net/netfilter/xt_qtaguid.c

   1 /*
   2  * Kernel iptables module to track stats for packets based on user tags.
   3  *
   4  * (C) 2011 Google, Inc
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 /*
  12  * There are run-time debug flags enabled via the debug_mask module param, or
  13  * via the DEFAULT_DEBUG_MASK. See xt_qtaguid_internal.h.
  14  */
  15 #define DEBUG
  16
  17 #include <linux/file.h>
  18 #include <linux/inetdevice.h>
  19 #include <linux/module.h>
  20 #include <linux/netfilter/x_tables.h>
  21 #include <linux/netfilter/xt_qtaguid.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/workqueue.h>
  24 #include <net/addrconf.h>
  25 #include <net/sock.h>
  26 #include <net/tcp.h>
  27 #include <net/udp.h>
  28
  29 #include <linux/netfilter/xt_socket.h>
  30 #include "xt_qtaguid_internal.h"
  31 #include "xt_qtaguid_print.h"
  32
  33 /*
  34  * We only use the xt_socket funcs within a similar context to avoid unexpected
  35  * return values.
  36  */
  37 #define XT_SOCKET_SUPPORTED_HOOKS \
  38         ((1 << NF_INET_PRE_ROUTING) | (1 << NF_INET_LOCAL_IN))
  39
  40
  41 static const char *module_procdirname = "xt_qtaguid";
  42 static struct proc_dir_entry *xt_qtaguid_procdir;
  43
  44 static unsigned int proc_iface_perms = S_IRUGO;
  45 module_param_named(iface_perms, proc_iface_perms, uint, S_IRUGO | S_IWUSR);
  46
  47 static struct proc_dir_entry *xt_qtaguid_stats_file;
  48 static unsigned int proc_stats_perms = S_IRUGO;
  49 module_param_named(stats_perms, proc_stats_perms, uint, S_IRUGO | S_IWUSR);
  50
  51 static struct proc_dir_entry *xt_qtaguid_ctrl_file;
  52 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
  53 static unsigned int proc_ctrl_perms = S_IRUGO | S_IWUGO;
  54 #else
  55 static unsigned int proc_ctrl_perms = S_IRUGO | S_IWUSR;
  56 #endif
  57 module_param_named(ctrl_perms, proc_ctrl_perms, uint, S_IRUGO | S_IWUSR);
  58
  59 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
  60 #include <linux/android_aid.h>
  61 static gid_t proc_stats_readall_gid = AID_NET_BW_STATS;
  62 static gid_t proc_ctrl_write_gid = AID_NET_BW_ACCT;
  63 #else
  64 /* 0 means, don't limit anybody */
  65 static gid_t proc_stats_readall_gid;
  66 static gid_t proc_ctrl_write_gid;
  67 #endif
  68 module_param_named(stats_readall_gid, proc_stats_readall_gid, uint,
  69                    S_IRUGO | S_IWUSR);
  70 module_param_named(ctrl_write_gid, proc_ctrl_write_gid, uint,
  71                    S_IRUGO | S_IWUSR);
  72
  73 /*
  74  * Limit the number of active tags (via socket tags) for a given UID.
  75  * Multiple processes could share the UID.
  76  */
  77 static int max_sock_tags = DEFAULT_MAX_SOCK_TAGS;
  78 module_param(max_sock_tags, int, S_IRUGO | S_IWUSR);
  79
  80 /*
  81  * After the kernel has initiallized this module, it is still possible
  82  * to make it passive.
  83  * Setting passive to Y:
  84  *  - the iface stats handling will not act on notifications.
  85  *  - iptables matches will never match.
  86  *  - ctrl commands silently succeed.
  87  *  - stats are always empty.
  88  * This is mostly usefull when a bug is suspected.
  89  */
  90 static bool module_passive;
  91 module_param_named(passive, module_passive, bool, S_IRUGO | S_IWUSR);
  92
  93 /*
  94  * Control how qtaguid data is tracked per proc/uid.
  95  * Setting tag_tracking_passive to Y:
  96  *  - don't create proc specific structs to track tags
  97  *  - don't check that active tag stats exceed some limits.
  98  *  - don't clean up socket tags on process exits.
  99  * This is mostly usefull when a bug is suspected.
 100  */
 101 static bool qtu_proc_handling_passive;
 102 module_param_named(tag_tracking_passive, qtu_proc_handling_passive, bool,
 103                    S_IRUGO | S_IWUSR);
 104
 105 #define QTU_DEV_NAME "xt_qtaguid"
 106
 107 uint qtaguid_debug_mask = DEFAULT_DEBUG_MASK;
 108 module_param_named(debug_mask, qtaguid_debug_mask, uint, S_IRUGO | S_IWUSR);
 109
 110 /*---------------------------------------------------------------------------*/
 111 static const char *iface_stat_procdirname = "iface_stat";
 112 static struct proc_dir_entry *iface_stat_procdir;
 113 static const char *iface_stat_all_procfilename = "iface_stat_all";
 114 static struct proc_dir_entry *iface_stat_all_procfile;
 115
 116 /*
 117  * Ordering of locks:
 118  *  outer locks:
 119  *    iface_stat_list_lock
 120  *    sock_tag_list_lock
 121  *  inner locks:
 122  *    uid_tag_data_tree_lock
 123  *    tag_counter_set_list_lock
 124  * Notice how sock_tag_list_lock is held sometimes when uid_tag_data_tree_lock
 125  * is acquired.
 126  *
 127  * Call tree with all lock holders as of 2011-09-25:
 128  *
 129  * iface_stat_all_proc_read()
 130  *   iface_stat_list_lock
 131  *     (struct iface_stat)
 132  *
 133  * qtaguid_ctrl_proc_read()
 134  *   sock_tag_list_lock
 135  *     (sock_tag_tree)
 136  *     (struct proc_qtu_data->sock_tag_list)
 137  *   prdebug_full_state()
 138  *     sock_tag_list_lock
 139  *       (sock_tag_tree)
 140  *     uid_tag_data_tree_lock
 141  *       (uid_tag_data_tree)
 142  *       (proc_qtu_data_tree)
 143  *     iface_stat_list_lock
 144  *
 145  * qtaguid_stats_proc_read()
 146  *   iface_stat_list_lock
 147  *     struct iface_stat->tag_stat_list_lock
 148  *
 149  * qtudev_open()
 150  *   uid_tag_data_tree_lock
 151  *
 152  * qtudev_release()
 153  *   sock_tag_data_list_lock
 154  *     uid_tag_data_tree_lock
 155  *   prdebug_full_state()
 156  *     sock_tag_list_lock
 157  *     uid_tag_data_tree_lock
 158  *     iface_stat_list_lock
 159  *
 160  * iface_netdev_event_handler()
 161  *   iface_stat_create()
 162  *     iface_stat_list_lock
 163  *   iface_stat_update()
 164  *     iface_stat_list_lock
 165  *
 166  * iface_inetaddr_event_handler()
 167  *   iface_stat_create()
 168  *     iface_stat_list_lock
 169  *   iface_stat_update()
 170  *     iface_stat_list_lock
 171  *
 172  * iface_inet6addr_event_handler()
 173  *   iface_stat_create_ipv6()
 174  *     iface_stat_list_lock
 175  *   iface_stat_update()
 176  *     iface_stat_list_lock
 177  *
 178  * qtaguid_mt()
 179  *   account_for_uid()
 180  *     if_tag_stat_update()
 181  *       get_sock_stat()
 182  *         sock_tag_list_lock
 183  *       struct iface_stat->tag_stat_list_lock
 184  *         tag_stat_update()
 185  *           get_active_counter_set()
 186  *             tag_counter_set_list_lock
 187  *         tag_stat_update()
 188  *           get_active_counter_set()
 189  *             tag_counter_set_list_lock
 190  *
 191  *
 192  * qtaguid_ctrl_parse()
 193  *   ctrl_cmd_delete()
 194  *     sock_tag_list_lock
 195  *     tag_counter_set_list_lock
 196  *     iface_stat_list_lock
 197  *       struct iface_stat->tag_stat_list_lock
 198  *     uid_tag_data_tree_lock
 199  *   ctrl_cmd_counter_set()
 200  *     tag_counter_set_list_lock
 201  *   ctrl_cmd_tag()
 202  *     sock_tag_list_lock
 203  *       (sock_tag_tree)
 204  *       get_tag_ref()
 205  *         uid_tag_data_tree_lock
 206  *           (uid_tag_data_tree)
 207  *       uid_tag_data_tree_lock
 208  *         (proc_qtu_data_tree)
 209  *   ctrl_cmd_untag()
 210  *     sock_tag_list_lock
 211  *     uid_tag_data_tree_lock
 212  *
 213  */
 214 static LIST_HEAD(iface_stat_list);
 215 static DEFINE_SPINLOCK(iface_stat_list_lock);
 216
 217 static struct rb_root sock_tag_tree = RB_ROOT;
 218 static DEFINE_SPINLOCK(sock_tag_list_lock);
 219
 220 static struct rb_root tag_counter_set_tree = RB_ROOT;
 221 static DEFINE_SPINLOCK(tag_counter_set_list_lock);
 222
 223 static struct rb_root uid_tag_data_tree = RB_ROOT;
 224 static DEFINE_SPINLOCK(uid_tag_data_tree_lock);
 225
 226 static struct rb_root proc_qtu_data_tree = RB_ROOT;
 227 /* No proc_qtu_data_tree_lock; use uid_tag_data_tree_lock */
 228
 229 static struct qtaguid_event_counts qtu_events;
 230 /*----------------------------------------------*/
 231 static bool can_manipulate_uids(void)
 232 {
 233         /* root pwnd */
 234         return unlikely(!current_fsuid()) || unlikely(!proc_ctrl_write_gid)
 235                 || in_egroup_p(proc_ctrl_write_gid);
 236 }
 237
 238 static bool can_impersonate_uid(uid_t uid)
 239 {
 240         return uid == current_fsuid() || can_manipulate_uids();
 241 }
 242
 243 static bool can_read_other_uid_stats(uid_t uid)
 244 {
 245         /* root pwnd */
 246         return unlikely(!current_fsuid()) || uid == current_fsuid()
 247                 || unlikely(!proc_stats_readall_gid)
 248                 || in_egroup_p(proc_stats_readall_gid);
 249 }
 250
 251 static inline void dc_add_byte_packets(struct data_counters *counters, int set,
 252                                   enum ifs_tx_rx direction,
 253                                   enum ifs_proto ifs_proto,
 254                                   int bytes,
 255                                   int packets)
 256 {
 257         counters->bpc[set][direction][ifs_proto].bytes += bytes;
 258         counters->bpc[set][direction][ifs_proto].packets += packets;
 259 }
 260
 261 static inline uint64_t dc_sum_bytes(struct data_counters *counters,
 262                                     int set,
 263                                     enum ifs_tx_rx direction)
 264 {
 265         return counters->bpc[set][direction][IFS_TCP].bytes
 266                 + counters->bpc[set][direction][IFS_UDP].bytes
 267                 + counters->bpc[set][direction][IFS_PROTO_OTHER].bytes;
 268 }
 269
 270 static inline uint64_t dc_sum_packets(struct data_counters *counters,
 271                                       int set,
 272                                       enum ifs_tx_rx direction)
 273 {
 274         return counters->bpc[set][direction][IFS_TCP].packets
 275                 + counters->bpc[set][direction][IFS_UDP].packets
 276                 + counters->bpc[set][direction][IFS_PROTO_OTHER].packets;
 277 }
 278
 279 static struct tag_node *tag_node_tree_search(struct rb_root *root, tag_t tag)
 280 {
 281         struct rb_node *node = root->rb_node;
 282
 283         while (node) {
 284                 struct tag_node *data = rb_entry(node, struct tag_node, node);
 285                 int result;
 286                 RB_DEBUG("qtaguid: tag_node_tree_search(0x%llx): "
 287                          " node=%p data=%p\n", tag, node, data);
 288                 result = tag_compare(tag, data->tag);
 289                 RB_DEBUG("qtaguid: tag_node_tree_search(0x%llx): "
 290                          " data.tag=0x%llx (uid=%u) res=%d\n",
 291                          tag, data->tag, get_uid_from_tag(data->tag), result);
 292                 if (result < 0)
 293                         node = node->rb_left;
 294                 else if (result > 0)
 295                         node = node->rb_right;
 296                 else
 297                         return data;
 298         }
 299         return NULL;
 300 }
 301
 302 static void tag_node_tree_insert(struct tag_node *data, struct rb_root *root)
 303 {
 304         struct rb_node **new = &(root->rb_node), *parent = NULL;
 305
 306         /* Figure out where to put new node */
 307         while (*new) {
 308                 struct tag_node *this = rb_entry(*new, struct tag_node,
 309                                                  node);
 310                 int result = tag_compare(data->tag, this->tag);
 311                 RB_DEBUG("qtaguid: %s(): tag=0x%llx"
 312                          " (uid=%u)\n", __func__,
 313                          this->tag,
 314                          get_uid_from_tag(this->tag));
 315                 parent = *new;
 316                 if (result < 0)
 317                         new = &((*new)->rb_left);
 318                 else if (result > 0)
 319                         new = &((*new)->rb_right);
 320                 else
 321                         BUG();
 322         }
 323
 324         /* Add new node and rebalance tree. */
 325         rb_link_node(&data->node, parent, new);
 326         rb_insert_color(&data->node, root);
 327 }
 328
 329 static void tag_stat_tree_insert(struct tag_stat *data, struct rb_root *root)
 330 {
 331         tag_node_tree_insert(&data->tn, root);
 332 }
 333
 334 static struct tag_stat *tag_stat_tree_search(struct rb_root *root, tag_t tag)
 335 {
 336         struct tag_node *node = tag_node_tree_search(root, tag);
 337         if (!node)
 338                 return NULL;
 339         return rb_entry(&node->node, struct tag_stat, tn.node);
 340 }
 341
 342 static void tag_counter_set_tree_insert(struct tag_counter_set *data,
 343                                         struct rb_root *root)
 344 {
 345         tag_node_tree_insert(&data->tn, root);
 346 }
 347
 348 static struct tag_counter_set *tag_counter_set_tree_search(struct rb_root *root,
 349                                                            tag_t tag)
 350 {
 351         struct tag_node *node = tag_node_tree_search(root, tag);
 352         if (!node)
 353                 return NULL;
 354         return rb_entry(&node->node, struct tag_counter_set, tn.node);
 355
 356 }
 357
 358 static void tag_ref_tree_insert(struct tag_ref *data, struct rb_root *root)
 359 {
 360         tag_node_tree_insert(&data->tn, root);
 361 }
 362
 363 static struct tag_ref *tag_ref_tree_search(struct rb_root *root, tag_t tag)
 364 {
 365         struct tag_node *node = tag_node_tree_search(root, tag);
 366         if (!node)
 367                 return NULL;
 368         return rb_entry(&node->node, struct tag_ref, tn.node);
 369 }
 370
 371 static struct sock_tag *sock_tag_tree_search(struct rb_root *root,
 372                                              const struct sock *sk)
 373 {
 374         struct rb_node *node = root->rb_node;
 375
 376         while (node) {
 377                 struct sock_tag *data = rb_entry(node, struct sock_tag,
 378                                                  sock_node);
 379                 if (sk < data->sk)
 380                         node = node->rb_left;
 381                 else if (sk > data->sk)
 382                         node = node->rb_right;
 383                 else
 384                         return data;
 385         }
 386         return NULL;
 387 }
 388
 389 static void sock_tag_tree_insert(struct sock_tag *data, struct rb_root *root)
 390 {
 391         struct rb_node **new = &(root->rb_node), *parent = NULL;
 392
