net/ipv4/ipvs/ip_vs_app.c

   1 /*
   2  * ip_vs_app.c: Application module support for IPVS
   3  *
   4  * Version:     $Id: ip_vs_app.c,v 1.17 2003/03/22 06:31:21 wensong Exp $
   5  *
   6  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   7  *
   8  *              This program is free software; you can redistribute it and/or
   9  *              modify it under the terms of the GNU General Public License
  10  *              as published by the Free Software Foundation; either version
  11  *              2 of the License, or (at your option) any later version.
  12  *
  13  * Most code here is taken from ip_masq_app.c in kernel 2.2. The difference
  14  * is that ip_vs_app module handles the reverse direction (incoming requests
  15  * and outgoing responses).
  16  *
  17  *              IP_MASQ_APP application masquerading module
  18  *
  19  * Author:      Juan Jose Ciarlante, <jjciarla@raiz.uncu.edu.ar>
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/kernel.h>
  25 #include <linux/skbuff.h>
  26 #include <linux/in.h>
  27 #include <linux/ip.h>
  28 #include <linux/netfilter.h>
  29 #include <net/net_namespace.h>
  30 #include <net/protocol.h>
  31 #include <net/tcp.h>
  32 #include <asm/system.h>
  33 #include <linux/stat.h>
  34 #include <linux/proc_fs.h>
  35 #include <linux/seq_file.h>
  36 #include <linux/mutex.h>
  37
  38 #include <net/ip_vs.h>
  39
  40 EXPORT_SYMBOL(register_ip_vs_app);
  41 EXPORT_SYMBOL(unregister_ip_vs_app);
  42 EXPORT_SYMBOL(register_ip_vs_app_inc);
  43
  44 /* ipvs application list head */
  45 static LIST_HEAD(ip_vs_app_list);
  46 static DEFINE_MUTEX(__ip_vs_app_mutex);
  47
  48
  49 /*
  50  *      Get an ip_vs_app object
  51  */
  52 static inline int ip_vs_app_get(struct ip_vs_app *app)
  53 {
  54         /* test and get the module atomically */
  55         if (app->module)
  56                 return try_module_get(app->module);
  57         else
  58                 return 1;
  59 }
  60
  61
  62 static inline void ip_vs_app_put(struct ip_vs_app *app)
  63 {
  64         if (app->module)
  65                 module_put(app->module);
  66 }
  67
  68
  69 /*
  70  *      Allocate/initialize app incarnation and register it in proto apps.
  71  */
  72 static int
  73 ip_vs_app_inc_new(struct ip_vs_app *app, __u16 proto, __u16 port)
  74 {
  75         struct ip_vs_protocol *pp;
  76         struct ip_vs_app *inc;
  77         int ret;
  78
  79         if (!(pp = ip_vs_proto_get(proto)))
  80                 return -EPROTONOSUPPORT;
  81
  82         if (!pp->unregister_app)
  83                 return -EOPNOTSUPP;
  84
  85         inc = kmemdup(app, sizeof(*inc), GFP_KERNEL);
  86         if (!inc)
  87                 return -ENOMEM;
  88         INIT_LIST_HEAD(&inc->p_list);
  89         INIT_LIST_HEAD(&inc->incs_list);
  90         inc->app = app;
  91         inc->port = htons(port);
  92         atomic_set(&inc->usecnt, 0);
  93
  94         if (app->timeouts) {
  95                 inc->timeout_table =
  96                         ip_vs_create_timeout_table(app->timeouts,
  97                                                    app->timeouts_size);
  98                 if (!inc->timeout_table) {
  99                         ret = -ENOMEM;
 100                         goto out;
 101                 }
 102         }
 103
 104         ret = pp->register_app(inc);
 105         if (ret)
 106                 goto out;
 107
 108         list_add(&inc->a_list, &app->incs_list);
 109         IP_VS_DBG(9, "%s application %s:%u registered\n",
 110                   pp->name, inc->name, inc->port);
 111
 112         return 0;
 113
 114   out:
 115         kfree(inc->timeout_table);
 116         kfree(inc);
 117         return ret;
 118 }
 119
 120
 121 /*
 122  *      Release app incarnation
