~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/netfilter/ipvs/ip_vs_sync.c

Version: ~ [ linux-5.5-rc1 ] ~ [ linux-5.4.2 ] ~ [ linux-5.3.15 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.88 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.158 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.206 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.206 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.78 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * IPVS         An implementation of the IP virtual server support for the
  3  *              LINUX operating system.  IPVS is now implemented as a module
  4  *              over the NetFilter framework. IPVS can be used to build a
  5  *              high-performance and highly available server based on a
  6  *              cluster of servers.
  7  *
  8  * Version 1,   is capable of handling both version 0 and 1 messages.
  9  *              Version 0 is the plain old format.
 10  *              Note Version 0 receivers will just drop Ver 1 messages.
 11  *              Version 1 is capable of handle IPv6, Persistence data,
 12  *              time-outs, and firewall marks.
 13  *              In ver.1 "ip_vs_sync_conn_options" will be sent in netw. order.
 14  *              Ver. 0 can be turned on by sysctl -w net.ipv4.vs.sync_version=0
 15  *
 16  * Definitions  Message: is a complete datagram
 17  *              Sync_conn: is a part of a Message
 18  *              Param Data is an option to a Sync_conn.
 19  *
 20  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
 21  *
 22  * ip_vs_sync:  sync connection info from master load balancer to backups
 23  *              through multicast
 24  *
 25  * Changes:
 26  *      Alexandre Cassen        :       Added master & backup support at a time.
 27  *      Alexandre Cassen        :       Added SyncID support for incoming sync
 28  *                                      messages filtering.
 29  *      Justin Ossevoort        :       Fix endian problem on sync message size.
 30  *      Hans Schillstrom        :       Added Version 1: i.e. IPv6,
 31  *                                      Persistence support, fwmark and time-out.
 32  */
 33 
 34 #define KMSG_COMPONENT "IPVS"
 35 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
 36 
 37 #include <linux/module.h>
 38 #include <linux/slab.h>
 39 #include <linux/inetdevice.h>
 40 #include <linux/net.h>
 41 #include <linux/completion.h>
 42 #include <linux/delay.h>
 43 #include <linux/skbuff.h>
 44 #include <linux/in.h>
 45 #include <linux/igmp.h>                 /* for ip_mc_join_group */
 46 #include <linux/udp.h>
 47 #include <linux/err.h>
 48 #include <linux/kthread.h>
 49 #include <linux/wait.h>
 50 #include <linux/kernel.h>
 51 
 52 #include <asm/unaligned.h>              /* Used for ntoh_seq and hton_seq */
 53 
 54 #include <net/ip.h>
 55 #include <net/sock.h>
 56 
 57 #include <net/ip_vs.h>
 58 
 59 #define IP_VS_SYNC_GROUP 0xe0000051    /* multicast addr - 224.0.0.81 */
 60 #define IP_VS_SYNC_PORT  8848          /* multicast port */
 61 
 62 #define SYNC_PROTO_VER  1               /* Protocol version in header */
 63 
 64 static struct lock_class_key __ipvs_sync_key;
 65 /*
 66  *      IPVS sync connection entry
 67  *      Version 0, i.e. original version.
 68  */
 69 struct ip_vs_sync_conn_v0 {
 70         __u8                    reserved;
 71 
 72         /* Protocol, addresses and port numbers */
 73         __u8                    protocol;       /* Which protocol (TCP/UDP) */
 74         __be16                  cport;
 75         __be16                  vport;
 76         __be16                  dport;
 77         __be32                  caddr;          /* client address */
 78         __be32                  vaddr;          /* virtual address */
 79         __be32                  daddr;          /* destination address */
 80 
 81         /* Flags and state transition */
 82         __be16                  flags;          /* status flags */
 83         __be16                  state;          /* state info */
 84 
 85         /* The sequence options start here */
 86 };
 87 
 88 struct ip_vs_sync_conn_options {
 89         struct ip_vs_seq        in_seq;         /* incoming seq. struct */
 90         struct ip_vs_seq        out_seq;        /* outgoing seq. struct */
 91 };
 92 
 93 /*
 94      Sync Connection format (sync_conn)
 95 
 96        0                   1                   2                   3
 97        0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
 98       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 99       |    Type       |    Protocol   | Ver.  |        Size           |
100       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
101       |                             Flags                             |
102       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
103       |            State              |         cport                 |
104       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
105       |            vport              |         dport                 |
106       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
107       |                             fwmark                            |
108       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
109       |                             timeout  (in sec.)                |
110       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
111       |                              ...                              |
112       |                        IP-Addresses  (v4 or v6)               |
113       |                              ...                              |
114       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
115   Optional Parameters.
116       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
117       | Param. Type    | Param. Length |   Param. data                |
118       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+                               |
119       |                              ...                              |
120       |                               +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
121       |                               | Param Type    | Param. Length |
122       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
123       |                           Param  data                         |
124       |         Last Param data should be padded for 32 bit alignment |
125       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
126 */
127 
128 /*
129  *  Type 0, IPv4 sync connection format
130  */
131 struct ip_vs_sync_v4 {
132         __u8                    type;
133         __u8                    protocol;       /* Which protocol (TCP/UDP) */
134         __be16                  ver_size;       /* Version msb 4 bits */
135         /* Flags and state transition */
136         __be32                  flags;          /* status flags */
137         __be16                  state;          /* state info   */
138         /* Protocol, addresses and port numbers */
139         __be16                  cport;
140         __be16                  vport;
141         __be16                  dport;
142         __be32                  fwmark;         /* Firewall mark from skb */
143         __be32                  timeout;        /* cp timeout */
144         __be32                  caddr;          /* client address */
145         __be32                  vaddr;          /* virtual address */
146         __be32                  daddr;          /* destination address */
147         /* The sequence options start here */
148         /* PE data padded to 32bit alignment after seq. options */
149 };
150 /*
151  * Type 2 messages IPv6
152  */
153 struct ip_vs_sync_v6 {
154         __u8                    type;
155         __u8                    protocol;       /* Which protocol (TCP/UDP) */
156         __be16                  ver_size;       /* Version msb 4 bits */
157         /* Flags and state transition */
158         __be32                  flags;          /* status flags */
159         __be16                  state;          /* state info   */
160         /* Protocol, addresses and port numbers */
161         __be16                  cport;
162         __be16                  vport;
163         __be16                  dport;
164         __be32                  fwmark;         /* Firewall mark from skb */
165         __be32                  timeout;        /* cp timeout */
166         struct in6_addr         caddr;          /* client address */
167         struct in6_addr         vaddr;          /* virtual address */
168         struct in6_addr         daddr;          /* destination address */
169         /* The sequence options start here */
170         /* PE data padded to 32bit alignment after seq. options */
171 };
172 
173 union ip_vs_sync_conn {
174         struct ip_vs_sync_v4    v4;
175         struct ip_vs_sync_v6    v6;
176 };
177 
178 /* Bits in Type field in above */
179 #define STYPE_INET6             0
180 #define STYPE_F_INET6           (1 << STYPE_INET6)
181 
182 #define SVER_SHIFT              12              /* Shift to get version */
183 #define SVER_MASK               0x0fff          /* Mask to strip version */
184 
185 #define IPVS_OPT_SEQ_DATA       1
186 #define IPVS_OPT_PE_DATA        2
187 #define IPVS_OPT_PE_NAME        3
188 #define IPVS_OPT_PARAM          7
189 
190 #define IPVS_OPT_F_SEQ_DATA     (1 << (IPVS_OPT_SEQ_DATA-1))
191 #define IPVS_OPT_F_PE_DATA      (1 << (IPVS_OPT_PE_DATA-1))
192 #define IPVS_OPT_F_PE_NAME      (1 << (IPVS_OPT_PE_NAME-1))
193 #define IPVS_OPT_F_PARAM        (1 << (IPVS_OPT_PARAM-1))
194 
195 struct ip_vs_sync_thread_data {
196         struct net *net;
197         struct socket *sock;
198         char *buf;
199         int id;
200 };
201 
202 /* Version 0 definition of packet sizes */
203 #define SIMPLE_CONN_SIZE  (sizeof(struct ip_vs_sync_conn_v0))
204 #define FULL_CONN_SIZE  \
205 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options))
206 
207 
208 /*
209   The master mulitcasts messages (Datagrams) to the backup load balancers
210   in the following format.
