
1218 if (!strcmp(name, "flag")) {
1219 char f[32];
1220 memset(f, 0, 32);
1221 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1222 if (len < 0)
1223 return len;
1225 if (copy_from_user(f, &user_buffer[i], len))
1226 return -EFAULT;
1227 i += len;
1228 if (strcmp(f, "IPSRC_RND") == 0)
1229 pkt_dev->flags |= F_IPSRC_RND;
1231 else if (strcmp(f, "!IPSRC_RND") == 0)
1232 pkt_dev->flags &= ~F_IPSRC_RND;
1234 else if (strcmp(f, "TXSIZE_RND") == 0)
1235 pkt_dev->flags |= F_TXSIZE_RND;
1237 else if (strcmp(f, "!TXSIZE_RND") == 0)
1238 pkt_dev->flags &= ~F_TXSIZE_RND;
1240 else if (strcmp(f, "IPDST_RND") == 0)
1241 pkt_dev->flags |= F_IPDST_RND;
1243 else if (strcmp(f, "!IPDST_RND") == 0)
1244 pkt_dev->flags &= ~F_IPDST_RND;
1246 else if (strcmp(f, "UDPSRC_RND") == 0)
1247 pkt_dev->flags |= F_UDPSRC_RND;
1249 else if (strcmp(f, "!UDPSRC_RND") == 0)
1250 pkt_dev->flags &= ~F_UDPSRC_RND;
1252 else if (strcmp(f, "UDPDST_RND") == 0)
1253 pkt_dev->flags |= F_UDPDST_RND;
1255 else if (strcmp(f, "!UDPDST_RND") == 0)
1256 pkt_dev->flags &= ~F_UDPDST_RND;
1258 else if (strcmp(f, "MACSRC_RND") == 0)
1259 pkt_dev->flags |= F_MACSRC_RND;
1261 else if (strcmp(f, "!MACSRC_RND") == 0)
1262 pkt_dev->flags &= ~F_MACSRC_RND;
1264 else if (strcmp(f, "MACDST_RND") == 0)
1265 pkt_dev->flags |= F_MACDST_RND;
1267 else if (strcmp(f, "!MACDST_RND") == 0)
1268 pkt_dev->flags &= ~F_MACDST_RND;
1270 else if (strcmp(f, "MPLS_RND") == 0)
1271 pkt_dev->flags |= F_MPLS_RND;
1273 else if (strcmp(f, "!MPLS_RND") == 0)
1274 pkt_dev->flags &= ~F_MPLS_RND;
1276 else if (strcmp(f, "VID_RND") == 0)
1277 pkt_dev->flags |= F_VID_RND;
1279 else if (strcmp(f, "!VID_RND") == 0)
1280 pkt_dev->flags &= ~F_VID_RND;
1282 else if (strcmp(f, "SVID_RND") == 0)
1283 pkt_dev->flags |= F_SVID_RND;
1285 else if (strcmp(f, "!SVID_RND") == 0)
1286 pkt_dev->flags &= ~F_SVID_RND;
1288 else if (strcmp(f, "FLOW_SEQ") == 0)
1289 pkt_dev->flags |= F_FLOW_SEQ;
1291 else if (strcmp(f, "QUEUE_MAP_RND") == 0)
1292 pkt_dev->flags |= F_QUEUE_MAP_RND;
1294 else if (strcmp(f, "!QUEUE_MAP_RND") == 0)
1295 pkt_dev->flags &= ~F_QUEUE_MAP_RND;
1297 else if (strcmp(f, "QUEUE_MAP_CPU") == 0)
1298 pkt_dev->flags |= F_QUEUE_MAP_CPU;
1300 else if (strcmp(f, "!QUEUE_MAP_CPU") == 0)
1301 pkt_dev->flags &= ~F_QUEUE_MAP_CPU;
1302 #ifdef CONFIG_XFRM
1303 else if (strcmp(f, "IPSEC") == 0)
1304 pkt_dev->flags |= F_IPSEC_ON;
1305 #endif
1307 else if (strcmp(f, "!IPV6") == 0)
1308 pkt_dev->flags &= ~F_IPV6;
1310 else if (strcmp(f, "NODE_ALLOC") == 0)
1311 pkt_dev->flags |= F_NODE;
1313 else if (strcmp(f, "!NODE_ALLOC") == 0)
1314 pkt_dev->flags &= ~F_NODE;
1316 else if (strcmp(f, "UDPCSUM") == 0)
1317 pkt_dev->flags |= F_UDPCSUM;
1319 else if (strcmp(f, "!UDPCSUM") == 0)
1320 pkt_dev->flags &= ~F_UDPCSUM;
1322 else if (strcmp(f, "NO_TIMESTAMP") == 0)
1323 pkt_dev->flags |= F_NO_TIMESTAMP;
1325 else if (strcmp(f, "!NO_TIMESTAMP") == 0)
1326 pkt_dev->flags &= ~F_NO_TIMESTAMP;
1328 else {
1329 sprintf(pg_result,
1330 "Flag -:%s:- unknown\nAvailable flags, (prepend ! to un-set flag):\n%s",
1332 "IPSRC_RND, IPDST_RND, UDPSRC_RND, UDPDST_RND, "
1333 "MACSRC_RND, MACDST_RND, TXSIZE_RND, IPV6, "
1334 "MPLS_RND, VID_RND, SVID_RND, FLOW_SEQ, "
1335 "QUEUE_MAP_RND, QUEUE_MAP_CPU, UDPCSUM, "
1336 "NO_TIMESTAMP, "
1337 #ifdef CONFIG_XFRM
1338 "IPSEC, "
1339 #endif
1340 "NODE_ALLOC\n");
1341 return count;
1343 sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags);
1344 return count;
1346 if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) {
1347 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_min) - 1);
1348 if (len < 0)
1349 return len;
1351 if (copy_from_user(buf, &user_buffer[i], len))
1352 return -EFAULT;
1353 buf[len] = 0;
1354 if (strcmp(buf, pkt_dev->dst_min) != 0) {
1355 memset(pkt_dev->dst_min, 0, sizeof(pkt_dev->dst_min));
1356 strncpy(pkt_dev->dst_min, buf, len);
1357 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
1358 pkt_dev->cur_daddr = pkt_dev->daddr_min;
1360 if (debug)
1361 pr_debug("dst_min set to: %s\n", pkt_dev->dst_min);
1362 i += len;
1363 sprintf(pg_result, "OK: dst_min=%s", pkt_dev->dst_min);
1364 return count;
1366 if (!strcmp(name, "dst_max")) {
1367 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_max) - 1);
1368 if (len < 0)
1369 return len;
1372 if (copy_from_user(buf, &user_buffer[i], len))
1373 return -EFAULT;
1375 buf[len] = 0;
1376 if (strcmp(buf, pkt_dev->dst_max) != 0) {
1377 memset(pkt_dev->dst_max, 0, sizeof(pkt_dev->dst_max));
1378 strncpy(pkt_dev->dst_max, buf, len);
1379 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
1380 pkt_dev->cur_daddr = pkt_dev->daddr_max;
1382 if (debug)
1383 pr_debug("dst_max set to: %s\n", pkt_dev->dst_max);
1384 i += len;
1385 sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max);
1386 return count;
1388 if (!strcmp(name, "dst6")) {
1389 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1390 if (len < 0)
1391 return len;
1393 pkt_dev->flags |= F_IPV6;
1395 if (copy_from_user(buf, &user_buffer[i], len))
1396 return -EFAULT;
1397 buf[len] = 0;
1399 in6_pton(buf, -1, pkt_dev->in6_daddr.s6_addr, -1, NULL);
1400 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_daddr);
1402 pkt_dev->cur_in6_daddr = pkt_dev->in6_daddr;
1404 if (debug)
1405 pr_debug("dst6 set to: %s\n", buf);
1407 i += len;
1408 sprintf(pg_result, "OK: dst6=%s", buf);
1409 return count;
1411 if (!strcmp(name, "dst6_min")) {
1412 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1413 if (len < 0)
1414 return len;
1416 pkt_dev->flags |= F_IPV6;
1418 if (copy_from_user(buf, &user_buffer[i], len))
1419 return -EFAULT;
1420 buf[len] = 0;
1422 in6_pton(buf, -1, pkt_dev->min_in6_daddr.s6_addr, -1, NULL);
1423 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->min_in6_daddr);
1425 pkt_dev->cur_in6_daddr = pkt_dev->min_in6_daddr;
1426 if (debug)
1427 pr_debug("dst6_min set to: %s\n", buf);
1429 i += len;
1430 sprintf(pg_result, "OK: dst6_min=%s", buf);
1431 return count;
1433 if (!strcmp(name, "dst6_max")) {
1434 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1435 if (len < 0)
1436 return len;
1438 pkt_dev->flags |= F_IPV6;
1440 if (copy_from_user(buf, &user_buffer[i], len))
1441 return -EFAULT;
1442 buf[len] = 0;
1444 in6_pton(buf, -1, pkt_dev->max_in6_daddr.s6_addr, -1, NULL);
1445 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->max_in6_daddr);
1447 if (debug)
1448 pr_debug("dst6_max set to: %s\n", buf);
1450 i += len;
1451 sprintf(pg_result, "OK: dst6_max=%s", buf);
1452 return count;
1454 if (!strcmp(name, "src6")) {
1455 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1456 if (len < 0)
1457 return len;
1459 pkt_dev->flags |= F_IPV6;
1461 if (copy_from_user(buf, &user_buffer[i], len))
1462 return -EFAULT;
1463 buf[len] = 0;
1465 in6_pton(buf, -1, pkt_dev->in6_saddr.s6_addr, -1, NULL);
1466 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_saddr);
1468 pkt_dev->cur_in6_saddr = pkt_dev->in6_saddr;
1470 if (debug)
1471 pr_debug("src6 set to: %s\n", buf);
1473 i += len;
1474 sprintf(pg_result, "OK: src6=%s", buf);
1475 return count;
1477 if (!strcmp(name, "src_min")) {
1478 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_min) - 1);
1479 if (len < 0)
1480 return len;
1482 if (copy_from_user(buf, &user_buffer[i], len))
1483 return -EFAULT;
1484 buf[len] = 0;
1485 if (strcmp(buf, pkt_dev->src_min) != 0) {
1486 memset(pkt_dev->src_min, 0, sizeof(pkt_dev->src_min));
1487 strncpy(pkt_dev->src_min, buf, len);
1488 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
1489 pkt_dev->cur_saddr = pkt_dev->saddr_min;
1491 if (debug)
1492 pr_debug("src_min set to: %s\n", pkt_dev->src_min);
1493 i += len;
1494 sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min);
1495 return count;
1497 if (!strcmp(name, "src_max")) {
1498 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_max) - 1);
1499 if (len < 0)
1500 return len;
1502 if (copy_from_user(buf, &user_buffer[i], len))
1503 return -EFAULT;
1504 buf[len] = 0;
1505 if (strcmp(buf, pkt_dev->src_max) != 0) {
1506 memset(pkt_dev->src_max, 0, sizeof(pkt_dev->src_max));
1507 strncpy(pkt_dev->src_max, buf, len);
1508 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