 123  */
 124 static void
 125 ip_vs_app_inc_release(struct ip_vs_app *inc)
 126 {
 127         struct ip_vs_protocol *pp;
 128
 129         if (!(pp = ip_vs_proto_get(inc->protocol)))
 130                 return;
 131
 132         if (pp->unregister_app)
 133                 pp->unregister_app(inc);
 134
 135         IP_VS_DBG(9, "%s App %s:%u unregistered\n",
 136                   pp->name, inc->name, inc->port);
 137
 138         list_del(&inc->a_list);
 139
 140         kfree(inc->timeout_table);
 141         kfree(inc);
 142 }
 143
 144
 145 /*
 146  *      Get reference to app inc (only called from softirq)
 147  *
 148  */
 149 int ip_vs_app_inc_get(struct ip_vs_app *inc)
 150 {
 151         int result;
 152
 153         atomic_inc(&inc->usecnt);
 154         if (unlikely((result = ip_vs_app_get(inc->app)) != 1))
 155                 atomic_dec(&inc->usecnt);
 156         return result;
 157 }
 158
 159
 160 /*
 161  *      Put the app inc (only called from timer or net softirq)
 162  */
 163 void ip_vs_app_inc_put(struct ip_vs_app *inc)
 164 {
 165         ip_vs_app_put(inc->app);
 166         atomic_dec(&inc->usecnt);
 167 }
 168
 169
 170 /*
 171  *      Register an application incarnation in protocol applications
 172  */
 173 int
 174 register_ip_vs_app_inc(struct ip_vs_app *app, __u16 proto, __u16 port)
 175 {
 176         int result;
 177
 178         mutex_lock(&__ip_vs_app_mutex);
 179
 180         result = ip_vs_app_inc_new(app, proto, port);
 181
 182         mutex_unlock(&__ip_vs_app_mutex);
 183
 184         return result;
 185 }
 186
 187
 188 /*
 189  *      ip_vs_app registration routine
 190  */
 191 int register_ip_vs_app(struct ip_vs_app *app)
 192 {
 193         /* increase the module use count */
 194         ip_vs_use_count_inc();
 195
 196         mutex_lock(&__ip_vs_app_mutex);
 197
 198         list_add(&app->a_list, &ip_vs_app_list);
 199
 200         mutex_unlock(&__ip_vs_app_mutex);
 201
 202         return 0;
 203 }
 204
 205
 206 /*
 207  *      ip_vs_app unregistration routine
 208  *      We are sure there are no app incarnations attached to services
 209  */
 210 void unregister_ip_vs_app(struct ip_vs_app *app)
 211 {
 212         struct ip_vs_app *inc, *nxt;
 213
 214         mutex_lock(&__ip_vs_app_mutex);
 215
 216         list_for_each_entry_safe(inc, nxt, &app->incs_list, a_list) {
 217                 ip_vs_app_inc_release(inc);
 218         }
 219
 220         list_del(&app->a_list);
 221
 222         mutex_unlock(&__ip_vs_app_mutex);
 223
 224         /* decrease the module use count */
 225         ip_vs_use_count_dec();
 226 }
 227
 228
 229 /*
 230  *      Bind ip_vs_conn to its ip_vs_app (called by cp constructor)
 231  */
 232 int ip_vs_bind_app(struct ip_vs_conn *cp, struct ip_vs_protocol *pp)
 233 {
 234         return pp->app_conn_bind(cp);
 235 }
 236
 237
 238 /*
 239  *      Unbind cp from application incarnation (called by cp destructor)
 240  */
 241 void ip_vs_unbind_app(struct ip_vs_conn *cp)
 242 {
 243         struct ip_vs_app *inc = cp->app;
 244
 245         if (!inc)
 246                 return;
 247
 248         if (inc->unbind_conn)
 249                 inc->unbind_conn(inc, cp);
 250         if (inc->done_conn)
 251                 inc->done_conn(inc, cp);
 252         ip_vs_app_inc_put(inc);
 253         cp->app = NULL;
 254 }
 255
 256
 257 /*
 258  *      Fixes th->seq based on ip_vs_seq info.
 259  */
 260 static inline void vs_fix_seq(const struct ip_vs_seq *vseq, struct tcphdr *th)
 261 {
 262         __u32 seq = ntohl(th->seq);
 263
 264         /*
 265          *      Adjust seq with delta-offset for all packets after
 266          *      the most recent resized pkt seq and with previous_delta offset
 267          *      for all packets before most recent resized pkt seq.