 393         /* Figure out where to put new node */
 394         while (*new) {
 395                 struct sock_tag *this = rb_entry(*new, struct sock_tag,
 396                                                  sock_node);
 397                 parent = *new;
 398                 if (data->sk < this->sk)
 399                         new = &((*new)->rb_left);
 400                 else if (data->sk > this->sk)
 401                         new = &((*new)->rb_right);
 402                 else
 403                         BUG();
 404         }
 405
 406         /* Add new node and rebalance tree. */
 407         rb_link_node(&data->sock_node, parent, new);
 408         rb_insert_color(&data->sock_node, root);
 409 }
 410
 411 static void sock_tag_tree_erase(struct rb_root *st_to_free_tree)
 412 {
 413         struct rb_node *node;
 414         struct sock_tag *st_entry;
 415
 416         node = rb_first(st_to_free_tree);
 417         while (node) {
 418                 st_entry = rb_entry(node, struct sock_tag, sock_node);
 419                 node = rb_next(node);
 420                 CT_DEBUG("qtaguid: %s(): "
 421                          "erase st: sk=%p tag=0x%llx (uid=%u)\n", __func__,
 422                          st_entry->sk,
 423                          st_entry->tag,
 424                          get_uid_from_tag(st_entry->tag));
 425                 rb_erase(&st_entry->sock_node, st_to_free_tree);
 426                 sockfd_put(st_entry->socket);
 427                 kfree(st_entry);
 428         }
 429 }
 430
 431 static struct proc_qtu_data *proc_qtu_data_tree_search(struct rb_root *root,
 432                                                        const pid_t pid)
 433 {
 434         struct rb_node *node = root->rb_node;
 435
 436         while (node) {
 437                 struct proc_qtu_data *data = rb_entry(node,
 438                                                       struct proc_qtu_data,
 439                                                       node);
 440                 if (pid < data->pid)
 441                         node = node->rb_left;
 442                 else if (pid > data->pid)
 443                         node = node->rb_right;
 444                 else
 445                         return data;
 446         }
 447         return NULL;
 448 }
 449
 450 static void proc_qtu_data_tree_insert(struct proc_qtu_data *data,
 451                                       struct rb_root *root)
 452 {
 453         struct rb_node **new = &(root->rb_node), *parent = NULL;
 454
 455         /* Figure out where to put new node */
 456         while (*new) {
 457                 struct proc_qtu_data *this = rb_entry(*new,
 458                                                       struct proc_qtu_data,
 459                                                       node);
 460                 parent = *new;
 461                 if (data->pid < this->pid)
 462                         new = &((*new)->rb_left);
 463                 else if (data->pid > this->pid)
 464                         new = &((*new)->rb_right);
 465                 else
 466                         BUG();
 467         }
 468
 469         /* Add new node and rebalance tree. */
 470         rb_link_node(&data->node, parent, new);
 471         rb_insert_color(&data->node, root);
 472 }
 473
 474 static void uid_tag_data_tree_insert(struct uid_tag_data *data,
 475                                      struct rb_root *root)
 476 {
 477         struct rb_node **new = &(root->rb_node), *parent = NULL;
 478
 479         /* Figure out where to put new node */
 480         while (*new) {
 481                 struct uid_tag_data *this = rb_entry(*new,
 482                                                      struct uid_tag_data,
 483                                                      node);
 484                 parent = *new;
 485                 if (data->uid < this->uid)
 486                         new = &((*new)->rb_left);
 487                 else if (data->uid > this->uid)
 488                         new = &((*new)->rb_right);
 489                 else
 490                         BUG();
 491         }
 492
 493         /* Add new node and rebalance tree. */
 494         rb_link_node(&data->node, parent, new);
 495         rb_insert_color(&data->node, root);
 496 }
 497
 498 static struct uid_tag_data *uid_tag_data_tree_search(struct rb_root *root,
 499                                                      uid_t uid)
 500 {
 501         struct rb_node *node = root->rb_node;
 502
 503         while (node) {
 504                 struct uid_tag_data *data = rb_entry(node,
 505                                                      struct uid_tag_data,
 506                                                      node);
 507                 if (uid < data->uid)
 508                         node = node->rb_left;
 509                 else if (uid > data->uid)
 510                         node = node->rb_right;
 511                 else
 512                         return data;
 513         }
 514         return NULL;
 515 }
 516
 517 /*
 518  * Allocates a new uid_tag_data struct if needed.
 519  * Returns a pointer to the found or allocated uid_tag_data.
 520  * Returns a PTR_ERR on failures, and lock is not held.
 521  * If found is not NULL:
 522  *   sets *found to true if not allocated.
 523  *   sets *found to false if allocated.
 524  */
 525 struct uid_tag_data *get_uid_data(uid_t uid, bool *found_res)
 526 {
 527         struct uid_tag_data *utd_entry;
 528
 529         /* Look for top level uid_tag_data for the UID */
 530         utd_entry = uid_tag_data_tree_search(&uid_tag_data_tree, uid);
 531         DR_DEBUG("qtaguid: get_uid_data(%u) utd=%p\n", uid, utd_entry);
 532
 533         if (found_res)
 534                 *found_res = utd_entry;
 535         if (utd_entry)
 536                 return utd_entry;
 537
 538         utd_entry = kzalloc(sizeof(*utd_entry), GFP_ATOMIC);
 539         if (!utd_entry) {
 540                 pr_err("qtaguid: get_uid_data(%u): "
 541                        "tag data alloc failed\n", uid);
 542                 return ERR_PTR(-ENOMEM);
 543         }
 544
 545         utd_entry->uid = uid;
 546         utd_entry->tag_ref_tree = RB_ROOT;
 547         uid_tag_data_tree_insert(utd_entry, &uid_tag_data_tree);
 548         DR_DEBUG("qtaguid: get_uid_data(%u) new utd=%p\n", uid, utd_entry);
 549         return utd_entry;
 550 }
 551
 552 /* Never returns NULL. Either PTR_ERR or a valid ptr. */
 553 static struct tag_ref *new_tag_ref(tag_t new_tag,
 554                                    struct uid_tag_data *utd_entry)
 555 {
 556         struct tag_ref *tr_entry;
 557         int res;
 558
 559         if (utd_entry->num_active_tags + 1 > max_sock_tags) {
 560                 pr_info("qtaguid: new_tag_ref(0x%llx): "
 561                         "tag ref alloc quota exceeded. max=%d\n",
 562                         new_tag, max_sock_tags);
 563                 res = -EMFILE;
 564                 goto err_res;
 565
 566         }
 567
 568         tr_entry = kzalloc(sizeof(*tr_entry), GFP_ATOMIC);
 569         if (!tr_entry) {
 570                 pr_err("qtaguid: new_tag_ref(0x%llx): "
 571                        "tag ref alloc failed\n",
 572                        new_tag);
 573                 res = -ENOMEM;
 574                 goto err_res;
 575         }
 576         tr_entry->tn.tag = new_tag;
 577         /* tr_entry->num_sock_tags  handled by caller */
 578         utd_entry->num_active_tags++;
 579         tag_ref_tree_insert(tr_entry, &utd_entry->tag_ref_tree);
 580         DR_DEBUG("qtaguid: new_tag_ref(0x%llx): "
 581                  " inserted new tag ref %p\n",
 582                  new_tag, tr_entry);
 583         return tr_entry;
 584
 585 err_res:
 586         return ERR_PTR(res);
 587 }
 588
 589 static struct tag_ref *lookup_tag_ref(tag_t full_tag,
 590                                       struct uid_tag_data **utd_res)
 591 {
 592         struct uid_tag_data *utd_entry;
 593         struct tag_ref *tr_entry;
 594         bool found_utd;
 595         uid_t uid = get_uid_from_tag(full_tag);
 596
 597         DR_DEBUG("qtaguid: lookup_tag_ref(tag=0x%llx (uid=%u))\n",
 598                  full_tag, uid);
 599
 600         utd_entry = get_uid_data(uid, &found_utd);
 601         if (IS_ERR_OR_NULL(utd_entry)) {
 602                 if (utd_res)
 603                         *utd_res = utd_entry;
 604                 return NULL;
 605         }
 606
 607         tr_entry = tag_ref_tree_search(&utd_entry->tag_ref_tree, full_tag);
 608         if (utd_res)
 609                 *utd_res = utd_entry;
 610         DR_DEBUG("qtaguid: lookup_tag_ref(0x%llx) utd_entry=%p tr_entry=%p\n",
 611                  full_tag, utd_entry, tr_entry);
 612         return tr_entry;
 613 }
 614
 615 /* Never returns NULL. Either PTR_ERR or a valid ptr. */
 616 static struct tag_ref *get_tag_ref(tag_t full_tag,
 617                                    struct uid_tag_data **utd_res)
 618 {
 619         struct uid_tag_data *utd_entry;
 620         struct tag_ref *tr_entry;
 621
 622         DR_DEBUG("qtaguid: get_tag_ref(0x%llx)\n",
 623                  full_tag);
 624         spin_lock_bh(&uid_tag_data_tree_lock);
 625         tr_entry = lookup_tag_ref(full_tag, &utd_entry);
 626         BUG_ON(IS_ERR_OR_NULL(utd_entry));
 627         if (!tr_entry)
 628                 tr_entry = new_tag_ref(full_tag, utd_entry);
 629
 630         spin_unlock_bh(&uid_tag_data_tree_lock);
 631         if (utd_res)
 632                 *utd_res = utd_entry;
 633         DR_DEBUG("qtaguid: get_tag_ref(0x%llx) utd=%p tr=%p\n",
 634                  full_tag, utd_entry, tr_entry);
 635         return tr_entry;
 636 }
 637
 638 /* Checks and maybe frees the UID Tag Data entry */
 639 static void put_utd_entry(struct uid_tag_data *utd_entry)
 640 {
 641         /* Are we done with the UID tag data entry? */
 642         if (RB_EMPTY_ROOT(&utd_entry->tag_ref_tree) &&
 643                 !utd_entry->num_pqd) {
 644                 DR_DEBUG("qtaguid: %s(): "
 645                          "erase utd_entry=%p uid=%u "
 646                          "by pid=%u tgid=%u uid=%u\n", __func__,
 647                          utd_entry, utd_entry->uid,
 648                          current->pid, current->tgid, current_fsuid());
 649                 BUG_ON(utd_entry->num_active_tags);
 650                 rb_erase(&utd_entry->node, &uid_tag_data_tree);
 651                 kfree(utd_entry);
 652         } else {
 653                 DR_DEBUG("qtaguid: %s(): "
 654                          "utd_entry=%p still has %d tags %d proc_qtu_data\n",
 655                          __func__, utd_entry, utd_entry->num_active_tags,
 656                          utd_entry->num_pqd);
 657                 BUG_ON(!(utd_entry->num_active_tags ||
 658                          utd_entry->num_pqd));
 659         }
 660 }
 661
 662 /*
 663  * If no sock_tags are using this tag_ref,
 664  * decrements refcount of utd_entry, removes tr_entry
 665  * from utd_entry->tag_ref_tree and frees.
 666  */
 667 static void free_tag_ref_from_utd_entry(struct tag_ref *tr_entry,
 668                                         struct uid_tag_data *utd_entry)
 669 {
 670         DR_DEBUG("qtaguid: %s(): %p tag=0x%llx (uid=%u)\n", __func__,
 671                  tr_entry, tr_entry->tn.tag,
 672                  get_uid_from_tag(tr_entry->tn.tag));
 673         if (!tr_entry->num_sock_tags) {
 674                 BUG_ON(!utd_entry->num_active_tags);
 675                 utd_entry->num_active_tags--;
 676                 rb_erase(&tr_entry->tn.node, &utd_entry->tag_ref_tree);
 677                 DR_DEBUG("qtaguid: %s(): erased %p\n", __func__, tr_entry);
 678                 kfree(tr_entry);
 679         }
 680 }
 681
 682 static void put_tag_ref_tree(tag_t full_tag, struct uid_tag_data *utd_entry)
 683 {
 684         struct rb_node *node;
 685         struct tag_ref *tr_entry;
 686         tag_t acct_tag;
 687
 688         DR_DEBUG("qtaguid: %s(tag=0x%llx (uid=%u))\n", __func__,
 689                  full_tag, get_uid_from_tag(full_tag));
 690         acct_tag = get_atag_from_tag(full_tag);
 691         node = rb_first(&utd_entry->tag_ref_tree);
 692         while (node) {
 693                 tr_entry = rb_entry(node, struct tag_ref, tn.node);
 694                 node = rb_next(node);
 695                 if (!acct_tag || tr_entry->tn.tag == full_tag)
 696                         free_tag_ref_from_utd_entry(tr_entry, utd_entry);
 697         }
 698 }
 699
 700 static int read_proc_u64(char *page, char **start, off_t off,
 701                         int count, int *eof, void *data)
 702 {
 703         int len;
 704         uint64_t value;
 705         char *p = page;
 706         uint64_t *iface_entry = data;
 707
 708         if (!data)
 709                 return 0;
 710
 711         value = *iface_entry;
 712         p += sprintf(p, "%llu\n", value);
 713         len = (p - page) - off;
 714         *eof = (len <= count) ? 1 : 0;
 715         *start = page + off;
 716         return len;
 717 }
 718
 719 static int read_proc_bool(char *page, char **start, off_t off,
 720                         int count, int *eof, void *data)
 721 {
 722         int len;
 723         bool value;
 724         char *p = page;
 725         bool *bool_entry = data;
 726
 727         if (!data)
 728                 return 0;
 729
 730         value = *bool_entry;
 731         p += sprintf(p, "%u\n", value);
 732         len = (p - page) - off;
 733         *eof = (len <= count) ? 1 : 0;
 734         *start = page + off;
 735         return len;
 736 }
 737
 738 static int get_active_counter_set(tag_t tag)
 739 {
 740         int active_set = 0;
 741         struct tag_counter_set *tcs;
 742
 743         MT_DEBUG("qtaguid: get_active_counter_set(tag=0x%llx)"
 744                  " (uid=%u)\n",
 745                  tag, get_uid_from_tag(tag));
 746         /* For now we only handle UID tags for active sets */
 747         tag = get_utag_from_tag(tag);
 748         spin_lock_bh(&tag_counter_set_list_lock);
 749         tcs = tag_counter_set_tree_search(&tag_counter_set_tree, tag);
 750         if (tcs)
 751                 active_set = tcs->active_set;
 752         spin_unlock_bh(&tag_counter_set_list_lock);
 753         return active_set;
 754 }
 755
 756 /*
 757  * Find the entry for tracking the specified interface.
