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
  50 #include <linux/audit.h>
  51
  52 #include <net/sock.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netlink.h>
  55
  56 /* No auditing will take place until audit_initialized != 0.
  57  * (Initialization happens after skb_init is called.) */
  58 static int      audit_initialized;
  59
  60 /* No syscall auditing will take place unless audit_enabled != 0. */
  61 int             audit_enabled;
  62
  63 /* Default state when kernel boots without any parameters. */
  64 static int      audit_default;
  65
  66 /* If auditing cannot proceed, audit_failure selects what happens. */
  67 static int      audit_failure = AUDIT_FAIL_PRINTK;
  68
  69 /* If audit records are to be written to the netlink socket, audit_pid
  70  * contains the (non-zero) pid. */
  71 static int      audit_pid;
  72
  73 /* If audit_limit is non-zero, limit the rate of sending audit records
  74  * to that number per second.  This prevents DoS attacks, but results in
  75  * audit records being dropped. */
  76 static int      audit_rate_limit;
  77
  78 /* Number of outstanding audit_buffers allowed. */
  79 static int      audit_backlog_limit = 64;
  80 static atomic_t audit_backlog       = ATOMIC_INIT(0);
  81
  82 /* Records can be lost in several ways:
  83    0) [suppressed in audit_alloc]
  84    1) out of memory in audit_log_start [kmalloc of struct audit_buffer]
  85    2) out of memory in audit_log_move [alloc_skb]
  86    3) suppressed due to audit_rate_limit
  87    4) suppressed due to audit_backlog_limit
  88 */
  89 static atomic_t    audit_lost = ATOMIC_INIT(0);
  90
  91 /* The netlink socket. */
  92 static struct sock *audit_sock;
  93
  94 /* There are two lists of audit buffers.  The txlist contains audit
  95  * buffers that cannot be sent immediately to the netlink device because
  96  * we are in an irq context (these are sent later in a tasklet).
  97  *
  98  * The second list is a list of pre-allocated audit buffers (if more
  99  * than AUDIT_MAXFREE are in use, the audit buffer is freed instead of
 100  * being placed on the freelist). */
 101 static DEFINE_SPINLOCK(audit_txlist_lock);
 102 static DEFINE_SPINLOCK(audit_freelist_lock);
 103 static int         audit_freelist_count = 0;
 104 static LIST_HEAD(audit_txlist);
 105 static LIST_HEAD(audit_freelist);
 106
 107 /* There are three lists of rules -- one to search at task creation
 108  * time, one to search at syscall entry time, and another to search at
 109  * syscall exit time. */
 110 static LIST_HEAD(audit_tsklist);
 111 static LIST_HEAD(audit_entlist);
 112 static LIST_HEAD(audit_extlist);
 113
 114 /* The netlink socket is only to be read by 1 CPU, which lets us assume
 115  * that list additions and deletions never happen simultaneiously in
 116  * auditsc.c */
 117 static DECLARE_MUTEX(audit_netlink_sem);
 118
 119 /* AUDIT_BUFSIZ is the size of the temporary buffer used for formatting
 120  * audit records.  Since printk uses a 1024 byte buffer, this buffer
 121  * should be at least that large. */
 122 #define AUDIT_BUFSIZ 1024
 123
 124 /* AUDIT_MAXFREE is the number of empty audit_buffers we keep on the
 125  * audit_freelist.  Doing so eliminates many kmalloc/kfree calls. */
 126 #define AUDIT_MAXFREE  (2*NR_CPUS)
 127
 128 /* The audit_buffer is used when formatting an audit record.  The caller
 129  * locks briefly to get the record off the freelist or to allocate the
 130  * buffer, and locks briefly to send the buffer to the netlink layer or
 131  * to place it on a transmit queue.  Multiple audit_buffers can be in
 132  * use simultaneously. */
 133 struct audit_buffer {
 134         struct list_head     list;
 135         struct sk_buff_head  sklist;    /* formatted skbs ready to send */
 136         struct audit_context *ctx;      /* NULL or associated context */
 137         int                  len;       /* used area of tmp */
 138         char                 tmp[AUDIT_BUFSIZ];