211 
212  Version 1:
213   Note, first byte should be Zero, so ver 0 receivers will drop the packet.
214 
215        0                   1                   2                   3
216        0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
217       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
218       |      0        |    SyncID     |            Size               |
219       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
220       |  Count Conns  |    Version    |    Reserved, set to Zero      |
221       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
222       |                                                               |
223       |                    IPVS Sync Connection (1)                   |
224       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
225       |                            .                                  |
226       ~                            .                                  ~
227       |                            .                                  |
228       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
229       |                                                               |
230       |                    IPVS Sync Connection (n)                   |
231       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
232 
233  Version 0 Header
234        0                   1                   2                   3
235        0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
236       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
237       |  Count Conns  |    SyncID     |            Size               |
238       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
239       |                    IPVS Sync Connection (1)                   |
240 */
241 
242 #define SYNC_MESG_HEADER_LEN    4
243 #define MAX_CONNS_PER_SYNCBUFF  255 /* nr_conns in ip_vs_sync_mesg is 8 bit */
244 
245 /* Version 0 header */
246 struct ip_vs_sync_mesg_v0 {
247         __u8                    nr_conns;
248         __u8                    syncid;
249         __u16                   size;
250 
251         /* ip_vs_sync_conn entries start here */
252 };
253 
254 /* Version 1 header */
255 struct ip_vs_sync_mesg {
256         __u8                    reserved;       /* must be zero */
257         __u8                    syncid;
258         __u16                   size;
259         __u8                    nr_conns;
260         __s8                    version;        /* SYNC_PROTO_VER  */
261         __u16                   spare;
262         /* ip_vs_sync_conn entries start here */
263 };
264 
265 struct ip_vs_sync_buff {
266         struct list_head        list;
267         unsigned long           firstuse;
268 
269         /* pointers for the message data */
270         struct ip_vs_sync_mesg  *mesg;
271         unsigned char           *head;
272         unsigned char           *end;
273 };
274 
275 /*
276  * Copy of struct ip_vs_seq
277  * From unaligned network order to aligned host order
278  */
279 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho)
280 {
281         ho->init_seq       = get_unaligned_be32(&no->init_seq);
282         ho->delta          = get_unaligned_be32(&no->delta);
283         ho->previous_delta = get_unaligned_be32(&no->previous_delta);
284 }
285 
286 /*
287  * Copy of struct ip_vs_seq
288  * From Aligned host order to unaligned network order
289  */
290 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no)
291 {
292         put_unaligned_be32(ho->init_seq, &no->init_seq);
293         put_unaligned_be32(ho->delta, &no->delta);
294         put_unaligned_be32(ho->previous_delta, &no->previous_delta);
295 }
296 
297 static inline struct ip_vs_sync_buff *
298 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
299 {
300         struct ip_vs_sync_buff *sb;
301 
302         spin_lock_bh(&ipvs->sync_lock);
303         if (list_empty(&ms->sync_queue)) {
304                 sb = NULL;
305                 __set_current_state(TASK_INTERRUPTIBLE);
306         } else {
307                 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff,
308                                 list);
309                 list_del(&sb->list);
310                 ms->sync_queue_len--;
311                 if (!ms->sync_queue_len)
312                         ms->sync_queue_delay = 0;
313         }
314         spin_unlock_bh(&ipvs->sync_lock);
315 
316         return sb;
317 }
318 
319 /*
320  * Create a new sync buffer for Version 1 proto.
321  */
322 static inline struct ip_vs_sync_buff *
323 ip_vs_sync_buff_create(struct netns_ipvs *ipvs)
324 {
325         struct ip_vs_sync_buff *sb;
326 
327         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
328                 return NULL;
329 
330         sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
331         if (!sb->mesg) {
332                 kfree(sb);
333                 return NULL;
334         }
335         sb->mesg->reserved = 0;  /* old nr_conns i.e. must be zero now */
336         sb->mesg->version = SYNC_PROTO_VER;
337         sb->mesg->syncid = ipvs->master_syncid;
338         sb->mesg->size = sizeof(struct ip_vs_sync_mesg);
339         sb->mesg->nr_conns = 0;
340         sb->mesg->spare = 0;
341         sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg);
342         sb->end = (unsigned char *)sb->mesg + ipvs->send_mesg_maxlen;
343 
344         sb->firstuse = jiffies;
345         return sb;
346 }
347 
348 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb)
349 {
350         kfree(sb->mesg);
351         kfree(sb);
352 }
353 
354 static inline void sb_queue_tail(struct netns_ipvs *ipvs,
355                                  struct ipvs_master_sync_state *ms)
356 {
357         struct ip_vs_sync_buff *sb = ms->sync_buff;
358 
359         spin_lock(&ipvs->sync_lock);
360         if (ipvs->sync_state & IP_VS_STATE_MASTER &&
361             ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) {
362                 if (!ms->sync_queue_len)
363                         schedule_delayed_work(&ms->master_wakeup_work,
364                                               max(IPVS_SYNC_SEND_DELAY, 1));
365                 ms->sync_queue_len++;
366                 list_add_tail(&sb->list, &ms->sync_queue);
367                 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE)
368                         wake_up_process(ms->master_thread);
369         } else
370                 ip_vs_sync_buff_release(sb);
371         spin_unlock(&ipvs->sync_lock);
372 }
373 
374 /*
375  *      Get the current sync buffer if it has been created for more
376  *      than the specified time or the specified time is zero.
377  */
378 static inline struct ip_vs_sync_buff *
379 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms,
380                    unsigned long time)
381 {
382         struct ip_vs_sync_buff *sb;
383 
384         spin_lock_bh(&ipvs->sync_buff_lock);
385         sb = ms->sync_buff;
386         if (sb && time_after_eq(jiffies - sb->firstuse, time)) {
387                 ms->sync_buff = NULL;
388                 __set_current_state(TASK_RUNNING);
389         } else
390                 sb = NULL;
391         spin_unlock_bh(&ipvs->sync_buff_lock);
392         return sb;
393 }
394 
395 static inline int
396 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp)
397 {
398         return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask;
399 }
400 
401 /*
402  * Create a new sync buffer for Version 0 proto.
403  */
404 static inline struct ip_vs_sync_buff *
405 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs)
406 {
407         struct ip_vs_sync_buff *sb;
408         struct ip_vs_sync_mesg_v0 *mesg;
409 
410         if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC)))
411                 return NULL;
412 
413         sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC);
414         if (!sb->mesg) {
415                 kfree(sb);
416                 return NULL;
417         }
418         mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg;
419         mesg->nr_conns = 0;
420         mesg->syncid = ipvs->master_syncid;
421         mesg->size = sizeof(struct ip_vs_sync_mesg_v0);
422         sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0);
423         sb->end = (unsigned char *)mesg + ipvs->send_mesg_maxlen;
424         sb->firstuse = jiffies;
425         return sb;
426 }
427 
428 /* Check if conn should be synced.
429  * pkts: conn packets, use sysctl_sync_threshold to avoid packet check
430  * - (1) sync_refresh_period: reduce sync rate. Additionally, retry
431  *      sync_retries times with period of sync_refresh_period/8
432  * - (2) if both sync_refresh_period and sync_period are 0 send sync only
433  *      for state changes or only once when pkts matches sync_threshold
434  * - (3) templates: rate can be reduced only with sync_refresh_period or
435  *      with (2)
436  */
437 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs,
438                                   struct ip_vs_conn *cp, int pkts)
439 {
440         unsigned long orig = ACCESS_ONCE(cp->sync_endtime);
441         unsigned long now = jiffies;
442         unsigned long n = (now + cp->timeout) & ~3UL;
443         unsigned int sync_refresh_period;
444         int sync_period;
445         int force;
446 
447         /* Check if we sync in current state */
448         if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE))
449                 force = 0;
450         else if (likely(cp->protocol == IPPROTO_TCP)) {
451                 if (!((1 << cp->state) &
452                       ((1 << IP_VS_TCP_S_ESTABLISHED) |
453                        (1 << IP_VS_TCP_S_FIN_WAIT) |
454                        (1 << IP_VS_TCP_S_CLOSE) |
455                        (1 << IP_VS_TCP_S_CLOSE_WAIT) |
456                        (1 << IP_VS_TCP_S_TIME_WAIT))))
457                         return 0;
458                 force = cp->state != cp->old_state;
459                 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED)
460                         goto set;
461         } else if (unlikely(cp->protocol == IPPROTO_SCTP)) {
462                 if (!((1 << cp->state) &
463                       ((1 << IP_VS_SCTP_S_ESTABLISHED) |
464                        (1 << IP_VS_SCTP_S_CLOSED) |
465                        (1 << IP_VS_SCTP_S_SHUT_ACK_CLI) |
466                        (1 << IP_VS_SCTP_S_SHUT_ACK_SER))))
467                         return 0;
468                 force = cp->state != cp->old_state;
469                 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED)
470                         goto set;
471         } else {
472                 /* UDP or another protocol with single state */
473                 force = 0;
474         }
475 
476         sync_refresh_period = sysctl_sync_refresh_period(ipvs);
477         if (sync_refresh_period > 0) {
478                 long diff = n - orig;
479                 long min_diff = max(cp->timeout >> 1, 10UL * HZ);
480 
481                 /* Avoid sync if difference is below sync_refresh_period
482                  * and below the half timeout.