1509 pkt_dev->cur_saddr = pkt_dev->saddr_max;
1511 if (debug)
1512 pr_debug("src_max set to: %s\n", pkt_dev->src_max);
1513 i += len;
1514 sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max);
1515 return count;
1517 if (!strcmp(name, "dst_mac")) {
1518 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1519 if (len < 0)
1520 return len;
1522 memset(valstr, 0, sizeof(valstr));
1523 if (copy_from_user(valstr, &user_buffer[i], len))
1524 return -EFAULT;
1526 if (!mac_pton(valstr, pkt_dev->dst_mac))
1527 return -EINVAL;
1528 /* Set up Dest MAC */
1529 ether_addr_copy(&pkt_dev->hh[0], pkt_dev->dst_mac);
1531 sprintf(pg_result, "OK: dstmac %pM", pkt_dev->dst_mac);
1532 return count;
1534 if (!strcmp(name, "src_mac")) {
1535 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1536 if (len < 0)
1537 return len;
1539 memset(valstr, 0, sizeof(valstr));
1540 if (copy_from_user(valstr, &user_buffer[i], len))
1541 return -EFAULT;
1543 if (!mac_pton(valstr, pkt_dev->src_mac))
1544 return -EINVAL;
1545 /* Set up Src MAC */
1546 ether_addr_copy(&pkt_dev->hh[6], pkt_dev->src_mac);
1548 sprintf(pg_result, "OK: srcmac %pM", pkt_dev->src_mac);
1549 return count;
1552 if (!strcmp(name, "clear_counters")) {
1553 pktgen_clear_counters(pkt_dev);
1554 sprintf(pg_result, "OK: Clearing counters.\n");
1555 return count;
1558 if (!strcmp(name, "flows")) {
1559 len = num_arg(&user_buffer[i], 10, &value);
1560 if (len < 0)
1561 return len;
1563 i += len;
1564 if (value > MAX_CFLOWS)
1565 value = MAX_CFLOWS;
1567 pkt_dev->cflows = value;
1568 sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows);
1569 return count;
1571 #ifdef CONFIG_XFRM
1572 if (!strcmp(name, "spi")) {
1573 len = num_arg(&user_buffer[i], 10, &value);
1574 if (len < 0)
1575 return len;
1577 i += len;
1578 pkt_dev->spi = value;
1579 sprintf(pg_result, "OK: spi=%u", pkt_dev->spi);
1580 return count;
1582 #endif
1583 if (!strcmp(name, "flowlen")) {
1584 len = num_arg(&user_buffer[i], 10, &value);
1585 if (len < 0)
1586 return len;
1588 i += len;
1589 pkt_dev->lflow = value;
1590 sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow);
1591 return count;
1594 if (!strcmp(name, "queue_map_min")) {
1595 len = num_arg(&user_buffer[i], 5, &value);
1596 if (len < 0)
1597 return len;
1599 i += len;
1600 pkt_dev->queue_map_min = value;
1601 sprintf(pg_result, "OK: queue_map_min=%u", pkt_dev->queue_map_min);
1602 return count;
1605 if (!strcmp(name, "queue_map_max")) {
1606 len = num_arg(&user_buffer[i], 5, &value);
1607 if (len < 0)
1608 return len;
1610 i += len;
1611 pkt_dev->queue_map_max = value;
1612 sprintf(pg_result, "OK: queue_map_max=%u", pkt_dev->queue_map_max);
1613 return count;
1616 if (!strcmp(name, "mpls")) {
1617 unsigned int n, cnt;
1619 len = get_labels(&user_buffer[i], pkt_dev);
1620 if (len < 0)
1621 return len;
1622 i += len;
1623 cnt = sprintf(pg_result, "OK: mpls=");
1624 for (n = 0; n < pkt_dev->nr_labels; n++)
1625 cnt += sprintf(pg_result + cnt,
1626 "%08x%s", ntohl(pkt_dev->labels[n]),
1627 n == pkt_dev->nr_labels-1 ? "" : ",");
1629 if (pkt_dev->nr_labels && pkt_dev->vlan_id != 0xffff) {
1630 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1631 pkt_dev->svlan_id = 0xffff;
1633 if (debug)
1634 pr_debug("VLAN/SVLAN auto turned off\n");
1636 return count;
1639 if (!strcmp(name, "vlan_id")) {
1640 len = num_arg(&user_buffer[i], 4, &value);
1641 if (len < 0)
1642 return len;
1644 i += len;
1645 if (value <= 4095) {
1646 pkt_dev->vlan_id = value; /* turn on VLAN */
1648 if (debug)
1649 pr_debug("VLAN turned on\n");
1651 if (debug && pkt_dev->nr_labels)
1652 pr_debug("MPLS auto turned off\n");
1654 pkt_dev->nr_labels = 0; /* turn off MPLS */
1655 sprintf(pg_result, "OK: vlan_id=%u", pkt_dev->vlan_id);
1656 } else {
1657 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1658 pkt_dev->svlan_id = 0xffff;
1660 if (debug)
1661 pr_debug("VLAN/SVLAN turned off\n");
1663 return count;
1666 if (!strcmp(name, "vlan_p")) {
1667 len = num_arg(&user_buffer[i], 1, &value);
1668 if (len < 0)
1669 return len;
1671 i += len;
1672 if ((value <= 7) && (pkt_dev->vlan_id != 0xffff)) {
1673 pkt_dev->vlan_p = value;
1674 sprintf(pg_result, "OK: vlan_p=%u", pkt_dev->vlan_p);
1675 } else {
1676 sprintf(pg_result, "ERROR: vlan_p must be 0-7");
1678 return count;
1681 if (!strcmp(name, "vlan_cfi")) {
1682 len = num_arg(&user_buffer[i], 1, &value);
1683 if (len < 0)
1684 return len;
1686 i += len;
1687 if ((value <= 1) && (pkt_dev->vlan_id != 0xffff)) {
1688 pkt_dev->vlan_cfi = value;
1689 sprintf(pg_result, "OK: vlan_cfi=%u", pkt_dev->vlan_cfi);
1690 } else {
1691 sprintf(pg_result, "ERROR: vlan_cfi must be 0-1");
1693 return count;
1696 if (!strcmp(name, "svlan_id")) {
1697 len = num_arg(&user_buffer[i], 4, &value);
1698 if (len < 0)
1699 return len;
1701 i += len;
1702 if ((value <= 4095) && ((pkt_dev->vlan_id != 0xffff))) {
1703 pkt_dev->svlan_id = value; /* turn on SVLAN */
1705 if (debug)
1706 pr_debug("SVLAN turned on\n");
1708 if (debug && pkt_dev->nr_labels)
1709 pr_debug("MPLS auto turned off\n");
1711 pkt_dev->nr_labels = 0; /* turn off MPLS */
1712 sprintf(pg_result, "OK: svlan_id=%u", pkt_dev->svlan_id);
1713 } else {
1714 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1715 pkt_dev->svlan_id = 0xffff;
1717 if (debug)
1718 pr_debug("VLAN/SVLAN turned off\n");
1720 return count;
1723 if (!strcmp(name, "svlan_p")) {
1724 len = num_arg(&user_buffer[i], 1, &value);
1725 if (len < 0)
1726 return len;
1728 i += len;
1729 if ((value <= 7) && (pkt_dev->svlan_id != 0xffff)) {
1730 pkt_dev->svlan_p = value;
1731 sprintf(pg_result, "OK: svlan_p=%u", pkt_dev->svlan_p);
1732 } else {
1733 sprintf(pg_result, "ERROR: svlan_p must be 0-7");
1735 return count;
1738 if (!strcmp(name, "svlan_cfi")) {
1739 len = num_arg(&user_buffer[i], 1, &value);
1740 if (len < 0)
1741 return len;
1743 i += len;
1744 if ((value <= 1) && (pkt_dev->svlan_id != 0xffff)) {
1745 pkt_dev->svlan_cfi = value;
1746 sprintf(pg_result, "OK: svlan_cfi=%u", pkt_dev->svlan_cfi);
1747 } else {
1748 sprintf(pg_result, "ERROR: svlan_cfi must be 0-1");
1750 return count;
1753 if (!strcmp(name, "tos")) {
1754 __u32 tmp_value = 0;
1755 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1756 if (len < 0)
1757 return len;
1759 i += len;
1760 if (len == 2) {
1761 pkt_dev->tos = tmp_value;
1762 sprintf(pg_result, "OK: tos=0x%02x", pkt_dev->tos);
1763 } else {
1764 sprintf(pg_result, "ERROR: tos must be 00-ff");
1766 return count;
1769 if (!strcmp(name, "traffic_class")) {
1770 __u32 tmp_value = 0;
1771 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1772 if (len < 0)
1773 return len;
1775 i += len;
1776 if (len == 2) {
1777 pkt_dev->traffic_class = tmp_value;
1778 sprintf(pg_result, "OK: traffic_class=0x%02x", pkt_dev->traffic_class);
1779 } else {
1780 sprintf(pg_result, "ERROR: traffic_class must be 00-ff");
1782 return count;
1785 if (!strcmp(name, "skb_priority")) {
1786 len = num_arg(&user_buffer[i], 9, &value);
1787 if (len < 0)
1788 return len;
1790 i += len;
1791 pkt_dev->skb_priority = value;
1792 sprintf(pg_result, "OK: skb_priority=%i",
1793 pkt_dev->skb_priority);
1794 return count;
1797 sprintf(pkt_dev->result, "No such parameter \"%s\"", name);
1798 return -EINVAL;
1801 static int pktgen_if_open(struct inode *inode, struct file *file)
1803 return single_open(file, pktgen_if_show, PDE_DATA(inode));
1806 static const struct file_operations pktgen_if_fops = {
1807 .owner = THIS_MODULE,
1808 .open = pktgen_if_open,
1809 .read = seq_read,
1810 .llseek = seq_lseek,
1811 .write = pktgen_if_write,
1812 .release = single_release,
1815 static int pktgen_thread_show(struct seq_file *seq, void *v)
1817 struct pktgen_thread *t = seq->private;
1818 const struct pktgen_dev *pkt_dev;
1820 BUG_ON(!t);
1822 seq_puts(seq, "Running: ");
1824 rcu_read_lock();
1825 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1826 if (pkt_dev->running)
1827 seq_printf(seq, "%s ", pkt_dev->odevname);
1829 seq_puts(seq, "\nStopped: ");
1831 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1832 if (!pkt_dev->running)
1833 seq_printf(seq, "%s ", pkt_dev->odevname);
1835 if (t->result[0])
1836 seq_printf(seq, "\nResult: %s\n", t->result);