 139
 140                                 /* Pointer to header and contents */
 141         struct nlmsghdr      *nlh;
 142         int                  total;
 143         int                  type;
 144         int                  pid;
 145 };
 146
 147 void audit_set_type(struct audit_buffer *ab, int type)
 148 {
 149         ab->type = type;
 150 }
 151
 152 struct audit_entry {
 153         struct list_head  list;
 154         struct audit_rule rule;
 155 };
 156
 157 static void audit_log_end_irq(struct audit_buffer *ab);
 158 static void audit_log_end_fast(struct audit_buffer *ab);
 159
 160 static void audit_panic(const char *message)
 161 {
 162         switch (audit_failure)
 163         {
 164         case AUDIT_FAIL_SILENT:
 165                 break;
 166         case AUDIT_FAIL_PRINTK:
 167                 printk(KERN_ERR "audit: %s\n", message);
 168                 break;
 169         case AUDIT_FAIL_PANIC:
 170                 panic("audit: %s\n", message);
 171                 break;
 172         }
 173 }
 174
 175 static inline int audit_rate_check(void)
 176 {
 177         static unsigned long    last_check = 0;
 178         static int              messages   = 0;
 179         static DEFINE_SPINLOCK(lock);
 180         unsigned long           flags;
 181         unsigned long           now;
 182         unsigned long           elapsed;
 183         int                     retval     = 0;
 184
 185         if (!audit_rate_limit) return 1;
 186
 187         spin_lock_irqsave(&lock, flags);
 188         if (++messages < audit_rate_limit) {
 189                 retval = 1;
 190         } else {
 191                 now     = jiffies;
 192                 elapsed = now - last_check;
 193                 if (elapsed > HZ) {
 194                         last_check = now;
 195                         messages   = 0;
 196                         retval     = 1;
 197                 }
 198         }
 199         spin_unlock_irqrestore(&lock, flags);
 200
 201         return retval;
 202 }
 203
 204 /* Emit at least 1 message per second, even if audit_rate_check is
 205  * throttling. */
 206 void audit_log_lost(const char *message)
 207 {
 208         static unsigned long    last_msg = 0;
 209         static DEFINE_SPINLOCK(lock);
 210         unsigned long           flags;
 211         unsigned long           now;
 212         int                     print;
 213
 214         atomic_inc(&audit_lost);
 215
 216         print = (audit_failure == AUDIT_FAIL_PANIC || !audit_rate_limit);
 217
 218         if (!print) {
 219                 spin_lock_irqsave(&lock, flags);
 220                 now = jiffies;
 221                 if (now - last_msg > HZ) {
 222                         print = 1;
 223                         last_msg = now;
 224                 }
 225                 spin_unlock_irqrestore(&lock, flags);
 226         }
 227
 228         if (print) {
 229                 printk(KERN_WARNING
 230                        "audit: audit_lost=%d audit_backlog=%d"
 231                        " audit_rate_limit=%d audit_backlog_limit=%d\n",
 232                        atomic_read(&audit_lost),
 233                        atomic_read(&audit_backlog),
 234                        audit_rate_limit,
 235                        audit_backlog_limit);
 236                 audit_panic(message);
 237         }
 238
 239 }
 240
 241 static int audit_set_rate_limit(int limit, uid_t loginuid)
 242 {
 243         int old          = audit_rate_limit;
 244         audit_rate_limit = limit;
 245         audit_log(NULL, "audit_rate_limit=%d old=%d by auid %u",
 246                         audit_rate_limit, old, loginuid);
 247         return old;
 248 }
 249
 250 static int audit_set_backlog_limit(int limit, uid_t loginuid)
 251 {
 252         int old          = audit_backlog_limit;
 253         audit_backlog_limit = limit;
 254         audit_log(NULL, "audit_backlog_limit=%d old=%d by auid %u",