483                  */
484                 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) {
485                         int retries = orig & 3;
486 
487                         if (retries >= sysctl_sync_retries(ipvs))
488                                 return 0;
489                         if (time_before(now, orig - cp->timeout +
490                                         (sync_refresh_period >> 3)))
491                                 return 0;
492                         n |= retries + 1;
493                 }
494         }
495         sync_period = sysctl_sync_period(ipvs);
496         if (sync_period > 0) {
497                 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) &&
498                     pkts % sync_period != sysctl_sync_threshold(ipvs))
499                         return 0;
500         } else if (sync_refresh_period <= 0 &&
501                    pkts != sysctl_sync_threshold(ipvs))
502                 return 0;
503 
504 set:
505         cp->old_state = cp->state;
506         n = cmpxchg(&cp->sync_endtime, orig, n);
507         return n == orig || force;
508 }
509 
510 /*
511  *      Version 0 , could be switched in by sys_ctl.
512  *      Add an ip_vs_conn information into the current sync_buff.
513  */
514 static void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp,
515                                int pkts)
516 {
517         struct netns_ipvs *ipvs = net_ipvs(net);
518         struct ip_vs_sync_mesg_v0 *m;
519         struct ip_vs_sync_conn_v0 *s;
520         struct ip_vs_sync_buff *buff;
521         struct ipvs_master_sync_state *ms;
522         int id;
523         int len;
524 
525         if (unlikely(cp->af != AF_INET))
526                 return;
527         /* Do not sync ONE PACKET */
528         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
529                 return;
530 
531         if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
532                 return;
533 
534         spin_lock(&ipvs->sync_buff_lock);
535         if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
536                 spin_unlock(&ipvs->sync_buff_lock);
537                 return;
538         }
539 
540         id = select_master_thread_id(ipvs, cp);
541         ms = &ipvs->ms[id];
542         buff = ms->sync_buff;
543         if (buff) {
544                 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
545                 /* Send buffer if it is for v1 */
546                 if (!m->nr_conns) {
547                         sb_queue_tail(ipvs, ms);
548                         ms->sync_buff = NULL;
549                         buff = NULL;
550                 }
551         }
552         if (!buff) {
553                 buff = ip_vs_sync_buff_create_v0(ipvs);
554                 if (!buff) {
555                         spin_unlock(&ipvs->sync_buff_lock);
556                         pr_err("ip_vs_sync_buff_create failed.\n");
557                         return;
558                 }
559                 ms->sync_buff = buff;
560         }
561 
562         len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE :
563                 SIMPLE_CONN_SIZE;
564         m = (struct ip_vs_sync_mesg_v0 *) buff->mesg;
565         s = (struct ip_vs_sync_conn_v0 *) buff->head;
566 
567         /* copy members */
568         s->reserved = 0;
569         s->protocol = cp->protocol;
570         s->cport = cp->cport;
571         s->vport = cp->vport;
572         s->dport = cp->dport;
573         s->caddr = cp->caddr.ip;
574         s->vaddr = cp->vaddr.ip;
575         s->daddr = cp->daddr.ip;
576         s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED);
577         s->state = htons(cp->state);
578         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
579                 struct ip_vs_sync_conn_options *opt =
580                         (struct ip_vs_sync_conn_options *)&s[1];
581                 memcpy(opt, &cp->in_seq, sizeof(*opt));
582         }
583 
584         m->nr_conns++;
585         m->size += len;
586         buff->head += len;
587 
588         /* check if there is a space for next one */
589         if (buff->head + FULL_CONN_SIZE > buff->end) {
590                 sb_queue_tail(ipvs, ms);
591                 ms->sync_buff = NULL;
592         }
593         spin_unlock(&ipvs->sync_buff_lock);
594 
595         /* synchronize its controller if it has */
596         cp = cp->control;
597         if (cp) {
598                 if (cp->flags & IP_VS_CONN_F_TEMPLATE)
599                         pkts = atomic_add_return(1, &cp->in_pkts);
600                 else
601                         pkts = sysctl_sync_threshold(ipvs);
602                 ip_vs_sync_conn(net, cp->control, pkts);
603         }
604 }
605 
606 /*
607  *      Add an ip_vs_conn information into the current sync_buff.
608  *      Called by ip_vs_in.
609  *      Sending Version 1 messages
610  */
611 void ip_vs_sync_conn(struct net *net, struct ip_vs_conn *cp, int pkts)
612 {
613         struct netns_ipvs *ipvs = net_ipvs(net);
614         struct ip_vs_sync_mesg *m;
615         union ip_vs_sync_conn *s;
616         struct ip_vs_sync_buff *buff;
617         struct ipvs_master_sync_state *ms;
618         int id;
619         __u8 *p;
620         unsigned int len, pe_name_len, pad;
621 
622         /* Handle old version of the protocol */
623         if (sysctl_sync_ver(ipvs) == 0) {
624                 ip_vs_sync_conn_v0(net, cp, pkts);
625                 return;
626         }
627         /* Do not sync ONE PACKET */
628         if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
629                 goto control;
630 sloop:
631         if (!ip_vs_sync_conn_needed(ipvs, cp, pkts))
632                 goto control;
633 
634         /* Sanity checks */
635         pe_name_len = 0;
636         if (cp->pe_data_len) {
637                 if (!cp->pe_data || !cp->dest) {
638                         IP_VS_ERR_RL("SYNC, connection pe_data invalid\n");
639                         return;
640                 }
641                 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN);
642         }
643 
644         spin_lock(&ipvs->sync_buff_lock);
645         if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
646                 spin_unlock(&ipvs->sync_buff_lock);
647                 return;
648         }
649 
650         id = select_master_thread_id(ipvs, cp);
651         ms = &ipvs->ms[id];
652 
653 #ifdef CONFIG_IP_VS_IPV6
654         if (cp->af == AF_INET6)
655                 len = sizeof(struct ip_vs_sync_v6);
656         else
657 #endif
658                 len = sizeof(struct ip_vs_sync_v4);
659 
660         if (cp->flags & IP_VS_CONN_F_SEQ_MASK)
661                 len += sizeof(struct ip_vs_sync_conn_options) + 2;
662 
663         if (cp->pe_data_len)
664                 len += cp->pe_data_len + 2;     /* + Param hdr field */
665         if (pe_name_len)
666                 len += pe_name_len + 2;
667 
668         /* check if there is a space for this one  */
669         pad = 0;
670         buff = ms->sync_buff;
671         if (buff) {
672                 m = buff->mesg;
673                 pad = (4 - (size_t) buff->head) & 3;
674                 /* Send buffer if it is for v0 */
675                 if (buff->head + len + pad > buff->end || m->reserved) {
676                         sb_queue_tail(ipvs, ms);
677                         ms->sync_buff = NULL;
678                         buff = NULL;
679                         pad = 0;
680                 }
681         }
682 
683         if (!buff) {
684                 buff = ip_vs_sync_buff_create(ipvs);
685                 if (!buff) {
686                         spin_unlock(&ipvs->sync_buff_lock);
687                         pr_err("ip_vs_sync_buff_create failed.\n");
688                         return;
689                 }
690                 ms->sync_buff = buff;
691                 m = buff->mesg;
692         }
693 
694         p = buff->head;
695         buff->head += pad + len;
696         m->size += pad + len;
697         /* Add ev. padding from prev. sync_conn */
698         while (pad--)
699                 *(p++) = 0;
700 
701         s = (union ip_vs_sync_conn *)p;
702 
703         /* Set message type  & copy members */
704         s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0);
705         s->v4.ver_size = htons(len & SVER_MASK);        /* Version 0 */