 268          */
 269         if (vseq->delta || vseq->previous_delta) {
 270                 if(after(seq, vseq->init_seq)) {
 271                         th->seq = htonl(seq + vseq->delta);
 272                         IP_VS_DBG(9, "vs_fix_seq(): added delta (%d) to seq\n",
 273                                   vseq->delta);
 274                 } else {
 275                         th->seq = htonl(seq + vseq->previous_delta);
 276                         IP_VS_DBG(9, "vs_fix_seq(): added previous_delta "
 277                                   "(%d) to seq\n", vseq->previous_delta);
 278                 }
 279         }
 280 }
 281
 282
 283 /*
 284  *      Fixes th->ack_seq based on ip_vs_seq info.
 285  */
 286 static inline void
 287 vs_fix_ack_seq(const struct ip_vs_seq *vseq, struct tcphdr *th)
 288 {
 289         __u32 ack_seq = ntohl(th->ack_seq);
 290
 291         /*
 292          * Adjust ack_seq with delta-offset for
 293          * the packets AFTER most recent resized pkt has caused a shift
 294          * for packets before most recent resized pkt, use previous_delta
 295          */
 296         if (vseq->delta || vseq->previous_delta) {
 297                 /* since ack_seq is the number of octet that is expected
 298                    to receive next, so compare it with init_seq+delta */
 299                 if(after(ack_seq, vseq->init_seq+vseq->delta)) {
 300                         th->ack_seq = htonl(ack_seq - vseq->delta);
 301                         IP_VS_DBG(9, "vs_fix_ack_seq(): subtracted delta "
 302                                   "(%d) from ack_seq\n", vseq->delta);
 303
 304                 } else {
 305                         th->ack_seq = htonl(ack_seq - vseq->previous_delta);
 306                         IP_VS_DBG(9, "vs_fix_ack_seq(): subtracted "
 307                                   "previous_delta (%d) from ack_seq\n",
 308                                   vseq->previous_delta);
 309                 }
 310         }
 311 }
 312
 313
 314 /*
 315  *      Updates ip_vs_seq if pkt has been resized
 316  *      Assumes already checked proto==IPPROTO_TCP and diff!=0.
 317  */
 318 static inline void vs_seq_update(struct ip_vs_conn *cp, struct ip_vs_seq *vseq,
 319                                  unsigned flag, __u32 seq, int diff)
 320 {
 321         /* spinlock is to keep updating cp->flags atomic */
 322         spin_lock(&cp->lock);
 323         if (!(cp->flags & flag) || after(seq, vseq->init_seq)) {
 324                 vseq->previous_delta = vseq->delta;
 325                 vseq->delta += diff;
 326                 vseq->init_seq = seq;
 327                 cp->flags |= flag;
 328         }
 329         spin_unlock(&cp->lock);
 330 }
 331
 332 static inline int app_tcp_pkt_out(struct ip_vs_conn *cp, struct sk_buff *skb,
 333                                   struct ip_vs_app *app)
 334 {
 335         int diff;
 336         const unsigned int tcp_offset = ip_hdrlen(skb);
 337         struct tcphdr *th;
 338         __u32 seq;
 339
 340         if (!skb_make_writable(skb, tcp_offset + sizeof(*th)))
 341                 return 0;
 342
 343         th = (struct tcphdr *)(skb_network_header(skb) + tcp_offset);
 344
 345         /*
 346          *      Remember seq number in case this pkt gets resized
 347          */
 348         seq = ntohl(th->seq);
 349
 350         /*
 351          *      Fix seq stuff if flagged as so.
 352          */
 353         if (cp->flags & IP_VS_CONN_F_OUT_SEQ)
 354                 vs_fix_seq(&cp->out_seq, th);
 355         if (cp->flags & IP_VS_CONN_F_IN_SEQ)
 356                 vs_fix_ack_seq(&cp->in_seq, th);
 357
 358         /*
 359          *      Call private output hook function
 360          */
 361         if (app->pkt_out == NULL)
 362                 return 1;
 363
 364         if (!app->pkt_out(app, cp, skb, &diff))
 365                 return 0;
 366
 367         /*
 368          *      Update ip_vs seq stuff if len has changed.
 369          */
 370         if (diff != 0)
 371                 vs_seq_update(cp, &cp->out_seq,
 372                               IP_VS_CONN_F_OUT_SEQ, seq, diff);
 373
 374         return 1;
 375 }
 376
 377 /*
 378  *      Output pkt hook. Will call bound ip_vs_app specific function
 379  *      called by ipvs packet handler, assumes previously checked cp!=NULL
 380  *      returns false if it can't handle packet (oom)
 381  */
 382 int ip_vs_app_pkt_out(struct ip_vs_conn *cp, struct sk_buff *skb)
 383 {
 384         struct ip_vs_app *app;
 385
 386         /*
 387          *      check if application module is bound to
 388          *      this ip_vs_conn.