 255                         audit_backlog_limit, old, loginuid);
 256         return old;
 257 }
 258
 259 static int audit_set_enabled(int state, uid_t loginuid)
 260 {
 261         int old          = audit_enabled;
 262         if (state != 0 && state != 1)
 263                 return -EINVAL;
 264         audit_enabled = state;
 265         audit_log(NULL, "audit_enabled=%d old=%d by auid %u",
 266                   audit_enabled, old, loginuid);
 267         return old;
 268 }
 269
 270 static int audit_set_failure(int state, uid_t loginuid)
 271 {
 272         int old          = audit_failure;
 273         if (state != AUDIT_FAIL_SILENT
 274             && state != AUDIT_FAIL_PRINTK
 275             && state != AUDIT_FAIL_PANIC)
 276                 return -EINVAL;
 277         audit_failure = state;
 278         audit_log(NULL, "audit_failure=%d old=%d by auid %u",
 279                   audit_failure, old, loginuid);
 280         return old;
 281 }
 282
 283 #ifdef CONFIG_NET
 284 void audit_send_reply(int pid, int seq, int type, int done, int multi,
 285                       void *payload, int size)
 286 {
 287         struct sk_buff  *skb;
 288         struct nlmsghdr *nlh;
 289         int             len = NLMSG_SPACE(size);
 290         void            *data;
 291         int             flags = multi ? NLM_F_MULTI : 0;
 292         int             t     = done  ? NLMSG_DONE  : type;
 293
 294         skb = alloc_skb(len, GFP_KERNEL);
 295         if (!skb)
 296                 goto nlmsg_failure;
 297
 298         nlh              = NLMSG_PUT(skb, pid, seq, t, len - sizeof(*nlh));
 299         nlh->nlmsg_flags = flags;
 300         data             = NLMSG_DATA(nlh);
 301         memcpy(data, payload, size);
 302         netlink_unicast(audit_sock, skb, pid, MSG_DONTWAIT);
 303         return;
 304
 305 nlmsg_failure:                  /* Used by NLMSG_PUT */
 306         if (skb)
 307                 kfree_skb(skb);
 308 }
 309
 310 /*
 311  * Check for appropriate CAP_AUDIT_ capabilities on incoming audit
 312  * control messages.
 313  */
 314 static int audit_netlink_ok(kernel_cap_t eff_cap, u16 msg_type)
 315 {
 316         int err = 0;
 317
 318         switch (msg_type) {
 319         case AUDIT_GET:
 320         case AUDIT_LIST:
 321         case AUDIT_SET:
 322         case AUDIT_ADD:
 323         case AUDIT_DEL:
 324                 if (!cap_raised(eff_cap, CAP_AUDIT_CONTROL))
 325                         err = -EPERM;
 326                 break;
 327         case AUDIT_USER:
 328                 if (!cap_raised(eff_cap, CAP_AUDIT_WRITE))
 329                         err = -EPERM;
 330                 break;
 331         default:  /* bad msg */
 332                 err = -EINVAL;
 333         }
 334
 335         return err;
 336 }
 337
 338 static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
 339 {
 340         u32                     uid, pid, seq;
 341         void                    *data;
 342         struct audit_status     *status_get, status_set;
 343         int                     err;
 344         struct audit_buffer     *ab;
 345         u16                     msg_type = nlh->nlmsg_type;
 346         uid_t                   loginuid; /* loginuid of sender */
 347
 348         err = audit_netlink_ok(NETLINK_CB(skb).eff_cap, msg_type);
 349         if (err)
 350                 return err;
 351
 352         pid  = NETLINK_CREDS(skb)->pid;
 353         uid  = NETLINK_CREDS(skb)->uid;
 354         loginuid = NETLINK_CB(skb).loginuid;
 355         seq  = nlh->nlmsg_seq;
 356         data = NLMSG_DATA(nlh);
 357
 358         switch (msg_type) {
 359         case AUDIT_GET:
 360                 status_set.enabled       = audit_enabled;
 361                 status_set.failure       = audit_failure;
 362                 status_set.pid           = audit_pid;
 363                 status_set.rate_limit    = audit_rate_limit;
 364                 status_set.backlog_limit = audit_backlog_limit;