706         s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED);
707         s->v4.state = htons(cp->state);
708         s->v4.protocol = cp->protocol;
709         s->v4.cport = cp->cport;
710         s->v4.vport = cp->vport;
711         s->v4.dport = cp->dport;
712         s->v4.fwmark = htonl(cp->fwmark);
713         s->v4.timeout = htonl(cp->timeout / HZ);
714         m->nr_conns++;
715 
716 #ifdef CONFIG_IP_VS_IPV6
717         if (cp->af == AF_INET6) {
718                 p += sizeof(struct ip_vs_sync_v6);
719                 s->v6.caddr = cp->caddr.in6;
720                 s->v6.vaddr = cp->vaddr.in6;
721                 s->v6.daddr = cp->daddr.in6;
722         } else
723 #endif
724         {
725                 p += sizeof(struct ip_vs_sync_v4);      /* options ptr */
726                 s->v4.caddr = cp->caddr.ip;
727                 s->v4.vaddr = cp->vaddr.ip;
728                 s->v4.daddr = cp->daddr.ip;
729         }
730         if (cp->flags & IP_VS_CONN_F_SEQ_MASK) {
731                 *(p++) = IPVS_OPT_SEQ_DATA;
732                 *(p++) = sizeof(struct ip_vs_sync_conn_options);
733                 hton_seq((struct ip_vs_seq *)p, &cp->in_seq);
734                 p += sizeof(struct ip_vs_seq);
735                 hton_seq((struct ip_vs_seq *)p, &cp->out_seq);
736                 p += sizeof(struct ip_vs_seq);
737         }
738         /* Handle pe data */
739         if (cp->pe_data_len && cp->pe_data) {
740                 *(p++) = IPVS_OPT_PE_DATA;
741                 *(p++) = cp->pe_data_len;
742                 memcpy(p, cp->pe_data, cp->pe_data_len);
743                 p += cp->pe_data_len;
744                 if (pe_name_len) {
745                         /* Add PE_NAME */
746                         *(p++) = IPVS_OPT_PE_NAME;
747                         *(p++) = pe_name_len;
748                         memcpy(p, cp->pe->name, pe_name_len);
749                         p += pe_name_len;
750                 }
751         }
752 
753         spin_unlock(&ipvs->sync_buff_lock);
754 
755 control:
756         /* synchronize its controller if it has */
757         cp = cp->control;
758         if (!cp)
759                 return;
760         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
761                 pkts = atomic_add_return(1, &cp->in_pkts);
762         else
763                 pkts = sysctl_sync_threshold(ipvs);
764         goto sloop;
765 }
766 
767 /*
768  *  fill_param used by version 1
769  */
770 static inline int
771 ip_vs_conn_fill_param_sync(struct net *net, int af, union ip_vs_sync_conn *sc,
772                            struct ip_vs_conn_param *p,
773                            __u8 *pe_data, unsigned int pe_data_len,
774                            __u8 *pe_name, unsigned int pe_name_len)
775 {
776 #ifdef CONFIG_IP_VS_IPV6
777         if (af == AF_INET6)
778                 ip_vs_conn_fill_param(net, af, sc->v6.protocol,
779                                       (const union nf_inet_addr *)&sc->v6.caddr,
780                                       sc->v6.cport,
781                                       (const union nf_inet_addr *)&sc->v6.vaddr,
782                                       sc->v6.vport, p);
783         else
784 #endif
785                 ip_vs_conn_fill_param(net, af, sc->v4.protocol,
786                                       (const union nf_inet_addr *)&sc->v4.caddr,
787                                       sc->v4.cport,
788                                       (const union nf_inet_addr *)&sc->v4.vaddr,
789                                       sc->v4.vport, p);
790         /* Handle pe data */
791         if (pe_data_len) {
792                 if (pe_name_len) {
793                         char buff[IP_VS_PENAME_MAXLEN+1];
794 
795                         memcpy(buff, pe_name, pe_name_len);
796                         buff[pe_name_len]=0;
797                         p->pe = __ip_vs_pe_getbyname(buff);
798                         if (!p->pe) {
799                                 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n",
800                                              buff);
801                                 return 1;
802                         }
803                 } else {
804                         IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n");
805                         return 1;
806                 }
807 
808                 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC);
809                 if (!p->pe_data) {
810                         if (p->pe->module)
811                                 module_put(p->pe->module);
812                         return -ENOMEM;
813                 }
814                 p->pe_data_len = pe_data_len;
815         }
816         return 0;
817 }
818 
819 /*
820  *  Connection Add / Update.
821  *  Common for version 0 and 1 reception of backup sync_conns.
822  *  Param: ...
823  *         timeout is in sec.
824  */
825 static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param,
826                             unsigned int flags, unsigned int state,
827                             unsigned int protocol, unsigned int type,
828                             const union nf_inet_addr *daddr, __be16 dport,
829                             unsigned long timeout, __u32 fwmark,
830                             struct ip_vs_sync_conn_options *opt)
831 {
832         struct ip_vs_dest *dest;
833         struct ip_vs_conn *cp;
834         struct netns_ipvs *ipvs = net_ipvs(net);
835 
836         if (!(flags & IP_VS_CONN_F_TEMPLATE))
837                 cp = ip_vs_conn_in_get(param);
838         else
839                 cp = ip_vs_ct_in_get(param);
840 
841         if (cp) {
842                 /* Free pe_data */
843                 kfree(param->pe_data);
844 
845                 dest = cp->dest;
846                 spin_lock(&cp->lock);
847                 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE &&
848                     !(flags & IP_VS_CONN_F_TEMPLATE) && dest) {
849                         if (flags & IP_VS_CONN_F_INACTIVE) {
850                                 atomic_dec(&dest->activeconns);
851                                 atomic_inc(&dest->inactconns);
852                         } else {
853                                 atomic_inc(&dest->activeconns);
854                                 atomic_dec(&dest->inactconns);
855                         }
856                 }
857                 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK;
858                 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK;
859                 cp->flags = flags;
860                 spin_unlock(&cp->lock);
861                 if (!dest) {
862                         dest = ip_vs_try_bind_dest(cp);
863                         if (dest)
864                                 atomic_dec(&dest->refcnt);
865                 }
866         } else {
867                 /*
868                  * Find the appropriate destination for the connection.
869                  * If it is not found the connection will remain unbound
870                  * but still handled.
871                  */
872                 dest = ip_vs_find_dest(net, type, daddr, dport, param->vaddr,
873                                        param->vport, protocol, fwmark, flags);
874 
875                 cp = ip_vs_conn_new(param, daddr, dport, flags, dest, fwmark);
876                 if (dest)
877                         atomic_dec(&dest->refcnt);
878                 if (!cp) {
879                         if (param->pe_data)
880                                 kfree(param->pe_data);
881                         IP_VS_DBG(2, "BACKUP, add new conn. failed\n");
882                         return;
883                 }
884         }
885 
886         if (opt)
887                 memcpy(&cp->in_seq, opt, sizeof(*opt));
888         atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs));
889         cp->state = state;
890         cp->old_state = cp->state;
891         /*
892          * For Ver 0 messages style
893          *  - Not possible to recover the right timeout for templates
894          *  - can not find the right fwmark
895          *    virtual service. If needed, we can do it for
896          *    non-fwmark persistent services.
897          * Ver 1 messages style.
898          *  - No problem.