 758  * Caller must hold iface_stat_list_lock
 759  */
 760 static struct iface_stat *get_iface_entry(const char *ifname)
 761 {
 762         struct iface_stat *iface_entry;
 763
 764         /* Find the entry for tracking the specified tag within the interface */
 765         if (ifname == NULL) {
 766                 pr_info("qtaguid: iface_stat: get() NULL device name\n");
 767                 return NULL;
 768         }
 769
 770         /* Iterate over interfaces */
 771         list_for_each_entry(iface_entry, &iface_stat_list, list) {
 772                 if (!strcmp(ifname, iface_entry->ifname))
 773                         goto done;
 774         }
 775         iface_entry = NULL;
 776 done:
 777         return iface_entry;
 778 }
 779
 780 static int iface_stat_all_proc_read(char *page, char **num_items_returned,
 781                                     off_t items_to_skip, int char_count,
 782                                     int *eof, void *data)
 783 {
 784         char *outp = page;
 785         int item_index = 0;
 786         int len;
 787         struct iface_stat *iface_entry;
 788         struct rtnl_link_stats64 dev_stats, *stats;
 789         struct rtnl_link_stats64 no_dev_stats = {0};
 790
 791         if (unlikely(module_passive)) {
 792                 *eof = 1;
 793                 return 0;
 794         }
 795
 796         CT_DEBUG("qtaguid:proc iface_stat_all "
 797                  "page=%p *num_items_returned=%p off=%ld "
 798                  "char_count=%d *eof=%d\n", page, *num_items_returned,
 799                  items_to_skip, char_count, *eof);
 800
 801         if (*eof)
 802                 return 0;
 803
 804         /*
 805          * This lock will prevent iface_stat_update() from changing active,
 806          * and in turn prevent an interface from unregistering itself.
 807          */
 808         spin_lock_bh(&iface_stat_list_lock);
 809         list_for_each_entry(iface_entry, &iface_stat_list, list) {
 810                 if (item_index++ < items_to_skip)
 811                         continue;
 812
 813                 if (iface_entry->active) {
 814                         stats = dev_get_stats(iface_entry->net_dev,
 815                                               &dev_stats);
 816                 } else {
 817                         stats = &no_dev_stats;
 818                 }
 819                 len = snprintf(outp, char_count,
 820                                "%s %d "
 821                                "%llu %llu %llu %llu "
 822                                "%llu %llu %llu %llu\n",
 823                                iface_entry->ifname,
 824                                iface_entry->active,
 825                                iface_entry->totals[IFS_RX].bytes,
 826                                iface_entry->totals[IFS_RX].packets,
 827                                iface_entry->totals[IFS_TX].bytes,
 828                                iface_entry->totals[IFS_TX].packets,
 829                                stats->rx_bytes, stats->rx_packets,
 830                                stats->tx_bytes, stats->tx_packets);
 831                 if (len >= char_count) {
 832                         spin_unlock_bh(&iface_stat_list_lock);
 833                         *outp = '\0';
 834                         return outp - page;
 835                 }
 836                 outp += len;
 837                 char_count -= len;
 838                 (*num_items_returned)++;
 839         }
 840         spin_unlock_bh(&iface_stat_list_lock);
 841
 842         *eof = 1;
 843         return outp - page;
 844 }
 845
 846 static void iface_create_proc_worker(struct work_struct *work)
 847 {
 848         struct proc_dir_entry *proc_entry;
 849         struct iface_stat_work *isw = container_of(work, struct iface_stat_work,
 850                                                    iface_work);
 851         struct iface_stat *new_iface  = isw->iface_entry;
 852
 853         /* iface_entries are not deleted, so safe to manipulate. */
 854         proc_entry = proc_mkdir(new_iface->ifname, iface_stat_procdir);
 855         if (IS_ERR_OR_NULL(proc_entry)) {
 856                 pr_err("qtaguid: iface_stat: create_proc(): alloc failed.\n");
 857                 kfree(isw);
 858                 return;
 859         }
 860
 861         new_iface->proc_ptr = proc_entry;
 862
 863         create_proc_read_entry("tx_bytes", proc_iface_perms, proc_entry,
 864                         read_proc_u64, &new_iface->totals[IFS_TX].bytes);
 865         create_proc_read_entry("rx_bytes", proc_iface_perms, proc_entry,
 866                         read_proc_u64, &new_iface->totals[IFS_RX].bytes);
 867         create_proc_read_entry("tx_packets", proc_iface_perms, proc_entry,
 868                         read_proc_u64, &new_iface->totals[IFS_TX].packets);
 869         create_proc_read_entry("rx_packets", proc_iface_perms, proc_entry,
 870                         read_proc_u64, &new_iface->totals[IFS_RX].packets);
 871         create_proc_read_entry("active", proc_iface_perms, proc_entry,
 872                         read_proc_bool, &new_iface->active);
 873
 874         IF_DEBUG("qtaguid: iface_stat: create_proc(): done "
 875                  "entry=%p dev=%s\n", new_iface, new_iface->ifname);
 876         kfree(isw);
 877 }
 878
 879 /*
 880  * Will set the entry's active state, and
 881  * update the net_dev accordingly also.
 882  */
 883 static void _iface_stat_set_active(struct iface_stat *entry,
 884                                    struct net_device *net_dev,
 885                                    bool activate)
 886 {
 887         if (activate) {
 888                 entry->net_dev = net_dev;
 889                 entry->active = true;
 890                 IF_DEBUG("qtaguid: %s(%s): "
 891                          "enable tracking. rfcnt=%d\n", __func__,
 892                          entry->ifname,
 893                          __this_cpu_read(*net_dev->pcpu_refcnt));
 894         } else {
 895                 entry->active = false;
 896                 entry->net_dev = NULL;
 897                 IF_DEBUG("qtaguid: %s(%s): "
 898                          "disable tracking. rfcnt=%d\n", __func__,
 899                          entry->ifname,
 900                          __this_cpu_read(*net_dev->pcpu_refcnt));
 901
 902         }
 903 }
 904
 905 /* Caller must hold iface_stat_list_lock */
 906 static struct iface_stat *iface_alloc(struct net_device *net_dev)
 907 {
 908         struct iface_stat *new_iface;
 909         struct iface_stat_work *isw;
 910
 911         new_iface = kzalloc(sizeof(*new_iface), GFP_ATOMIC);
 912         if (new_iface == NULL) {
 913                 pr_err("qtaguid: iface_stat: create(%s): "
 914                        "iface_stat alloc failed\n", net_dev->name);
 915                 return NULL;
 916         }
 917         new_iface->ifname = kstrdup(net_dev->name, GFP_ATOMIC);
 918         if (new_iface->ifname == NULL) {
 919                 pr_err("qtaguid: iface_stat: create(%s): "
 920                        "ifname alloc failed\n", net_dev->name);
 921                 kfree(new_iface);
 922                 return NULL;
 923         }
 924         spin_lock_init(&new_iface->tag_stat_list_lock);
 925         new_iface->tag_stat_tree = RB_ROOT;
 926         _iface_stat_set_active(new_iface, net_dev, true);
 927
 928         /*
 929          * ipv6 notifier chains are atomic :(
 930          * No create_proc_read_entry() for you!
 931          */
 932         isw = kmalloc(sizeof(*isw), GFP_ATOMIC);
 933         if (!isw) {
 934                 pr_err("qtaguid: iface_stat: create(%s): "
 935                        "work alloc failed\n", new_iface->ifname);
 936                 _iface_stat_set_active(new_iface, net_dev, false);
 937                 kfree(new_iface->ifname);
 938                 kfree(new_iface);
 939                 return NULL;
 940         }
 941         isw->iface_entry = new_iface;
 942         INIT_WORK(&isw->iface_work, iface_create_proc_worker);
 943         schedule_work(&isw->iface_work);
 944         list_add(&new_iface->list, &iface_stat_list);
 945         return new_iface;
 946 }
 947
 948 static void iface_check_stats_reset_and_adjust(struct net_device *net_dev,
 949                                                struct iface_stat *iface)
 950 {
 951         struct rtnl_link_stats64 dev_stats, *stats;
 952         bool stats_rewound;
 953
 954         stats = dev_get_stats(net_dev, &dev_stats);
 955         /* No empty packets */
 956         stats_rewound =
 957                 (stats->rx_bytes < iface->last_known[IFS_RX].bytes)
 958                 || (stats->tx_bytes < iface->last_known[IFS_TX].bytes);
 959
 960         IF_DEBUG("qtaguid: %s(%s): iface=%p netdev=%p "
 961                  "bytes rx/tx=%llu/%llu "
 962                  "active=%d last_known=%d "
 963                  "stats_rewound=%d\n", __func__,
 964                  net_dev ? net_dev->name : "?",
 965                  iface, net_dev,
 966                  stats->rx_bytes, stats->tx_bytes,
 967                  iface->active, iface->last_known_valid, stats_rewound);
 968
 969         if (iface->active && iface->last_known_valid && stats_rewound) {
 970                 pr_warn_once("qtaguid: iface_stat: %s(%s): "
 971                              "iface reset its stats unexpectedly\n", __func__,
 972                              net_dev->name);
 973
 974                 iface->totals[IFS_TX].bytes += iface->last_known[IFS_TX].bytes;
 975                 iface->totals[IFS_TX].packets +=
 976                         iface->last_known[IFS_TX].packets;
 977                 iface->totals[IFS_RX].bytes += iface->last_known[IFS_RX].bytes;
 978                 iface->totals[IFS_RX].packets +=
 979                         iface->last_known[IFS_RX].packets;
 980                 iface->last_known_valid = false;
 981                 IF_DEBUG("qtaguid: %s(%s): iface=%p "
 982                          "used last known bytes rx/tx=%llu/%llu\n", __func__,
 983                          iface->ifname, iface, iface->last_known[IFS_RX].bytes,
 984                          iface->last_known[IFS_TX].bytes);
 985         }
 986 }
 987
 988 /*
 989  * Create a new entry for tracking the specified interface.
 990  * Do nothing if the entry already exists.
 991  * Called when an interface is configured with a valid IP address.
 992  */
 993 static void iface_stat_create(struct net_device *net_dev,
 994                               struct in_ifaddr *ifa)
 995 {
 996         struct in_device *in_dev = NULL;
 997         const char *ifname;
 998         struct iface_stat *entry;
 999         __be32 ipaddr = 0;
1000         struct iface_stat *new_iface;
1001
1002         IF_DEBUG("qtaguid: iface_stat: create(%s): ifa=%p netdev=%p\n",
1003                  net_dev ? net_dev->name : "?",
1004                  ifa, net_dev);
1005         if (!net_dev) {
1006                 pr_err("qtaguid: iface_stat: create(): no net dev\n");
1007                 return;
1008         }
1009
1010         ifname = net_dev->name;
1011         if (!ifa) {
1012                 in_dev = in_dev_get(net_dev);
1013                 if (!in_dev) {
1014                         pr_err("qtaguid: iface_stat: create(%s): no inet dev\n",
1015                                ifname);
1016                         return;
1017                 }
1018                 IF_DEBUG("qtaguid: iface_stat: create(%s): in_dev=%p\n",
1019                          ifname, in_dev);
1020                 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
1021                         IF_DEBUG("qtaguid: iface_stat: create(%s): "
1022                                  "ifa=%p ifa_label=%s\n",
1023                                  ifname, ifa,
1024                                  ifa->ifa_label ? ifa->ifa_label : "(null)");
1025                         if (ifa->ifa_label && !strcmp(ifname, ifa->ifa_label))
1026                                 break;
1027                 }
1028         }
1029
1030         if (!ifa) {
1031                 IF_DEBUG("qtaguid: iface_stat: create(%s): no matching IP\n",
1032                          ifname);
1033                 goto done_put;
1034         }
1035         ipaddr = ifa->ifa_local;
1036
1037         spin_lock_bh(&iface_stat_list_lock);
1038         entry = get_iface_entry(ifname);
1039         if (entry != NULL) {
1040                 bool activate = !ipv4_is_loopback(ipaddr);
1041                 IF_DEBUG("qtaguid: iface_stat: create(%s): entry=%p\n",
1042                          ifname, entry);
1043                 iface_check_stats_reset_and_adjust(net_dev, entry);
1044                 _iface_stat_set_active(entry, net_dev, activate);
1045                 IF_DEBUG("qtaguid: %s(%s): "
1046                          "tracking now %d on ip=%pI4\n", __func__,
1047                          entry->ifname, activate, &ipaddr);
1048                 goto done_unlock_put;
1049         } else if (ipv4_is_loopback(ipaddr)) {
1050                 IF_DEBUG("qtaguid: iface_stat: create(%s): "
1051                          "ignore loopback dev. ip=%pI4\n", ifname, &ipaddr);
1052                 goto done_unlock_put;
1053         }
1054
1055         new_iface = iface_alloc(net_dev);
1056         IF_DEBUG("qtaguid: iface_stat: create(%s): done "
1057                  "entry=%p ip=%pI4\n", ifname, new_iface, &ipaddr);
1058 done_unlock_put:
1059         spin_unlock_bh(&iface_stat_list_lock);
1060 done_put:
1061         if (in_dev)
1062                 in_dev_put(in_dev);
1063 }
1064
1065 static void iface_stat_create_ipv6(struct net_device *net_dev,
1066                                    struct inet6_ifaddr *ifa)
1067 {
1068         struct in_device *in_dev;
1069         const char *ifname;
1070         struct iface_stat *entry;
1071         struct iface_stat *new_iface;
1072         int addr_type;
1073
1074         IF_DEBUG("qtaguid: iface_stat: create6(): ifa=%p netdev=%p->name=%s\n",
1075                  ifa, net_dev, net_dev ? net_dev->name : "");
1076         if (!net_dev) {
1077                 pr_err("qtaguid: iface_stat: create6(): no net dev!\n");
1078                 return;
1079         }
1080         ifname = net_dev->name;
1081
1082         in_dev = in_dev_get(net_dev);
1083         if (!in_dev) {
1084                 pr_err("qtaguid: iface_stat: create6(%s): no inet dev\n",
1085                        ifname);
1086                 return;
1087         }
1088
1089         IF_DEBUG("qtaguid: iface_stat: create6(%s): in_dev=%p\n",
1090                  ifname, in_dev);
1091
1092         if (!ifa) {
1093                 IF_DEBUG("qtaguid: iface_stat: create6(%s): no matching IP\n",
1094                          ifname);
1095                 goto done_put;
1096         }
1097         addr_type = ipv6_addr_type(&ifa->addr);
1098
1099         spin_lock_bh(&iface_stat_list_lock);
1100         entry = get_iface_entry(ifname);
1101         if (entry != NULL) {
1102                 bool activate = !(addr_type & IPV6_ADDR_LOOPBACK);
1103                 IF_DEBUG("qtaguid: %s(%s): entry=%p\n", __func__,
1104                          ifname, entry);
1105                 iface_check_stats_reset_and_adjust(net_dev, entry);
1106                 _iface_stat_set_active(entry, net_dev, activate);
1107                 IF_DEBUG("qtaguid: %s(%s): "
1108                          "tracking now %d on ip=%pI6c\n", __func__,
1109                          entry->ifname, activate, &ifa->addr);
1110                 goto done_unlock_put;
1111         } else if (addr_type & IPV6_ADDR_LOOPBACK) {
1112                 IF_DEBUG("qtaguid: %s(%s): "
1113                          "ignore loopback dev. ip=%pI6c\n", __func__,
1114                          ifname, &ifa->addr);
1115                 goto done_unlock_put;
1116         }
1117
1118         new_iface = iface_alloc(net_dev);
1119         IF_DEBUG("qtaguid: iface_stat: create6(%s): done "
1120                  "entry=%p ip=%pI6c\n", ifname, new_iface, &ifa->addr);
1121
1122 done_unlock_put:
1123         spin_unlock_bh(&iface_stat_list_lock);
1124 done_put:
1125         in_dev_put(in_dev);
1126 }
1127
1128 static struct sock_tag *get_sock_stat_nl(const struct sock *sk)
1129 {
1130         MT_DEBUG("qtaguid: get_sock_stat_nl(sk=%p)\n", sk);
1131         return sock_tag_tree_search(&sock_tag_tree, sk);
1132 }
1133
1134 static struct sock_tag *get_sock_stat(const struct sock *sk)
1135 {
1136         struct sock_tag *sock_tag_entry;
1137         MT_DEBUG("qtaguid: get_sock_stat(sk=%p)\n", sk);
1138         if (!sk)
1139                 return NULL;
1140         spin_lock_bh(&sock_tag_list_lock);
1141         sock_tag_entry = get_sock_stat_nl(sk);
1142         spin_unlock_bh(&sock_tag_list_lock);
1143         return sock_tag_entry;
1144 }
1145
1146 static void
1147 data_counters_update(struct data_counters *dc, int set,
1148                      enum ifs_tx_rx direction, int proto, int bytes)
1149 {
1150         switch (proto) {
1151         case IPPROTO_TCP:
1152                 dc_add_byte_packets(dc, set, direction, IFS_TCP, bytes, 1);
1153                 break;
1154         case IPPROTO_UDP:
1155                 dc_add_byte_packets(dc, set, direction, IFS_UDP, bytes, 1);
1156                 break;
1157         case IPPROTO_IP:
1158         default:
1159                 dc_add_byte_packets(dc, set, direction, IFS_PROTO_OTHER, bytes,
1160                                     1);
1161                 break;
1162         }
1163 }
1164
1165 /*
1166  * Update stats for the specified interface. Do nothing if the entry
1167  * does not exist (when a device was never configured with an IP address).