1837 else
1838 seq_puts(seq, "\nResult: NA\n");
1840 rcu_read_unlock();
1842 return 0;
1845 static ssize_t pktgen_thread_write(struct file *file,
1846 const char __user * user_buffer,
1847 size_t count, loff_t * offset)
1849 struct seq_file *seq = file->private_data;
1850 struct pktgen_thread *t = seq->private;
1851 int i, max, len, ret;
1852 char name[40];
1853 char *pg_result;
1855 if (count < 1) {
1856 // sprintf(pg_result, "Wrong command format");
1857 return -EINVAL;
1860 max = count;
1861 len = count_trail_chars(user_buffer, max);
1862 if (len < 0)
1863 return len;
1865 i = len;
1867 /* Read variable name */
1869 len = strn_len(&user_buffer[i], sizeof(name) - 1);
1870 if (len < 0)
1871 return len;
1873 memset(name, 0, sizeof(name));
1874 if (copy_from_user(name, &user_buffer[i], len))
1875 return -EFAULT;
1876 i += len;
1878 max = count - i;
1879 len = count_trail_chars(&user_buffer[i], max);
1880 if (len < 0)
1881 return len;
1883 i += len;
1885 if (debug)
1886 pr_debug("t=%s, count=%lu\n", name, (unsigned long)count);
1888 if (!t) {
1889 pr_err("ERROR: No thread\n");
1890 ret = -EINVAL;
1891 goto out;
1894 pg_result = &(t->result[0]);
1896 if (!strcmp(name, "add_device")) {
1897 char f[32];
1898 memset(f, 0, 32);
1899 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1900 if (len < 0) {
1901 ret = len;
1902 goto out;
1904 if (copy_from_user(f, &user_buffer[i], len))
1905 return -EFAULT;
1906 i += len;
1907 mutex_lock(&pktgen_thread_lock);
1908 ret = pktgen_add_device(t, f);
1909 mutex_unlock(&pktgen_thread_lock);
1910 if (!ret) {
1911 ret = count;
1912 sprintf(pg_result, "OK: add_device=%s", f);
1913 } else
1914 sprintf(pg_result, "ERROR: can not add device %s", f);
1915 goto out;
1918 if (!strcmp(name, "rem_device_all")) {
1919 mutex_lock(&pktgen_thread_lock);
1920 t->control |= T_REMDEVALL;
1921 mutex_unlock(&pktgen_thread_lock);
1922 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */
1923 ret = count;
1924 sprintf(pg_result, "OK: rem_device_all");
1925 goto out;
1928 if (!strcmp(name, "max_before_softirq")) {
1929 sprintf(pg_result, "OK: Note! max_before_softirq is obsoleted -- Do not use");
1930 ret = count;
1931 goto out;
1934 ret = -EINVAL;
1935 out:
1936 return ret;
1939 static int pktgen_thread_open(struct inode *inode, struct file *file)
1941 return single_open(file, pktgen_thread_show, PDE_DATA(inode));
1944 static const struct file_operations pktgen_thread_fops = {
1945 .owner = THIS_MODULE,
1946 .open = pktgen_thread_open,
1947 .read = seq_read,
1948 .llseek = seq_lseek,
1949 .write = pktgen_thread_write,
1950 .release = single_release,
1953 /* Think find or remove for NN */
1954 static struct pktgen_dev *__pktgen_NN_threads(const struct pktgen_net *pn,
1955 const char *ifname, int remove)
1957 struct pktgen_thread *t;
1958 struct pktgen_dev *pkt_dev = NULL;
1959 bool exact = (remove == FIND);
1961 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
1962 pkt_dev = pktgen_find_dev(t, ifname, exact);
1963 if (pkt_dev) {
1964 if (remove) {
1965 pkt_dev->removal_mark = 1;
1966 t->control |= T_REMDEV;
1968 break;
1971 return pkt_dev;
1975 * mark a device for removal
1977 static void pktgen_mark_device(const struct pktgen_net *pn, const char *ifname)
1979 struct pktgen_dev *pkt_dev = NULL;
1980 const int max_tries = 10, msec_per_try = 125;
1981 int i = 0;
1983 mutex_lock(&pktgen_thread_lock);
1984 pr_debug("%s: marking %s for removal\n", __func__, ifname);
1986 while (1) {
1988 pkt_dev = __pktgen_NN_threads(pn, ifname, REMOVE);
1989 if (pkt_dev == NULL)
1990 break; /* success */
1992 mutex_unlock(&pktgen_thread_lock);
1993 pr_debug("%s: waiting for %s to disappear....\n",
1994 __func__, ifname);
1995 schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try));
1996 mutex_lock(&pktgen_thread_lock);
1998 if (++i >= max_tries) {
1999 pr_err("%s: timed out after waiting %d msec for device %s to be removed\n",
2000 __func__, msec_per_try * i, ifname);
2001 break;
2006 mutex_unlock(&pktgen_thread_lock);
2009 static void pktgen_change_name(const struct pktgen_net *pn, struct net_device *dev)
2011 struct pktgen_thread *t;
2013 mutex_lock(&pktgen_thread_lock);
2015 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
2016 struct pktgen_dev *pkt_dev;
2018 if_lock(t);
2019 list_for_each_entry(pkt_dev, &t->if_list, list) {
2020 if (pkt_dev->odev != dev)
2021 continue;
2023 proc_remove(pkt_dev->entry);
2025 pkt_dev->entry = proc_create_data(dev->name, 0600,
2026 pn->proc_dir,
2027 &pktgen_if_fops,
2028 pkt_dev);
2029 if (!pkt_dev->entry)
2030 pr_err("can't move proc entry for '%s'\n",
2031 dev->name);
2032 break;
2034 if_unlock(t);
2036 mutex_unlock(&pktgen_thread_lock);
2039 static int pktgen_device_event(struct notifier_block *unused,
2040 unsigned long event, void *ptr)
2042 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2043 struct pktgen_net *pn = net_generic(dev_net(dev), pg_net_id);
2045 if (pn->pktgen_exiting)
2046 return NOTIFY_DONE;
2048 /* It is OK that we do not hold the group lock right now,
2049 * as we run under the RTNL lock.
2052 switch (event) {
2053 case NETDEV_CHANGENAME:
2054 pktgen_change_name(pn, dev);
2055 break;
2057 case NETDEV_UNREGISTER:
2058 pktgen_mark_device(pn, dev->name);
2059 break;
2062 return NOTIFY_DONE;
2065 static struct net_device *pktgen_dev_get_by_name(const struct pktgen_net *pn,
2066 struct pktgen_dev *pkt_dev,
2067 const char *ifname)
2069 char b[IFNAMSIZ+5];
2070 int i;
2072 for (i = 0; ifname[i] != '@'; i++) {
2073 if (i == IFNAMSIZ)
2074 break;
2076 b[i] = ifname[i];
2078 b[i] = 0;
2080 return dev_get_by_name(pn->net, b);
2084 /* Associate pktgen_dev with a device. */
2086 static int pktgen_setup_dev(const struct pktgen_net *pn,
2087 struct pktgen_dev *pkt_dev, const char *ifname)
2089 struct net_device *odev;
2090 int err;
2092 /* Clean old setups */
2093 if (pkt_dev->odev) {
2094 dev_put(pkt_dev->odev);
2095 pkt_dev->odev = NULL;
2098 odev = pktgen_dev_get_by_name(pn, pkt_dev, ifname);
2099 if (!odev) {
2100 pr_err("no such netdevice: \"%s\"\n", ifname);
2101 return -ENODEV;
2104 if (odev->type != ARPHRD_ETHER) {
2105 pr_err("not an ethernet device: \"%s\"\n", ifname);
2106 err = -EINVAL;
2107 } else if (!netif_running(odev)) {
2108 pr_err("device is down: \"%s\"\n", ifname);
2109 err = -ENETDOWN;
2110 } else {
2111 pkt_dev->odev = odev;
2112 return 0;
2115 dev_put(odev);
2116 return err;
2119 /* Read pkt_dev from the interface and set up internal pktgen_dev
2120 * structure to have the right information to create/send packets
2122 static void pktgen_setup_inject(struct pktgen_dev *pkt_dev)
2124 int ntxq;
2126 if (!pkt_dev->odev) {
2127 pr_err("ERROR: pkt_dev->odev == NULL in setup_inject\n");
2128 sprintf(pkt_dev->result,
2129 "ERROR: pkt_dev->odev == NULL in setup_inject.\n");
2130 return;
2133 /* make sure that we don't pick a non-existing transmit queue */
2134 ntxq = pkt_dev->odev->real_num_tx_queues;
2136 if (ntxq <= pkt_dev->queue_map_min) {
2137 pr_warn("WARNING: Requested queue_map_min (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2138 pkt_dev->queue_map_min, (ntxq ?: 1) - 1, ntxq,
2139 pkt_dev->odevname);
2140 pkt_dev->queue_map_min = (ntxq ?: 1) - 1;
2142 if (pkt_dev->queue_map_max >= ntxq) {
2143 pr_warn("WARNING: Requested queue_map_max (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2144 pkt_dev->queue_map_max, (ntxq ?: 1) - 1, ntxq,
2145 pkt_dev->odevname);
2146 pkt_dev->queue_map_max = (ntxq ?: 1) - 1;
2149 /* Default to the interface's mac if not explicitly set. */
2151 if (is_zero_ether_addr(pkt_dev->src_mac))
2152 ether_addr_copy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr);
2154 /* Set up Dest MAC */
2155 ether_addr_copy(&(pkt_dev->hh[0]), pkt_dev->dst_mac);
2157 if (pkt_dev->flags & F_IPV6) {
2158 int i, set = 0, err = 1;
2159 struct inet6_dev *idev;
2161 if (pkt_dev->min_pkt_size == 0) {
2162 pkt_dev->min_pkt_size = 14 + sizeof(struct ipv6hdr)
2163 + sizeof(struct udphdr)
2164 + sizeof(struct pktgen_hdr)
2165 + pkt_dev->pkt_overhead;
2168 for (i = 0; i < IN6_ADDR_HSIZE; i++)
2169 if (pkt_dev->cur_in6_saddr.s6_addr[i]) {
2170 set = 1;
2171 break;
2174 if (!set) {
2177 * Use linklevel address if unconfigured.
