kernel/audit.c

   1 /* audit.c -- Auditing support
   2  * Gateway between the kernel (e.g., selinux) and the user-space audit daemon.
   3  * System-call specific features have moved to auditsc.c
   4  *
   5  * Copyright 2003-2004 Red Hat Inc., Durham, North Carolina.
   6  * All Rights Reserved.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 2 of the License, or
  11  * (at your option) any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * along with this program; if not, write to the Free Software
  20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  21  *
  22  * Written by Rickard E. (Rik) Faith <faith@redhat.com>
  23  *
  24  * Goals: 1) Integrate fully with SELinux.
  25  *        2) Minimal run-time overhead:
  26  *           a) Minimal when syscall auditing is disabled (audit_enable=0).
  27  *           b) Small when syscall auditing is enabled and no audit record
  28  *              is generated (defer as much work as possible to record
  29  *              generation time):
  30  *              i) context is allocated,
  31  *              ii) names from getname are stored without a copy, and
  32  *              iii) inode information stored from path_lookup.
  33  *        3) Ability to disable syscall auditing at boot time (audit=0).
  34  *        4) Usable by other parts of the kernel (if audit_log* is called,
  35  *           then a syscall record will be generated automatically for the
  36  *           current syscall).
  37  *        5) Netlink interface to user-space.
  38  *        6) Support low-overhead kernel-based filtering to minimize the
  39  *           information that must be passed to user-space.
  40  *
  41  * Example user-space utilities: http://people.redhat.com/sgrubb/audit/
  42  */
  43
  44 #include <linux/init.h>
  45 #include <asm/atomic.h>
  46 #include <asm/types.h>
  47 #include <linux/mm.h>
  48 #include <linux/module.h>
  49 #include <linux/err.h>
  50 #include <linux/kthread.h>
  51
  52 #include <linux/audit.h>
  53
  54 #include <net/sock.h>
  55 #include <linux/skbuff.h>
  56 #include <linux/netlink.h>
  57
  58 /* No auditing will take place until audit_initialized != 0.
  59  * (Initialization happens after skb_init is called.) */
  60 static int      audit_initialized;
  61
  62 /* No syscall auditing will take place unless audit_enabled != 0. */
  63 int             audit_enabled;
  64
  65 /* Default state when kernel boots without any parameters. */
  66 static int      audit_default;
  67
  68 /* If auditing cannot proceed, audit_failure selects what happens. */
  69 static int      audit_failure = AUDIT_FAIL_PRINTK;
  70
  71 /* If audit records are to be written to the netlink socket, audit_pid
  72  * contains the (non-zero) pid. */
  73 int             audit_pid;
  74
  75 /* If audit_limit is non-zero, limit the rate of sending audit records
  76  * to that number per second.  This prevents DoS attacks, but results in
  77  * audit records being dropped. */
  78 static int      audit_rate_limit;
  79
  80 /* Number of outstanding audit_buffers allowed. */
  81 static int      audit_backlog_limit = 64;
  82
  83 /* The identity of the user shutting down the audit system. */
  84 uid_t           audit_sig_uid = -1;
  85 pid_t           audit_sig_pid = -1;
  86
  87 /* Records can be lost in several ways:
  88    0) [suppressed in audit_alloc]
  89    1) out of memory in audit_log_start [kmalloc of struct audit_buffer]
  90    2) out of memory in audit_log_move [alloc_skb]
  91    3) suppressed due to audit_rate_limit
  92    4) suppressed due to audit_backlog_limit
  93 */
  94 static atomic_t    audit_lost = ATOMIC_INIT(0);
  95
  96 /* The netlink socket. */
  97 static struct sock *audit_sock;
  98
  99 /* The audit_freelist is a list of pre-allocated audit buffers (if more
 100  * than AUDIT_MAXFREE are in use, the audit buffer is freed instead of
 101  * being placed on the freelist). */
 102 static DEFINE_SPINLOCK(audit_freelist_lock);
 103 static int         audit_freelist_count = 0;
 104 static LIST_HEAD(audit_freelist);
 105
 106 static struct sk_buff_head audit_skb_queue;
 107 static struct task_struct *kauditd_task;
 108 static DECLARE_WAIT_QUEUE_HEAD(kauditd_wait);
 109
 110 /* There are three lists of rules -- one to search at task creation
 111  * time, one to search at syscall entry time, and another to search at