 389          */
 390         if ((app = cp->app) == NULL)
 391                 return 1;
 392
 393         /* TCP is complicated */
 394         if (cp->protocol == IPPROTO_TCP)
 395                 return app_tcp_pkt_out(cp, skb, app);
 396
 397         /*
 398          *      Call private output hook function
 399          */
 400         if (app->pkt_out == NULL)
 401                 return 1;
 402
 403         return app->pkt_out(app, cp, skb, NULL);
 404 }
 405
 406
 407 static inline int app_tcp_pkt_in(struct ip_vs_conn *cp, struct sk_buff *skb,
 408                                  struct ip_vs_app *app)
 409 {
 410         int diff;
 411         const unsigned int tcp_offset = ip_hdrlen(skb);
 412         struct tcphdr *th;
 413         __u32 seq;
 414
 415         if (!skb_make_writable(skb, tcp_offset + sizeof(*th)))
 416                 return 0;
 417
 418         th = (struct tcphdr *)(skb_network_header(skb) + tcp_offset);
 419
 420         /*
 421          *      Remember seq number in case this pkt gets resized
 422          */
 423         seq = ntohl(th->seq);
 424
 425         /*
 426          *      Fix seq stuff if flagged as so.
 427          */
 428         if (cp->flags & IP_VS_CONN_F_IN_SEQ)
 429                 vs_fix_seq(&cp->in_seq, th);
 430         if (cp->flags & IP_VS_CONN_F_OUT_SEQ)
 431                 vs_fix_ack_seq(&cp->out_seq, th);
 432
 433         /*
 434          *      Call private input hook function
 435          */
 436         if (app->pkt_in == NULL)
 437                 return 1;
 438
 439         if (!app->pkt_in(app, cp, skb, &diff))
 440                 return 0;
 441
 442         /*
 443          *      Update ip_vs seq stuff if len has changed.
 444          */
 445         if (diff != 0)
 446                 vs_seq_update(cp, &cp->in_seq,
 447                               IP_VS_CONN_F_IN_SEQ, seq, diff);
 448
 449         return 1;
 450 }
 451
 452 /*
 453  *      Input pkt hook. Will call bound ip_vs_app specific function
 454  *      called by ipvs packet handler, assumes previously checked cp!=NULL.
 455  *      returns false if can't handle packet (oom).
 456  */
 457 int ip_vs_app_pkt_in(struct ip_vs_conn *cp, struct sk_buff *skb)
 458 {
 459         struct ip_vs_app *app;
 460
 461         /*
 462          *      check if application module is bound to
 463          *      this ip_vs_conn.
 464          */
 465         if ((app = cp->app) == NULL)
 466                 return 1;
 467
 468         /* TCP is complicated */
 469         if (cp->protocol == IPPROTO_TCP)
 470                 return app_tcp_pkt_in(cp, skb, app);
 471
 472         /*
 473          *      Call private input hook function
 474          */
 475         if (app->pkt_in == NULL)
 476                 return 1;
 477
 478         return app->pkt_in(app, cp, skb, NULL);
 479 }
 480
 481
 482 #ifdef CONFIG_PROC_FS
 483 /*
 484  *      /proc/net/ip_vs_app entry function
 485  */
 486
 487 static struct ip_vs_app *ip_vs_app_idx(loff_t pos)
 488 {
 489         struct ip_vs_app *app, *inc;
 490
 491         list_for_each_entry(app, &ip_vs_app_list, a_list) {
 492                 list_for_each_entry(inc, &app->incs_list, a_list) {
 493                         if (pos-- == 0)
 494                                 return inc;
 495                 }
 496         }
 497         return NULL;
 498
 499 }
 500
 501 static void *ip_vs_app_seq_start(struct seq_file *seq, loff_t *pos)
 502 {
 503         mutex_lock(&__ip_vs_app_mutex);
 504
 505         return *pos ? ip_vs_app_idx(*pos - 1) : SEQ_START_TOKEN;
 506 }
 507
 508 static void *ip_vs_app_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 509 {
 510         struct ip_vs_app *inc, *app;
 511         struct list_head *e;
 512
 513         ++*pos;
 514         if (v == SEQ_START_TOKEN)
 515                 return ip_vs_app_idx(0);