 365                 status_set.lost          = atomic_read(&audit_lost);
 366                 status_set.backlog       = atomic_read(&audit_backlog);
 367                 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_GET, 0, 0,
 368                                  &status_set, sizeof(status_set));
 369                 break;
 370         case AUDIT_SET:
 371                 if (nlh->nlmsg_len < sizeof(struct audit_status))
 372                         return -EINVAL;
 373                 status_get   = (struct audit_status *)data;
 374                 if (status_get->mask & AUDIT_STATUS_ENABLED) {
 375                         err = audit_set_enabled(status_get->enabled, loginuid);
 376                         if (err < 0) return err;
 377                 }
 378                 if (status_get->mask & AUDIT_STATUS_FAILURE) {
 379                         err = audit_set_failure(status_get->failure, loginuid);
 380                         if (err < 0) return err;
 381                 }
 382                 if (status_get->mask & AUDIT_STATUS_PID) {
 383                         int old   = audit_pid;
 384                         audit_pid = status_get->pid;
 385                         audit_log(NULL, "audit_pid=%d old=%d by auid %u",
 386                                   audit_pid, old, loginuid);
 387                 }
 388                 if (status_get->mask & AUDIT_STATUS_RATE_LIMIT)
 389                         audit_set_rate_limit(status_get->rate_limit, loginuid);
 390                 if (status_get->mask & AUDIT_STATUS_BACKLOG_LIMIT)
 391                         audit_set_backlog_limit(status_get->backlog_limit,
 392                                                         loginuid);
 393                 break;
 394         case AUDIT_USER:
 395                 ab = audit_log_start(NULL);
 396                 if (!ab)
 397                         break;  /* audit_panic has been called */
 398                 audit_log_format(ab,
 399                                  "user pid=%d uid=%d length=%d loginuid=%u"
 400                                  " msg='%.1024s'",
 401                                  pid, uid,
 402                                  (int)(nlh->nlmsg_len
 403                                        - ((char *)data - (char *)nlh)),
 404                                  loginuid, (char *)data);
 405                 ab->type = AUDIT_USER;
 406                 ab->pid  = pid;
 407                 audit_log_end(ab);
 408                 break;
 409         case AUDIT_ADD:
 410         case AUDIT_DEL:
 411                 if (nlh->nlmsg_len < sizeof(struct audit_rule))
 412                         return -EINVAL;
 413                 /* fallthrough */
 414         case AUDIT_LIST:
 415 #ifdef CONFIG_AUDITSYSCALL
 416                 err = audit_receive_filter(nlh->nlmsg_type, NETLINK_CB(skb).pid,
 417                                            uid, seq, data, loginuid);
 418 #else
 419                 err = -EOPNOTSUPP;
 420 #endif
 421                 break;
 422         default:
 423                 err = -EINVAL;
 424                 break;
 425         }
 426
 427         return err < 0 ? err : 0;
 428 }
 429
 430 /* Get message from skb (based on rtnetlink_rcv_skb).  Each message is
 431  * processed by audit_receive_msg.  Malformed skbs with wrong length are
 432  * discarded silently.  */
 433 static void audit_receive_skb(struct sk_buff *skb)
 434 {
 435         int             err;
 436         struct nlmsghdr *nlh;
 437         u32             rlen;
 438
 439         while (skb->len >= NLMSG_SPACE(0)) {
 440                 nlh = (struct nlmsghdr *)skb->data;
 441                 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
 442                         return;
 443                 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
 444                 if (rlen > skb->len)
 445                         rlen = skb->len;
 446                 if ((err = audit_receive_msg(skb, nlh))) {