1168  * Called when an device is being unregistered.
1169  */
1170 static void iface_stat_update(struct net_device *net_dev, bool stash_only)
1171 {
1172         struct rtnl_link_stats64 dev_stats, *stats;
1173         struct iface_stat *entry;
1174
1175         stats = dev_get_stats(net_dev, &dev_stats);
1176         spin_lock_bh(&iface_stat_list_lock);
1177         entry = get_iface_entry(net_dev->name);
1178         if (entry == NULL) {
1179                 IF_DEBUG("qtaguid: iface_stat: update(%s): not tracked\n",
1180                          net_dev->name);
1181                 spin_unlock_bh(&iface_stat_list_lock);
1182                 return;
1183         }
1184
1185         IF_DEBUG("qtaguid: %s(%s): entry=%p\n", __func__,
1186                  net_dev->name, entry);
1187         if (!entry->active) {
1188                 IF_DEBUG("qtaguid: %s(%s): already disabled\n", __func__,
1189                          net_dev->name);
1190                 spin_unlock_bh(&iface_stat_list_lock);
1191                 return;
1192         }
1193
1194         if (stash_only) {
1195                 entry->last_known[IFS_TX].bytes = stats->tx_bytes;
1196                 entry->last_known[IFS_TX].packets = stats->tx_packets;
1197                 entry->last_known[IFS_RX].bytes = stats->rx_bytes;
1198                 entry->last_known[IFS_RX].packets = stats->rx_packets;
1199                 entry->last_known_valid = true;
1200                 IF_DEBUG("qtaguid: %s(%s): "
1201                          "dev stats stashed rx/tx=%llu/%llu\n", __func__,
1202                          net_dev->name, stats->rx_bytes, stats->tx_bytes);
1203                 spin_unlock_bh(&iface_stat_list_lock);
1204                 return;
1205         }
1206         entry->totals[IFS_TX].bytes += stats->tx_bytes;
1207         entry->totals[IFS_TX].packets += stats->tx_packets;
1208         entry->totals[IFS_RX].bytes += stats->rx_bytes;
1209         entry->totals[IFS_RX].packets += stats->rx_packets;
1210         /* We don't need the last_known[] anymore */
1211         entry->last_known_valid = false;
1212         _iface_stat_set_active(entry, net_dev, false);
1213         IF_DEBUG("qtaguid: %s(%s): "
1214                  "disable tracking. rx/tx=%llu/%llu\n", __func__,
1215                  net_dev->name, stats->rx_bytes, stats->tx_bytes);
1216         spin_unlock_bh(&iface_stat_list_lock);
1217 }
1218
1219 static void tag_stat_update(struct tag_stat *tag_entry,
1220                         enum ifs_tx_rx direction, int proto, int bytes)
1221 {
1222         int active_set;
1223         active_set = get_active_counter_set(tag_entry->tn.tag);
1224         MT_DEBUG("qtaguid: tag_stat_update(tag=0x%llx (uid=%u) set=%d "
1225                  "dir=%d proto=%d bytes=%d)\n",
1226                  tag_entry->tn.tag, get_uid_from_tag(tag_entry->tn.tag),
1227                  active_set, direction, proto, bytes);
1228         data_counters_update(&tag_entry->counters, active_set, direction,
1229                              proto, bytes);
1230         if (tag_entry->parent_counters)
1231                 data_counters_update(tag_entry->parent_counters, active_set,
1232                                      direction, proto, bytes);
1233 }
1234
1235 /*
1236  * Create a new entry for tracking the specified {acct_tag,uid_tag} within
1237  * the interface.
1238  * iface_entry->tag_stat_list_lock should be held.
1239  */
1240 static struct tag_stat *create_if_tag_stat(struct iface_stat *iface_entry,
1241                                            tag_t tag)
1242 {
1243         struct tag_stat *new_tag_stat_entry = NULL;
1244         IF_DEBUG("qtaguid: iface_stat: %s(): ife=%p tag=0x%llx"
1245                  " (uid=%u)\n", __func__,
1246                  iface_entry, tag, get_uid_from_tag(tag));
1247         new_tag_stat_entry = kzalloc(sizeof(*new_tag_stat_entry), GFP_ATOMIC);
1248         if (!new_tag_stat_entry) {
1249                 pr_err("qtaguid: iface_stat: tag stat alloc failed\n");
1250                 goto done;
1251         }
1252         new_tag_stat_entry->tn.tag = tag;
1253         tag_stat_tree_insert(new_tag_stat_entry, &iface_entry->tag_stat_tree);
1254 done:
1255         return new_tag_stat_entry;
1256 }
1257
1258 static void if_tag_stat_update(const char *ifname, uid_t uid,
1259                                const struct sock *sk, enum ifs_tx_rx direction,
1260                                int proto, int bytes)
1261 {
1262         struct tag_stat *tag_stat_entry;
1263         tag_t tag, acct_tag;
1264         tag_t uid_tag;
1265         struct data_counters *uid_tag_counters;
1266         struct sock_tag *sock_tag_entry;
1267         struct iface_stat *iface_entry;
1268         struct tag_stat *new_tag_stat;
1269         MT_DEBUG("qtaguid: if_tag_stat_update(ifname=%s "
1270                 "uid=%u sk=%p dir=%d proto=%d bytes=%d)\n",
1271                  ifname, uid, sk, direction, proto, bytes);
1272
1273
1274         iface_entry = get_iface_entry(ifname);
1275         if (!iface_entry) {
1276                 pr_err("qtaguid: iface_stat: stat_update() %s not found\n",
1277                        ifname);
1278                 return;
1279         }
1280         /* It is ok to process data when an iface_entry is inactive */
1281
1282         MT_DEBUG("qtaguid: iface_stat: stat_update() dev=%s entry=%p\n",
1283                  ifname, iface_entry);
1284
1285         /*
1286          * Look for a tagged sock.
1287          * It will have an acct_uid.
1288          */
1289         sock_tag_entry = get_sock_stat(sk);
1290         if (sock_tag_entry) {
1291                 tag = sock_tag_entry->tag;
1292                 acct_tag = get_atag_from_tag(tag);
1293                 uid_tag = get_utag_from_tag(tag);
1294         } else {
1295                 acct_tag = make_atag_from_value(0);
1296                 tag = combine_atag_with_uid(acct_tag, uid);
1297                 uid_tag = make_tag_from_uid(uid);
1298         }
1299         MT_DEBUG("qtaguid: iface_stat: stat_update(): "
1300                  " looking for tag=0x%llx (uid=%u) in ife=%p\n",
1301                  tag, get_uid_from_tag(tag), iface_entry);
1302         /* Loop over tag list under this interface for {acct_tag,uid_tag} */
1303         spin_lock_bh(&iface_entry->tag_stat_list_lock);
1304
1305         tag_stat_entry = tag_stat_tree_search(&iface_entry->tag_stat_tree,
1306                                               tag);
1307         if (tag_stat_entry) {
1308                 /*
1309                  * Updating the {acct_tag, uid_tag} entry handles both stats:
1310                  * {0, uid_tag} will also get updated.
1311                  */
1312                 tag_stat_update(tag_stat_entry, direction, proto, bytes);
1313                 spin_unlock_bh(&iface_entry->tag_stat_list_lock);
1314                 return;
1315         }
1316
1317         /* Loop over tag list under this interface for {0,uid_tag} */
1318         tag_stat_entry = tag_stat_tree_search(&iface_entry->tag_stat_tree,
1319                                               uid_tag);
1320         if (!tag_stat_entry) {
1321                 /* Here: the base uid_tag did not exist */
1322                 /*
1323                  * No parent counters. So
1324                  *  - No {0, uid_tag} stats and no {acc_tag, uid_tag} stats.
1325                  */
1326                 new_tag_stat = create_if_tag_stat(iface_entry, uid_tag);
1327                 uid_tag_counters = &new_tag_stat->counters;
1328         } else {
1329                 uid_tag_counters = &tag_stat_entry->counters;
1330         }
1331
1332         if (acct_tag) {
1333                 new_tag_stat = create_if_tag_stat(iface_entry, tag);
1334                 new_tag_stat->parent_counters = uid_tag_counters;
1335         }
1336         tag_stat_update(new_tag_stat, direction, proto, bytes);
1337         spin_unlock_bh(&iface_entry->tag_stat_list_lock);
1338 }
1339
1340 static int iface_netdev_event_handler(struct notifier_block *nb,
1341                                       unsigned long event, void *ptr) {
1342         struct net_device *dev = ptr;
1343
1344         if (unlikely(module_passive))
1345                 return NOTIFY_DONE;
1346
1347         IF_DEBUG("qtaguid: iface_stat: netdev_event(): "
1348                  "ev=0x%lx/%s netdev=%p->name=%s\n",
1349                  event, netdev_evt_str(event), dev, dev ? dev->name : "");
1350
1351         switch (event) {
1352         case NETDEV_UP:
1353                 iface_stat_create(dev, NULL);
1354                 atomic64_inc(&qtu_events.iface_events);
1355                 break;
1356         case NETDEV_DOWN:
1357         case NETDEV_UNREGISTER:
1358                 iface_stat_update(dev, event == NETDEV_DOWN);
1359                 atomic64_inc(&qtu_events.iface_events);
1360                 break;
1361         }
1362         return NOTIFY_DONE;
1363 }
1364
1365 static int iface_inet6addr_event_handler(struct notifier_block *nb,
1366                                          unsigned long event, void *ptr)
1367 {
1368         struct inet6_ifaddr *ifa = ptr;
1369         struct net_device *dev;
1370
1371         if (unlikely(module_passive))
1372                 return NOTIFY_DONE;
1373
1374         IF_DEBUG("qtaguid: iface_stat: inet6addr_event(): "
1375                  "ev=0x%lx/%s ifa=%p\n",
1376                  event, netdev_evt_str(event), ifa);
1377
1378         switch (event) {
1379         case NETDEV_UP:
1380                 BUG_ON(!ifa || !ifa->idev);
1381                 dev = (struct net_device *)ifa->idev->dev;
1382                 iface_stat_create_ipv6(dev, ifa);
1383                 atomic64_inc(&qtu_events.iface_events);
1384                 break;
1385         case NETDEV_DOWN:
1386         case NETDEV_UNREGISTER:
1387                 BUG_ON(!ifa || !ifa->idev);
1388                 dev = (struct net_device *)ifa->idev->dev;
1389                 iface_stat_update(dev, event == NETDEV_DOWN);
1390                 atomic64_inc(&qtu_events.iface_events);
1391                 break;
1392         }
1393         return NOTIFY_DONE;
1394 }
1395
1396 static int iface_inetaddr_event_handler(struct notifier_block *nb,
1397                                         unsigned long event, void *ptr)
1398 {
1399         struct in_ifaddr *ifa = ptr;
1400         struct net_device *dev;
1401
1402         if (unlikely(module_passive))
1403                 return NOTIFY_DONE;
1404
1405         IF_DEBUG("qtaguid: iface_stat: inetaddr_event(): "
1406                  "ev=0x%lx/%s ifa=%p\n",
1407                  event, netdev_evt_str(event), ifa);
1408
1409         switch (event) {
1410         case NETDEV_UP:
1411                 BUG_ON(!ifa || !ifa->ifa_dev);
1412                 dev = ifa->ifa_dev->dev;
1413                 iface_stat_create(dev, ifa);
1414                 atomic64_inc(&qtu_events.iface_events);
1415                 break;
1416         case NETDEV_DOWN:
1417         case NETDEV_UNREGISTER:
1418                 BUG_ON(!ifa || !ifa->ifa_dev);
1419                 dev = ifa->ifa_dev->dev;
1420                 iface_stat_update(dev, event == NETDEV_DOWN);
1421                 atomic64_inc(&qtu_events.iface_events);
1422                 break;
1423         }
1424         return NOTIFY_DONE;
1425 }
1426
1427 static struct notifier_block iface_netdev_notifier_blk = {
1428         .notifier_call = iface_netdev_event_handler,
1429 };
1430
1431 static struct notifier_block iface_inetaddr_notifier_blk = {
1432         .notifier_call = iface_inetaddr_event_handler,
1433 };
1434
1435 static struct notifier_block iface_inet6addr_notifier_blk = {
1436         .notifier_call = iface_inet6addr_event_handler,
1437 };
1438
1439 static int __init iface_stat_init(struct proc_dir_entry *parent_procdir)
1440 {
1441         int err;
1442
1443         iface_stat_procdir = proc_mkdir(iface_stat_procdirname, parent_procdir);
1444         if (!iface_stat_procdir) {
1445                 pr_err("qtaguid: iface_stat: init failed to create proc entry\n");
1446                 err = -1;
1447                 goto err;
1448         }
1449
1450         iface_stat_all_procfile = create_proc_entry(iface_stat_all_procfilename,
1451                                                     proc_iface_perms,
1452                                                     parent_procdir);
1453         if (!iface_stat_all_procfile) {
1454                 pr_err("qtaguid: iface_stat: init "
1455                        " failed to create stat_all proc entry\n");
1456                 err = -1;
1457                 goto err_zap_entry;
1458         }
1459         iface_stat_all_procfile->read_proc = iface_stat_all_proc_read;
1460
1461
1462         err = register_netdevice_notifier(&iface_netdev_notifier_blk);
1463         if (err) {
1464                 pr_err("qtaguid: iface_stat: init "
1465                        "failed to register dev event handler\n");
1466                 goto err_zap_all_stats_entry;
1467         }
1468         err = register_inetaddr_notifier(&iface_inetaddr_notifier_blk);
1469         if (err) {
1470                 pr_err("qtaguid: iface_stat: init "
1471                        "failed to register ipv4 dev event handler\n");
1472                 goto err_unreg_nd;
1473         }
1474
1475         err = register_inet6addr_notifier(&iface_inet6addr_notifier_blk);
1476         if (err) {
1477                 pr_err("qtaguid: iface_stat: init "
1478                        "failed to register ipv6 dev event handler\n");
1479                 goto err_unreg_ip4_addr;
1480         }
1481         return 0;
1482
1483 err_unreg_ip4_addr:
1484         unregister_inetaddr_notifier(&iface_inetaddr_notifier_blk);
1485 err_unreg_nd:
1486         unregister_netdevice_notifier(&iface_netdev_notifier_blk);
1487 err_zap_all_stats_entry:
1488         remove_proc_entry(iface_stat_all_procfilename, parent_procdir);
1489 err_zap_entry:
1490         remove_proc_entry(iface_stat_procdirname, parent_procdir);
1491 err:
1492         return err;
1493 }
1494
1495 static struct sock *qtaguid_find_sk(const struct sk_buff *skb,
1496                                     struct xt_action_param *par)
1497 {
1498         struct sock *sk;
1499         unsigned int hook_mask = (1 << par->hooknum);
1500
1501         MT_DEBUG("qtaguid: find_sk(skb=%p) hooknum=%d family=%d\n", skb,
1502                  par->hooknum, par->family);
1503
1504         /*
1505          * Let's not abuse the the xt_socket_get*_sk(), or else it will
1506          * return garbage SKs.