 112  * syscall exit time. */
 113 static LIST_HEAD(audit_tsklist);
 114 static LIST_HEAD(audit_entlist);
 115 static LIST_HEAD(audit_extlist);
 116
 117 /* The netlink socket is only to be read by 1 CPU, which lets us assume
 118  * that list additions and deletions never happen simultaneously in
 119  * auditsc.c */
 120 static DECLARE_MUTEX(audit_netlink_sem);
 121
 122 /* AUDIT_BUFSIZ is the size of the temporary buffer used for formatting
 123  * audit records.  Since printk uses a 1024 byte buffer, this buffer
 124  * should be at least that large. */
 125 #define AUDIT_BUFSIZ 1024
 126
 127 /* AUDIT_MAXFREE is the number of empty audit_buffers we keep on the
 128  * audit_freelist.  Doing so eliminates many kmalloc/kfree calls. */
 129 #define AUDIT_MAXFREE  (2*NR_CPUS)
 130
 131 /* The audit_buffer is used when formatting an audit record.  The caller
 132  * locks briefly to get the record off the freelist or to allocate the
 133  * buffer, and locks briefly to send the buffer to the netlink layer or
 134  * to place it on a transmit queue.  Multiple audit_buffers can be in
 135  * use simultaneously. */
 136 struct audit_buffer {
 137         struct list_head     list;
 138         struct sk_buff       *skb;      /* formatted skb ready to send */
 139         struct audit_context *ctx;      /* NULL or associated context */
 140 };
 141
 142 static void audit_set_pid(struct audit_buffer *ab, pid_t pid)
 143 {
 144         struct nlmsghdr *nlh = (struct nlmsghdr *)ab->skb->data;
 145         nlh->nlmsg_pid = pid;
 146 }
 147
 148 struct audit_entry {
 149         struct list_head  list;
 150         struct audit_rule rule;
 151 };
 152
 153 static void audit_panic(const char *message)
 154 {
 155         switch (audit_failure)
 156         {
 157         case AUDIT_FAIL_SILENT:
 158                 break;
 159         case AUDIT_FAIL_PRINTK:
 160                 printk(KERN_ERR "audit: %s\n", message);
 161                 break;
 162         case AUDIT_FAIL_PANIC:
 163                 panic("audit: %s\n", message);
 164                 break;
 165         }
 166 }
 167
 168 static inline int audit_rate_check(void)
 169 {
 170         static unsigned long    last_check = 0;
 171         static int              messages   = 0;
 172         static DEFINE_SPINLOCK(lock);
 173         unsigned long           flags;
 174         unsigned long           now;
 175         unsigned long           elapsed;
 176         int                     retval     = 0;
 177
 178         if (!audit_rate_limit) return 1;
 179
 180         spin_lock_irqsave(&lock, flags);
 181         if (++messages < audit_rate_limit) {
 182                 retval = 1;
 183         } else {
 184                 now     = jiffies;
 185                 elapsed = now - last_check;
 186                 if (elapsed > HZ) {
 187                         last_check = now;
 188                         messages   = 0;
 189                         retval     = 1;
 190                 }
 191         }
 192         spin_unlock_irqrestore(&lock, flags);
 193
 194         return retval;
 195 }
 196
 197 /* Emit at least 1 message per second, even if audit_rate_check is
 198  * throttling. */
 199 void audit_log_lost(const char *message)
 200 {
 201         static unsigned long    last_msg = 0;
 202         static DEFINE_SPINLOCK(lock);
 203         unsigned long           flags;
 204         unsigned long           now;
 205         int                     print;
 206
 207         atomic_inc(&audit_lost);
 208
 209         print = (audit_failure == AUDIT_FAIL_PANIC || !audit_rate_limit);
 210
 211         if (!print) {
 212                 spin_lock_irqsave(&lock, flags);
 213                 now = jiffies;
 214                 if (now - last_msg > HZ) {
 215                         print = 1;
 216                         last_msg = now;
 217                 }
 218                 spin_unlock_irqrestore(&lock, flags);
 219         }
 220
 221         if (print) {
 222                 printk(KERN_WARNING
 223                        "audit: audit_lost=%d audit_rate_limit=%d audit_backlog_limit=%d\n",
 224                        atomic_read(&audit_lost),
 225                        audit_rate_limit,
 226                        audit_backlog_limit);
 227                 audit_panic(message);
 228         }
 229
 230 }
 231