 447                         netlink_ack(skb, nlh, err);
 448                 } else if (nlh->nlmsg_flags & NLM_F_ACK)
 449                         netlink_ack(skb, nlh, 0);
 450                 skb_pull(skb, rlen);
 451         }
 452 }
 453
 454 /* Receive messages from netlink socket. */
 455 static void audit_receive(struct sock *sk, int length)
 456 {
 457         struct sk_buff  *skb;
 458         unsigned int qlen;
 459
 460         down(&audit_netlink_sem);
 461
 462         for (qlen = skb_queue_len(&sk->sk_receive_queue); qlen; qlen--) {
 463                 skb = skb_dequeue(&sk->sk_receive_queue);
 464                 audit_receive_skb(skb);
 465                 kfree_skb(skb);
 466         }
 467         up(&audit_netlink_sem);
 468 }
 469
 470 /* Move data from tmp buffer into an skb.  This is an extra copy, and
 471  * that is unfortunate.  However, the copy will only occur when a record
 472  * is being written to user space, which is already a high-overhead
 473  * operation.  (Elimination of the copy is possible, for example, by
 474  * writing directly into a pre-allocated skb, at the cost of wasting
 475  * memory. */
 476 static void audit_log_move(struct audit_buffer *ab)
 477 {
 478         struct sk_buff  *skb;
 479         char            *start;
 480         int             extra = ab->nlh ? 0 : NLMSG_SPACE(0);
 481
 482         /* possible resubmission */
 483         if (ab->len == 0)
 484                 return;
 485
 486         skb = skb_peek_tail(&ab->sklist);
 487         if (!skb || skb_tailroom(skb) <= ab->len + extra) {
 488                 skb = alloc_skb(2 * ab->len + extra, GFP_ATOMIC);
 489                 if (!skb) {
 490                         ab->len = 0; /* Lose information in ab->tmp */
 491                         audit_log_lost("out of memory in audit_log_move");
 492                         return;
 493                 }
 494                 __skb_queue_tail(&ab->sklist, skb);
 495                 if (!ab->nlh)
 496                         ab->nlh = (struct nlmsghdr *)skb_put(skb,
 497                                                              NLMSG_SPACE(0));
 498         }
 499         start = skb_put(skb, ab->len);
 500         memcpy(start, ab->tmp, ab->len);
 501         ab->len = 0;
 502 }
 503
 504 /* Iterate over the skbuff in the audit_buffer, sending their contents
 505  * to user space. */
 506 static inline int audit_log_drain(struct audit_buffer *ab)
 507 {
 508         struct sk_buff *skb;
 509
 510         while ((skb = skb_dequeue(&ab->sklist))) {
 511                 int retval = 0;
 512
 513                 if (audit_pid) {
 514                         if (ab->nlh) {
 515                                 ab->nlh->nlmsg_len   = ab->total;
 516                                 ab->nlh->nlmsg_type  = ab->type;
 517                                 ab->nlh->nlmsg_flags = 0;
 518                                 ab->nlh->nlmsg_seq   = 0;
 519                                 ab->nlh->nlmsg_pid   = ab->pid;
 520                         }
 521                         skb_get(skb); /* because netlink_* frees */
 522                         retval = netlink_unicast(audit_sock, skb, audit_pid,
 523                                                  MSG_DONTWAIT);
 524                 }
 525                 if (retval == -EAGAIN &&
 526                     (atomic_read(&audit_backlog)) < audit_backlog_limit) {
 527                         skb_queue_head(&ab->sklist, skb);
 528                         audit_log_end_irq(ab);
 529                         return 1;
 530                 }
 531                 if (retval < 0) {
 532                         if (retval == -ECONNREFUSED) {
 533                                 printk(KERN_ERR
 534                                        "audit: *NO* daemon at audit_pid=%d\n",
 535                                        audit_pid);
 536                                 audit_pid = 0;
 537                         } else