1507          */
1508         if (!(hook_mask & XT_SOCKET_SUPPORTED_HOOKS))
1509                 return NULL;
1510
1511         switch (par->family) {
1512         case NFPROTO_IPV6:
1513                 sk = xt_socket_get6_sk(skb, par);
1514                 break;
1515         case NFPROTO_IPV4:
1516                 sk = xt_socket_get4_sk(skb, par);
1517                 break;
1518         default:
1519                 return NULL;
1520         }
1521
1522         /*
1523          * Seems to be issues on the file ptr for TCP_TIME_WAIT SKs.
1524          * http://kerneltrap.org/mailarchive/linux-netdev/2010/10/21/6287959
1525          * Not fixed in 3.0-r3 :(
1526          */
1527         if (sk) {
1528                 MT_DEBUG("qtaguid: %p->sk_proto=%u "
1529                          "->sk_state=%d\n", sk, sk->sk_protocol, sk->sk_state);
1530                 if (sk->sk_state  == TCP_TIME_WAIT) {
1531                         xt_socket_put_sk(sk);
1532                         sk = NULL;
1533                 }
1534         }
1535         return sk;
1536 }
1537
1538 static void account_for_uid(const struct sk_buff *skb,
1539                             const struct sock *alternate_sk, uid_t uid,
1540                             struct xt_action_param *par)
1541 {
1542         const struct net_device *el_dev;
1543
1544         if (!skb->dev) {
1545                 MT_DEBUG("qtaguid[%d]: no skb->dev\n", par->hooknum);
1546                 el_dev = par->in ? : par->out;
1547         } else {
1548                 const struct net_device *other_dev;
1549                 el_dev = skb->dev;
1550                 other_dev = par->in ? : par->out;
1551                 if (el_dev != other_dev) {
1552                         MT_DEBUG("qtaguid[%d]: skb->dev=%p %s vs "
1553                                 "par->(in/out)=%p %s\n",
1554                                 par->hooknum, el_dev, el_dev->name, other_dev,
1555                                 other_dev->name);
1556                 }
1557         }
1558
1559         if (unlikely(!el_dev)) {
1560                 pr_info("qtaguid[%d]: no par->in/out?!!\n", par->hooknum);
1561         } else if (unlikely(!el_dev->name)) {
1562                 pr_info("qtaguid[%d]: no dev->name?!!\n", par->hooknum);
1563         } else {
1564                 MT_DEBUG("qtaguid[%d]: dev name=%s type=%d\n",
1565                          par->hooknum,
1566                          el_dev->name,
1567                          el_dev->type);
1568
1569                 if_tag_stat_update(el_dev->name, uid,
1570                                 skb->sk ? skb->sk : alternate_sk,
1571                                 par->in ? IFS_RX : IFS_TX,
1572                                 ip_hdr(skb)->protocol, skb->len);
1573         }
1574 }
1575
1576 static bool qtaguid_mt(const struct sk_buff *skb, struct xt_action_param *par)
1577 {
1578         const struct xt_qtaguid_match_info *info = par->matchinfo;
1579         const struct file *filp;
1580         bool got_sock = false;
1581         struct sock *sk;
1582         uid_t sock_uid;
1583         bool res;
1584
1585         if (unlikely(module_passive))
1586                 return (info->match ^ info->invert) == 0;
1587
1588         MT_DEBUG("qtaguid[%d]: entered skb=%p par->in=%p/out=%p fam=%d\n",
1589                  par->hooknum, skb, par->in, par->out, par->family);
1590
1591         atomic64_inc(&qtu_events.match_calls);
1592         if (skb == NULL) {
1593                 res = (info->match ^ info->invert) == 0;
1594                 goto ret_res;
1595         }
1596
1597         sk = skb->sk;
1598
1599         if (sk == NULL) {
1600                 /*
1601                  * A missing sk->sk_socket happens when packets are in-flight
1602                  * and the matching socket is already closed and gone.
1603                  */
1604                 sk = qtaguid_find_sk(skb, par);
1605                 /*
1606                  * If we got the socket from the find_sk(), we will need to put
1607                  * it back, as nf_tproxy_get_sock_v4() got it.
1608                  */
1609                 got_sock = sk;
1610                 if (sk)
1611                         atomic64_inc(&qtu_events.match_found_sk_in_ct);
1612                 else
1613                         atomic64_inc(&qtu_events.match_found_no_sk_in_ct);
1614         } else {
1615                 atomic64_inc(&qtu_events.match_found_sk);
1616         }
1617         MT_DEBUG("qtaguid[%d]: sk=%p got_sock=%d proto=%d\n",
1618                 par->hooknum, sk, got_sock, ip_hdr(skb)->protocol);
1619         if (sk != NULL) {
1620                 MT_DEBUG("qtaguid[%d]: sk=%p->sk_socket=%p->file=%p\n",
1621                         par->hooknum, sk, sk->sk_socket,
1622                         sk->sk_socket ? sk->sk_socket->file : (void *)-1LL);
1623                 filp = sk->sk_socket ? sk->sk_socket->file : NULL;
1624                 MT_DEBUG("qtaguid[%d]: filp...uid=%u\n",
1625                         par->hooknum, filp ? filp->f_cred->fsuid : -1);
1626         }
1627
1628         if (sk == NULL || sk->sk_socket == NULL) {
1629                 /*
1630                  * Here, the qtaguid_find_sk() using connection tracking
1631                  * couldn't find the owner, so for now we just count them
1632                  * against the system.
1633                  */
1634                 /*
1635                  * TODO: unhack how to force just accounting.
1636                  * For now we only do iface stats when the uid-owner is not
1637                  * requested.
1638                  */
1639                 if (!(info->match & XT_QTAGUID_UID))
1640                         account_for_uid(skb, sk, 0, par);
1641                 MT_DEBUG("qtaguid[%d]: leaving (sk?sk->sk_socket)=%p\n",
1642                         par->hooknum,
1643                         sk ? sk->sk_socket : NULL);
1644                 res = (info->match ^ info->invert) == 0;
1645                 atomic64_inc(&qtu_events.match_no_sk);
1646                 goto put_sock_ret_res;
1647         } else if (info->match & info->invert & XT_QTAGUID_SOCKET) {
1648                 res = false;
1649                 goto put_sock_ret_res;
1650         }
1651         filp = sk->sk_socket->file;
1652         if (filp == NULL) {
1653                 MT_DEBUG("qtaguid[%d]: leaving filp=NULL\n", par->hooknum);
1654                 account_for_uid(skb, sk, 0, par);
1655                 res = ((info->match ^ info->invert) &
1656                         (XT_QTAGUID_UID | XT_QTAGUID_GID)) == 0;
1657                 atomic64_inc(&qtu_events.match_no_sk_file);
1658                 goto put_sock_ret_res;
1659         }
1660         sock_uid = filp->f_cred->fsuid;
1661         /*
1662          * TODO: unhack how to force just accounting.
1663          * For now we only do iface stats when the uid-owner is not requested
1664          */
1665         if (!(info->match & XT_QTAGUID_UID))
1666                 account_for_uid(skb, sk, sock_uid, par);
1667
1668         /*
1669          * The following two tests fail the match when:
1670          *    id not in range AND no inverted condition requested
1671          * or id     in range AND    inverted condition requested
1672          * Thus (!a && b) || (a && !b) == a ^ b
1673          */
1674         if (info->match & XT_QTAGUID_UID)
1675                 if ((filp->f_cred->fsuid >= info->uid_min &&
1676                      filp->f_cred->fsuid <= info->uid_max) ^
1677                     !(info->invert & XT_QTAGUID_UID)) {
1678                         MT_DEBUG("qtaguid[%d]: leaving uid not matching\n",
1679                                  par->hooknum);
1680                         res = false;
1681                         goto put_sock_ret_res;
1682                 }
1683         if (info->match & XT_QTAGUID_GID)
1684                 if ((filp->f_cred->fsgid >= info->gid_min &&
1685                                 filp->f_cred->fsgid <= info->gid_max) ^
1686                         !(info->invert & XT_QTAGUID_GID)) {
1687                         MT_DEBUG("qtaguid[%d]: leaving gid not matching\n",
1688                                 par->hooknum);
1689                         res = false;
1690                         goto put_sock_ret_res;
1691                 }
1692
1693         MT_DEBUG("qtaguid[%d]: leaving matched\n", par->hooknum);
1694         res = true;
1695
1696 put_sock_ret_res:
1697         if (got_sock)
1698                 xt_socket_put_sk(sk);
1699 ret_res:
1700         MT_DEBUG("qtaguid[%d]: left %d\n", par->hooknum, res);
1701         return res;
1702 }
1703
1704 #ifdef DDEBUG
1705 /* This function is not in xt_qtaguid_print.c because of locks visibility */
1706 static void prdebug_full_state(int indent_level, const char *fmt, ...)