2179 * use ipv6_get_lladdr if/when it's get exported
2182 rcu_read_lock();
2183 idev = __in6_dev_get(pkt_dev->odev);
2184 if (idev) {
2185 struct inet6_ifaddr *ifp;
2187 read_lock_bh(&idev->lock);
2188 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2189 if ((ifp->scope & IFA_LINK) &&
2190 !(ifp->flags & IFA_F_TENTATIVE)) {
2191 pkt_dev->cur_in6_saddr = ifp->addr;
2192 err = 0;
2193 break;
2196 read_unlock_bh(&idev->lock);
2198 rcu_read_unlock();
2199 if (err)
2200 pr_err("ERROR: IPv6 link address not available\n");
2202 } else {
2203 if (pkt_dev->min_pkt_size == 0) {
2204 pkt_dev->min_pkt_size = 14 + sizeof(struct iphdr)
2205 + sizeof(struct udphdr)
2206 + sizeof(struct pktgen_hdr)
2207 + pkt_dev->pkt_overhead;
2210 pkt_dev->saddr_min = 0;
2211 pkt_dev->saddr_max = 0;
2212 if (strlen(pkt_dev->src_min) == 0) {
2214 struct in_device *in_dev;
2216 rcu_read_lock();
2217 in_dev = __in_dev_get_rcu(pkt_dev->odev);
2218 if (in_dev) {
2219 if (in_dev->ifa_list) {
2220 pkt_dev->saddr_min =
2221 in_dev->ifa_list->ifa_address;
2222 pkt_dev->saddr_max = pkt_dev->saddr_min;
2225 rcu_read_unlock();
2226 } else {
2227 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
2228 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
2231 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
2232 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
2234 /* Initialize current values. */
2235 pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size;
2236 if (pkt_dev->min_pkt_size > pkt_dev->max_pkt_size)
2237 pkt_dev->max_pkt_size = pkt_dev->min_pkt_size;
2239 pkt_dev->cur_dst_mac_offset = 0;
2240 pkt_dev->cur_src_mac_offset = 0;
2241 pkt_dev->cur_saddr = pkt_dev->saddr_min;
2242 pkt_dev->cur_daddr = pkt_dev->daddr_min;
2243 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2244 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2245 pkt_dev->nflows = 0;
2249 static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until)
2251 ktime_t start_time, end_time;
2252 s64 remaining;
2253 struct hrtimer_sleeper t;
2255 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
2256 hrtimer_set_expires(&t.timer, spin_until);
2258 remaining = ktime_to_ns(hrtimer_expires_remaining(&t.timer));
2259 if (remaining <= 0)
2260 goto out;
2262 start_time = ktime_get();
2263 if (remaining < 100000) {
2264 /* for small delays (<100us), just loop until limit is reached */
2265 do {
2266 end_time = ktime_get();
2267 } while (ktime_compare(end_time, spin_until) < 0);
2268 } else {
2269 /* see do_nanosleep */
2270 hrtimer_init_sleeper(&t, current);
2271 do {
2272 set_current_state(TASK_INTERRUPTIBLE);
2273 hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS);
2275 if (likely(t.task))
2276 schedule();
2278 hrtimer_cancel(&t.timer);
2279 } while (t.task && pkt_dev->running && !signal_pending(current));
2280 __set_current_state(TASK_RUNNING);
2281 end_time = ktime_get();
2284 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(end_time, start_time));
2285 out:
2286 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay);
2287 destroy_hrtimer_on_stack(&t.timer);
2290 static inline void set_pkt_overhead(struct pktgen_dev *pkt_dev)
2292 pkt_dev->pkt_overhead = 0;
2293 pkt_dev->pkt_overhead += pkt_dev->nr_labels*sizeof(u32);
2294 pkt_dev->pkt_overhead += VLAN_TAG_SIZE(pkt_dev);
2295 pkt_dev->pkt_overhead += SVLAN_TAG_SIZE(pkt_dev);
2298 static inline int f_seen(const struct pktgen_dev *pkt_dev, int flow)
2300 return !!(pkt_dev->flows[flow].flags & F_INIT);
2303 static inline int f_pick(struct pktgen_dev *pkt_dev)
2305 int flow = pkt_dev->curfl;
2307 if (pkt_dev->flags & F_FLOW_SEQ) {
2308 if (pkt_dev->flows[flow].count >= pkt_dev->lflow) {
2309 /* reset time */
2310 pkt_dev->flows[flow].count = 0;
2311 pkt_dev->flows[flow].flags = 0;
2312 pkt_dev->curfl += 1;
2313 if (pkt_dev->curfl >= pkt_dev->cflows)
2314 pkt_dev->curfl = 0; /*reset */
2316 } else {
2317 flow = prandom_u32() % pkt_dev->cflows;
2318 pkt_dev->curfl = flow;
2320 if (pkt_dev->flows[flow].count > pkt_dev->lflow) {
2321 pkt_dev->flows[flow].count = 0;
2322 pkt_dev->flows[flow].flags = 0;
2326 return pkt_dev->curfl;
2330 #ifdef CONFIG_XFRM
2331 /* If there was already an IPSEC SA, we keep it as is, else
2332 * we go look for it ...
2334 #define DUMMY_MARK 0
2335 static void get_ipsec_sa(struct pktgen_dev *pkt_dev, int flow)
2337 struct xfrm_state *x = pkt_dev->flows[flow].x;
2338 struct pktgen_net *pn = net_generic(dev_net(pkt_dev->odev), pg_net_id);
2339 if (!x) {
2341 if (pkt_dev->spi) {
2342 /* We need as quick as possible to find the right SA
2343 * Searching with minimum criteria to archieve this.
2345 x = xfrm_state_lookup_byspi(pn->net, htonl(pkt_dev->spi), AF_INET);
2346 } else {
2347 /* slow path: we dont already have xfrm_state */
2348 x = xfrm_stateonly_find(pn->net, DUMMY_MARK,
2349 (xfrm_address_t *)&pkt_dev->cur_daddr,
2350 (xfrm_address_t *)&pkt_dev->cur_saddr,
2351 AF_INET,
2352 pkt_dev->ipsmode,
2353 pkt_dev->ipsproto, 0);
2355 if (x) {
2356 pkt_dev->flows[flow].x = x;
2357 set_pkt_overhead(pkt_dev);
2358 pkt_dev->pkt_overhead += x->props.header_len;
2363 #endif
2364 static void set_cur_queue_map(struct pktgen_dev *pkt_dev)
2367 if (pkt_dev->flags & F_QUEUE_MAP_CPU)
2368 pkt_dev->cur_queue_map = smp_processor_id();
2370 else if (pkt_dev->queue_map_min <= pkt_dev->queue_map_max) {
2371 __u16 t;
2372 if (pkt_dev->flags & F_QUEUE_MAP_RND) {
2373 t = prandom_u32() %
2374 (pkt_dev->queue_map_max -
2375 pkt_dev->queue_map_min + 1)
2376 + pkt_dev->queue_map_min;
2377 } else {
2378 t = pkt_dev->cur_queue_map + 1;
2379 if (t > pkt_dev->queue_map_max)
2380 t = pkt_dev->queue_map_min;
2382 pkt_dev->cur_queue_map = t;
2384 pkt_dev->cur_queue_map = pkt_dev->cur_queue_map % pkt_dev->odev->real_num_tx_queues;
2387 /* Increment/randomize headers according to flags and current values
2388 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst
2390 static void mod_cur_headers(struct pktgen_dev *pkt_dev)
2392 __u32 imn;
2393 __u32 imx;
2394 int flow = 0;
2396 if (pkt_dev->cflows)
2397 flow = f_pick(pkt_dev);
2399 /* Deal with source MAC */
2400 if (pkt_dev->src_mac_count > 1) {
2401 __u32 mc;
2402 __u32 tmp;
2404 if (pkt_dev->flags & F_MACSRC_RND)
2405 mc = prandom_u32() % pkt_dev->src_mac_count;
2406 else {
2407 mc = pkt_dev->cur_src_mac_offset++;
2408 if (pkt_dev->cur_src_mac_offset >=
2409 pkt_dev->src_mac_count)
2410 pkt_dev->cur_src_mac_offset = 0;
2413 tmp = pkt_dev->src_mac[5] + (mc & 0xFF);
2414 pkt_dev->hh[11] = tmp;
2415 tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2416 pkt_dev->hh[10] = tmp;
2417 tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2418 pkt_dev->hh[9] = tmp;
2419 tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2420 pkt_dev->hh[8] = tmp;
2421 tmp = (pkt_dev->src_mac[1] + (tmp >> 8));
2422 pkt_dev->hh[7] = tmp;
2425 /* Deal with Destination MAC */
2426 if (pkt_dev->dst_mac_count > 1) {
2427 __u32 mc;
2428 __u32 tmp;
2430 if (pkt_dev->flags & F_MACDST_RND)
2431 mc = prandom_u32() % pkt_dev->dst_mac_count;
2433 else {
2434 mc = pkt_dev->cur_dst_mac_offset++;
2435 if (pkt_dev->cur_dst_mac_offset >=
2436 pkt_dev->dst_mac_count) {
2437 pkt_dev->cur_dst_mac_offset = 0;
2441 tmp = pkt_dev->dst_mac[5] + (mc & 0xFF);
2442 pkt_dev->hh[5] = tmp;
2443 tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2444 pkt_dev->hh[4] = tmp;
2445 tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2446 pkt_dev->hh[3] = tmp;
2447 tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2448 pkt_dev->hh[2] = tmp;
2449 tmp = (pkt_dev->dst_mac[1] + (tmp >> 8));
2450 pkt_dev->hh[1] = tmp;
2453 if (pkt_dev->flags & F_MPLS_RND) {
2454 unsigned int i;
2455 for (i = 0; i < pkt_dev->nr_labels; i++)
2456 if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM)
2457 pkt_dev->labels[i] = MPLS_STACK_BOTTOM |
2458 ((__force __be32)prandom_u32() &
2459 htonl(0x000fffff));
2462 if ((pkt_dev->flags & F_VID_RND) && (pkt_dev->vlan_id != 0xffff)) {
2463 pkt_dev->vlan_id = prandom_u32() & (4096 - 1);
2466 if ((pkt_dev->flags & F_SVID_RND) && (pkt_dev->svlan_id != 0xffff)) {
2467 pkt_dev->svlan_id = prandom_u32() & (4096 - 1);
2470 if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) {
2471 if (pkt_dev->flags & F_UDPSRC_RND)
2472 pkt_dev->cur_udp_src = prandom_u32() %
2473 (pkt_dev->udp_src_max - pkt_dev->udp_src_min)
2474 + pkt_dev->udp_src_min;
2476 else {
2477 pkt_dev->cur_udp_src++;
2478 if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max)
2479 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2483 if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) {
2484 if (pkt_dev->flags & F_UDPDST_RND) {
2485 pkt_dev->cur_udp_dst = prandom_u32() %
2486 (pkt_dev->udp_dst_max - pkt_dev->udp_dst_min)
2487 + pkt_dev->udp_dst_min;
2488 } else {
2489 pkt_dev->cur_udp_dst++;
2490 if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max)
2491 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2495 if (!(pkt_dev->flags & F_IPV6)) {
2497 imn = ntohl(pkt_dev->saddr_min);
2498 imx = ntohl(pkt_dev->saddr_max);
2499 if (imn < imx) {
2500 __u32 t;
2501 if (pkt_dev->flags & F_IPSRC_RND)
2502 t = prandom_u32() % (imx - imn) + imn;
2503 else {
2504 t = ntohl(pkt_dev->cur_saddr);
2505 t++;
2506 if (t > imx)
2507 t = imn;
2510 pkt_dev->cur_saddr = htonl(t);
2513 if (pkt_dev->cflows && f_seen(pkt_dev, flow)) {
2514 pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr;
2515 } else {
2516 imn = ntohl(pkt_dev->daddr_min);
2517 imx = ntohl(pkt_dev->daddr_max);
2518 if (imn < imx) {
2519 __u32 t;
2520 __be32 s;
2521 if (pkt_dev->flags & F_IPDST_RND) {
2523 do {
2524 t = prandom_u32() %
2525 (imx - imn) + imn;
2526 s = htonl(t);
2527 } while (ipv4_is_loopback(s) ||
2528 ipv4_is_multicast(s) ||
2529 ipv4_is_lbcast(s) ||
2530 ipv4_is_zeronet(s) ||
2531 ipv4_is_local_multicast(s));
2532 pkt_dev->cur_daddr = s;
2533 } else {
2534 t = ntohl(pkt_dev->cur_daddr);
2535 t++;
2536 if (t > imx) {
2537 t = imn;
2539 pkt_dev->cur_daddr = htonl(t);
2542 if (pkt_dev->cflows) {
2543 pkt_dev->flows[flow].flags |= F_INIT;
2544 pkt_dev->flows[flow].cur_daddr =
2545 pkt_dev->cur_daddr;
2546 #ifdef CONFIG_XFRM
2547 if (pkt_dev->flags & F_IPSEC_ON)
2548 get_ipsec_sa(pkt_dev, flow);
2549 #endif
2550 pkt_dev->nflows++;
2553 } else { /* IPV6 * */
2555 if (!ipv6_addr_any(&pkt_dev->min_in6_daddr)) {
2556 int i;
2558 /* Only random destinations yet */
2560 for (i = 0; i < 4; i++) {
2561 pkt_dev->cur_in6_daddr.s6_addr32[i] =
2562 (((__force __be32)prandom_u32() |
2563 pkt_dev->min_in6_daddr.s6_addr32[i]) &
2564 pkt_dev->max_in6_daddr.s6_addr32[i]);
2569 if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) {
2570 __u32 t;
2571 if (pkt_dev->flags & F_TXSIZE_RND) {
2572 t = prandom_u32() %
2573 (pkt_dev->max_pkt_size - pkt_dev->min_pkt_size)
2574 + pkt_dev->min_pkt_size;
2575 } else {
2576 t = pkt_dev->cur_pkt_size + 1;
2577 if (t > pkt_dev->max_pkt_size)
2578 t = pkt_dev->min_pkt_size;
2580 pkt_dev->cur_pkt_size = t;
2583 set_cur_queue_map(pkt_dev);
2585 pkt_dev->flows[flow].count++;
2589 #ifdef CONFIG_XFRM
2590 static u32 pktgen_dst_metrics[RTAX_MAX + 1] = {
2592 [RTAX_HOPLIMIT] = 0x5, /* Set a static hoplimit */
2595 static int pktgen_output_ipsec(struct sk_buff *skb, struct pktgen_dev *pkt_dev)
2597 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2598 int err = 0;
2599 struct net *net = dev_net(pkt_dev->odev);
2601 if (!x)
2602 return 0;
2603 /* XXX: we dont support tunnel mode for now until
2604 * we resolve the dst issue */
2605 if ((x->props.mode != XFRM_MODE_TRANSPORT) && (pkt_dev->spi == 0))
2606 return 0;
2608 /* But when user specify an valid SPI, transformation
2609 * supports both transport/tunnel mode + ESP/AH type.