 538                                 audit_log_lost("netlink socket too busy");
 539                 }
 540                 if (!audit_pid) { /* No daemon */
 541                         int offset = ab->nlh ? NLMSG_SPACE(0) : 0;
 542                         int len    = skb->len - offset;
 543                         skb->data[offset + len] = '\0';
 544                         printk(KERN_ERR "%s\n", skb->data + offset);
 545                 }
 546                 kfree_skb(skb);
 547                 ab->nlh = NULL;
 548         }
 549         return 0;
 550 }
 551
 552 /* Initialize audit support at boot time. */
 553 static int __init audit_init(void)
 554 {
 555         printk(KERN_INFO "audit: initializing netlink socket (%s)\n",
 556                audit_default ? "enabled" : "disabled");
 557         audit_sock = netlink_kernel_create(NETLINK_AUDIT, audit_receive);
 558         if (!audit_sock)
 559                 audit_panic("cannot initialize netlink socket");
 560
 561         audit_initialized = 1;
 562         audit_enabled = audit_default;
 563         audit_log(NULL, "initialized");
 564         return 0;
 565 }
 566
 567 #else
 568 /* Without CONFIG_NET, we have no skbuffs.  For now, print what we have
 569  * in the buffer. */
 570 static void audit_log_move(struct audit_buffer *ab)
 571 {
 572         printk(KERN_ERR "%*.*s\n", ab->len, ab->len, ab->tmp);
 573         ab->len = 0;
 574 }
 575
 576 static inline int audit_log_drain(struct audit_buffer *ab)
 577 {
 578         return 0;
 579 }
 580
 581 /* Initialize audit support at boot time. */
 582 int __init audit_init(void)
 583 {
 584         printk(KERN_INFO "audit: initializing WITHOUT netlink support\n");
 585         audit_sock = NULL;
 586         audit_pid  = 0;
 587
 588         audit_initialized = 1;
 589         audit_enabled = audit_default;
 590         audit_log(NULL, "initialized");
 591         return 0;
 592 }
 593 #endif
 594
 595 __initcall(audit_init);
 596
 597 /* Process kernel command-line parameter at boot time.  audit=0 or audit=1. */
 598 static int __init audit_enable(char *str)
 599 {
 600         audit_default = !!simple_strtol(str, NULL, 0);
 601         printk(KERN_INFO "audit: %s%s\n",
 602                audit_default ? "enabled" : "disabled",
 603                audit_initialized ? "" : " (after initialization)");
 604         if (audit_initialized)
 605                 audit_enabled = audit_default;
 606         return 0;
 607 }
 608
 609 __setup("audit=", audit_enable);
 610
 611
 612 /* Obtain an audit buffer.  This routine does locking to obtain the
 613  * audit buffer, but then no locking is required for calls to
 614  * audit_log_*format.  If the tsk is a task that is currently in a
 615  * syscall, then the syscall is marked as auditable and an audit record
 616  * will be written at syscall exit.  If there is no associated task, tsk
 617  * should be NULL. */
 618 struct audit_buffer *audit_log_start(struct audit_context *ctx)
 619 {
 620         struct audit_buffer     *ab     = NULL;
 621         unsigned long           flags;
 622         struct timespec         t;
 623         unsigned int            serial;
 624
 625         if (!audit_initialized)
 626                 return NULL;
 627
 628         if (audit_backlog_limit
 629             && atomic_read(&audit_backlog) > audit_backlog_limit) {
 630                 if (audit_rate_check())
 631                         printk(KERN_WARNING
 632                                "audit: audit_backlog=%d > "
 633                                "audit_backlog_limit=%d\n",
 634                                atomic_read(&audit_backlog),
 635                                audit_backlog_limit);
 636                 audit_log_lost("backlog limit exceeded");
 637                 return NULL;
 638         }
 639
 640         spin_lock_irqsave(&audit_freelist_lock, flags);
 641         if (!list_empty(&audit_freelist)) {
 642                 ab = list_entry(audit_freelist.next,