899          */
900         if (timeout) {
901                 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ)
902                         timeout = MAX_SCHEDULE_TIMEOUT / HZ;
903                 cp->timeout = timeout*HZ;
904         } else {
905                 struct ip_vs_proto_data *pd;
906 
907                 pd = ip_vs_proto_data_get(net, protocol);
908                 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table)
909                         cp->timeout = pd->timeout_table[state];
910                 else
911                         cp->timeout = (3*60*HZ);
912         }
913         ip_vs_conn_put(cp);
914 }
915 
916 /*
917  *  Process received multicast message for Version 0
918  */
919 static void ip_vs_process_message_v0(struct net *net, const char *buffer,
920                                      const size_t buflen)
921 {
922         struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer;
923         struct ip_vs_sync_conn_v0 *s;
924         struct ip_vs_sync_conn_options *opt;
925         struct ip_vs_protocol *pp;
926         struct ip_vs_conn_param param;
927         char *p;
928         int i;
929 
930         p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0);
931         for (i=0; i<m->nr_conns; i++) {
932                 unsigned int flags, state;
933 
934                 if (p + SIMPLE_CONN_SIZE > buffer+buflen) {
935                         IP_VS_ERR_RL("BACKUP v0, bogus conn\n");
936                         return;
937                 }
938                 s = (struct ip_vs_sync_conn_v0 *) p;
939                 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC;
940                 flags &= ~IP_VS_CONN_F_HASHED;
941                 if (flags & IP_VS_CONN_F_SEQ_MASK) {
942                         opt = (struct ip_vs_sync_conn_options *)&s[1];
943                         p += FULL_CONN_SIZE;
944                         if (p > buffer+buflen) {
945                                 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n");
946                                 return;
947                         }
948                 } else {
949                         opt = NULL;
950                         p += SIMPLE_CONN_SIZE;
951                 }
952 
953                 state = ntohs(s->state);
954                 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
955                         pp = ip_vs_proto_get(s->protocol);
956                         if (!pp) {
957                                 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n",
958                                         s->protocol);
959                                 continue;
960                         }
961                         if (state >= pp->num_states) {
962                                 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n",
963                                         pp->name, state);
964                                 continue;
965                         }
966                 } else {
967                         /* protocol in templates is not used for state/timeout */
968                         if (state > 0) {
969                                 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n",
970                                         state);
971                                 state = 0;
972                         }
973                 }
974 
975                 ip_vs_conn_fill_param(net, AF_INET, s->protocol,
976                                       (const union nf_inet_addr *)&s->caddr,
977                                       s->cport,
978                                       (const union nf_inet_addr *)&s->vaddr,
979                                       s->vport, &param);
980 
981                 /* Send timeout as Zero */
982                 ip_vs_proc_conn(net, &param, flags, state, s->protocol, AF_INET,
983                                 (union nf_inet_addr *)&s->daddr, s->dport,
984                                 0, 0, opt);
985         }
986 }
987 
988 /*
989  * Handle options
990  */
991 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen,
992                                     __u32 *opt_flags,
993                                     struct ip_vs_sync_conn_options *opt)
994 {
995         struct ip_vs_sync_conn_options *topt;
996 
997         topt = (struct ip_vs_sync_conn_options *)p;
998 
999         if (plen != sizeof(struct ip_vs_sync_conn_options)) {
1000                 IP_VS_DBG(2, "BACKUP, bogus conn options length\n");
1001                 return -EINVAL;
1002         }
1003         if (*opt_flags & IPVS_OPT_F_SEQ_DATA) {
1004                 IP_VS_DBG(2, "BACKUP, conn options found twice\n");
1005                 return -EINVAL;
1006         }
1007         ntoh_seq(&topt->in_seq, &opt->in_seq);
1008         ntoh_seq(&topt->out_seq, &opt->out_seq);
1009         *opt_flags |= IPVS_OPT_F_SEQ_DATA;
1010         return 0;
1011 }
1012 
1013 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len,
1014                           __u8 **data, unsigned int maxlen,
1015                           __u32 *opt_flags, __u32 flag)
1016 {
1017         if (plen > maxlen) {
1018                 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen);
1019                 return -EINVAL;
1020         }
1021         if (*opt_flags & flag) {
1022                 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag);
1023                 return -EINVAL;
1024         }
1025         *data_len = plen;
1026         *data = p;
1027         *opt_flags |= flag;
1028         return 0;
1029 }
1030 /*
1031  *   Process a Version 1 sync. connection
1032  */
1033 static inline int ip_vs_proc_sync_conn(struct net *net, __u8 *p, __u8 *msg_end)
1034 {
1035         struct ip_vs_sync_conn_options opt;
1036         union  ip_vs_sync_conn *s;
1037         struct ip_vs_protocol *pp;
1038         struct ip_vs_conn_param param;
1039         __u32 flags;
1040         unsigned int af, state, pe_data_len=0, pe_name_len=0;
1041         __u8 *pe_data=NULL, *pe_name=NULL;
1042         __u32 opt_flags=0;
1043         int retc=0;
1044 
1045         s = (union ip_vs_sync_conn *) p;
1046 
1047         if (s->v6.type & STYPE_F_INET6) {
1048 #ifdef CONFIG_IP_VS_IPV6
1049                 af = AF_INET6;
1050                 p += sizeof(struct ip_vs_sync_v6);
1051 #else
1052                 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n");
1053                 retc = 10;
1054                 goto out;
1055 #endif
1056         } else if (!s->v4.type) {
1057                 af = AF_INET;
1058                 p += sizeof(struct ip_vs_sync_v4);
1059         } else {
1060                 return -10;
1061         }
1062         if (p > msg_end)
1063                 return -20;
1064 
1065         /* Process optional params check Type & Len. */
1066         while (p < msg_end) {
1067                 int ptype;
1068                 int plen;
1069 
1070                 if (p+2 > msg_end)
1071                         return -30;
1072                 ptype = *(p++);
1073                 plen  = *(p++);
1074 
1075                 if (!plen || ((p + plen) > msg_end))
1076                         return -40;
1077                 /* Handle seq option  p = param data */
1078                 switch (ptype & ~IPVS_OPT_F_PARAM) {
1079                 case IPVS_OPT_SEQ_DATA:
1080                         if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt))
1081                                 return -50;
1082                         break;
1083 
1084                 case IPVS_OPT_PE_DATA:
1085                         if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data,
1086                                            IP_VS_PEDATA_MAXLEN, &opt_flags,
1087                                            IPVS_OPT_F_PE_DATA))
1088                                 return -60;
1089                         break;
1090 
1091                 case IPVS_OPT_PE_NAME:
1092                         if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name,
1093                                            IP_VS_PENAME_MAXLEN, &opt_flags,
1094                                            IPVS_OPT_F_PE_NAME))
1095                                 return -70;
1096                         break;
1097 
1098                 default:
1099                         /* Param data mandatory ? */
1100                         if (!(ptype & IPVS_OPT_F_PARAM)) {
1101                                 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n",
1102                                           ptype & ~IPVS_OPT_F_PARAM);
1103                                 retc = 20;
1104                                 goto out;
1105                         }
1106                 }
1107                 p += plen;  /* Next option */
1108         }
1109 
1110         /* Get flags and Mask off unsupported */
1111         flags  = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK;
1112         flags |= IP_VS_CONN_F_SYNC;
1113         state = ntohs(s->v4.state);
1114 
1115         if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
1116                 pp = ip_vs_proto_get(s->v4.protocol);
1117                 if (!pp) {
1118                         IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n",
1119                                 s->v4.protocol);
1120                         retc = 30;
1121                         goto out;
1122                 }
1123                 if (state >= pp->num_states) {
1124                         IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n",
1125                                 pp->name, state);
1126                         retc = 40;
1127                         goto out;
1128                 }
1129         } else {
1130                 /* protocol in templates is not used for state/timeout */
1131                 if (state > 0) {
1132                         IP_VS_DBG(3, "BACKUP, Invalid template state %u\n",
1133                                 state);
1134                         state = 0;
1135                 }
1136         }
1137         if (ip_vs_conn_fill_param_sync(net, af, s, &param, pe_data,
1138                                        pe_data_len, pe_name, pe_name_len)) {
1139                 retc = 50;
1140                 goto out;
1141         }
1142         /* If only IPv4, just silent skip IPv6 */
1143         if (af == AF_INET)
1144                 ip_vs_proc_conn(net, &param, flags, state, s->v4.protocol, af,
1145                                 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport,
1146                                 ntohl(s->v4.timeout), ntohl(s->v4.fwmark),
1147                                 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1148                                 );
1149 #ifdef CONFIG_IP_VS_IPV6
1150         else
1151                 ip_vs_proc_conn(net, &param, flags, state, s->v6.protocol, af,
1152                                 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport,
1153                                 ntohl(s->v6.timeout), ntohl(s->v6.fwmark),
1154                                 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL)
1155                                 );
1156 #endif
1157         return 0;
1158         /* Error exit */
1159 out:
1160         IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc);
1161         return retc;
1162 
1163 }
1164 /*
1165  *      Process received multicast message and create the corresponding
1166  *      ip_vs_conn entries.