2611 if ((x->props.mode == XFRM_MODE_TUNNEL) && (pkt_dev->spi != 0))
2612 skb->_skb_refdst = (unsigned long)&pkt_dev->dst | SKB_DST_NOREF;
2614 rcu_read_lock_bh();
2615 err = x->outer_mode->output(x, skb);
2616 rcu_read_unlock_bh();
2617 if (err) {
2618 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
2619 goto error;
2621 err = x->type->output(x, skb);
2622 if (err) {
2623 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
2624 goto error;
2626 spin_lock_bh(&x->lock);
2627 x->curlft.bytes += skb->len;
2628 x->curlft.packets++;
2629 spin_unlock_bh(&x->lock);
2630 error:
2631 return err;
2634 static void free_SAs(struct pktgen_dev *pkt_dev)
2636 if (pkt_dev->cflows) {
2637 /* let go of the SAs if we have them */
2638 int i;
2639 for (i = 0; i < pkt_dev->cflows; i++) {
2640 struct xfrm_state *x = pkt_dev->flows[i].x;
2641 if (x) {
2642 xfrm_state_put(x);
2643 pkt_dev->flows[i].x = NULL;
2649 static int process_ipsec(struct pktgen_dev *pkt_dev,
2650 struct sk_buff *skb, __be16 protocol)
2652 if (pkt_dev->flags & F_IPSEC_ON) {
2653 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2654 int nhead = 0;
2655 if (x) {
2656 struct ethhdr *eth;
2657 struct iphdr *iph;
2658 int ret;
2660 nhead = x->props.header_len - skb_headroom(skb);
2661 if (nhead > 0) {
2662 ret = pskb_expand_head(skb, nhead, 0, GFP_ATOMIC);
2663 if (ret < 0) {
2664 pr_err("Error expanding ipsec packet %d\n",
2665 ret);
2666 goto err;
2670 /* ipsec is not expecting ll header */
2671 skb_pull(skb, ETH_HLEN);
2672 ret = pktgen_output_ipsec(skb, pkt_dev);
2673 if (ret) {
2674 pr_err("Error creating ipsec packet %d\n", ret);
2675 goto err;
2677 /* restore ll */
2678 eth = (struct ethhdr *)skb_push(skb, ETH_HLEN);
2679 memcpy(eth, pkt_dev->hh, 2 * ETH_ALEN);
2680 eth->h_proto = protocol;
2682 /* Update IPv4 header len as well as checksum value */
2683 iph = ip_hdr(skb);
2684 iph->tot_len = htons(skb->len - ETH_HLEN);
2685 ip_send_check(iph);
2688 return 1;
2689 err:
2690 kfree_skb(skb);
2691 return 0;
2693 #endif
2695 static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev)
2697 unsigned int i;
2698 for (i = 0; i < pkt_dev->nr_labels; i++)
2699 *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM;
2701 mpls--;
2702 *mpls |= MPLS_STACK_BOTTOM;
2705 static inline __be16 build_tci(unsigned int id, unsigned int cfi,
2706 unsigned int prio)
2708 return htons(id | (cfi << 12) | (prio << 13));
2711 static void pktgen_finalize_skb(struct pktgen_dev *pkt_dev, struct sk_buff *skb,
2712 int datalen)
2714 struct timeval timestamp;
2715 struct pktgen_hdr *pgh;
2717 pgh = (struct pktgen_hdr *)skb_put(skb, sizeof(*pgh));
2718 datalen -= sizeof(*pgh);
2720 if (pkt_dev->nfrags <= 0) {
2721 memset(skb_put(skb, datalen), 0, datalen);
2722 } else {
2723 int frags = pkt_dev->nfrags;
2724 int i, len;
2725 int frag_len;
2728 if (frags > MAX_SKB_FRAGS)
2729 frags = MAX_SKB_FRAGS;
2730 len = datalen - frags * PAGE_SIZE;
2731 if (len > 0) {
2732 memset(skb_put(skb, len), 0, len);
2733 datalen = frags * PAGE_SIZE;
2736 i = 0;
2737 frag_len = (datalen/frags) < PAGE_SIZE ?
2738 (datalen/frags) : PAGE_SIZE;
2739 while (datalen > 0) {
2740 if (unlikely(!pkt_dev->page)) {
2741 int node = numa_node_id();
2743 if (pkt_dev->node >= 0 && (pkt_dev->flags & F_NODE))
2744 node = pkt_dev->node;
2745 pkt_dev->page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0);
2746 if (!pkt_dev->page)
2747 break;
2749 get_page(pkt_dev->page);
2750 skb_frag_set_page(skb, i, pkt_dev->page);
2751 skb_shinfo(skb)->frags[i].page_offset = 0;
2752 /*last fragment, fill rest of data*/
2753 if (i == (frags - 1))
2754 skb_frag_size_set(&skb_shinfo(skb)->frags[i],
2755 (datalen < PAGE_SIZE ? datalen : PAGE_SIZE));
2756 else
2757 skb_frag_size_set(&skb_shinfo(skb)->frags[i], frag_len);
2758 datalen -= skb_frag_size(&skb_shinfo(skb)->frags[i]);
2759 skb->len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2760 skb->data_len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2761 i++;
2762 skb_shinfo(skb)->nr_frags = i;
2766 /* Stamp the time, and sequence number,
2767 * convert them to network byte order
2769 pgh->pgh_magic = htonl(PKTGEN_MAGIC);
2770 pgh->seq_num = htonl(pkt_dev->seq_num);
2772 if (pkt_dev->flags & F_NO_TIMESTAMP) {
2773 pgh->tv_sec = 0;
2774 pgh->tv_usec = 0;
2775 } else {
2776 do_gettimeofday(&timestamp);
2777 pgh->tv_sec = htonl(timestamp.tv_sec);
2778 pgh->tv_usec = htonl(timestamp.tv_usec);
2782 static struct sk_buff *pktgen_alloc_skb(struct net_device *dev,
2783 struct pktgen_dev *pkt_dev)
2785 unsigned int extralen = LL_RESERVED_SPACE(dev);
2786 struct sk_buff *skb = NULL;
2787 unsigned int size;
2789 size = pkt_dev->cur_pkt_size + 64 + extralen + pkt_dev->pkt_overhead;
2790 if (pkt_dev->flags & F_NODE) {
2791 int node = pkt_dev->node >= 0 ? pkt_dev->node : numa_node_id();
2793 skb = __alloc_skb(NET_SKB_PAD + size, GFP_NOWAIT, 0, node);
2794 if (likely(skb)) {
2795 skb_reserve(skb, NET_SKB_PAD);
2796 skb->dev = dev;
2798 } else {
2799 skb = __netdev_alloc_skb(dev, size, GFP_NOWAIT);
2802 /* the caller pre-fetches from skb->data and reserves for the mac hdr */
2803 if (likely(skb))
2804 skb_reserve(skb, extralen - 16);
2806 return skb;
2809 static struct sk_buff *fill_packet_ipv4(struct net_device *odev,
2810 struct pktgen_dev *pkt_dev)
2812 struct sk_buff *skb = NULL;
2813 __u8 *eth;
2814 struct udphdr *udph;
2815 int datalen, iplen;
2816 struct iphdr *iph;
2817 __be16 protocol = htons(ETH_P_IP);
2818 __be32 *mpls;
2819 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2820 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2821 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2822 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2823 u16 queue_map;
2825 if (pkt_dev->nr_labels)
2826 protocol = htons(ETH_P_MPLS_UC);
2828 if (pkt_dev->vlan_id != 0xffff)
2829 protocol = htons(ETH_P_8021Q);
2831 /* Update any of the values, used when we're incrementing various
2832 * fields.
2834 mod_cur_headers(pkt_dev);
2835 queue_map = pkt_dev->cur_queue_map;
2837 skb = pktgen_alloc_skb(odev, pkt_dev);
2838 if (!skb) {
2839 sprintf(pkt_dev->result, "No memory");
2840 return NULL;
2843 prefetchw(skb->data);
2844 skb_reserve(skb, 16);
2846 /* Reserve for ethernet and IP header */
2847 eth = (__u8 *) skb_push(skb, 14);
2848 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32));
2849 if (pkt_dev->nr_labels)
2850 mpls_push(mpls, pkt_dev);
2852 if (pkt_dev->vlan_id != 0xffff) {
2853 if (pkt_dev->svlan_id != 0xffff) {
2854 svlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2855 *svlan_tci = build_tci(pkt_dev->svlan_id,
2856 pkt_dev->svlan_cfi,
2857 pkt_dev->svlan_p);
2858 svlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2859 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2861 vlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2862 *vlan_tci = build_tci(pkt_dev->vlan_id,
2863 pkt_dev->vlan_cfi,
2864 pkt_dev->vlan_p);
2865 vlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2866 *vlan_encapsulated_proto = htons(ETH_P_IP);
2869 skb_reset_mac_header(skb);
2870 skb_set_network_header(skb, skb->len);
2871 iph = (struct iphdr *) skb_put(skb, sizeof(struct iphdr));
2873 skb_set_transport_header(skb, skb->len);
2874 udph = (struct udphdr *) skb_put(skb, sizeof(struct udphdr));
2875 skb_set_queue_mapping(skb, queue_map);
2876 skb->priority = pkt_dev->skb_priority;
2878 memcpy(eth, pkt_dev->hh, 12);
2879 *(__be16 *) & eth[12] = protocol;
2881 /* Eth + IPh + UDPh + mpls */
2882 datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 -
2883 pkt_dev->pkt_overhead;
2884 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr))
2885 datalen = sizeof(struct pktgen_hdr);
2887 udph->source = htons(pkt_dev->cur_udp_src);
2888 udph->dest = htons(pkt_dev->cur_udp_dst);
2889 udph->len = htons(datalen + 8); /* DATA + udphdr */
2890 udph->check = 0;
2892 iph->ihl = 5;
2893 iph->version = 4;
2894 iph->ttl = 32;
2895 iph->tos = pkt_dev->tos;
2896 iph->protocol = IPPROTO_UDP; /* UDP */
2897 iph->saddr = pkt_dev->cur_saddr;
2898 iph->daddr = pkt_dev->cur_daddr;
2899 iph->id = htons(pkt_dev->ip_id);
2900 pkt_dev->ip_id++;
2901 iph->frag_off = 0;
2902 iplen = 20 + 8 + datalen;
2903 iph->tot_len = htons(iplen);
2904 ip_send_check(iph);
2905 skb->protocol = protocol;
2906 skb->dev = odev;
2907 skb->pkt_type = PACKET_HOST;
2909 pktgen_finalize_skb(pkt_dev, skb, datalen);
2911 if (!(pkt_dev->flags & F_UDPCSUM)) {
2912 skb->ip_summed = CHECKSUM_NONE;
2913 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM)) {
2914 skb->ip_summed = CHECKSUM_PARTIAL;
2915 skb->csum = 0;
2916 udp4_hwcsum(skb, iph->saddr, iph->daddr);
2917 } else {
2918 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), datalen + 8, 0);
2920 /* add protocol-dependent pseudo-header */
2921 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
2922 datalen + 8, IPPROTO_UDP, csum);
2924 if (udph->check == 0)
2925 udph->check = CSUM_MANGLED_0;
2928 #ifdef CONFIG_XFRM
2929 if (!process_ipsec(pkt_dev, skb, protocol))
2930 return NULL;
2931 #endif
2933 return skb;
2936 static struct sk_buff *fill_packet_ipv6(struct net_device *odev,
2937 struct pktgen_dev *pkt_dev)
2939 struct sk_buff *skb = NULL;
2940 __u8 *eth;
2941 struct udphdr *udph;
2942 int datalen, udplen;
2943 struct ipv6hdr *iph;
2944 __be16 protocol = htons(ETH_P_IPV6);
2945 __be32 *mpls;
2946 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2947 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2948 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2949 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2950 u16 queue_map;
2952 if (pkt_dev->nr_labels)
2953 protocol = htons(ETH_P_MPLS_UC);
2955 if (pkt_dev->vlan_id != 0xffff)
2956 protocol = htons(ETH_P_8021Q);
2958 /* Update any of the values, used when we're incrementing various
2959 * fields.