 232 static int audit_set_rate_limit(int limit, uid_t loginuid)
 233 {
 234         int old          = audit_rate_limit;
 235         audit_rate_limit = limit;
 236         audit_log(NULL, AUDIT_CONFIG_CHANGE,
 237                         "audit_rate_limit=%d old=%d by auid=%u",
 238                         audit_rate_limit, old, loginuid);
 239         return old;
 240 }
 241
 242 static int audit_set_backlog_limit(int limit, uid_t loginuid)
 243 {
 244         int old          = audit_backlog_limit;
 245         audit_backlog_limit = limit;
 246         audit_log(NULL, AUDIT_CONFIG_CHANGE,
 247                         "audit_backlog_limit=%d old=%d by auid=%u",
 248                         audit_backlog_limit, old, loginuid);
 249         return old;
 250 }
 251
 252 static int audit_set_enabled(int state, uid_t loginuid)
 253 {
 254         int old          = audit_enabled;
 255         if (state != 0 && state != 1)
 256                 return -EINVAL;
 257         audit_enabled = state;
 258         audit_log(NULL, AUDIT_CONFIG_CHANGE,
 259                         "audit_enabled=%d old=%d by auid=%u",
 260                         audit_enabled, old, loginuid);
 261         return old;
 262 }
 263
 264 static int audit_set_failure(int state, uid_t loginuid)
 265 {
 266         int old          = audit_failure;
 267         if (state != AUDIT_FAIL_SILENT
 268             && state != AUDIT_FAIL_PRINTK
 269             && state != AUDIT_FAIL_PANIC)
 270                 return -EINVAL;
 271         audit_failure = state;
 272         audit_log(NULL, AUDIT_CONFIG_CHANGE,
 273                         "audit_failure=%d old=%d by auid=%u",
 274                         audit_failure, old, loginuid);
 275         return old;
 276 }
 277
 278 int kauditd_thread(void *dummy)
 279 {
 280         struct sk_buff *skb;
 281
 282         while (1) {
 283                 skb = skb_dequeue(&audit_skb_queue);
 284                 if (skb) {
 285                         if (audit_pid) {
 286                                 int err = netlink_unicast(audit_sock, skb, audit_pid, 0);
 287                                 if (err < 0) {
 288                                         BUG_ON(err != -ECONNREFUSED); /* Shoudn't happen */
 289                                         printk(KERN_ERR "audit: *NO* daemon at audit_pid=%d\n", audit_pid);
 290                                         audit_pid = 0;
 291                                 }
 292                         } else {
 293                                 printk(KERN_ERR "%s\n", skb->data + NLMSG_SPACE(0));
 294                                 kfree_skb(skb);
 295                         }
 296                 } else {
 297                         DECLARE_WAITQUEUE(wait, current);
 298                         set_current_state(TASK_INTERRUPTIBLE);
 299                         add_wait_queue(&kauditd_wait, &wait);
 300
 301                         if (!skb_queue_len(&audit_skb_queue))
 302                                 schedule();
 303
 304                         __set_current_state(TASK_RUNNING);
 305                         remove_wait_queue(&kauditd_wait, &wait);
 306                 }
 307         }
 308 }
 309
 310 void audit_send_reply(int pid, int seq, int type, int done, int multi,
 311                       void *payload, int size)
 312 {
 313         struct sk_buff  *skb;
 314         struct nlmsghdr *nlh;
 315         int             len = NLMSG_SPACE(size);
 316         void            *data;
 317         int             flags = multi ? NLM_F_MULTI : 0;
 318         int             t     = done  ? NLMSG_DONE  : type;
 319
 320         skb = alloc_skb(len, GFP_KERNEL);
 321         if (!skb)
 322                 return;
 323
 324         nlh              = NLMSG_PUT(skb, pid, seq, t, size);
 325         nlh->nlmsg_flags = flags;
 326         data             = NLMSG_DATA(nlh);
 327         memcpy(data, payload, size);
 328
 329         /* Ignore failure. It'll only happen if the sender goes away,
 330            because our timeout is set to infinite. */
 331         netlink_unicast(audit_sock, skb, pid, 0);
 332         return;
 333
 334 nlmsg_failure:                  /* Used by NLMSG_PUT */
 335         if (skb)
 336                 kfree_skb(skb);
 337 }
 338
 339 /*
 340  * Check for appropriate CAP_AUDIT_ capabilities on incoming audit
 341  * control messages.