1167  *      Handles Version 0 & 1
1168  */
1169 static void ip_vs_process_message(struct net *net, __u8 *buffer,
1170                                   const size_t buflen)
1171 {
1172         struct netns_ipvs *ipvs = net_ipvs(net);
1173         struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer;
1174         __u8 *p, *msg_end;
1175         int i, nr_conns;
1176 
1177         if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) {
1178                 IP_VS_DBG(2, "BACKUP, message header too short\n");
1179                 return;
1180         }
1181         /* Convert size back to host byte order */
1182         m2->size = ntohs(m2->size);
1183 
1184         if (buflen != m2->size) {
1185                 IP_VS_DBG(2, "BACKUP, bogus message size\n");
1186                 return;
1187         }
1188         /* SyncID sanity check */
1189         if (ipvs->backup_syncid != 0 && m2->syncid != ipvs->backup_syncid) {
1190                 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid);
1191                 return;
1192         }
1193         /* Handle version 1  message */
1194         if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0)
1195             && (m2->spare == 0)) {
1196 
1197                 msg_end = buffer + sizeof(struct ip_vs_sync_mesg);
1198                 nr_conns = m2->nr_conns;
1199 
1200                 for (i=0; i<nr_conns; i++) {
1201                         union ip_vs_sync_conn *s;
1202                         unsigned int size;
1203                         int retc;
1204 
1205                         p = msg_end;
1206                         if (p + sizeof(s->v4) > buffer+buflen) {
1207                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n");
1208                                 return;
1209                         }
1210                         s = (union ip_vs_sync_conn *)p;
1211                         size = ntohs(s->v4.ver_size) & SVER_MASK;
1212                         msg_end = p + size;
1213                         /* Basic sanity checks */
1214                         if (msg_end  > buffer+buflen) {
1215                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n");
1216                                 return;
1217                         }
1218                         if (ntohs(s->v4.ver_size) >> SVER_SHIFT) {
1219                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n",
1220                                               ntohs(s->v4.ver_size) >> SVER_SHIFT);
1221                                 return;
1222                         }
1223                         /* Process a single sync_conn */
1224                         retc = ip_vs_proc_sync_conn(net, p, msg_end);
1225                         if (retc < 0) {
1226                                 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n",
1227                                              retc);
1228                                 return;
1229                         }
1230                         /* Make sure we have 32 bit alignment */
1231                         msg_end = p + ((size + 3) & ~3);
1232                 }
1233         } else {
1234                 /* Old type of message */
1235                 ip_vs_process_message_v0(net, buffer, buflen);
1236                 return;
1237         }
1238 }
1239 
1240 
1241 /*
1242  *      Setup sndbuf (mode=1) or rcvbuf (mode=0)
1243  */
1244 static void set_sock_size(struct sock *sk, int mode, int val)
1245 {
1246         /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */
1247         /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */
1248         lock_sock(sk);
1249         if (mode) {
1250                 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2,
1251                               sysctl_wmem_max);
1252                 sk->sk_sndbuf = val * 2;
1253                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1254         } else {
1255                 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2,
1256                               sysctl_rmem_max);
1257                 sk->sk_rcvbuf = val * 2;
1258                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1259         }
1260         release_sock(sk);
1261 }
1262 
1263 /*
1264  *      Setup loopback of outgoing multicasts on a sending socket
1265  */
1266 static void set_mcast_loop(struct sock *sk, u_char loop)
1267 {
1268         struct inet_sock *inet = inet_sk(sk);
1269 
1270         /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */
1271         lock_sock(sk);
1272         inet->mc_loop = loop ? 1 : 0;
1273         release_sock(sk);
1274 }
1275 
1276 /*
1277  *      Specify TTL for outgoing multicasts on a sending socket
1278  */
1279 static void set_mcast_ttl(struct sock *sk, u_char ttl)
1280 {
1281         struct inet_sock *inet = inet_sk(sk);
1282 
1283         /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */
1284         lock_sock(sk);
1285         inet->mc_ttl = ttl;
1286         release_sock(sk);
1287 }
1288 
1289 /*
1290  *      Specifiy default interface for outgoing multicasts
1291  */
1292 static int set_mcast_if(struct sock *sk, char *ifname)
1293 {
1294         struct net_device *dev;
1295         struct inet_sock *inet = inet_sk(sk);
1296         struct net *net = sock_net(sk);
1297 
1298         dev = __dev_get_by_name(net, ifname);
1299         if (!dev)
1300                 return -ENODEV;
1301 
1302         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1303                 return -EINVAL;
1304 
1305         lock_sock(sk);
1306         inet->mc_index = dev->ifindex;
1307         /*  inet->mc_addr  = 0; */
1308         release_sock(sk);
1309 
1310         return 0;
1311 }
1312 
1313 
1314 /*
1315  *      Set the maximum length of sync message according to the
1316  *      specified interface's MTU.
1317  */
1318 static int set_sync_mesg_maxlen(struct net *net, int sync_state)
1319 {
1320         struct netns_ipvs *ipvs = net_ipvs(net);
1321         struct net_device *dev;
1322         int num;
1323 
1324         if (sync_state == IP_VS_STATE_MASTER) {
1325                 dev = __dev_get_by_name(net, ipvs->master_mcast_ifn);
1326                 if (!dev)
1327                         return -ENODEV;
1328 
1329                 num = (dev->mtu - sizeof(struct iphdr) -
1330                        sizeof(struct udphdr) -
1331                        SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE;
1332                 ipvs->send_mesg_maxlen = SYNC_MESG_HEADER_LEN +
1333                         SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF);
1334                 IP_VS_DBG(7, "setting the maximum length of sync sending "
1335                           "message %d.\n", ipvs->send_mesg_maxlen);
1336         } else if (sync_state == IP_VS_STATE_BACKUP) {
1337                 dev = __dev_get_by_name(net, ipvs->backup_mcast_ifn);
1338                 if (!dev)
1339                         return -ENODEV;
1340 
1341                 ipvs->recv_mesg_maxlen = dev->mtu -
1342                         sizeof(struct iphdr) - sizeof(struct udphdr);
1343                 IP_VS_DBG(7, "setting the maximum length of sync receiving "
1344                           "message %d.\n", ipvs->recv_mesg_maxlen);
1345         }
1346 
1347         return 0;
1348 }
1349 
1350 
1351 /*
1352  *      Join a multicast group.
1353  *      the group is specified by a class D multicast address 224.0.0.0/8
1354  *      in the in_addr structure passed in as a parameter.
1355  */
1356 static int
1357 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname)
1358 {
1359         struct net *net = sock_net(sk);
1360         struct ip_mreqn mreq;
1361         struct net_device *dev;
1362         int ret;
1363 
1364         memset(&mreq, 0, sizeof(mreq));
1365         memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr));
1366 
1367         dev = __dev_get_by_name(net, ifname);
1368         if (!dev)
1369                 return -ENODEV;
1370         if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if)
1371                 return -EINVAL;
1372 
1373         mreq.imr_ifindex = dev->ifindex;
1374 
1375         lock_sock(sk);
1376         ret = ip_mc_join_group(sk, &mreq);
1377         release_sock(sk);
1378 
1379         return ret;
1380 }
1381 
1382 
1383 static int bind_mcastif_addr(struct socket *sock, char *ifname)
1384 {
1385         struct net *net = sock_net(sock->sk);
1386         struct net_device *dev;
1387         __be32 addr;
1388         struct sockaddr_in sin;
1389 
1390         dev = __dev_get_by_name(net, ifname);
1391         if (!dev)
1392                 return -ENODEV;
1393 
1394         addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
1395         if (!addr)
1396                 pr_err("You probably need to specify IP address on "
1397                        "multicast interface.\n");
1398 
1399         IP_VS_DBG(7, "binding socket with (%s) %pI4\n",
1400                   ifname, &addr);
1401 
1402         /* Now bind the socket with the address of multicast interface */
1403         sin.sin_family       = AF_INET;
1404         sin.sin_addr.s_addr  = addr;
1405         sin.sin_port         = 0;
1406 
1407         return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin));
1408 }
1409 
1410 /*
1411  *      Set up sending multicast socket over UDP
1412  */
1413 static struct socket *make_send_sock(struct net *net, int id)
1414 {
1415         struct netns_ipvs *ipvs = net_ipvs(net);
1416         /* multicast addr */
1417         struct sockaddr_in mcast_addr = {
1418                 .sin_family             = AF_INET,
1419                 .sin_port               = cpu_to_be16(IP_VS_SYNC_PORT + id),
1420                 .sin_addr.s_addr        = cpu_to_be32(IP_VS_SYNC_GROUP),
1421         };
1422         struct socket *sock;
1423         int result;
1424 
1425         /* First create a socket move it to right name space later */
1426         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1427         if (result < 0) {
1428                 pr_err("Error during creation of socket; terminating\n");
1429                 return ERR_PTR(result);
1430         }
1431         /*
1432          * Kernel sockets that are a part of a namespace, should not
1433          * hold a reference to a namespace in order to allow to stop it.
1434          * After sk_change_net should be released using sk_release_kernel.