2961 mod_cur_headers(pkt_dev);
2962 queue_map = pkt_dev->cur_queue_map;
2964 skb = pktgen_alloc_skb(odev, pkt_dev);
2965 if (!skb) {
2966 sprintf(pkt_dev->result, "No memory");
2967 return NULL;
2970 prefetchw(skb->data);
2971 skb_reserve(skb, 16);
2973 /* Reserve for ethernet and IP header */
2974 eth = (__u8 *) skb_push(skb, 14);
2975 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32));
2976 if (pkt_dev->nr_labels)
2977 mpls_push(mpls, pkt_dev);
2979 if (pkt_dev->vlan_id != 0xffff) {
2980 if (pkt_dev->svlan_id != 0xffff) {
2981 svlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2982 *svlan_tci = build_tci(pkt_dev->svlan_id,
2983 pkt_dev->svlan_cfi,
2984 pkt_dev->svlan_p);
2985 svlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2986 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2988 vlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2989 *vlan_tci = build_tci(pkt_dev->vlan_id,
2990 pkt_dev->vlan_cfi,
2991 pkt_dev->vlan_p);
2992 vlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2993 *vlan_encapsulated_proto = htons(ETH_P_IPV6);
2996 skb_reset_mac_header(skb);
2997 skb_set_network_header(skb, skb->len);
2998 iph = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
3000 skb_set_transport_header(skb, skb->len);
3001 udph = (struct udphdr *) skb_put(skb, sizeof(struct udphdr));
3002 skb_set_queue_mapping(skb, queue_map);
3003 skb->priority = pkt_dev->skb_priority;
3005 memcpy(eth, pkt_dev->hh, 12);
3006 *(__be16 *) &eth[12] = protocol;
3008 /* Eth + IPh + UDPh + mpls */
3009 datalen = pkt_dev->cur_pkt_size - 14 -
3010 sizeof(struct ipv6hdr) - sizeof(struct udphdr) -
3011 pkt_dev->pkt_overhead;
3013 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) {
3014 datalen = sizeof(struct pktgen_hdr);
3015 net_info_ratelimited("increased datalen to %d\n", datalen);
3018 udplen = datalen + sizeof(struct udphdr);
3019 udph->source = htons(pkt_dev->cur_udp_src);
3020 udph->dest = htons(pkt_dev->cur_udp_dst);
3021 udph->len = htons(udplen);
3022 udph->check = 0;
3024 *(__be32 *) iph = htonl(0x60000000); /* Version + flow */
3026 if (pkt_dev->traffic_class) {
3027 /* Version + traffic class + flow (0) */
3028 *(__be32 *)iph |= htonl(0x60000000 | (pkt_dev->traffic_class << 20));
3031 iph->hop_limit = 32;
3033 iph->payload_len = htons(udplen);
3034 iph->nexthdr = IPPROTO_UDP;
3036 iph->daddr = pkt_dev->cur_in6_daddr;
3037 iph->saddr = pkt_dev->cur_in6_saddr;
3039 skb->protocol = protocol;
3040 skb->dev = odev;
3041 skb->pkt_type = PACKET_HOST;
3043 pktgen_finalize_skb(pkt_dev, skb, datalen);
3045 if (!(pkt_dev->flags & F_UDPCSUM)) {
3046 skb->ip_summed = CHECKSUM_NONE;
3047 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM)) {
3048 skb->ip_summed = CHECKSUM_PARTIAL;
3049 skb->csum_start = skb_transport_header(skb) - skb->head;
3050 skb->csum_offset = offsetof(struct udphdr, check);
3051 udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0);
3052 } else {
3053 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), udplen, 0);
3055 /* add protocol-dependent pseudo-header */
3056 udph->check = csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, csum);
3058 if (udph->check == 0)
3059 udph->check = CSUM_MANGLED_0;
3062 return skb;
3065 static struct sk_buff *fill_packet(struct net_device *odev,
3066 struct pktgen_dev *pkt_dev)
3068 if (pkt_dev->flags & F_IPV6)
3069 return fill_packet_ipv6(odev, pkt_dev);
3070 else
3071 return fill_packet_ipv4(odev, pkt_dev);
3074 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev)
3076 pkt_dev->seq_num = 1;
3077 pkt_dev->idle_acc = 0;
3078 pkt_dev->sofar = 0;
3079 pkt_dev->tx_bytes = 0;
3080 pkt_dev->errors = 0;
3083 /* Set up structure for sending pkts, clear counters */
3085 static void pktgen_run(struct pktgen_thread *t)
3087 struct pktgen_dev *pkt_dev;
3088 int started = 0;
3090 func_enter();
3092 rcu_read_lock();
3093 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3096 * setup odev and create initial packet.
3098 pktgen_setup_inject(pkt_dev);
3100 if (pkt_dev->odev) {
3101 pktgen_clear_counters(pkt_dev);
3102 pkt_dev->skb = NULL;
3103 pkt_dev->started_at = pkt_dev->next_tx = ktime_get();
3105 set_pkt_overhead(pkt_dev);
3107 strcpy(pkt_dev->result, "Starting");
3108 pkt_dev->running = 1; /* Cranke yeself! */
3109 started++;
3110 } else
3111 strcpy(pkt_dev->result, "Error starting");
3113 rcu_read_unlock();
3114 if (started)
3115 t->control &= ~(T_STOP);
3118 static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn)
3120 struct pktgen_thread *t;
3122 func_enter();
3124 mutex_lock(&pktgen_thread_lock);
3126 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3127 t->control |= T_STOP;
3129 mutex_unlock(&pktgen_thread_lock);
3132 static int thread_is_running(const struct pktgen_thread *t)
3134 const struct pktgen_dev *pkt_dev;
3136 rcu_read_lock();
3137 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
3138 if (pkt_dev->running) {
3139 rcu_read_unlock();
3140 return 1;
3142 rcu_read_unlock();
3143 return 0;
3146 static int pktgen_wait_thread_run(struct pktgen_thread *t)
3148 while (thread_is_running(t)) {
3150 msleep_interruptible(100);
3152 if (signal_pending(current))
3153 goto signal;
3155 return 1;
3156 signal:
3157 return 0;
3160 static int pktgen_wait_all_threads_run(struct pktgen_net *pn)
3162 struct pktgen_thread *t;
3163 int sig = 1;
3165 mutex_lock(&pktgen_thread_lock);
3167 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
3168 sig = pktgen_wait_thread_run(t);
3169 if (sig == 0)
3170 break;
3173 if (sig == 0)
3174 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3175 t->control |= (T_STOP);
3177 mutex_unlock(&pktgen_thread_lock);
3178 return sig;
3181 static void pktgen_run_all_threads(struct pktgen_net *pn)
3183 struct pktgen_thread *t;
3185 func_enter();
3187 mutex_lock(&pktgen_thread_lock);
3189 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3190 t->control |= (T_RUN);
3192 mutex_unlock(&pktgen_thread_lock);
3194 /* Propagate thread->control */
3195 schedule_timeout_interruptible(msecs_to_jiffies(125));
3197 pktgen_wait_all_threads_run(pn);
3200 static void pktgen_reset_all_threads(struct pktgen_net *pn)
3202 struct pktgen_thread *t;
3204 func_enter();
3206 mutex_lock(&pktgen_thread_lock);
3208 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3209 t->control |= (T_REMDEVALL);
3211 mutex_unlock(&pktgen_thread_lock);
3213 /* Propagate thread->control */
3214 schedule_timeout_interruptible(msecs_to_jiffies(125));
3216 pktgen_wait_all_threads_run(pn);
3219 static void show_results(struct pktgen_dev *pkt_dev, int nr_frags)
3221 __u64 bps, mbps, pps;
3222 char *p = pkt_dev->result;
3223 ktime_t elapsed = ktime_sub(pkt_dev->stopped_at,
3224 pkt_dev->started_at);
3225 ktime_t idle = ns_to_ktime(pkt_dev->idle_acc);
3227 p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n",
3228 (unsigned long long)ktime_to_us(elapsed),
3229 (unsigned long long)ktime_to_us(ktime_sub(elapsed, idle)),
3230 (unsigned long long)ktime_to_us(idle),
3231 (unsigned long long)pkt_dev->sofar,
3232 pkt_dev->cur_pkt_size, nr_frags);
3234 pps = div64_u64(pkt_dev->sofar * NSEC_PER_SEC,
3235 ktime_to_ns(elapsed));
3237 bps = pps * 8 * pkt_dev->cur_pkt_size;
3239 mbps = bps;
3240 do_div(mbps, 1000000);
3241 p += sprintf(p, " %llupps %lluMb/sec (%llubps) errors: %llu",
3242 (unsigned long long)pps,
3243 (unsigned long long)mbps,
3244 (unsigned long long)bps,
3245 (unsigned long long)pkt_dev->errors);
3248 /* Set stopped-at timer, remove from running list, do counters & statistics */
3249 static int pktgen_stop_device(struct pktgen_dev *pkt_dev)
3251 int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1;
3253 if (!pkt_dev->running) {
3254 pr_warn("interface: %s is already stopped\n",
3255 pkt_dev->odevname);
3256 return -EINVAL;
3259 pkt_dev->running = 0;
3260 kfree_skb(pkt_dev->skb);
3261 pkt_dev->skb = NULL;
3262 pkt_dev->stopped_at = ktime_get();
3264 show_results(pkt_dev, nr_frags);
3266 return 0;
3269 static struct pktgen_dev *next_to_run(struct pktgen_thread *t)
3271 struct pktgen_dev *pkt_dev, *best = NULL;
3273 rcu_read_lock();
3274 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3275 if (!pkt_dev->running)
3276 continue;
3277 if (best == NULL)
3278 best = pkt_dev;