 342  */
 343 static int audit_netlink_ok(kernel_cap_t eff_cap, u16 msg_type)
 344 {
 345         int err = 0;
 346
 347         switch (msg_type) {
 348         case AUDIT_GET:
 349         case AUDIT_LIST:
 350         case AUDIT_SET:
 351         case AUDIT_ADD:
 352         case AUDIT_DEL:
 353         case AUDIT_SIGNAL_INFO:
 354                 if (!cap_raised(eff_cap, CAP_AUDIT_CONTROL))
 355                         err = -EPERM;
 356                 break;
 357         case AUDIT_USER:
 358         case AUDIT_FIRST_USER_MSG...AUDIT_LAST_USER_MSG:
 359                 if (!cap_raised(eff_cap, CAP_AUDIT_WRITE))
 360                         err = -EPERM;
 361                 break;
 362         default:  /* bad msg */
 363                 err = -EINVAL;
 364         }
 365
 366         return err;
 367 }
 368
 369 static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
 370 {
 371         u32                     uid, pid, seq;
 372         void                    *data;
 373         struct audit_status     *status_get, status_set;
 374         int                     err;
 375         struct audit_buffer     *ab;
 376         u16                     msg_type = nlh->nlmsg_type;
 377         uid_t                   loginuid; /* loginuid of sender */
 378         struct audit_sig_info   sig_data;
 379
 380         err = audit_netlink_ok(NETLINK_CB(skb).eff_cap, msg_type);
 381         if (err)
 382                 return err;
 383
 384         /* As soon as there's any sign of userspace auditd, start kauditd to talk to it */
 385         if (!kauditd_task)
 386                 kauditd_task = kthread_run(kauditd_thread, NULL, "kauditd");
 387         if (IS_ERR(kauditd_task)) {
 388                 err = PTR_ERR(kauditd_task);
 389                 kauditd_task = NULL;
 390                 return err;
 391         }
 392
 393         pid  = NETLINK_CREDS(skb)->pid;
 394         uid  = NETLINK_CREDS(skb)->uid;
 395         loginuid = NETLINK_CB(skb).loginuid;
 396         seq  = nlh->nlmsg_seq;
 397         data = NLMSG_DATA(nlh);
 398
 399         switch (msg_type) {
 400         case AUDIT_GET:
 401                 status_set.enabled       = audit_enabled;
 402                 status_set.failure       = audit_failure;
 403                 status_set.pid           = audit_pid;
 404                 status_set.rate_limit    = audit_rate_limit;
 405                 status_set.backlog_limit = audit_backlog_limit;
 406                 status_set.lost          = atomic_read(&audit_lost);
 407                 status_set.backlog       = skb_queue_len(&audit_skb_queue);
 408                 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_GET, 0, 0,
 409                                  &status_set, sizeof(status_set));
 410                 break;
 411         case AUDIT_SET:
 412                 if (nlh->nlmsg_len < sizeof(struct audit_status))
 413                         return -EINVAL;
 414                 status_get   = (struct audit_status *)data;
 415                 if (status_get->mask & AUDIT_STATUS_ENABLED) {
 416                         err = audit_set_enabled(status_get->enabled, loginuid);
 417                         if (err < 0) return err;
 418                 }
 419                 if (status_get->mask & AUDIT_STATUS_FAILURE) {
 420                         err = audit_set_failure(status_get->failure, loginuid);
 421                         if (err < 0) return err;
 422                 }
 423                 if (status_get->mask & AUDIT_STATUS_PID) {
 424                         int old   = audit_pid;
 425                         audit_pid = status_get->pid;
 426                         audit_log(NULL, AUDIT_CONFIG_CHANGE,
 427                                 "audit_pid=%d old=%d by auid=%u",
 428                                   audit_pid, old, loginuid);
 429                 }
 430                 if (status_get->mask & AUDIT_STATUS_RATE_LIMIT)
 431                         audit_set_rate_limit(status_get->rate_limit, loginuid);
 432                 if (status_get->mask & AUDIT_STATUS_BACKLOG_LIMIT)