 643                                 struct audit_buffer, list);
 644                 list_del(&ab->list);
 645                 --audit_freelist_count;
 646         }
 647         spin_unlock_irqrestore(&audit_freelist_lock, flags);
 648
 649         if (!ab)
 650                 ab = kmalloc(sizeof(*ab), GFP_ATOMIC);
 651         if (!ab) {
 652                 audit_log_lost("out of memory in audit_log_start");
 653                 return NULL;
 654         }
 655
 656         atomic_inc(&audit_backlog);
 657         skb_queue_head_init(&ab->sklist);
 658
 659         ab->ctx   = ctx;
 660         ab->len   = 0;
 661         ab->nlh   = NULL;
 662         ab->total = 0;
 663         ab->type  = AUDIT_KERNEL;
 664         ab->pid   = 0;
 665
 666 #ifdef CONFIG_AUDITSYSCALL
 667         if (ab->ctx)
 668                 audit_get_stamp(ab->ctx, &t, &serial);
 669         else
 670 #endif
 671         {
 672                 t = CURRENT_TIME;
 673                 serial = 0;
 674         }
 675         audit_log_format(ab, "audit(%lu.%03lu:%u): ",
 676                          t.tv_sec, t.tv_nsec/1000000, serial);
 677         return ab;
 678 }
 679
 680
 681 /* Format an audit message into the audit buffer.  If there isn't enough
 682  * room in the audit buffer, more room will be allocated and vsnprint
 683  * will be called a second time.  Currently, we assume that a printk
 684  * can't format message larger than 1024 bytes, so we don't either. */
 685 static void audit_log_vformat(struct audit_buffer *ab, const char *fmt,
 686                               va_list args)
 687 {
 688         int len, avail;
 689
 690         if (!ab)
 691                 return;
 692
 693         avail = sizeof(ab->tmp) - ab->len;
 694         if (avail <= 0) {
 695                 audit_log_move(ab);
 696                 avail = sizeof(ab->tmp) - ab->len;
 697         }
 698         len   = vsnprintf(ab->tmp + ab->len, avail, fmt, args);
 699         if (len >= avail) {
 700                 /* The printk buffer is 1024 bytes long, so if we get
 701                  * here and AUDIT_BUFSIZ is at least 1024, then we can
 702                  * log everything that printk could have logged. */
 703                 audit_log_move(ab);
 704                 avail = sizeof(ab->tmp) - ab->len;
 705                 len   = vsnprintf(ab->tmp + ab->len, avail, fmt, args);
 706         }
 707         ab->len   += (len < avail) ? len : avail;
 708         ab->total += (len < avail) ? len : avail;
 709 }
 710
 711 /* Format a message into the audit buffer.  All the work is done in
 712  * audit_log_vformat. */
 713 void audit_log_format(struct audit_buffer *ab, const char *fmt, ...)
 714 {
 715         va_list args;
 716
 717         if (!ab)
 718                 return;
 719         va_start(args, fmt);
 720         audit_log_vformat(ab, fmt, args);
 721         va_end(args);
 722 }
 723
 724 void audit_log_hex(struct audit_buffer *ab, const unsigned char *buf, size_t len)
 725 {
 726         int i;
 727
 728         for (i=0; i<len; i++)
 729                 audit_log_format(ab, "%02x", buf[i]);
 730 }
 731
 732 void audit_log_untrustedstring(struct audit_buffer *ab, const char *string)
 733 {
 734         const unsigned char *p = string;
 735
 736         while (*p) {
 737                 if (*p == '"' || *p == ' ' || *p < 0x20 || *p > 0x7f) {
 738                         audit_log_hex(ab, string, strlen(string));
 739                         return;
 740                 }
 741                 p++;
 742         }
 743         audit_log_format(ab, "\"%s\"", string);
 744 }
 745
 746
 747 /* This is a helper-function to print the d_path without using a static
 748  * buffer or allocating another buffer in addition to the one in
 749  * audit_buffer. */
 750 void audit_log_d_path(struct audit_buffer *ab, const char *prefix,
 751                       struct dentry *dentry, struct vfsmount *vfsmnt)
 752 {
 753         char *p;
 754         int  len, avail;
 755