 516
 517         inc = v;
 518         app = inc->app;
 519
 520         if ((e = inc->a_list.next) != &app->incs_list)
 521                 return list_entry(e, struct ip_vs_app, a_list);
 522
 523         /* go on to next application */
 524         for (e = app->a_list.next; e != &ip_vs_app_list; e = e->next) {
 525                 app = list_entry(e, struct ip_vs_app, a_list);
 526                 list_for_each_entry(inc, &app->incs_list, a_list) {
 527                         return inc;
 528                 }
 529         }
 530         return NULL;
 531 }
 532
 533 static void ip_vs_app_seq_stop(struct seq_file *seq, void *v)
 534 {
 535         mutex_unlock(&__ip_vs_app_mutex);
 536 }
 537
 538 static int ip_vs_app_seq_show(struct seq_file *seq, void *v)
 539 {
 540         if (v == SEQ_START_TOKEN)
 541                 seq_puts(seq, "prot port    usecnt name\n");
 542         else {
 543                 const struct ip_vs_app *inc = v;
 544
 545                 seq_printf(seq, "%-3s  %-7u %-6d %-17s\n",
 546                            ip_vs_proto_name(inc->protocol),
 547                            ntohs(inc->port),
 548                            atomic_read(&inc->usecnt),
 549                            inc->name);
 550         }
 551         return 0;
 552 }
 553
 554 static const struct seq_operations ip_vs_app_seq_ops = {
 555         .start = ip_vs_app_seq_start,
 556         .next  = ip_vs_app_seq_next,
 557         .stop  = ip_vs_app_seq_stop,
 558         .show  = ip_vs_app_seq_show,
 559 };
 560
 561 static int ip_vs_app_open(struct inode *inode, struct file *file)
 562 {
 563         return seq_open(file, &ip_vs_app_seq_ops);
 564 }
 565
 566 static const struct file_operations ip_vs_app_fops = {
 567         .owner   = THIS_MODULE,
 568         .open    = ip_vs_app_open,
 569         .read    = seq_read,
 570         .llseek  = seq_lseek,
 571         .release = seq_release,
 572 };
 573 #endif
 574
 575
 576 /*
 577  *      Replace a segment of data with a new segment
 578  */
 579 int ip_vs_skb_replace(struct sk_buff *skb, gfp_t pri,
 580                       char *o_buf, int o_len, char *n_buf, int n_len)
 581 {
 582         int diff;
 583         int o_offset;
 584         int o_left;
 585
 586         EnterFunction(9);
 587
 588         diff = n_len - o_len;
 589         o_offset = o_buf - (char *)skb->data;
 590         /* The length of left data after o_buf+o_len in the skb data */
 591         o_left = skb->len - (o_offset + o_len);
 592
 593         if (diff <= 0) {
 594                 memmove(o_buf + n_len, o_buf + o_len, o_left);
 595                 memcpy(o_buf, n_buf, n_len);
 596                 skb_trim(skb, skb->len + diff);
 597         } else if (diff <= skb_tailroom(skb)) {
 598                 skb_put(skb, diff);
 599                 memmove(o_buf + n_len, o_buf + o_len, o_left);
 600                 memcpy(o_buf, n_buf, n_len);
 601         } else {
 602                 if (pskb_expand_head(skb, skb_headroom(skb), diff, pri))
 603                         return -ENOMEM;
 604                 skb_put(skb, diff);
 605                 memmove(skb->data + o_offset + n_len,
 606                         skb->data + o_offset + o_len, o_left);
 607                 skb_copy_to_linear_data_offset(skb, o_offset, n_buf, n_len);
 608         }
 609
 610         /* must update the iph total length here */
 611         ip_hdr(skb)->tot_len = htons(skb->len);
 612
 613         LeaveFunction(9);
 614         return 0;
 615 }
 616
 617
 618 int ip_vs_app_init(void)
 619 {
 620         /* we will replace it with proc_net_ipvs_create() soon */
 621         proc_net_fops_create(&init_net, "ip_vs_app", 0, &ip_vs_app_fops);
 622         return 0;
 623 }
 624
 625
 626 void ip_vs_app_cleanup(void)
 627 {
 628         proc_net_remove(&init_net, "ip_vs_app");
 629 }