 433                         audit_set_backlog_limit(status_get->backlog_limit,
 434                                                         loginuid);
 435                 break;
 436         case AUDIT_USER:
 437         case AUDIT_FIRST_USER_MSG...AUDIT_LAST_USER_MSG:
 438                 ab = audit_log_start(NULL, msg_type);
 439                 if (!ab)
 440                         break;  /* audit_panic has been called */
 441                 audit_log_format(ab,
 442                                  "user pid=%d uid=%u auid=%u"
 443                                  " msg='%.1024s'",
 444                                  pid, uid, loginuid, (char *)data);
 445                 audit_set_pid(ab, pid);
 446                 audit_log_end(ab);
 447                 break;
 448         case AUDIT_ADD:
 449         case AUDIT_DEL:
 450                 if (nlh->nlmsg_len < sizeof(struct audit_rule))
 451                         return -EINVAL;
 452                 /* fallthrough */
 453         case AUDIT_LIST:
 454                 err = audit_receive_filter(nlh->nlmsg_type, NETLINK_CB(skb).pid,
 455                                            uid, seq, data, loginuid);
 456                 break;
 457         case AUDIT_SIGNAL_INFO:
 458                 sig_data.uid = audit_sig_uid;
 459                 sig_data.pid = audit_sig_pid;
 460                 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_SIGNAL_INFO,
 461                                 0, 0, &sig_data, sizeof(sig_data));
 462                 break;
 463         default:
 464                 err = -EINVAL;
 465                 break;
 466         }
 467
 468         return err < 0 ? err : 0;
 469 }
 470
 471 /* Get message from skb (based on rtnetlink_rcv_skb).  Each message is
 472  * processed by audit_receive_msg.  Malformed skbs with wrong length are
 473  * discarded silently.  */
 474 static void audit_receive_skb(struct sk_buff *skb)
 475 {
 476         int             err;
 477         struct nlmsghdr *nlh;
 478         u32             rlen;
 479
 480         while (skb->len >= NLMSG_SPACE(0)) {
 481                 nlh = (struct nlmsghdr *)skb->data;
 482                 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
 483                         return;
 484                 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
 485                 if (rlen > skb->len)
 486                         rlen = skb->len;
 487                 if ((err = audit_receive_msg(skb, nlh))) {
 488                         netlink_ack(skb, nlh, err);
 489                 } else if (nlh->nlmsg_flags & NLM_F_ACK)
 490                         netlink_ack(skb, nlh, 0);
 491                 skb_pull(skb, rlen);
 492         }
 493 }
 494
 495 /* Receive messages from netlink socket. */
 496 static void audit_receive(struct sock *sk, int length)
 497 {
 498         struct sk_buff  *skb;
 499         unsigned int qlen;
 500
 501         down(&audit_netlink_sem);
 502
 503         for (qlen = skb_queue_len(&sk->sk_receive_queue); qlen; qlen--) {
 504                 skb = skb_dequeue(&sk->sk_receive_queue);
 505                 audit_receive_skb(skb);
 506                 kfree_skb(skb);
 507         }
 508         up(&audit_netlink_sem);
 509 }
 510
 511
 512 /* Initialize audit support at boot time. */
 513 static int __init audit_init(void)
 514 {
 515         printk(KERN_INFO "audit: initializing netlink socket (%s)\n",
 516                audit_default ? "enabled" : "disabled");
 517         audit_sock = netlink_kernel_create(NETLINK_AUDIT, audit_receive);
 518         if (!audit_sock)
 519                 audit_panic("cannot initialize netlink socket");
 520
 521         audit_sock->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
 522         skb_queue_head_init(&audit_skb_queue);
 523         audit_initialized = 1;
 524         audit_enabled = audit_default;
 525         audit_log(NULL, AUDIT_KERNEL, "initialized");
 526         return 0;
 527 }
 528 __initcall(audit_init);
 529
 530 /* Process kernel command-line parameter at boot time.  audit=0 or audit=1. */