1707 {
1708         va_list args;
1709         char *fmt_buff;
1710         char *buff;
1711
1712         if (!unlikely(qtaguid_debug_mask & DDEBUG_MASK))
1713                 return;
1714
1715         fmt_buff = kasprintf(GFP_ATOMIC,
1716                              "qtaguid: %s(): %s {\n", __func__, fmt);
1717         BUG_ON(!fmt_buff);
1718         va_start(args, fmt);
1719         buff = kvasprintf(GFP_ATOMIC,
1720                           fmt_buff, args);
1721         BUG_ON(!buff);
1722         pr_debug("%s", buff);
1723         kfree(fmt_buff);
1724         kfree(buff);
1725         va_end(args);
1726
1727         spin_lock_bh(&sock_tag_list_lock);
1728         prdebug_sock_tag_tree(indent_level, &sock_tag_tree);
1729         spin_unlock_bh(&sock_tag_list_lock);
1730
1731         spin_lock_bh(&sock_tag_list_lock);
1732         spin_lock_bh(&uid_tag_data_tree_lock);
1733         prdebug_uid_tag_data_tree(indent_level, &uid_tag_data_tree);
1734         prdebug_proc_qtu_data_tree(indent_level, &proc_qtu_data_tree);
1735         spin_unlock_bh(&uid_tag_data_tree_lock);
1736         spin_unlock_bh(&sock_tag_list_lock);
1737
1738         spin_lock_bh(&iface_stat_list_lock);
1739         prdebug_iface_stat_list(indent_level, &iface_stat_list);
1740         spin_unlock_bh(&iface_stat_list_lock);
1741
1742         pr_debug("qtaguid: %s(): }\n", __func__);
1743 }
1744 #else
1745 static void prdebug_full_state(int indent_level, const char *fmt, ...) {}
1746 #endif
1747
1748 /*
1749  * Procfs reader to get all active socket tags using style "1)" as described in
1750  * fs/proc/generic.c
1751  */
1752 static int qtaguid_ctrl_proc_read(char *page, char **num_items_returned,
1753                                   off_t items_to_skip, int char_count, int *eof,
1754                                   void *data)
1755 {
1756         char *outp = page;
1757         int len;
1758         uid_t uid;
1759         struct rb_node *node;
1760         struct sock_tag *sock_tag_entry;
1761         int item_index = 0;
1762         int indent_level = 0;
1763         long f_count;
1764
1765         if (unlikely(module_passive)) {
1766                 *eof = 1;
1767                 return 0;
1768         }
1769
1770         if (*eof)
1771                 return 0;
1772
1773         CT_DEBUG("qtaguid: proc ctrl page=%p off=%ld char_count=%d *eof=%d\n",
1774                 page, items_to_skip, char_count, *eof);
1775
1776         spin_lock_bh(&sock_tag_list_lock);
1777         for (node = rb_first(&sock_tag_tree);
1778              node;
1779              node = rb_next(node)) {
1780                 if (item_index++ < items_to_skip)
1781                         continue;
1782                 sock_tag_entry = rb_entry(node, struct sock_tag, sock_node);
1783                 uid = get_uid_from_tag(sock_tag_entry->tag);
1784                 CT_DEBUG("qtaguid: proc_read(): sk=%p tag=0x%llx (uid=%u) "
1785                          "pid=%u\n",
1786                          sock_tag_entry->sk,
1787                          sock_tag_entry->tag,
1788                          uid,
1789                          sock_tag_entry->pid
1790                         );
1791                 f_count = atomic_long_read(
1792                         &sock_tag_entry->socket->file->f_count);
1793                 len = snprintf(outp, char_count,
1794                                "sock=%p tag=0x%llx (uid=%u) pid=%u "
1795                                "f_count=%lu\n",
1796                                sock_tag_entry->sk,
1797                                sock_tag_entry->tag, uid,
1798                                sock_tag_entry->pid, f_count);
1799                 if (len >= char_count) {
1800                         spin_unlock_bh(&sock_tag_list_lock);
1801                         *outp = '\0';
1802                         return outp - page;
1803                 }
1804                 outp += len;
1805                 char_count -= len;
1806                 (*num_items_returned)++;
1807         }
1808         spin_unlock_bh(&sock_tag_list_lock);
1809
1810         if (item_index++ >= items_to_skip) {
1811                 len = snprintf(outp, char_count,
1812                                "events: sockets_tagged=%llu "
1813                                "sockets_untagged=%llu "
1814                                "counter_set_changes=%llu "
1815                                "delete_cmds=%llu "
1816                                "iface_events=%llu "
1817                                "match_calls=%llu "
1818                                "match_found_sk=%llu "
1819                                "match_found_sk_in_ct=%llu "
1820                                "match_found_no_sk_in_ct=%llu "
1821                                "match_no_sk=%llu "
1822                                "match_no_sk_file=%llu\n",
1823                                atomic64_read(&qtu_events.sockets_tagged),
1824                                atomic64_read(&qtu_events.sockets_untagged),
1825                                atomic64_read(&qtu_events.counter_set_changes),
1826                                atomic64_read(&qtu_events.delete_cmds),
1827                                atomic64_read(&qtu_events.iface_events),
1828                                atomic64_read(&qtu_events.match_calls),
1829                                atomic64_read(&qtu_events.match_found_sk),
1830                                atomic64_read(&qtu_events.match_found_sk_in_ct),
1831                                atomic64_read(
1832                                        &qtu_events.match_found_no_sk_in_ct),
1833                                atomic64_read(&qtu_events.match_no_sk),
1834                                atomic64_read(&qtu_events.match_no_sk_file));
1835                 if (len >= char_count) {
1836                         *outp = '\0';
1837                         return outp - page;
1838                 }
1839                 outp += len;
1840                 char_count -= len;
1841                 (*num_items_returned)++;
1842         }
1843
1844         /* Count the following as part of the last item_index */
1845         if (item_index > items_to_skip) {
1846                 prdebug_full_state(indent_level, "proc ctrl");
1847         }
1848
1849         *eof = 1;
1850         return outp - page;
1851 }
1852
1853 /*
1854  * Delete socket tags, and stat tags associated with a given
1855  * accouting tag and uid.
1856  */
1857 static int ctrl_cmd_delete(const char *input)
1858 {
1859         char cmd;
1860         uid_t uid;
1861         uid_t entry_uid;
1862         tag_t acct_tag;
1863         tag_t tag;
1864         int res, argc;
1865         struct iface_stat *iface_entry;
1866         struct rb_node *node;
1867         struct sock_tag *st_entry;
1868         struct rb_root st_to_free_tree = RB_ROOT;
1869         struct tag_stat *ts_entry;
1870         struct tag_counter_set *tcs_entry;
1871         struct tag_ref *tr_entry;
1872         struct uid_tag_data *utd_entry;
1873
1874         argc = sscanf(input, "%c %llu %u", &cmd, &acct_tag, &uid);
1875         CT_DEBUG("qtaguid: ctrl_delete(%s): argc=%d cmd=%c "
1876                  "user_tag=0x%llx uid=%u\n", input, argc, cmd,
1877                  acct_tag, uid);
1878         if (argc < 2) {
1879                 res = -EINVAL;
1880                 goto err;
1881         }
1882         if (!valid_atag(acct_tag)) {
1883                 pr_info("qtaguid: ctrl_delete(%s): invalid tag\n", input);
1884                 res = -EINVAL;
1885                 goto err;
1886         }
1887         if (argc < 3) {
1888                 uid = current_fsuid();
1889         } else if (!can_impersonate_uid(uid)) {
1890                 pr_info("qtaguid: ctrl_delete(%s): "
1891                         "insufficient priv from pid=%u tgid=%u uid=%u\n",
1892                         input, current->pid, current->tgid, current_fsuid());
1893                 res = -EPERM;
1894                 goto err;
1895         }
1896
1897         tag = combine_atag_with_uid(acct_tag, uid);
1898         CT_DEBUG("qtaguid: ctrl_delete(%s): "
1899                  "looking for tag=0x%llx (uid=%u)\n",
1900                  input, tag, uid);
1901
1902         /* Delete socket tags */
1903         spin_lock_bh(&sock_tag_list_lock);
1904         node = rb_first(&sock_tag_tree);
1905         while (node) {
1906                 st_entry = rb_entry(node, struct sock_tag, sock_node);
1907                 entry_uid = get_uid_from_tag(st_entry->tag);
1908                 node = rb_next(node);
1909                 if (entry_uid != uid)
1910                         continue;
1911
1912                 CT_DEBUG("qtaguid: ctrl_delete(%s): st tag=0x%llx (uid=%u)\n",
1913                          input, st_entry->tag, entry_uid);
1914
1915                 if (!acct_tag || st_entry->tag == tag) {
1916                         rb_erase(&st_entry->sock_node, &sock_tag_tree);
1917                         /* Can't sockfd_put() within spinlock, do it later. */
1918                         sock_tag_tree_insert(st_entry, &st_to_free_tree);
1919                         tr_entry = lookup_tag_ref(st_entry->tag, NULL);
1920                         BUG_ON(tr_entry->num_sock_tags <= 0);
1921                         tr_entry->num_sock_tags--;
1922                         /*
1923                          * TODO: remove if, and start failing.
1924                          * This is a hack to work around the fact that in some
1925                          * places we have "if (IS_ERR_OR_NULL(pqd_entry))"
1926                          * and are trying to work around apps
1927                          * that didn't open the /dev/xt_qtaguid.
1928                          */
1929                         if (st_entry->list.next && st_entry->list.prev)
1930                                 list_del(&st_entry->list);
1931                 }
1932         }
1933         spin_unlock_bh(&sock_tag_list_lock);
1934
1935         sock_tag_tree_erase(&st_to_free_tree);
1936
1937         /* Delete tag counter-sets */
1938         spin_lock_bh(&tag_counter_set_list_lock);
1939         /* Counter sets are only on the uid tag, not full tag */
1940         tcs_entry = tag_counter_set_tree_search(&tag_counter_set_tree, tag);
1941         if (tcs_entry) {
1942                 CT_DEBUG("qtaguid: ctrl_delete(%s): "
1943                          "erase tcs: tag=0x%llx (uid=%u) set=%d\n",
1944                          input,
1945                          tcs_entry->tn.tag,
1946                          get_uid_from_tag(tcs_entry->tn.tag),
1947                          tcs_entry->active_set);
1948                 rb_erase(&tcs_entry->tn.node, &tag_counter_set_tree);
1949                 kfree(tcs_entry);
1950         }
1951         spin_unlock_bh(&tag_counter_set_list_lock);
1952
1953         /*
1954          * If acct_tag is 0, then all entries belonging to uid are
1955          * erased.
1956          */
1957         spin_lock_bh(&iface_stat_list_lock);
1958         list_for_each_entry(iface_entry, &iface_stat_list, list) {
1959                 spin_lock_bh(&iface_entry->tag_stat_list_lock);
1960                 node = rb_first(&iface_entry->tag_stat_tree);
1961                 while (node) {
1962                         ts_entry = rb_entry(node, struct tag_stat, tn.node);
1963                         entry_uid = get_uid_from_tag(ts_entry->tn.tag);
1964                         node = rb_next(node);
1965
1966                         CT_DEBUG("qtaguid: ctrl_delete(%s): "
1967                                  "ts tag=0x%llx (uid=%u)\n",
1968                                  input, ts_entry->tn.tag, entry_uid);
1969
1970                         if (entry_uid != uid)
1971                                 continue;
1972                         if (!acct_tag || ts_entry->tn.tag == tag) {
1973                                 CT_DEBUG("qtaguid: ctrl_delete(%s): "
1974                                          "erase ts: %s 0x%llx %u\n",
1975                                          input, iface_entry->ifname,
1976                                          get_atag_from_tag(ts_entry->tn.tag),
1977                                          entry_uid);
1978                                 rb_erase(&ts_entry->tn.node,
1979                                          &iface_entry->tag_stat_tree);
1980                                 kfree(ts_entry);
1981                         }
1982                 }
1983                 spin_unlock_bh(&iface_entry->tag_stat_list_lock);
1984         }
1985         spin_unlock_bh(&iface_stat_list_lock);
1986
1987         /* Cleanup the uid_tag_data */
1988         spin_lock_bh(&uid_tag_data_tree_lock);
1989         node = rb_first(&uid_tag_data_tree);
1990         while (node) {
1991                 utd_entry = rb_entry(node, struct uid_tag_data, node);
1992                 entry_uid = utd_entry->uid;
1993                 node = rb_next(node);
1994
1995                 CT_DEBUG("qtaguid: ctrl_delete(%s): "
1996                          "utd uid=%u\n",
1997                          input, entry_uid);
1998
1999                 if (entry_uid != uid)
2000                         continue;
2001                 /*
2002                  * Go over the tag_refs, and those that don't have
2003                  * sock_tags using them are freed.
2004                  */
2005                 put_tag_ref_tree(tag, utd_entry);
2006                 put_utd_entry(utd_entry);
2007         }
2008         spin_unlock_bh(&uid_tag_data_tree_lock);
2009
2010         atomic64_inc(&qtu_events.delete_cmds);
2011         res = 0;
2012
2013 err:
2014         return res;
2015 }
2016
2017 static int ctrl_cmd_counter_set(const char *input)
2018 {
2019         char cmd;
2020         uid_t uid = 0;
2021         tag_t tag;
2022         int res, argc;
2023         struct tag_counter_set *tcs;
2024         int counter_set;
2025
2026         argc = sscanf(input, "%c %d %u", &cmd, &counter_set, &uid);
2027         CT_DEBUG("qtaguid: ctrl_counterset(%s): argc=%d cmd=%c "
2028                  "set=%d uid=%u\n", input, argc, cmd,
2029                  counter_set, uid);
2030         if (argc != 3) {
2031                 res = -EINVAL;
2032                 goto err;
2033         }
2034         if (counter_set < 0 || counter_set >= IFS_MAX_COUNTER_SETS) {
2035                 pr_info("qtaguid: ctrl_counterset(%s): invalid counter_set range\n",
2036                         input);
2037                 res = -EINVAL;
2038                 goto err;
2039         }
2040         if (!can_manipulate_uids()) {
2041                 pr_info("qtaguid: ctrl_counterset(%s): "
2042                         "insufficient priv from pid=%u tgid=%u uid=%u\n",
2043                         input, current->pid, current->tgid, current_fsuid());
2044                 res = -EPERM;
2045                 goto err;
2046         }
2047
2048         tag = make_tag_from_uid(uid);
2049         spin_lock_bh(&tag_counter_set_list_lock);
2050         tcs = tag_counter_set_tree_search(&tag_counter_set_tree, tag);
2051         if (!tcs) {
2052                 tcs = kzalloc(sizeof(*tcs), GFP_ATOMIC);
2053                 if (!tcs) {
2054                         spin_unlock_bh(&tag_counter_set_list_lock);
2055                         pr_err("qtaguid: ctrl_counterset(%s): "
2056                                "failed to alloc counter set\n",
2057                                input);
2058                         res = -ENOMEM;
2059                         goto err;
2060                 }
2061                 tcs->tn.tag = tag;
2062                 tag_counter_set_tree_insert(tcs, &tag_counter_set_tree);
2063                 CT_DEBUG("qtaguid: ctrl_counterset(%s): added tcs tag=0x%llx "
2064                          "(uid=%u) set=%d\n",
2065                          input, tag, get_uid_from_tag(tag), counter_set);
2066         }
2067         tcs->active_set = counter_set;
2068         spin_unlock_bh(&tag_counter_set_list_lock);
2069         atomic64_inc(&qtu_events.counter_set_changes);
2070         res = 0;
2071
2072 err:
2073         return res;
2074 }
2075
2076 static int ctrl_cmd_tag(const char *input)
2077 {
2078         char cmd;
2079         int sock_fd = 0;
2080         uid_t uid = 0;
2081         tag_t acct_tag = make_atag_from_value(0);
2082         tag_t full_tag;
2083         struct socket *el_socket;
2084         int res, argc;
2085         struct sock_tag *sock_tag_entry;
2086         struct tag_ref *tag_ref_entry;
2087         struct uid_tag_data *uid_tag_data_entry;
2088         struct proc_qtu_data *pqd_entry;
2089
2090         /* Unassigned args will get defaulted later. */
2091         argc = sscanf(input, "%c %d %llu %u", &cmd, &sock_fd, &acct_tag, &uid);
2092         CT_DEBUG("qtaguid: ctrl_tag(%s): argc=%d cmd=%c sock_fd=%d "
2093                  "acct_tag=0x%llx uid=%u\n", input, argc, cmd, sock_fd,
2094                  acct_tag, uid);
2095         if (argc < 2) {
2096                 res = -EINVAL;
2097                 goto err;
2098         }
2099         el_socket = sockfd_lookup(sock_fd, &res);  /* This locks the file */
2100         if (!el_socket) {
2101                 pr_info("qtaguid: ctrl_tag(%s): failed to lookup"
2102                         " sock_fd=%d err=%d\n", input, sock_fd, res);
2103                 goto err;
2104         }
2105         CT_DEBUG("qtaguid: ctrl_tag(%s): socket->...->f_count=%ld ->sk=%p\n",
2106                  input, atomic_long_read(&el_socket->file->f_count),
2107                  el_socket->sk);
2108         if (argc < 3) {
2109                 acct_tag = make_atag_from_value(0);
2110         } else if (!valid_atag(acct_tag)) {
2111                 pr_info("qtaguid: ctrl_tag(%s): invalid tag\n", input);
2112                 res = -EINVAL;
2113                 goto err_put;
2114         }
2115         CT_DEBUG("qtaguid: ctrl_tag(%s): "
2116                  "pid=%u tgid=%u uid=%u euid=%u fsuid=%u "
2117                  "in_group=%d in_egroup=%d\n",
2118                  input, current->pid, current->tgid, current_uid(),
2119                  current_euid(), current_fsuid(),
2120                  in_group_p(proc_ctrl_write_gid),
2121                  in_egroup_p(proc_ctrl_write_gid));
2122         if (argc < 4) {
2123                 uid = current_fsuid();
2124         } else if (!can_impersonate_uid(uid)) {
2125                 pr_info("qtaguid: ctrl_tag(%s): "
2126                         "insufficient priv from pid=%u tgid=%u uid=%u\n",
2127                         input, current->pid, current->tgid, current_fsuid());
2128                 res = -EPERM;
2129                 goto err_put;
2130         }
2131         full_tag = combine_atag_with_uid(acct_tag, uid);
2132
2133         spin_lock_bh(&sock_tag_list_lock);
2134         sock_tag_entry = get_sock_stat_nl(el_socket->sk);
2135         tag_ref_entry = get_tag_ref(full_tag, &uid_tag_data_entry);
2136         if (IS_ERR(tag_ref_entry)) {
2137                 res = PTR_ERR(tag_ref_entry);
2138                 spin_unlock_bh(&sock_tag_list_lock);
2139                 goto err_put;
2140         }
2141         tag_ref_entry->num_sock_tags++;
2142         if (sock_tag_entry) {
2143                 struct tag_ref *prev_tag_ref_entry;
2144
2145                 CT_DEBUG("qtaguid: ctrl_tag(%s): retag for sk=%p "
2146                          "st@%p ...->f_count=%ld\n",
2147                          input, el_socket->sk, sock_tag_entry,
2148                          atomic_long_read(&el_socket->file->f_count));
2149                 /*
2150                  * This is a re-tagging, so release the sock_fd that was
2151                  * locked at the time of the 1st tagging.