3279 else if (ktime_compare(pkt_dev->next_tx, best->next_tx) < 0)
3280 best = pkt_dev;
3282 rcu_read_unlock();
3284 return best;
3287 static void pktgen_stop(struct pktgen_thread *t)
3289 struct pktgen_dev *pkt_dev;
3291 func_enter();
3293 rcu_read_lock();
3295 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3296 pktgen_stop_device(pkt_dev);
3299 rcu_read_unlock();
3303 * one of our devices needs to be removed - find it
3304 * and remove it
3306 static void pktgen_rem_one_if(struct pktgen_thread *t)
3308 struct list_head *q, *n;
3309 struct pktgen_dev *cur;
3311 func_enter();
3313 list_for_each_safe(q, n, &t->if_list) {
3314 cur = list_entry(q, struct pktgen_dev, list);
3316 if (!cur->removal_mark)
3317 continue;
3319 kfree_skb(cur->skb);
3320 cur->skb = NULL;
3322 pktgen_remove_device(t, cur);
3324 break;
3328 static void pktgen_rem_all_ifs(struct pktgen_thread *t)
3330 struct list_head *q, *n;
3331 struct pktgen_dev *cur;
3333 func_enter();
3335 /* Remove all devices, free mem */
3337 list_for_each_safe(q, n, &t->if_list) {
3338 cur = list_entry(q, struct pktgen_dev, list);
3340 kfree_skb(cur->skb);
3341 cur->skb = NULL;
3343 pktgen_remove_device(t, cur);
3347 static void pktgen_rem_thread(struct pktgen_thread *t)
3349 /* Remove from the thread list */
3350 remove_proc_entry(t->tsk->comm, t->net->proc_dir);
3353 static void pktgen_resched(struct pktgen_dev *pkt_dev)
3355 ktime_t idle_start = ktime_get();
3356 schedule();
3357 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3360 static void pktgen_wait_for_skb(struct pktgen_dev *pkt_dev)
3362 ktime_t idle_start = ktime_get();
3364 while (atomic_read(&(pkt_dev->skb->users)) != 1) {
3365 if (signal_pending(current))
3366 break;
3368 if (need_resched())
3369 pktgen_resched(pkt_dev);
3370 else
3371 cpu_relax();
3373 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3376 static void pktgen_xmit(struct pktgen_dev *pkt_dev)
3378 unsigned int burst = ACCESS_ONCE(pkt_dev->burst);
3379 struct net_device *odev = pkt_dev->odev;
3380 struct netdev_queue *txq;
3381 struct sk_buff *skb;
3382 int ret;
3384 /* If device is offline, then don't send */
3385 if (unlikely(!netif_running(odev) || !netif_carrier_ok(odev))) {
3386 pktgen_stop_device(pkt_dev);
3387 return;
3390 /* This is max DELAY, this has special meaning of
3391 * "never transmit"
3393 if (unlikely(pkt_dev->delay == ULLONG_MAX)) {
3394 pkt_dev->next_tx = ktime_add_ns(ktime_get(), ULONG_MAX);
3395 return;
3398 /* If no skb or clone count exhausted then get new one */
3399 if (!pkt_dev->skb || (pkt_dev->last_ok &&
3400 ++pkt_dev->clone_count >= pkt_dev->clone_skb)) {
3401 /* build a new pkt */
3402 kfree_skb(pkt_dev->skb);
3404 pkt_dev->skb = fill_packet(odev, pkt_dev);
3405 if (pkt_dev->skb == NULL) {
3406 pr_err("ERROR: couldn't allocate skb in fill_packet\n");
3407 schedule();
3408 pkt_dev->clone_count--; /* back out increment, OOM */
3409 return;
3411 pkt_dev->last_pkt_size = pkt_dev->skb->len;
3412 pkt_dev->clone_count = 0; /* reset counter */
3415 if (pkt_dev->delay && pkt_dev->last_ok)
3416 spin(pkt_dev, pkt_dev->next_tx);
3418 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) {
3419 skb = pkt_dev->skb;
3420 skb->protocol = eth_type_trans(skb, skb->dev);
3421 atomic_add(burst, &skb->users);
3422 local_bh_disable();
3423 do {
3424 ret = netif_receive_skb(skb);
3425 if (ret == NET_RX_DROP)
3426 pkt_dev->errors++;
3427 pkt_dev->sofar++;
3428 pkt_dev->seq_num++;
3429 if (atomic_read(&skb->users) != burst) {
3430 /* skb was queued by rps/rfs or taps,
3431 * so cannot reuse this skb
3433 atomic_sub(burst - 1, &skb->users);
3434 /* get out of the loop and wait
3435 * until skb is consumed
3437 break;
3439 /* skb was 'freed' by stack, so clean few
3440 * bits and reuse it
3442 skb_reset_tc(skb);
3443 } while (--burst > 0);
3444 goto out; /* Skips xmit_mode M_START_XMIT */
3445 } else if (pkt_dev->xmit_mode == M_QUEUE_XMIT) {
3446 local_bh_disable();
3447 atomic_inc(&pkt_dev->skb->users);
3449 ret = dev_queue_xmit(pkt_dev->skb);
3450 switch (ret) {
3451 case NET_XMIT_SUCCESS:
3452 pkt_dev->sofar++;
3453 pkt_dev->seq_num++;
3454 pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3455 break;
3456 case NET_XMIT_DROP:
3457 case NET_XMIT_CN:
3458 /* These are all valid return codes for a qdisc but
3459 * indicate packets are being dropped or will likely
3460 * be dropped soon.
3462 case NETDEV_TX_BUSY:
3463 /* qdisc may call dev_hard_start_xmit directly in cases
3464 * where no queues exist e.g. loopback device, virtual
3465 * devices, etc. In this case we need to handle
3466 * NETDEV_TX_ codes.
3468 default:
3469 pkt_dev->errors++;
3470 net_info_ratelimited("%s xmit error: %d\n",
3471 pkt_dev->odevname, ret);
3472 break;
3474 goto out;
3477 txq = skb_get_tx_queue(odev, pkt_dev->skb);
3479 local_bh_disable();
3481 HARD_TX_LOCK(odev, txq, smp_processor_id());
3483 if (unlikely(netif_xmit_frozen_or_drv_stopped(txq))) {
3484 ret = NETDEV_TX_BUSY;
3485 pkt_dev->last_ok = 0;
3486 goto unlock;
3488 atomic_add(burst, &pkt_dev->skb->users);
3490 xmit_more:
3491 ret = netdev_start_xmit(pkt_dev->skb, odev, txq, --burst > 0);
3493 switch (ret) {
3494 case NETDEV_TX_OK:
3495 pkt_dev->last_ok = 1;
3496 pkt_dev->sofar++;
3497 pkt_dev->seq_num++;
3498 pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3499 if (burst > 0 && !netif_xmit_frozen_or_drv_stopped(txq))
3500 goto xmit_more;
3501 break;
3502 case NET_XMIT_DROP:
3503 case NET_XMIT_CN:
3504 /* skb has been consumed */
3505 pkt_dev->errors++;
3506 break;
3507 default: /* Drivers are not supposed to return other values! */
3508 net_info_ratelimited("%s xmit error: %d\n",
3509 pkt_dev->odevname, ret);
3510 pkt_dev->errors++;
3511 /* fallthru */
3512 case NETDEV_TX_BUSY:
3513 /* Retry it next time */
3514 atomic_dec(&(pkt_dev->skb->users));
3515 pkt_dev->last_ok = 0;
3517 if (unlikely(burst))
3518 atomic_sub(burst, &pkt_dev->skb->users);
3519 unlock:
3520 HARD_TX_UNLOCK(odev, txq);
3522 out:
3523 local_bh_enable();
3525 /* If pkt_dev->count is zero, then run forever */
3526 if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) {
3527 pktgen_wait_for_skb(pkt_dev);
3529 /* Done with this */
3530 pktgen_stop_device(pkt_dev);
3535 * Main loop of the thread goes here
3538 static int pktgen_thread_worker(void *arg)
3540 DEFINE_WAIT(wait);
3541 struct pktgen_thread *t = arg;
3542 struct pktgen_dev *pkt_dev = NULL;
3543 int cpu = t->cpu;
3545 BUG_ON(smp_processor_id() != cpu);
3547 init_waitqueue_head(&t->queue);
3548 complete(&t->start_done);
3550 pr_debug("starting pktgen/%d: pid=%d\n", cpu, task_pid_nr(current));
3552 set_freezable();
3554 while (!kthread_should_stop()) {
3555 pkt_dev = next_to_run(t);
3557 if (unlikely(!pkt_dev && t->control == 0)) {
3558 if (t->net->pktgen_exiting)
3559 break;
3560 wait_event_interruptible_timeout(t->queue,
3561 t->control != 0,
3562 HZ/10);
3563 try_to_freeze();
3564 continue;
3567 if (likely(pkt_dev)) {
3568 pktgen_xmit(pkt_dev);
3570 if (need_resched())
3571 pktgen_resched(pkt_dev);
3572 else
3573 cpu_relax();
3576 if (t->control & T_STOP) {
3577 pktgen_stop(t);
3578 t->control &= ~(T_STOP);
3581 if (t->control & T_RUN) {
3582 pktgen_run(t);
3583 t->control &= ~(T_RUN);
3586 if (t->control & T_REMDEVALL) {
3587 pktgen_rem_all_ifs(t);
3588 t->control &= ~(T_REMDEVALL);
3591 if (t->control & T_REMDEV) {
3592 pktgen_rem_one_if(t);
3593 t->control &= ~(T_REMDEV);
3596 try_to_freeze();
3599 pr_debug("%s stopping all device\n", t->tsk->comm);
3600 pktgen_stop(t);
3602 pr_debug("%s removing all device\n", t->tsk->comm);
3603 pktgen_rem_all_ifs(t);
3605 pr_debug("%s removing thread\n", t->tsk->comm);
3606 pktgen_rem_thread(t);
3608 return 0;
3611 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
3612 const char *ifname, bool exact)
3614 struct pktgen_dev *p, *pkt_dev = NULL;
3615 size_t len = strlen(ifname);
3617 rcu_read_lock();
3618 list_for_each_entry_rcu(p, &t->if_list, list)
3619 if (strncmp(p->odevname, ifname, len) == 0) {
3620 if (p->odevname[len]) {
3621 if (exact || p->odevname[len] != '@')
3622 continue;
3624 pkt_dev = p;
3625 break;
3628 rcu_read_unlock();
3629 pr_debug("find_dev(%s) returning %p\n", ifname, pkt_dev);
3630 return pkt_dev;
3634 * Adds a dev at front of if_list.