 531 static int __init audit_enable(char *str)
 532 {
 533         audit_default = !!simple_strtol(str, NULL, 0);
 534         printk(KERN_INFO "audit: %s%s\n",
 535                audit_default ? "enabled" : "disabled",
 536                audit_initialized ? "" : " (after initialization)");
 537         if (audit_initialized)
 538                 audit_enabled = audit_default;
 539         return 0;
 540 }
 541
 542 __setup("audit=", audit_enable);
 543
 544 static void audit_buffer_free(struct audit_buffer *ab)
 545 {
 546         unsigned long flags;
 547
 548         if (!ab)
 549                 return;
 550
 551         if (ab->skb)
 552                 kfree_skb(ab->skb);
 553
 554         spin_lock_irqsave(&audit_freelist_lock, flags);
 555         if (++audit_freelist_count > AUDIT_MAXFREE)
 556                 kfree(ab);
 557         else
 558                 list_add(&ab->list, &audit_freelist);
 559         spin_unlock_irqrestore(&audit_freelist_lock, flags);
 560 }
 561
 562 static struct audit_buffer * audit_buffer_alloc(struct audit_context *ctx,
 563                                                 int gfp_mask, int type)
 564 {
 565         unsigned long flags;
 566         struct audit_buffer *ab = NULL;
 567         struct nlmsghdr *nlh;
 568
 569         spin_lock_irqsave(&audit_freelist_lock, flags);
 570         if (!list_empty(&audit_freelist)) {
 571                 ab = list_entry(audit_freelist.next,
 572                                 struct audit_buffer, list);
 573                 list_del(&ab->list);
 574                 --audit_freelist_count;
 575         }
 576         spin_unlock_irqrestore(&audit_freelist_lock, flags);
 577
 578         if (!ab) {
 579                 ab = kmalloc(sizeof(*ab), gfp_mask);
 580                 if (!ab)
 581                         goto err;
 582         }
 583
 584         ab->skb = alloc_skb(AUDIT_BUFSIZ, gfp_mask);
 585         if (!ab->skb)
 586                 goto err;
 587
 588         ab->ctx = ctx;
 589         nlh = (struct nlmsghdr *)skb_put(ab->skb, NLMSG_SPACE(0));
 590         nlh->nlmsg_type = type;
 591         nlh->nlmsg_flags = 0;
 592         nlh->nlmsg_pid = 0;
 593         nlh->nlmsg_seq = 0;
 594         return ab;
 595 err:
 596         audit_buffer_free(ab);
 597         return NULL;
 598 }
 599
 600 /* Compute a serial number for the audit record.  Audit records are
 601  * written to user-space as soon as they are generated, so a complete
 602  * audit record may be written in several pieces.  The timestamp of the
 603  * record and this serial number are used by the user-space tools to
 604  * determine which pieces belong to the same audit record.  The
 605  * (timestamp,serial) tuple is unique for each syscall and is live from
 606  * syscall entry to syscall exit.
 607  *
 608  * Atomic values are only guaranteed to be 24-bit, so we count down.
 609  *
 610  * NOTE: Another possibility is to store the formatted records off the
 611  * audit context (for those records that have a context), and emit them
 612  * all at syscall exit.  However, this could delay the reporting of
 613  * significant errors until syscall exit (or never, if the system
 614  * halts). */
 615 unsigned int audit_serial(void)
 616 {
 617         static atomic_t serial = ATOMIC_INIT(0xffffff);
 618         unsigned int a, b;
 619
 620         do {
 621                 a = atomic_read(&serial);
 622                 if (atomic_dec_and_test(&serial))
 623                         atomic_set(&serial, 0xffffff);
 624                 b = atomic_read(&serial);
 625         } while (b != a - 1);
 626
 627         return 0xffffff - b;
 628 }
 629
 630 static inline void audit_get_stamp(struct audit_context *ctx,
 631                                    struct timespec *t, unsigned int *serial)
 632 {
 633         if (ctx)
 634                 auditsc_get_stamp(ctx, t, serial);
 635         else {
 636                 *t = CURRENT_TIME;
 637                 *serial = audit_serial();
 638         }
 639 }
 640