1435          */
1436         sk_change_net(sock->sk, net);
1437         result = set_mcast_if(sock->sk, ipvs->master_mcast_ifn);
1438         if (result < 0) {
1439                 pr_err("Error setting outbound mcast interface\n");
1440                 goto error;
1441         }
1442 
1443         set_mcast_loop(sock->sk, 0);
1444         set_mcast_ttl(sock->sk, 1);
1445         result = sysctl_sync_sock_size(ipvs);
1446         if (result > 0)
1447                 set_sock_size(sock->sk, 1, result);
1448 
1449         result = bind_mcastif_addr(sock, ipvs->master_mcast_ifn);
1450         if (result < 0) {
1451                 pr_err("Error binding address of the mcast interface\n");
1452                 goto error;
1453         }
1454 
1455         result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr,
1456                         sizeof(struct sockaddr), 0);
1457         if (result < 0) {
1458                 pr_err("Error connecting to the multicast addr\n");
1459                 goto error;
1460         }
1461 
1462         return sock;
1463 
1464 error:
1465         sk_release_kernel(sock->sk);
1466         return ERR_PTR(result);
1467 }
1468 
1469 
1470 /*
1471  *      Set up receiving multicast socket over UDP
1472  */
1473 static struct socket *make_receive_sock(struct net *net, int id)
1474 {
1475         struct netns_ipvs *ipvs = net_ipvs(net);
1476         /* multicast addr */
1477         struct sockaddr_in mcast_addr = {
1478                 .sin_family             = AF_INET,
1479                 .sin_port               = cpu_to_be16(IP_VS_SYNC_PORT + id),
1480                 .sin_addr.s_addr        = cpu_to_be32(IP_VS_SYNC_GROUP),
1481         };
1482         struct socket *sock;
1483         int result;
1484 
1485         /* First create a socket */
1486         result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1487         if (result < 0) {
1488                 pr_err("Error during creation of socket; terminating\n");
1489                 return ERR_PTR(result);
1490         }
1491         /*
1492          * Kernel sockets that are a part of a namespace, should not
1493          * hold a reference to a namespace in order to allow to stop it.
1494          * After sk_change_net should be released using sk_release_kernel.
1495          */
1496         sk_change_net(sock->sk, net);
1497         /* it is equivalent to the REUSEADDR option in user-space */
1498         sock->sk->sk_reuse = SK_CAN_REUSE;
1499         result = sysctl_sync_sock_size(ipvs);
1500         if (result > 0)
1501                 set_sock_size(sock->sk, 0, result);
1502 
1503         result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr,
1504                         sizeof(struct sockaddr));
1505         if (result < 0) {
1506                 pr_err("Error binding to the multicast addr\n");
1507                 goto error;
1508         }
1509 
1510         /* join the multicast group */
1511         result = join_mcast_group(sock->sk,
1512                         (struct in_addr *) &mcast_addr.sin_addr,
1513                         ipvs->backup_mcast_ifn);
1514         if (result < 0) {
1515                 pr_err("Error joining to the multicast group\n");
1516                 goto error;
1517         }
1518 
1519         return sock;
1520 
1521 error:
1522         sk_release_kernel(sock->sk);
1523         return ERR_PTR(result);
1524 }
1525 
1526 
1527 static int
1528 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length)
1529 {
1530         struct msghdr   msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL};
1531         struct kvec     iov;
1532         int             len;
1533 
1534         EnterFunction(7);
1535         iov.iov_base     = (void *)buffer;
1536         iov.iov_len      = length;
1537 
1538         len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length));
1539 
1540         LeaveFunction(7);
1541         return len;
1542 }
1543 
1544 static int
1545 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg)
1546 {
1547         int msize;
1548         int ret;
1549 
1550         msize = msg->size;
1551 
1552         /* Put size in network byte order */
1553         msg->size = htons(msg->size);
1554 
1555         ret = ip_vs_send_async(sock, (char *)msg, msize);
1556         if (ret >= 0 || ret == -EAGAIN)
1557                 return ret;
1558         pr_err("ip_vs_send_async error %d\n", ret);
1559         return 0;
1560 }
1561 
1562 static int
1563 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen)
1564 {
1565         struct msghdr           msg = {NULL,};
1566         struct kvec             iov;
1567         int                     len;
1568 
1569         EnterFunction(7);
1570 
1571         /* Receive a packet */
1572         iov.iov_base     = buffer;
1573         iov.iov_len      = (size_t)buflen;
1574 
1575         len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, MSG_DONTWAIT);
1576 
1577         if (len < 0)
1578                 return len;
1579 
1580         LeaveFunction(7);
1581         return len;
1582 }
1583 
1584 /* Wakeup the master thread for sending */
1585 static void master_wakeup_work_handler(struct work_struct *work)
1586 {
1587         struct ipvs_master_sync_state *ms =
1588                 container_of(work, struct ipvs_master_sync_state,
1589                              master_wakeup_work.work);
1590         struct netns_ipvs *ipvs = ms->ipvs;
1591 
1592         spin_lock_bh(&ipvs->sync_lock);
1593         if (ms->sync_queue_len &&
1594             ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) {
1595                 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE;
1596                 wake_up_process(ms->master_thread);
1597         }
1598         spin_unlock_bh(&ipvs->sync_lock);
1599 }
1600 
1601 /* Get next buffer to send */
1602 static inline struct ip_vs_sync_buff *
1603 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms)
1604 {
1605         struct ip_vs_sync_buff *sb;
1606 
1607         sb = sb_dequeue(ipvs, ms);
1608         if (sb)
1609                 return sb;
1610         /* Do not delay entries in buffer for more than 2 seconds */
1611         return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME);
1612 }
1613 
1614 static int sync_thread_master(void *data)
1615 {
1616         struct ip_vs_sync_thread_data *tinfo = data;
1617         struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1618         struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id];
1619         struct sock *sk = tinfo->sock->sk;
1620         struct ip_vs_sync_buff *sb;
1621 
1622         pr_info("sync thread started: state = MASTER, mcast_ifn = %s, "
1623                 "syncid = %d, id = %d\n",
1624                 ipvs->master_mcast_ifn, ipvs->master_syncid, tinfo->id);
1625 
1626         for (;;) {
1627                 sb = next_sync_buff(ipvs, ms);
1628                 if (unlikely(kthread_should_stop()))
1629                         break;
1630                 if (!sb) {
1631                         schedule_timeout(IPVS_SYNC_CHECK_PERIOD);
1632                         continue;
1633                 }
1634                 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) {
1635                         int ret = 0;
1636 
1637                         __wait_event_interruptible(*sk_sleep(sk),
1638                                                    sock_writeable(sk) ||
1639                                                    kthread_should_stop(),
1640                                                    ret);
1641                         if (unlikely(kthread_should_stop()))
1642                                 goto done;
1643                 }
1644                 ip_vs_sync_buff_release(sb);
1645         }
1646 
1647 done:
1648         __set_current_state(TASK_RUNNING);
1649         if (sb)
1650                 ip_vs_sync_buff_release(sb);
1651 
1652         /* clean up the sync_buff queue */
1653         while ((sb = sb_dequeue(ipvs, ms)))
1654                 ip_vs_sync_buff_release(sb);
1655         __set_current_state(TASK_RUNNING);
1656 
1657         /* clean up the current sync_buff */
1658         sb = get_curr_sync_buff(ipvs, ms, 0);
1659         if (sb)
1660                 ip_vs_sync_buff_release(sb);
1661 
1662         /* release the sending multicast socket */
1663         sk_release_kernel(tinfo->sock->sk);
1664         kfree(tinfo);
1665 
1666         return 0;
1667 }
1668 
1669 
1670 static int sync_thread_backup(void *data)
1671 {
1672         struct ip_vs_sync_thread_data *tinfo = data;
1673         struct netns_ipvs *ipvs = net_ipvs(tinfo->net);
1674         int len;
1675 
1676         pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, "
1677                 "syncid = %d, id = %d\n",
1678                 ipvs->backup_mcast_ifn, ipvs->backup_syncid, tinfo->id);
1679 
1680         while (!kthread_should_stop()) {
1681                 wait_event_interruptible(*sk_sleep(tinfo->sock->sk),
1682                          !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue)
1683                          || kthread_should_stop());
1684 
1685                 /* do we have data now? */
1686                 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) {
1687                         len = ip_vs_receive(tinfo->sock, tinfo->buf,
1688                                         ipvs->recv_mesg_maxlen);
1689                         if (len <= 0) {
1690                                 if (len != -EAGAIN)
1691                                         pr_err("receiving message error\n");
1692                                 break;
1693                         }
1694 
1695                         /* disable bottom half, because it accesses the data