3637 static int add_dev_to_thread(struct pktgen_thread *t,
3638 struct pktgen_dev *pkt_dev)
3640 int rv = 0;
3642 /* This function cannot be called concurrently, as its called
3643 * under pktgen_thread_lock mutex, but it can run from
3644 * userspace on another CPU than the kthread. The if_lock()
3645 * is used here to sync with concurrent instances of
3646 * _rem_dev_from_if_list() invoked via kthread, which is also
3647 * updating the if_list */
3648 if_lock(t);
3650 if (pkt_dev->pg_thread) {
3651 pr_err("ERROR: already assigned to a thread\n");
3652 rv = -EBUSY;
3653 goto out;
3656 pkt_dev->running = 0;
3657 pkt_dev->pg_thread = t;
3658 list_add_rcu(&pkt_dev->list, &t->if_list);
3660 out:
3661 if_unlock(t);
3662 return rv;
3665 /* Called under thread lock */
3667 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname)
3669 struct pktgen_dev *pkt_dev;
3670 int err;
3671 int node = cpu_to_node(t->cpu);
3673 /* We don't allow a device to be on several threads */
3675 pkt_dev = __pktgen_NN_threads(t->net, ifname, FIND);
3676 if (pkt_dev) {
3677 pr_err("ERROR: interface already used\n");
3678 return -EBUSY;
3681 pkt_dev = kzalloc_node(sizeof(struct pktgen_dev), GFP_KERNEL, node);
3682 if (!pkt_dev)
3683 return -ENOMEM;
3685 strcpy(pkt_dev->odevname, ifname);
3686 pkt_dev->flows = vzalloc_node(MAX_CFLOWS * sizeof(struct flow_state),
3687 node);
3688 if (pkt_dev->flows == NULL) {
3689 kfree(pkt_dev);
3690 return -ENOMEM;
3693 pkt_dev->removal_mark = 0;
3694 pkt_dev->nfrags = 0;
3695 pkt_dev->delay = pg_delay_d;
3696 pkt_dev->count = pg_count_d;
3697 pkt_dev->sofar = 0;
3698 pkt_dev->udp_src_min = 9; /* sink port */
3699 pkt_dev->udp_src_max = 9;
3700 pkt_dev->udp_dst_min = 9;
3701 pkt_dev->udp_dst_max = 9;
3702 pkt_dev->vlan_p = 0;
3703 pkt_dev->vlan_cfi = 0;
3704 pkt_dev->vlan_id = 0xffff;
3705 pkt_dev->svlan_p = 0;
3706 pkt_dev->svlan_cfi = 0;
3707 pkt_dev->svlan_id = 0xffff;
3708 pkt_dev->burst = 1;
3709 pkt_dev->node = -1;
3711 err = pktgen_setup_dev(t->net, pkt_dev, ifname);
3712 if (err)
3713 goto out1;
3714 if (pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)
3715 pkt_dev->clone_skb = pg_clone_skb_d;
3717 pkt_dev->entry = proc_create_data(ifname, 0600, t->net->proc_dir,
3718 &pktgen_if_fops, pkt_dev);
3719 if (!pkt_dev->entry) {
3720 pr_err("cannot create %s/%s procfs entry\n",
3721 PG_PROC_DIR, ifname);
3722 err = -EINVAL;
3723 goto out2;
3725 #ifdef CONFIG_XFRM
3726 pkt_dev->ipsmode = XFRM_MODE_TRANSPORT;
3727 pkt_dev->ipsproto = IPPROTO_ESP;
3729 /* xfrm tunnel mode needs additional dst to extract outter
3730 * ip header protocol/ttl/id field, here creat a phony one.
3731 * instead of looking for a valid rt, which definitely hurting
3732 * performance under such circumstance.
3734 pkt_dev->dstops.family = AF_INET;
3735 pkt_dev->dst.dev = pkt_dev->odev;
3736 dst_init_metrics(&pkt_dev->dst, pktgen_dst_metrics, false);
3737 pkt_dev->dst.child = &pkt_dev->dst;
3738 pkt_dev->dst.ops = &pkt_dev->dstops;
3739 #endif
3741 return add_dev_to_thread(t, pkt_dev);
3742 out2:
3743 dev_put(pkt_dev->odev);
3744 out1:
3745 #ifdef CONFIG_XFRM
3746 free_SAs(pkt_dev);
3747 #endif
3748 vfree(pkt_dev->flows);
3749 kfree(pkt_dev);
3750 return err;
3753 static int __net_init pktgen_create_thread(int cpu, struct pktgen_net *pn)
3755 struct pktgen_thread *t;
3756 struct proc_dir_entry *pe;
3757 struct task_struct *p;
3759 t = kzalloc_node(sizeof(struct pktgen_thread), GFP_KERNEL,
3760 cpu_to_node(cpu));
3761 if (!t) {
3762 pr_err("ERROR: out of memory, can't create new thread\n");
3763 return -ENOMEM;
3766 mutex_init(&t->if_lock);
3767 t->cpu = cpu;
3769 INIT_LIST_HEAD(&t->if_list);
3771 list_add_tail(&t->th_list, &pn->pktgen_threads);
3772 init_completion(&t->start_done);
3774 p = kthread_create_on_node(pktgen_thread_worker,
3776 cpu_to_node(cpu),
3777 "kpktgend_%d", cpu);
3778 if (IS_ERR(p)) {
3779 pr_err("kernel_thread() failed for cpu %d\n", t->cpu);
3780 list_del(&t->th_list);
3781 kfree(t);
3782 return PTR_ERR(p);
3784 kthread_bind(p, cpu);
3785 t->tsk = p;
3787 pe = proc_create_data(t->tsk->comm, 0600, pn->proc_dir,
3788 &pktgen_thread_fops, t);
3789 if (!pe) {
3790 pr_err("cannot create %s/%s procfs entry\n",
3791 PG_PROC_DIR, t->tsk->comm);
3792 kthread_stop(p);
3793 list_del(&t->th_list);
3794 kfree(t);
3795 return -EINVAL;
3798 t->net = pn;
3799 get_task_struct(p);
3800 wake_up_process(p);
3801 wait_for_completion(&t->start_done);
3803 return 0;
3807 * Removes a device from the thread if_list.
3809 static void _rem_dev_from_if_list(struct pktgen_thread *t,
3810 struct pktgen_dev *pkt_dev)
3812 struct list_head *q, *n;
3813 struct pktgen_dev *p;
3815 if_lock(t);
3816 list_for_each_safe(q, n, &t->if_list) {
3817 p = list_entry(q, struct pktgen_dev, list);
3818 if (p == pkt_dev)
3819 list_del_rcu(&p->list);
3821 if_unlock(t);
3824 static int pktgen_remove_device(struct pktgen_thread *t,
3825 struct pktgen_dev *pkt_dev)
3827 pr_debug("remove_device pkt_dev=%p\n", pkt_dev);
3829 if (pkt_dev->running) {
3830 pr_warn("WARNING: trying to remove a running interface, stopping it now\n");
3831 pktgen_stop_device(pkt_dev);
3834 /* Dis-associate from the interface */
3836 if (pkt_dev->odev) {
3837 dev_put(pkt_dev->odev);
3838 pkt_dev->odev = NULL;
3841 /* Remove proc before if_list entry, because add_device uses
3842 * list to determine if interface already exist, avoid race
3843 * with proc_create_data() */
3844 proc_remove(pkt_dev->entry);
3846 /* And update the thread if_list */
3847 _rem_dev_from_if_list(t, pkt_dev);
3849 #ifdef CONFIG_XFRM
3850 free_SAs(pkt_dev);
3851 #endif
3852 vfree(pkt_dev->flows);
3853 if (pkt_dev->page)
3854 put_page(pkt_dev->page);
3855 kfree_rcu(pkt_dev, rcu);
3856 return 0;
3859 static int __net_init pg_net_init(struct net *net)
3861 struct pktgen_net *pn = net_generic(net, pg_net_id);
3862 struct proc_dir_entry *pe;
3863 int cpu, ret = 0;
3865 pn->net = net;
3866 INIT_LIST_HEAD(&pn->pktgen_threads);
3867 pn->pktgen_exiting = false;
3868 pn->proc_dir = proc_mkdir(PG_PROC_DIR, pn->net->proc_net);
3869 if (!pn->proc_dir) {
3870 pr_warn("cannot create /proc/net/%s\n", PG_PROC_DIR);
3871 return -ENODEV;
3873 pe = proc_create(PGCTRL, 0600, pn->proc_dir, &pktgen_fops);
3874 if (pe == NULL) {
3875 pr_err("cannot create %s procfs entry\n", PGCTRL);
3876 ret = -EINVAL;
3877 goto remove;
3880 for_each_online_cpu(cpu) {
3881 int err;
3883 err = pktgen_create_thread(cpu, pn);
3884 if (err)
3885 pr_warn("Cannot create thread for cpu %d (%d)\n",
3886 cpu, err);
3889 if (list_empty(&pn->pktgen_threads)) {
3890 pr_err("Initialization failed for all threads\n");
3891 ret = -ENODEV;
3892 goto remove_entry;
3895 return 0;
3897 remove_entry:
3898 remove_proc_entry(PGCTRL, pn->proc_dir);
3899 remove:
3900 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3901 return ret;
3904 static void __net_exit pg_net_exit(struct net *net)
3906 struct pktgen_net *pn = net_generic(net, pg_net_id);
3907 struct pktgen_thread *t;
3908 struct list_head *q, *n;
3909 LIST_HEAD(list);
3911 /* Stop all interfaces & threads */
3912 pn->pktgen_exiting = true;
3914 mutex_lock(&pktgen_thread_lock);
3915 list_splice_init(&pn->pktgen_threads, &list);
3916 mutex_unlock(&pktgen_thread_lock);
3918 list_for_each_safe(q, n, &list) {
3919 t = list_entry(q, struct pktgen_thread, th_list);
3920 list_del(&t->th_list);
3921 kthread_stop(t->tsk);
3922 put_task_struct(t->tsk);
3923 kfree(t);
3926 remove_proc_entry(PGCTRL, pn->proc_dir);
3927 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3930 static struct pernet_operations pg_net_ops = {
3931 .init = pg_net_init,
3932 .exit = pg_net_exit,
3933 .id = &pg_net_id,
3934 .size = sizeof(struct pktgen_net),
3937 static int __init pg_init(void)
3939 int ret = 0;
3941 pr_info("%s", version);
3942 ret = register_pernet_subsys(&pg_net_ops);
3943 if (ret)
3944 return ret;
3945 ret = register_netdevice_notifier(&pktgen_notifier_block);
3946 if (ret)
3947 unregister_pernet_subsys(&pg_net_ops);
3949 return ret;
3952 static void __exit pg_cleanup(void)
3954 unregister_netdevice_notifier(&pktgen_notifier_block);
3955 unregister_pernet_subsys(&pg_net_ops);
3956 /* Don't need rcu_barrier() due to use of kfree_rcu() */
3959 module_init(pg_init);
3960 module_exit(pg_cleanup);
3962 MODULE_AUTHOR("Robert Olsson <robert.olsson@its.uu.se>");
3963 MODULE_DESCRIPTION("Packet Generator tool");
3964 MODULE_LICENSE("GPL");
3965 MODULE_VERSION(VERSION);
3966 module_param(pg_count_d, int, 0);
3967 MODULE_PARM_DESC(pg_count_d, "Default number of packets to inject");
3968 module_param(pg_delay_d, int, 0);
3969 MODULE_PARM_DESC(pg_delay_d, "Default delay between packets (nanoseconds)");
3970 module_param(pg_clone_skb_d, int, 0);
3971 MODULE_PARM_DESC(pg_clone_skb_d, "Default number of copies of the same packet");
3972 module_param(debug, int, 0);
3973 MODULE_PARM_DESC(debug, "Enable debugging of pktgen module");