2152                  * There is still the ref from this call's sockfd_lookup() so
2153                  * it can be done within the spinlock.
2154                  */
2155                 sockfd_put(sock_tag_entry->socket);
2156                 prev_tag_ref_entry = lookup_tag_ref(sock_tag_entry->tag,
2157                                                     &uid_tag_data_entry);
2158                 BUG_ON(IS_ERR_OR_NULL(prev_tag_ref_entry));
2159                 BUG_ON(prev_tag_ref_entry->num_sock_tags <= 0);
2160                 prev_tag_ref_entry->num_sock_tags--;
2161                 sock_tag_entry->tag = full_tag;
2162         } else {
2163                 CT_DEBUG("qtaguid: ctrl_tag(%s): newtag for sk=%p\n",
2164                          input, el_socket->sk);
2165                 sock_tag_entry = kzalloc(sizeof(*sock_tag_entry),
2166                                          GFP_ATOMIC);
2167                 if (!sock_tag_entry) {
2168                         pr_err("qtaguid: ctrl_tag(%s): "
2169                                "socket tag alloc failed\n",
2170                                input);
2171                         spin_unlock_bh(&sock_tag_list_lock);
2172                         res = -ENOMEM;
2173                         goto err_tag_unref_put;
2174                 }
2175                 sock_tag_entry->sk = el_socket->sk;
2176                 sock_tag_entry->socket = el_socket;
2177                 sock_tag_entry->pid = current->tgid;
2178                 sock_tag_entry->tag = combine_atag_with_uid(acct_tag,
2179                                                             uid);
2180                 spin_lock_bh(&uid_tag_data_tree_lock);
2181                 pqd_entry = proc_qtu_data_tree_search(
2182                         &proc_qtu_data_tree, current->tgid);
2183                 /*
2184                  * TODO: remove if, and start failing.
2185                  * At first, we want to catch user-space code that is not
2186                  * opening the /dev/xt_qtaguid.
2187                  */
2188                 if (IS_ERR_OR_NULL(pqd_entry))
2189                         pr_warn_once(
2190                                 "qtaguid: %s(): "
2191                                 "User space forgot to open /dev/xt_qtaguid? "
2192                                 "pid=%u tgid=%u uid=%u\n", __func__,
2193                                 current->pid, current->tgid,
2194                                 current_fsuid());
2195                 else
2196                         list_add(&sock_tag_entry->list,
2197                                  &pqd_entry->sock_tag_list);
2198                 spin_unlock_bh(&uid_tag_data_tree_lock);
2199
2200                 sock_tag_tree_insert(sock_tag_entry, &sock_tag_tree);
2201                 atomic64_inc(&qtu_events.sockets_tagged);
2202         }
2203         spin_unlock_bh(&sock_tag_list_lock);
2204         /* We keep the ref to the socket (file) until it is untagged */
2205         CT_DEBUG("qtaguid: ctrl_tag(%s): done st@%p ...->f_count=%ld\n",
2206                  input, sock_tag_entry,
2207                  atomic_long_read(&el_socket->file->f_count));
2208         return 0;
2209
2210 err_tag_unref_put:
2211         BUG_ON(tag_ref_entry->num_sock_tags <= 0);
2212         tag_ref_entry->num_sock_tags--;
2213         free_tag_ref_from_utd_entry(tag_ref_entry, uid_tag_data_entry);
2214 err_put:
2215         CT_DEBUG("qtaguid: ctrl_tag(%s): done. ...->f_count=%ld\n",
2216                  input, atomic_long_read(&el_socket->file->f_count) - 1);
2217         /* Release the sock_fd that was grabbed by sockfd_lookup(). */
2218         sockfd_put(el_socket);
2219         return res;
2220
2221 err:
2222         CT_DEBUG("qtaguid: ctrl_tag(%s): done.\n", input);
2223         return res;
2224 }
2225
2226 static int ctrl_cmd_untag(const char *input)
2227 {
2228         char cmd;
2229         int sock_fd = 0;
2230         struct socket *el_socket;
2231         int res, argc;
2232         struct sock_tag *sock_tag_entry;
2233         struct tag_ref *tag_ref_entry;
2234         struct uid_tag_data *utd_entry;
2235         struct proc_qtu_data *pqd_entry;
2236
2237         argc = sscanf(input, "%c %d", &cmd, &sock_fd);
2238         CT_DEBUG("qtaguid: ctrl_untag(%s): argc=%d cmd=%c sock_fd=%d\n",
2239                  input, argc, cmd, sock_fd);
2240         if (argc < 2) {
2241                 res = -EINVAL;
2242                 goto err;
2243         }
2244         el_socket = sockfd_lookup(sock_fd, &res);  /* This locks the file */
2245         if (!el_socket) {
2246                 pr_info("qtaguid: ctrl_untag(%s): failed to lookup"
2247                         " sock_fd=%d err=%d\n", input, sock_fd, res);
2248                 goto err;
2249         }
2250         CT_DEBUG("qtaguid: ctrl_untag(%s): socket->...->f_count=%ld ->sk=%p\n",
2251                  input, atomic_long_read(&el_socket->file->f_count),
2252                  el_socket->sk);
2253         spin_lock_bh(&sock_tag_list_lock);
2254         sock_tag_entry = get_sock_stat_nl(el_socket->sk);
2255         if (!sock_tag_entry) {
2256                 spin_unlock_bh(&sock_tag_list_lock);
2257                 res = -EINVAL;
2258                 goto err_put;
2259         }
2260         /*
2261          * The socket already belongs to the current process
2262          * so it can do whatever it wants to it.
2263          */
2264         rb_erase(&sock_tag_entry->sock_node, &sock_tag_tree);
2265
2266         tag_ref_entry = lookup_tag_ref(sock_tag_entry->tag, &utd_entry);
2267         BUG_ON(!tag_ref_entry);
2268         BUG_ON(tag_ref_entry->num_sock_tags <= 0);
2269         spin_lock_bh(&uid_tag_data_tree_lock);
2270         pqd_entry = proc_qtu_data_tree_search(
2271                 &proc_qtu_data_tree, current->tgid);
2272         /*
2273          * TODO: remove if, and start failing.
2274          * At first, we want to catch user-space code that is not
2275          * opening the /dev/xt_qtaguid.
2276          */
2277         if (IS_ERR_OR_NULL(pqd_entry))
2278                 pr_warn_once("qtaguid: %s(): "
2279                              "User space forgot to open /dev/xt_qtaguid? "
2280                              "pid=%u tgid=%u uid=%u\n", __func__,
2281                              current->pid, current->tgid, current_fsuid());
2282         else
2283                 list_del(&sock_tag_entry->list);
2284         spin_unlock_bh(&uid_tag_data_tree_lock);
2285         /*
2286          * We don't free tag_ref from the utd_entry here,
2287          * only during a cmd_delete().
2288          */
2289         tag_ref_entry->num_sock_tags--;
2290         spin_unlock_bh(&sock_tag_list_lock);
2291         /*
2292          * Release the sock_fd that was grabbed at tag time,
2293          * and once more for the sockfd_lookup() here.
2294          */
2295         sockfd_put(sock_tag_entry->socket);
2296         CT_DEBUG("qtaguid: ctrl_untag(%s): done. st@%p ...->f_count=%ld\n",
2297                  input, sock_tag_entry,
2298                  atomic_long_read(&el_socket->file->f_count) - 1);
2299         sockfd_put(el_socket);
2300
2301         kfree(sock_tag_entry);
2302         atomic64_inc(&qtu_events.sockets_untagged);
2303
2304         return 0;
2305
2306 err_put:
2307         CT_DEBUG("qtaguid: ctrl_untag(%s): done. socket->...->f_count=%ld\n",
2308                  input, atomic_long_read(&el_socket->file->f_count) - 1);
2309         /* Release the sock_fd that was grabbed by sockfd_lookup(). */
2310         sockfd_put(el_socket);
2311         return res;
2312
2313 err:
2314         CT_DEBUG("qtaguid: ctrl_untag(%s): done.\n", input);
2315         return res;
2316 }
2317
2318 static int qtaguid_ctrl_parse(const char *input, int count)
2319 {
2320         char cmd;
2321         int res;
2322
2323         cmd = input[0];
2324         /* Collect params for commands */
2325         switch (cmd) {
2326         case 'd':
2327                 res = ctrl_cmd_delete(input);
2328                 break;
2329
2330         case 's':
2331                 res = ctrl_cmd_counter_set(input);
2332                 break;
2333
2334         case 't':
2335                 res = ctrl_cmd_tag(input);
2336                 break;
2337
2338         case 'u':
2339                 res = ctrl_cmd_untag(input);
2340                 break;
2341
2342         default:
2343                 res = -EINVAL;
2344                 goto err;
2345         }
2346         if (!res)
2347                 res = count;
2348 err:
2349         CT_DEBUG("qtaguid: ctrl(%s): res=%d\n", input, res);
2350         return res;
2351 }
2352
2353 #define MAX_QTAGUID_CTRL_INPUT_LEN 255
2354 static int qtaguid_ctrl_proc_write(struct file *file, const char __user *buffer,
2355                         unsigned long count, void *data)
2356 {
2357         char input_buf[MAX_QTAGUID_CTRL_INPUT_LEN];
2358
2359         if (unlikely(module_passive))
2360                 return count;
2361
2362         if (count >= MAX_QTAGUID_CTRL_INPUT_LEN)
2363                 return -EINVAL;
2364
2365         if (copy_from_user(input_buf, buffer, count))
2366                 return -EFAULT;
2367
2368         input_buf[count] = '\0';
2369         return qtaguid_ctrl_parse(input_buf, count);
2370 }
2371
2372 struct proc_print_info {
2373         char *outp;
2374         char **num_items_returned;
2375         struct iface_stat *iface_entry;
2376         struct tag_stat *ts_entry;
2377         int item_index;
2378         int items_to_skip;
2379         int char_count;
2380 };
2381
2382 static int pp_stats_line(struct proc_print_info *ppi, int cnt_set)
2383 {
2384         int len;
2385         struct data_counters *cnts;
2386
2387         if (!ppi->item_index) {
2388                 if (ppi->item_index++ < ppi->items_to_skip)
2389                         return 0;
2390                 len = snprintf(ppi->outp, ppi->char_count,
2391                                "idx iface acct_tag_hex uid_tag_int cnt_set "
2392                                "rx_bytes rx_packets "
2393                                "tx_bytes tx_packets "
2394                                "rx_tcp_bytes rx_tcp_packets "
2395                                "rx_udp_bytes rx_udp_packets "
2396                                "rx_other_bytes rx_other_packets "
2397                                "tx_tcp_bytes tx_tcp_packets "
2398                                "tx_udp_bytes tx_udp_packets "
2399                                "tx_other_bytes tx_other_packets\n");
2400         } else {
2401                 tag_t tag = ppi->ts_entry->tn.tag;
2402                 uid_t stat_uid = get_uid_from_tag(tag);
2403
2404                 if (!can_read_other_uid_stats(stat_uid)) {
2405                         CT_DEBUG("qtaguid: stats line: "
2406                                  "%s 0x%llx %u: insufficient priv "
2407                                  "from pid=%u tgid=%u uid=%u\n",
2408                                  ppi->iface_entry->ifname,
2409                                  get_atag_from_tag(tag), stat_uid,
2410                                  current->pid, current->tgid, current_fsuid());
2411                         return 0;
2412                 }
2413                 if (ppi->item_index++ < ppi->items_to_skip)
2414                         return 0;
2415                 cnts = &ppi->ts_entry->counters;
2416                 len = snprintf(
2417                         ppi->outp, ppi->char_count,
2418                         "%d %s 0x%llx %u %u "
2419                         "%llu %llu "
2420                         "%llu %llu "
2421                         "%llu %llu "
2422                         "%llu %llu "
2423                         "%llu %llu "
2424                         "%llu %llu "
2425                         "%llu %llu "
2426                         "%llu %llu\n",
2427                         ppi->item_index,
2428                         ppi->iface_entry->ifname,
2429                         get_atag_from_tag(tag),
2430                         stat_uid,
2431                         cnt_set,
2432                         dc_sum_bytes(cnts, cnt_set, IFS_RX),
2433                         dc_sum_packets(cnts, cnt_set, IFS_RX),
2434                         dc_sum_bytes(cnts, cnt_set, IFS_TX),
2435                         dc_sum_packets(cnts, cnt_set, IFS_TX),
2436                         cnts->bpc[cnt_set][IFS_RX][IFS_TCP].bytes,
2437                         cnts->bpc[cnt_set][IFS_RX][IFS_TCP].packets,
2438                         cnts->bpc[cnt_set][IFS_RX][IFS_UDP].bytes,
2439                         cnts->bpc[cnt_set][IFS_RX][IFS_UDP].packets,
2440                         cnts->bpc[cnt_set][IFS_RX][IFS_PROTO_OTHER].bytes,
2441                         cnts->bpc[cnt_set][IFS_RX][IFS_PROTO_OTHER].packets,
2442                         cnts->bpc[cnt_set][IFS_TX][IFS_TCP].bytes,
2443                         cnts->bpc[cnt_set][IFS_TX][IFS_TCP].packets,
2444                         cnts->bpc[cnt_set][IFS_TX][IFS_UDP].bytes,
2445                         cnts->bpc[cnt_set][IFS_TX][IFS_UDP].packets,
2446                         cnts->bpc[cnt_set][IFS_TX][IFS_PROTO_OTHER].bytes,
2447                         cnts->bpc[cnt_set][IFS_TX][IFS_PROTO_OTHER].packets);
2448         }
2449         return len;
2450 }
2451
2452 static bool pp_sets(struct proc_print_info *ppi)
2453 {
2454         int len;
2455         int counter_set;
2456         for (counter_set = 0; counter_set < IFS_MAX_COUNTER_SETS;
2457              counter_set++) {
2458                 len = pp_stats_line(ppi, counter_set);
2459                 if (len >= ppi->char_count) {
2460                         *ppi->outp = '\0';
2461                         return false;
2462                 }
2463                 if (len) {
2464                         ppi->outp += len;
2465                         ppi->char_count -= len;
2466                         (*ppi->num_items_returned)++;
2467                 }
2468         }
2469         return true;
2470 }
2471
2472 /*
2473  * Procfs reader to get all tag stats using style "1)" as described in
2474  * fs/proc/generic.c
2475  * Groups all protocols tx/rx bytes.