 641 /* Obtain an audit buffer.  This routine does locking to obtain the
 642  * audit buffer, but then no locking is required for calls to
 643  * audit_log_*format.  If the tsk is a task that is currently in a
 644  * syscall, then the syscall is marked as auditable and an audit record
 645  * will be written at syscall exit.  If there is no associated task, tsk
 646  * should be NULL. */
 647 struct audit_buffer *audit_log_start(struct audit_context *ctx, int type)
 648 {
 649         struct audit_buffer     *ab     = NULL;
 650         struct timespec         t;
 651         unsigned int            serial;
 652
 653         if (!audit_initialized)
 654                 return NULL;
 655
 656         if (audit_backlog_limit
 657             && skb_queue_len(&audit_skb_queue) > audit_backlog_limit) {
 658                 if (audit_rate_check())
 659                         printk(KERN_WARNING
 660                                "audit: audit_backlog=%d > "
 661                                "audit_backlog_limit=%d\n",
 662                                skb_queue_len(&audit_skb_queue),
 663                                audit_backlog_limit);
 664                 audit_log_lost("backlog limit exceeded");
 665                 return NULL;
 666         }
 667
 668         ab = audit_buffer_alloc(ctx, GFP_ATOMIC, type);
 669         if (!ab) {
 670                 audit_log_lost("out of memory in audit_log_start");
 671                 return NULL;
 672         }
 673
 674         audit_get_stamp(ab->ctx, &t, &serial);
 675
 676         audit_log_format(ab, "audit(%lu.%03lu:%u): ",
 677                          t.tv_sec, t.tv_nsec/1000000, serial);
 678         return ab;
 679 }
 680
 681 /**
 682  * audit_expand - expand skb in the audit buffer
 683  * @ab: audit_buffer
 684  *
 685  * Returns 0 (no space) on failed expansion, or available space if
 686  * successful.
 687  */
 688 static inline int audit_expand(struct audit_buffer *ab, int extra)
 689 {
 690         struct sk_buff *skb = ab->skb;
 691         int ret = pskb_expand_head(skb, skb_headroom(skb), extra,
 692                                    GFP_ATOMIC);
 693         if (ret < 0) {
 694                 audit_log_lost("out of memory in audit_expand");
 695                 return 0;
 696         }
 697         return skb_tailroom(skb);
 698 }
 699
 700 /* Format an audit message into the audit buffer.  If there isn't enough
 701  * room in the audit buffer, more room will be allocated and vsnprint
 702  * will be called a second time.  Currently, we assume that a printk
 703  * can't format message larger than 1024 bytes, so we don't either. */
 704 static void audit_log_vformat(struct audit_buffer *ab, const char *fmt,
 705                               va_list args)
 706 {
 707         int len, avail;
 708         struct sk_buff *skb;
 709         va_list args2;
 710
 711         if (!ab)
 712                 return;
 713
 714         BUG_ON(!ab->skb);
 715         skb = ab->skb;
 716         avail = skb_tailroom(skb);
 717         if (avail == 0) {
 718                 avail = audit_expand(ab, AUDIT_BUFSIZ);
 719                 if (!avail)
 720                         goto out;
 721         }
 722         va_copy(args2, args);
 723         len = vsnprintf(skb->tail, avail, fmt, args);
 724         if (len >= avail) {
 725                 /* The printk buffer is 1024 bytes long, so if we get
 726                  * here and AUDIT_BUFSIZ is at least 1024, then we can
 727                  * log everything that printk could have logged. */
 728                 avail = audit_expand(ab, max_t(unsigned, AUDIT_BUFSIZ, 1+len-avail));
 729                 if (!avail)
 730                         goto out;
 731                 len = vsnprintf(skb->tail, avail, fmt, args2);
 732         }
 733         if (len > 0)
 734                 skb_put(skb, len);
 735 out:
 736         return;
 737 }
 738
 739 /* Format a message into the audit buffer.  All the work is done in
 740  * audit_log_vformat. */
 741 void audit_log_format(struct audit_buffer *ab, const char *fmt, ...)