 756         if (prefix) audit_log_format(ab, " %s", prefix);
 757
 758         if (ab->len > 128)
 759                 audit_log_move(ab);
 760         avail = sizeof(ab->tmp) - ab->len;
 761         p = d_path(dentry, vfsmnt, ab->tmp + ab->len, avail);
 762         if (IS_ERR(p)) {
 763                 /* FIXME: can we save some information here? */
 764                 audit_log_format(ab, "<toolong>");
 765         } else {
 766                                 /* path isn't at start of buffer */
 767                 len        = (ab->tmp + sizeof(ab->tmp) - 1) - p;
 768                 memmove(ab->tmp + ab->len, p, len);
 769                 ab->len   += len;
 770                 ab->total += len;
 771         }
 772 }
 773
 774 /* Remove queued messages from the audit_txlist and send them to userspace. */
 775 static void audit_tasklet_handler(unsigned long arg)
 776 {
 777         LIST_HEAD(list);
 778         struct audit_buffer *ab;
 779         unsigned long       flags;
 780
 781         spin_lock_irqsave(&audit_txlist_lock, flags);
 782         list_splice_init(&audit_txlist, &list);
 783         spin_unlock_irqrestore(&audit_txlist_lock, flags);
 784
 785         while (!list_empty(&list)) {
 786                 ab = list_entry(list.next, struct audit_buffer, list);
 787                 list_del(&ab->list);
 788                 audit_log_end_fast(ab);
 789         }
 790 }
 791
 792 static DECLARE_TASKLET(audit_tasklet, audit_tasklet_handler, 0);
 793
 794 /* The netlink_* functions cannot be called inside an irq context, so
 795  * the audit buffer is places on a queue and a tasklet is scheduled to
 796  * remove them from the queue outside the irq context.  May be called in
 797  * any context. */
 798 static void audit_log_end_irq(struct audit_buffer *ab)
 799 {
 800         unsigned long flags;
 801
 802         if (!ab)
 803                 return;
 804         spin_lock_irqsave(&audit_txlist_lock, flags);
 805         list_add_tail(&ab->list, &audit_txlist);
 806         spin_unlock_irqrestore(&audit_txlist_lock, flags);
 807
 808         tasklet_schedule(&audit_tasklet);
 809 }
 810
 811 /* Send the message in the audit buffer directly to user space.  May not
 812  * be called in an irq context. */
 813 static void audit_log_end_fast(struct audit_buffer *ab)
 814 {
 815         unsigned long flags;
 816
 817         BUG_ON(in_irq());
 818         if (!ab)
 819                 return;
 820         if (!audit_rate_check()) {
 821                 audit_log_lost("rate limit exceeded");
 822         } else {
 823                 audit_log_move(ab);
 824                 if (audit_log_drain(ab))
 825                         return;
 826         }
 827
 828         atomic_dec(&audit_backlog);
 829         spin_lock_irqsave(&audit_freelist_lock, flags);
 830         if (++audit_freelist_count > AUDIT_MAXFREE)
 831                 kfree(ab);
 832         else
 833                 list_add(&ab->list, &audit_freelist);
 834         spin_unlock_irqrestore(&audit_freelist_lock, flags);
 835 }
 836
 837 /* Send or queue the message in the audit buffer, depending on the
 838  * current context.  (A convenience function that may be called in any
 839  * context.) */
 840 void audit_log_end(struct audit_buffer *ab)
 841 {
 842         if (in_irq())
 843                 audit_log_end_irq(ab);
 844         else
 845                 audit_log_end_fast(ab);
 846 }
 847
 848 /* Log an audit record.  This is a convenience function that calls
 849  * audit_log_start, audit_log_vformat, and audit_log_end.  It may be
 850  * called in any context. */
 851 void audit_log(struct audit_context *ctx, const char *fmt, ...)
 852 {
 853         struct audit_buffer *ab;
 854         va_list args;
 855
 856         ab = audit_log_start(ctx);
 857         if (ab) {
 858                 va_start(args, fmt);
 859                 audit_log_vformat(ab, fmt, args);
 860                 va_end(args);
 861                 audit_log_end(ab);
 862         }
 863 }