2476  */
2477 static int qtaguid_stats_proc_read(char *page, char **num_items_returned,
2478                                 off_t items_to_skip, int char_count, int *eof,
2479                                 void *data)
2480 {
2481         struct proc_print_info ppi;
2482         int len;
2483
2484         ppi.outp = page;
2485         ppi.item_index = 0;
2486         ppi.char_count = char_count;
2487         ppi.num_items_returned = num_items_returned;
2488         ppi.items_to_skip = items_to_skip;
2489
2490         if (unlikely(module_passive)) {
2491                 len = pp_stats_line(&ppi, 0);
2492                 /* The header should always be shorter than the buffer. */
2493                 BUG_ON(len >= ppi.char_count);
2494                 (*num_items_returned)++;
2495                 *eof = 1;
2496                 return len;
2497         }
2498
2499         CT_DEBUG("qtaguid:proc stats page=%p *num_items_returned=%p off=%ld "
2500                 "char_count=%d *eof=%d\n", page, *num_items_returned,
2501                 items_to_skip, char_count, *eof);
2502
2503         if (*eof)
2504                 return 0;
2505
2506         /* The idx is there to help debug when things go belly up. */
2507         len = pp_stats_line(&ppi, 0);
2508         /* Don't advance the outp unless the whole line was printed */
2509         if (len >= ppi.char_count) {
2510                 *ppi.outp = '\0';
2511                 return ppi.outp - page;
2512         }
2513         if (len) {
2514                 ppi.outp += len;
2515                 ppi.char_count -= len;
2516                 (*num_items_returned)++;
2517         }
2518
2519         spin_lock_bh(&iface_stat_list_lock);
2520         list_for_each_entry(ppi.iface_entry, &iface_stat_list, list) {
2521                 struct rb_node *node;
2522                 spin_lock_bh(&ppi.iface_entry->tag_stat_list_lock);
2523                 for (node = rb_first(&ppi.iface_entry->tag_stat_tree);
2524                      node;
2525                      node = rb_next(node)) {
2526                         ppi.ts_entry = rb_entry(node, struct tag_stat, tn.node);
2527                         if (!pp_sets(&ppi)) {
2528                                 spin_unlock_bh(
2529                                         &ppi.iface_entry->tag_stat_list_lock);
2530                                 spin_unlock_bh(&iface_stat_list_lock);
2531                                 return ppi.outp - page;
2532                         }
2533                 }
2534                 spin_unlock_bh(&ppi.iface_entry->tag_stat_list_lock);
2535         }
2536         spin_unlock_bh(&iface_stat_list_lock);
2537
2538         *eof = 1;
2539         return ppi.outp - page;
2540 }
2541
2542 /*------------------------------------------*/
2543 static int qtudev_open(struct inode *inode, struct file *file)
2544 {
2545         struct uid_tag_data *utd_entry;
2546         struct proc_qtu_data  *pqd_entry;
2547         struct proc_qtu_data  *new_pqd_entry;
2548         int res;
2549         bool utd_entry_found;
2550
2551         if (unlikely(qtu_proc_handling_passive))
2552                 return 0;
2553
2554         DR_DEBUG("qtaguid: qtudev_open(): pid=%u tgid=%u uid=%u\n",
2555                  current->pid, current->tgid, current_fsuid());
2556
2557         spin_lock_bh(&uid_tag_data_tree_lock);
2558
2559         /* Look for existing uid data, or alloc one. */
2560         utd_entry = get_uid_data(current_fsuid(), &utd_entry_found);
2561         if (IS_ERR_OR_NULL(utd_entry)) {
2562                 res = PTR_ERR(utd_entry);
2563                 goto err;
2564         }
2565
2566         /* Look for existing PID based proc_data */
2567         pqd_entry = proc_qtu_data_tree_search(&proc_qtu_data_tree,
2568                                               current->tgid);
2569         if (pqd_entry) {
2570                 pr_err("qtaguid: qtudev_open(): %u/%u %u "
2571                        "%s already opened\n",
2572                        current->pid, current->tgid, current_fsuid(),
2573                        QTU_DEV_NAME);
2574                 res = -EBUSY;
2575                 goto err_unlock_free_utd;
2576         }
2577
2578         new_pqd_entry = kzalloc(sizeof(*new_pqd_entry), GFP_ATOMIC);
2579         if (!new_pqd_entry) {
2580                 pr_err("qtaguid: qtudev_open(): %u/%u %u: "
2581                        "proc data alloc failed\n",
2582                        current->pid, current->tgid, current_fsuid());
2583                 res = -ENOMEM;
2584                 goto err_unlock_free_utd;
2585         }
2586         new_pqd_entry->pid = current->tgid;
2587         INIT_LIST_HEAD(&new_pqd_entry->sock_tag_list);
2588         new_pqd_entry->parent_tag_data = utd_entry;
2589         utd_entry->num_pqd++;
2590
2591         proc_qtu_data_tree_insert(new_pqd_entry,
2592                                   &proc_qtu_data_tree);
2593
2594         spin_unlock_bh(&uid_tag_data_tree_lock);
2595         DR_DEBUG("qtaguid: tracking data for uid=%u in pqd=%p\n",
2596                  current_fsuid(), new_pqd_entry);
2597         file->private_data = new_pqd_entry;
2598         return 0;
2599
2600 err_unlock_free_utd:
2601         if (!utd_entry_found) {
2602                 rb_erase(&utd_entry->node, &uid_tag_data_tree);
2603                 kfree(utd_entry);
2604         }
2605         spin_unlock_bh(&uid_tag_data_tree_lock);
2606 err:
2607         return res;
2608 }
2609
2610 static int qtudev_release(struct inode *inode, struct file *file)
2611 {
2612         struct proc_qtu_data  *pqd_entry = file->private_data;
2613         struct uid_tag_data  *utd_entry = pqd_entry->parent_tag_data;
2614         struct sock_tag *st_entry;
2615         struct rb_root st_to_free_tree = RB_ROOT;
2616         struct list_head *entry, *next;
2617         struct tag_ref *tr;
2618
2619         if (unlikely(qtu_proc_handling_passive))
2620                 return 0;
2621
2622         /*
2623          * Do not trust the current->pid, it might just be a kworker cleaning
2624          * up after a dead proc.
2625          */
2626         DR_DEBUG("qtaguid: qtudev_release(): "
2627                  "pid=%u tgid=%u uid=%u "
2628                  "pqd_entry=%p->pid=%u utd_entry=%p->active_tags=%d\n",
2629                  current->pid, current->tgid, pqd_entry->parent_tag_data->uid,
2630                  pqd_entry, pqd_entry->pid, utd_entry,
2631                  utd_entry->num_active_tags);
2632
2633         spin_lock_bh(&sock_tag_list_lock);
2634         spin_lock_bh(&uid_tag_data_tree_lock);
2635
2636         list_for_each_safe(entry, next, &pqd_entry->sock_tag_list) {
2637                 st_entry = list_entry(entry, struct sock_tag, list);
2638                 DR_DEBUG("qtaguid: %s(): "
2639                          "erase sock_tag=%p->sk=%p pid=%u tgid=%u uid=%u\n",
2640                          __func__,
2641                          st_entry, st_entry->sk,
2642                          current->pid, current->tgid,
2643                          pqd_entry->parent_tag_data->uid);
2644
2645                 utd_entry = uid_tag_data_tree_search(
2646                         &uid_tag_data_tree,
2647                         get_uid_from_tag(st_entry->tag));
2648                 BUG_ON(IS_ERR_OR_NULL(utd_entry));
2649                 DR_DEBUG("qtaguid: %s(): "
2650                          "looking for tag=0x%llx in utd_entry=%p\n", __func__,
2651                          st_entry->tag, utd_entry);
2652                 tr = tag_ref_tree_search(&utd_entry->tag_ref_tree,
2653                                          st_entry->tag);
2654                 BUG_ON(!tr);
2655                 BUG_ON(tr->num_sock_tags <= 0);
2656                 tr->num_sock_tags--;
2657                 free_tag_ref_from_utd_entry(tr, utd_entry);
2658
2659                 rb_erase(&st_entry->sock_node, &sock_tag_tree);
2660                 list_del(&st_entry->list);
2661                 /* Can't sockfd_put() within spinlock, do it later. */
2662                 sock_tag_tree_insert(st_entry, &st_to_free_tree);
2663
2664                 /*
2665                  * Try to free the utd_entry if no other proc_qtu_data is
2666                  * using it (num_pqd is 0) and it doesn't have active tags
2667                  * (num_active_tags is 0).
2668                  */
2669                 put_utd_entry(utd_entry);
2670         }
2671
2672         rb_erase(&pqd_entry->node, &proc_qtu_data_tree);
2673         BUG_ON(pqd_entry->parent_tag_data->num_pqd < 1);
2674         pqd_entry->parent_tag_data->num_pqd--;
2675         put_utd_entry(pqd_entry->parent_tag_data);
2676         kfree(pqd_entry);
2677         file->private_data = NULL;
2678
2679         spin_unlock_bh(&uid_tag_data_tree_lock);
2680         spin_unlock_bh(&sock_tag_list_lock);
2681
2682
2683         sock_tag_tree_erase(&st_to_free_tree);
2684
2685         prdebug_full_state(0, "%s(): pid=%u tgid=%u", __func__,
2686                            current->pid, current->tgid);
2687         return 0;
2688 }
2689
2690 /*------------------------------------------*/
2691 static const struct file_operations qtudev_fops = {
2692         .owner = THIS_MODULE,
2693         .open = qtudev_open,
2694         .release = qtudev_release,
2695 };
2696
2697 static struct miscdevice qtu_device = {
2698         .minor = MISC_DYNAMIC_MINOR,
2699         .name = QTU_DEV_NAME,
2700         .fops = &qtudev_fops,
2701         /* How sad it doesn't allow for defaults: .mode = S_IRUGO | S_IWUSR */
2702 };
2703
2704 /*------------------------------------------*/
2705 static int __init qtaguid_proc_register(struct proc_dir_entry **res_procdir)
2706 {
2707         int ret;
2708         *res_procdir = proc_mkdir(module_procdirname, init_net.proc_net);
2709         if (!*res_procdir) {
2710                 pr_err("qtaguid: failed to create proc/.../xt_qtaguid\n");
2711                 ret = -ENOMEM;
2712                 goto no_dir;
2713         }
2714
2715         xt_qtaguid_ctrl_file = create_proc_entry("ctrl", proc_ctrl_perms,
2716                                                 *res_procdir);
2717         if (!xt_qtaguid_ctrl_file) {
2718                 pr_err("qtaguid: failed to create xt_qtaguid/ctrl "
2719                         " file\n");
2720                 ret = -ENOMEM;
2721                 goto no_ctrl_entry;
2722         }
2723         xt_qtaguid_ctrl_file->read_proc = qtaguid_ctrl_proc_read;
2724         xt_qtaguid_ctrl_file->write_proc = qtaguid_ctrl_proc_write;
2725
2726         xt_qtaguid_stats_file = create_proc_entry("stats", proc_stats_perms,
2727                                                 *res_procdir);
2728         if (!xt_qtaguid_stats_file) {
2729                 pr_err("qtaguid: failed to create xt_qtaguid/stats "
2730                         "file\n");
2731                 ret = -ENOMEM;
2732                 goto no_stats_entry;
2733         }
2734         xt_qtaguid_stats_file->read_proc = qtaguid_stats_proc_read;
2735         /*
2736          * TODO: add support counter hacking
2737          * xt_qtaguid_stats_file->write_proc = qtaguid_stats_proc_write;
2738          */
2739         return 0;
2740
2741 no_stats_entry:
2742         remove_proc_entry("ctrl", *res_procdir);
2743 no_ctrl_entry:
2744         remove_proc_entry("xt_qtaguid", NULL);
2745 no_dir:
2746         return ret;
2747 }
2748
2749 static struct xt_match qtaguid_mt_reg __read_mostly = {
2750         /*
2751          * This module masquerades as the "owner" module so that iptables
2752          * tools can deal with it.
2753          */
2754         .name       = "owner",
2755         .revision   = 1,
2756         .family     = NFPROTO_UNSPEC,
2757         .match      = qtaguid_mt,
2758         .matchsize  = sizeof(struct xt_qtaguid_match_info),
2759         .me         = THIS_MODULE,
2760 };
2761
2762 static int __init qtaguid_mt_init(void)
2763 {
2764         if (qtaguid_proc_register(&xt_qtaguid_procdir)
2765             || iface_stat_init(xt_qtaguid_procdir)
2766             || xt_register_match(&qtaguid_mt_reg)
2767             || misc_register(&qtu_device))
2768                 return -1;
2769         return 0;
2770 }
2771
2772 /*
2773  * TODO: allow unloading of the module.
2774  * For now stats are permanent.
2775  * Kconfig forces'y/n' and never an 'm'.
2776  */
2777
2778 module_init(qtaguid_mt_init);
2779 MODULE_AUTHOR("jpa <jpa@google.com>");
2780 MODULE_DESCRIPTION("Xtables: socket owner+tag matching and associated stats");
2781 MODULE_LICENSE("GPL");
2782 MODULE_ALIAS("ipt_owner");
2783 MODULE_ALIAS("ip6t_owner");
2784 MODULE_ALIAS("ipt_qtaguid");
2785 MODULE_ALIAS("ip6t_qtaguid");