 742 {
 743         va_list args;
 744
 745         if (!ab)
 746                 return;
 747         va_start(args, fmt);
 748         audit_log_vformat(ab, fmt, args);
 749         va_end(args);
 750 }
 751
 752 /* This function will take the passed buf and convert it into a string of
 753  * ascii hex digits. The new string is placed onto the skb. */
 754 void audit_log_hex(struct audit_buffer *ab, const unsigned char *buf,
 755                 size_t len)
 756 {
 757         int i, avail, new_len;
 758         unsigned char *ptr;
 759         struct sk_buff *skb;
 760         static const unsigned char *hex = "0123456789ABCDEF";
 761
 762         BUG_ON(!ab->skb);
 763         skb = ab->skb;
 764         avail = skb_tailroom(skb);
 765         new_len = len<<1;
 766         if (new_len >= avail) {
 767                 /* Round the buffer request up to the next multiple */
 768                 new_len = AUDIT_BUFSIZ*(((new_len-avail)/AUDIT_BUFSIZ) + 1);
 769                 avail = audit_expand(ab, new_len);
 770                 if (!avail)
 771                         return;
 772         }
 773
 774         ptr = skb->tail;
 775         for (i=0; i<len; i++) {
 776                 *ptr++ = hex[(buf[i] & 0xF0)>>4]; /* Upper nibble */
 777                 *ptr++ = hex[buf[i] & 0x0F];      /* Lower nibble */
 778         }
 779         *ptr = 0;
 780         skb_put(skb, len << 1); /* new string is twice the old string */
 781 }
 782
 783 /* This code will escape a string that is passed to it if the string
 784  * contains a control character, unprintable character, double quote mark,
 785  * or a space. Unescaped strings will start and end with a double quote mark.
 786  * Strings that are escaped are printed in hex (2 digits per char). */
 787 void audit_log_untrustedstring(struct audit_buffer *ab, const char *string)
 788 {
 789         const unsigned char *p = string;
 790
 791         while (*p) {
 792                 if (*p == '"' || *p < 0x21 || *p > 0x7f) {
 793                         audit_log_hex(ab, string, strlen(string));
 794                         return;
 795                 }
 796                 p++;
 797         }
 798         audit_log_format(ab, "\"%s\"", string);
 799 }
 800
 801 /* This is a helper-function to print the escaped d_path */
 802 void audit_log_d_path(struct audit_buffer *ab, const char *prefix,
 803                       struct dentry *dentry, struct vfsmount *vfsmnt)
 804 {
 805         char *p, *path;
 806
 807         if (prefix)
 808                 audit_log_format(ab, " %s", prefix);
 809
 810         /* We will allow 11 spaces for ' (deleted)' to be appended */
 811         path = kmalloc(PATH_MAX+11, GFP_KERNEL);
 812         if (!path) {
 813                 audit_log_format(ab, "<no memory>");
 814                 return;
 815         }
 816         p = d_path(dentry, vfsmnt, path, PATH_MAX+11);
 817         if (IS_ERR(p)) { /* Should never happen since we send PATH_MAX */
 818                 /* FIXME: can we save some information here? */
 819                 audit_log_format(ab, "<too long>");
 820         } else
 821                 audit_log_untrustedstring(ab, p);
 822         kfree(path);
 823 }
 824
 825 /* The netlink_* functions cannot be called inside an irq context, so
 826  * the audit buffer is places on a queue and a tasklet is scheduled to
 827  * remove them from the queue outside the irq context.  May be called in
 828  * any context. */
 829 void audit_log_end(struct audit_buffer *ab)
 830 {
 831         if (!ab)
 832                 return;
 833         if (!audit_rate_check()) {
 834                 audit_log_lost("rate limit exceeded");
 835         } else {
 836                 if (audit_pid) {
 837                         struct nlmsghdr *nlh = (struct nlmsghdr *)ab->skb->data;
 838                         nlh->nlmsg_len = ab->skb->len - NLMSG_SPACE(0);
 839                         skb_queue_tail(&audit_skb_queue, ab->skb);
 840                         ab->skb = NULL;
 841                         wake_up_interruptible(&kauditd_wait);
 842                 } else {
 843                         printk("%s\n", ab->skb->data + NLMSG_SPACE(0));
 844                 }
 845         }
 846         audit_buffer_free(ab);
 847 }
 848
 849 /* Log an audit record.  This is a convenience function that calls
 850  * audit_log_start, audit_log_vformat, and audit_log_end.  It may be
 851  * called in any context. */
 852 void audit_log(struct audit_context *ctx, int type, const char *fmt, ...)
 853 {
 854         struct audit_buffer *ab;
 855         va_list args;
 856
 857         ab = audit_log_start(ctx, type);
 858         if (ab) {
 859                 va_start(args, fmt);
 860                 audit_log_vformat(ab, fmt, args);
 861                 va_end(args);
 862                 audit_log_end(ab);
 863         }
 864 }