1696                            shared by softirq while getting/creating conns */
1697                         local_bh_disable();
1698                         ip_vs_process_message(tinfo->net, tinfo->buf, len);
1699                         local_bh_enable();
1700                 }
1701         }
1702 
1703         /* release the sending multicast socket */
1704         sk_release_kernel(tinfo->sock->sk);
1705         kfree(tinfo->buf);
1706         kfree(tinfo);
1707 
1708         return 0;
1709 }
1710 
1711 
1712 int start_sync_thread(struct net *net, int state, char *mcast_ifn, __u8 syncid)
1713 {
1714         struct ip_vs_sync_thread_data *tinfo;
1715         struct task_struct **array = NULL, *task;
1716         struct socket *sock;
1717         struct netns_ipvs *ipvs = net_ipvs(net);
1718         char *name;
1719         int (*threadfn)(void *data);
1720         int id, count;
1721         int result = -ENOMEM;
1722 
1723         IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1724         IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n",
1725                   sizeof(struct ip_vs_sync_conn_v0));
1726 
1727         if (!ipvs->sync_state) {
1728                 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX);
1729                 ipvs->threads_mask = count - 1;
1730         } else
1731                 count = ipvs->threads_mask + 1;
1732 
1733         if (state == IP_VS_STATE_MASTER) {
1734                 if (ipvs->ms)
1735                         return -EEXIST;
1736 
1737                 strlcpy(ipvs->master_mcast_ifn, mcast_ifn,
1738                         sizeof(ipvs->master_mcast_ifn));
1739                 ipvs->master_syncid = syncid;
1740                 name = "ipvs-m:%d:%d";
1741                 threadfn = sync_thread_master;
1742         } else if (state == IP_VS_STATE_BACKUP) {
1743                 if (ipvs->backup_threads)
1744                         return -EEXIST;
1745 
1746                 strlcpy(ipvs->backup_mcast_ifn, mcast_ifn,
1747                         sizeof(ipvs->backup_mcast_ifn));
1748                 ipvs->backup_syncid = syncid;
1749                 name = "ipvs-b:%d:%d";
1750                 threadfn = sync_thread_backup;
1751         } else {
1752                 return -EINVAL;
1753         }
1754 
1755         if (state == IP_VS_STATE_MASTER) {
1756                 struct ipvs_master_sync_state *ms;
1757 
1758                 ipvs->ms = kzalloc(count * sizeof(ipvs->ms[0]), GFP_KERNEL);
1759                 if (!ipvs->ms)
1760                         goto out;
1761                 ms = ipvs->ms;
1762                 for (id = 0; id < count; id++, ms++) {
1763                         INIT_LIST_HEAD(&ms->sync_queue);
1764                         ms->sync_queue_len = 0;
1765                         ms->sync_queue_delay = 0;
1766                         INIT_DELAYED_WORK(&ms->master_wakeup_work,
1767                                           master_wakeup_work_handler);
1768                         ms->ipvs = ipvs;
1769                 }
1770         } else {
1771                 array = kzalloc(count * sizeof(struct task_struct *),
1772                                 GFP_KERNEL);
1773                 if (!array)
1774                         goto out;
1775         }
1776         set_sync_mesg_maxlen(net, state);
1777 
1778         tinfo = NULL;
1779         for (id = 0; id < count; id++) {
1780                 if (state == IP_VS_STATE_MASTER)
1781                         sock = make_send_sock(net, id);
1782                 else
1783                         sock = make_receive_sock(net, id);
1784                 if (IS_ERR(sock)) {
1785                         result = PTR_ERR(sock);
1786                         goto outtinfo;
1787                 }
1788                 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL);
1789                 if (!tinfo)
1790                         goto outsocket;
1791                 tinfo->net = net;
1792                 tinfo->sock = sock;
1793                 if (state == IP_VS_STATE_BACKUP) {
1794                         tinfo->buf = kmalloc(ipvs->recv_mesg_maxlen,
1795                                              GFP_KERNEL);
1796                         if (!tinfo->buf)
1797                                 goto outtinfo;
1798                 } else {
1799                         tinfo->buf = NULL;
1800                 }
1801                 tinfo->id = id;
1802 
1803                 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id);
1804                 if (IS_ERR(task)) {
1805                         result = PTR_ERR(task);
1806                         goto outtinfo;
1807                 }
1808                 tinfo = NULL;
1809                 if (state == IP_VS_STATE_MASTER)
1810                         ipvs->ms[id].master_thread = task;
1811                 else
1812                         array[id] = task;
1813         }
1814 
1815         /* mark as active */
1816 
1817         if (state == IP_VS_STATE_BACKUP)
1818                 ipvs->backup_threads = array;
1819         spin_lock_bh(&ipvs->sync_buff_lock);
1820         ipvs->sync_state |= state;
1821         spin_unlock_bh(&ipvs->sync_buff_lock);
1822 
1823         /* increase the module use count */
1824         ip_vs_use_count_inc();
1825 
1826         return 0;
1827 
1828 outsocket:
1829         sk_release_kernel(sock->sk);
1830 
1831 outtinfo:
1832         if (tinfo) {
1833                 sk_release_kernel(tinfo->sock->sk);
1834                 kfree(tinfo->buf);
1835                 kfree(tinfo);
1836         }
1837         count = id;
1838         while (count-- > 0) {
1839                 if (state == IP_VS_STATE_MASTER)
1840                         kthread_stop(ipvs->ms[count].master_thread);
1841                 else
1842                         kthread_stop(array[count]);
1843         }
1844         kfree(array);
1845 
1846 out:
1847         if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) {
1848                 kfree(ipvs->ms);
1849                 ipvs->ms = NULL;
1850         }
1851         return result;
1852 }
1853 
1854 
1855 int stop_sync_thread(struct net *net, int state)
1856 {
1857         struct netns_ipvs *ipvs = net_ipvs(net);
1858         struct task_struct **array;
1859         int id;
1860         int retc = -EINVAL;
1861 
1862         IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current));
1863 
1864         if (state == IP_VS_STATE_MASTER) {
1865                 if (!ipvs->ms)
1866                         return -ESRCH;
1867 
1868                 /*
1869                  * The lock synchronizes with sb_queue_tail(), so that we don't
1870                  * add sync buffers to the queue, when we are already in
1871                  * progress of stopping the master sync daemon.
1872                  */
1873 
1874                 spin_lock_bh(&ipvs->sync_buff_lock);
1875                 spin_lock(&ipvs->sync_lock);
1876                 ipvs->sync_state &= ~IP_VS_STATE_MASTER;
1877                 spin_unlock(&ipvs->sync_lock);
1878                 spin_unlock_bh(&ipvs->sync_buff_lock);
1879 
1880                 retc = 0;
1881                 for (id = ipvs->threads_mask; id >= 0; id--) {
1882                         struct ipvs_master_sync_state *ms = &ipvs->ms[id];
1883                         int ret;
1884 
1885                         pr_info("stopping master sync thread %d ...\n",
1886                                 task_pid_nr(ms->master_thread));
1887                         cancel_delayed_work_sync(&ms->master_wakeup_work);
1888                         ret = kthread_stop(ms->master_thread);
1889                         if (retc >= 0)
1890                                 retc = ret;
1891                 }
1892                 kfree(ipvs->ms);
1893                 ipvs->ms = NULL;
1894         } else if (state == IP_VS_STATE_BACKUP) {
1895                 if (!ipvs->backup_threads)
1896                         return -ESRCH;
1897 
1898                 ipvs->sync_state &= ~IP_VS_STATE_BACKUP;
1899                 array = ipvs->backup_threads;
1900                 retc = 0;
1901                 for (id = ipvs->threads_mask; id >= 0; id--) {
1902                         int ret;
1903 
1904                         pr_info("stopping backup sync thread %d ...\n",
1905                                 task_pid_nr(array[id]));
1906                         ret = kthread_stop(array[id]);
1907                         if (retc >= 0)
1908                                 retc = ret;
1909                 }
1910                 kfree(array);
1911                 ipvs->backup_threads = NULL;
1912         }
1913 
1914         /* decrease the module use count */
1915         ip_vs_use_count_dec();
1916 
1917         return retc;
1918 }
1919 
1920 /*
1921  * Initialize data struct for each netns
1922  */
1923 int __net_init ip_vs_sync_net_init(struct net *net)
1924 {
1925         struct netns_ipvs *ipvs = net_ipvs(net);
1926 
1927         __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key);
1928         spin_lock_init(&ipvs->sync_lock);
1929         spin_lock_init(&ipvs->sync_buff_lock);
1930         return 0;
1931 }
1932 
1933 void ip_vs_sync_net_cleanup(struct net *net)
1934 {
1935         int retc;
1936         struct netns_ipvs *ipvs = net_ipvs(net);
1937 
1938         mutex_lock(&ipvs->sync_mutex);
1939         retc = stop_sync_thread(net, IP_VS_STATE_MASTER);
1940         if (retc && retc != -ESRCH)
1941                 pr_err("Failed to stop Master Daemon\n");
1942 
1943         retc = stop_sync_thread(net, IP_VS_STATE_BACKUP);
1944         if (retc && retc != -ESRCH)
1945                 pr_err("Failed to stop Backup Daemon\n");
1946         mutex_unlock(&ipvs->sync_mutex);
1947 }
1948 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp