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 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 /* The identity of the user shutting down the audit system. */
  83 uid_t           audit_sig_uid = -1;
  84 pid_t           audit_sig_pid = -1;
  85
  86 /* Records can be lost in several ways:
  87    0) [suppressed in audit_alloc]
  88    1) out of memory in audit_log_start [kmalloc of struct audit_buffer]
  89    2) out of memory in audit_log_move [alloc_skb]
  90    3) suppressed due to audit_rate_limit
  91    4) suppressed due to audit_backlog_limit
  92 */
  93 static atomic_t    audit_lost = ATOMIC_INIT(0);
  94
  95 /* The netlink socket. */
  96 static struct sock *audit_sock;
  97
  98 /* There are two lists of audit buffers.  The txlist contains audit
  99  * buffers that cannot be sent immediately to the netlink device because
 100  * we are in an irq context (these are sent later in a tasklet).
 101  *
 102  * The second list is a list of pre-allocated audit buffers (if more
 103  * than AUDIT_MAXFREE are in use, the audit buffer is freed instead of
 104  * being placed on the freelist). */
 105 static DEFINE_SPINLOCK(audit_txlist_lock);
 106 static DEFINE_SPINLOCK(audit_freelist_lock);
 107 static int         audit_freelist_count = 0;
 108 static LIST_HEAD(audit_txlist);
 109 static LIST_HEAD(audit_freelist);
 110
 111 /* There are three lists of rules -- one to search at task creation
 112  * time, one to search at syscall entry time, and another to search at
 113  * syscall exit time. */
 114 static LIST_HEAD(audit_tsklist);
 115 static LIST_HEAD(audit_entlist);
 116 static LIST_HEAD(audit_extlist);
 117
 118 /* The netlink socket is only to be read by 1 CPU, which lets us assume
 119  * that list additions and deletions never happen simultaneiously in
 120  * auditsc.c */
 121 static DECLARE_MUTEX(audit_netlink_sem);
 122
 123 /* AUDIT_BUFSIZ is the size of the temporary buffer used for formatting
 124  * audit records.  Since printk uses a 1024 byte buffer, this buffer
 125  * should be at least that large. */
 126 #define AUDIT_BUFSIZ 1024
 127
 128 /* AUDIT_MAXFREE is the number of empty audit_buffers we keep on the
 129  * audit_freelist.  Doing so eliminates many kmalloc/kfree calls. */
 130 #define AUDIT_MAXFREE  (2*NR_CPUS)
 131
 132 /* The audit_buffer is used when formatting an audit record.  The caller
 133  * locks briefly to get the record off the freelist or to allocate the
 134  * buffer, and locks briefly to send the buffer to the netlink layer or
 135  * to place it on a transmit queue.  Multiple audit_buffers can be in
 136  * use simultaneously. */
 137 struct audit_buffer {
 138         struct list_head     list;
 139         struct sk_buff_head  sklist;    /* formatted skbs ready to send */
 140         struct audit_context *ctx;      /* NULL or associated context */
 141         int                  len;       /* used area of tmp */
 142         char                 tmp[AUDIT_BUFSIZ];
 143
 144                                 /* Pointer to header and contents */
 145         struct nlmsghdr      *nlh;
 146         int                  total;
 147         int                  type;
 148         int                  pid;
 149 };
 150
 151 void audit_set_type(struct audit_buffer *ab, int type)
 152 {
 153         ab->type = type;
 154 }
 155
 156 struct audit_entry {
 157         struct list_head  list;
 158         struct audit_rule rule;
 159 };
 160
 161 static void audit_log_end_irq(struct audit_buffer *ab);
 162 static void audit_log_end_fast(struct audit_buffer *ab);
 163
 164 static void audit_panic(const char *message)
 165 {
 166         switch (audit_failure)
 167         {
 168         case AUDIT_FAIL_SILENT:
 169                 break;
 170         case AUDIT_FAIL_PRINTK:
 171                 printk(KERN_ERR "audit: %s\n", message);
 172                 break;
 173         case AUDIT_FAIL_PANIC:
 174                 panic("audit: %s\n", message);
 175                 break;
 176         }
 177 }
 178
 179 static inline int audit_rate_check(void)
 180 {
 181         static unsigned long    last_check = 0;
 182         static int              messages   = 0;
 183         static DEFINE_SPINLOCK(lock);
 184         unsigned long           flags;
 185         unsigned long           now;
 186         unsigned long           elapsed;
 187         int                     retval     = 0;
 188
 189         if (!audit_rate_limit) return 1;
 190
 191         spin_lock_irqsave(&lock, flags);
 192         if (++messages < audit_rate_limit) {
 193                 retval = 1;
 194         } else {
 195                 now     = jiffies;
 196                 elapsed = now - last_check;
 197                 if (elapsed > HZ) {
 198                         last_check = now;
 199                         messages   = 0;
 200                         retval     = 1;
 201                 }
 202         }
 203         spin_unlock_irqrestore(&lock, flags);
 204
 205         return retval;
 206 }
 207
 208 /* Emit at least 1 message per second, even if audit_rate_check is
 209  * throttling. */
 210 void audit_log_lost(const char *message)
 211 {
 212         static unsigned long    last_msg = 0;
 213         static DEFINE_SPINLOCK(lock);
 214         unsigned long           flags;
 215         unsigned long           now;
 216         int                     print;
 217
 218         atomic_inc(&audit_lost);
 219
 220         print = (audit_failure == AUDIT_FAIL_PANIC || !audit_rate_limit);
 221
 222         if (!print) {
 223                 spin_lock_irqsave(&lock, flags);
 224                 now = jiffies;
 225                 if (now - last_msg > HZ) {
 226                         print = 1;
 227                         last_msg = now;
 228                 }
 229                 spin_unlock_irqrestore(&lock, flags);
 230         }
 231
 232         if (print) {
 233                 printk(KERN_WARNING
 234                        "audit: audit_lost=%d audit_backlog=%d"
 235                        " audit_rate_limit=%d audit_backlog_limit=%d\n",
 236                        atomic_read(&audit_lost),
 237                        atomic_read(&audit_backlog),
 238                        audit_rate_limit,
 239                        audit_backlog_limit);
 240                 audit_panic(message);
 241         }
 242
 243 }
 244
 245 static int audit_set_rate_limit(int limit, uid_t loginuid)
 246 {
 247         int old          = audit_rate_limit;
 248         audit_rate_limit = limit;
 249         audit_log(NULL, "audit_rate_limit=%d old=%d by auid %u",
 250                         audit_rate_limit, old, loginuid);
 251         return old;
 252 }
 253
 254 static int audit_set_backlog_limit(int limit, uid_t loginuid)
 255 {
 256         int old          = audit_backlog_limit;
 257         audit_backlog_limit = limit;
 258         audit_log(NULL, "audit_backlog_limit=%d old=%d by auid %u",
 259                         audit_backlog_limit, old, loginuid);
 260         return old;
 261 }
 262
 263 static int audit_set_enabled(int state, uid_t loginuid)
 264 {
 265         int old          = audit_enabled;
 266         if (state != 0 && state != 1)
 267                 return -EINVAL;
 268         audit_enabled = state;
 269         audit_log(NULL, "audit_enabled=%d old=%d by auid %u",
 270                   audit_enabled, old, loginuid);
 271         return old;
 272 }
 273
 274 static int audit_set_failure(int state, uid_t loginuid)
 275 {
 276         int old          = audit_failure;
 277         if (state != AUDIT_FAIL_SILENT
 278             && state != AUDIT_FAIL_PRINTK
 279             && state != AUDIT_FAIL_PANIC)
 280                 return -EINVAL;
 281         audit_failure = state;
 282         audit_log(NULL, "audit_failure=%d old=%d by auid %u",
 283                   audit_failure, old, loginuid);
 284         return old;
 285 }
 286
 287 #ifdef CONFIG_NET
 288 void audit_send_reply(int pid, int seq, int type, int done, int multi,
 289                       void *payload, int size)
 290 {
 291         struct sk_buff  *skb;
 292         struct nlmsghdr *nlh;
 293         int             len = NLMSG_SPACE(size);
 294         void            *data;
 295         int             flags = multi ? NLM_F_MULTI : 0;
 296         int             t     = done  ? NLMSG_DONE  : type;
 297
 298         skb = alloc_skb(len, GFP_KERNEL);
 299         if (!skb)
 300                 goto nlmsg_failure;
 301
 302         nlh              = NLMSG_PUT(skb, pid, seq, t, len - sizeof(*nlh));
 303         nlh->nlmsg_flags = flags;
 304         data             = NLMSG_DATA(nlh);
 305         memcpy(data, payload, size);
 306         netlink_unicast(audit_sock, skb, pid, MSG_DONTWAIT);
 307         return;
 308
 309 nlmsg_failure:                  /* Used by NLMSG_PUT */
 310         if (skb)
 311                 kfree_skb(skb);
 312 }
 313
 314 /*
 315  * Check for appropriate CAP_AUDIT_ capabilities on incoming audit
 316  * control messages.
 317  */
 318 static int audit_netlink_ok(kernel_cap_t eff_cap, u16 msg_type)
 319 {
 320         int err = 0;
 321
 322         switch (msg_type) {
 323         case AUDIT_GET:
 324         case AUDIT_LIST:
 325         case AUDIT_SET:
 326         case AUDIT_ADD:
 327         case AUDIT_DEL:
 328         case AUDIT_SIGNAL_INFO:
 329                 if (!cap_raised(eff_cap, CAP_AUDIT_CONTROL))
 330                         err = -EPERM;
 331                 break;
 332         case AUDIT_USER:
 333                 if (!cap_raised(eff_cap, CAP_AUDIT_WRITE))
 334                         err = -EPERM;
 335                 break;
 336         default:  /* bad msg */
 337                 err = -EINVAL;
 338         }
 339
 340         return err;
 341 }
 342
 343 static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
 344 {
 345         u32                     uid, pid, seq;
 346         void                    *data;
 347         struct audit_status     *status_get, status_set;
 348         int                     err;
 349         struct audit_buffer     *ab;
 350         u16                     msg_type = nlh->nlmsg_type;
 351         uid_t                   loginuid; /* loginuid of sender */
 352         struct audit_sig_info   sig_data;
 353
 354         err = audit_netlink_ok(NETLINK_CB(skb).eff_cap, msg_type);
 355         if (err)
 356                 return err;
 357
 358         pid  = NETLINK_CREDS(skb)->pid;
 359         uid  = NETLINK_CREDS(skb)->uid;
 360         loginuid = NETLINK_CB(skb).loginuid;
 361         seq  = nlh->nlmsg_seq;
 362         data = NLMSG_DATA(nlh);
 363
 364         switch (msg_type) {
 365         case AUDIT_GET:
 366                 status_set.enabled       = audit_enabled;
 367                 status_set.failure       = audit_failure;
 368                 status_set.pid           = audit_pid;
 369                 status_set.rate_limit    = audit_rate_limit;
 370                 status_set.backlog_limit = audit_backlog_limit;
 371                 status_set.lost          = atomic_read(&audit_lost);
 372                 status_set.backlog       = atomic_read(&audit_backlog);
 373                 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_GET, 0, 0,
 374                                  &status_set, sizeof(status_set));
 375                 break;
 376         case AUDIT_SET:
 377                 if (nlh->nlmsg_len < sizeof(struct audit_status))
 378                         return -EINVAL;
 379                 status_get   = (struct audit_status *)data;
 380                 if (status_get->mask & AUDIT_STATUS_ENABLED) {
 381                         err = audit_set_enabled(status_get->enabled, loginuid);
 382                         if (err < 0) return err;
 383                 }
 384                 if (status_get->mask & AUDIT_STATUS_FAILURE) {
 385                         err = audit_set_failure(status_get->failure, loginuid);
 386                         if (err < 0) return err;
 387                 }
 388                 if (status_get->mask & AUDIT_STATUS_PID) {
 389                         int old   = audit_pid;
 390                         audit_pid = status_get->pid;
 391                         audit_log(NULL, "audit_pid=%d old=%d by auid %u",
 392                                   audit_pid, old, loginuid);
 393                 }
 394                 if (status_get->mask & AUDIT_STATUS_RATE_LIMIT)
 395                         audit_set_rate_limit(status_get->rate_limit, loginuid);
 396                 if (status_get->mask & AUDIT_STATUS_BACKLOG_LIMIT)
 397                         audit_set_backlog_limit(status_get->backlog_limit,
 398                                                         loginuid);
 399                 break;
 400         case AUDIT_USER:
 401                 ab = audit_log_start(NULL);
 402                 if (!ab)
 403                         break;  /* audit_panic has been called */
 404                 audit_log_format(ab,
 405                                  "user pid=%d uid=%d length=%d loginuid=%u"
 406                                  " msg='%.1024s'",
 407                                  pid, uid,
 408                                  (int)(nlh->nlmsg_len
 409                                        - ((char *)data - (char *)nlh)),
 410                                  loginuid, (char *)data);
 411                 ab->type = AUDIT_USER;
 412                 ab->pid  = pid;
 413                 audit_log_end(ab);
 414                 break;
 415         case AUDIT_ADD:
 416         case AUDIT_DEL:
 417                 if (nlh->nlmsg_len < sizeof(struct audit_rule))
 418                         return -EINVAL;
 419                 /* fallthrough */
 420         case AUDIT_LIST:
 421 #ifdef CONFIG_AUDITSYSCALL
 422                 err = audit_receive_filter(nlh->nlmsg_type, NETLINK_CB(skb).pid,
 423                                            uid, seq, data, loginuid);
 424 #else
 425                 err = -EOPNOTSUPP;
 426 #endif
 427                 break;
 428         case AUDIT_SIGNAL_INFO:
 429                 sig_data.uid = audit_sig_uid;
 430                 sig_data.pid = audit_sig_pid;
 431                 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_SIGNAL_INFO,
 432                                 0, 0, &sig_data, sizeof(sig_data));
 433                 break;
 434         default:
 435                 err = -EINVAL;
 436                 break;
 437         }
 438
 439         return err < 0 ? err : 0;
 440 }
 441
 442 /* Get message from skb (based on rtnetlink_rcv_skb).  Each message is
 443  * processed by audit_receive_msg.  Malformed skbs with wrong length are
 444  * discarded silently.  */
 445 static void audit_receive_skb(struct sk_buff *skb)
 446 {
 447         int             err;
 448         struct nlmsghdr *nlh;
 449         u32             rlen;
 450
 451         while (skb->len >= NLMSG_SPACE(0)) {
 452                 nlh = (struct nlmsghdr *)skb->data;
 453                 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
 454                         return;
 455                 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
 456                 if (rlen > skb->len)
 457                         rlen = skb->len;
 458                 if ((err = audit_receive_msg(skb, nlh))) {
 459                         netlink_ack(skb, nlh, err);
 460                 } else if (nlh->nlmsg_flags & NLM_F_ACK)
 461                         netlink_ack(skb, nlh, 0);
 462                 skb_pull(skb, rlen);
 463         }
 464 }
 465
 466 /* Receive messages from netlink socket. */
 467 static void audit_receive(struct sock *sk, int length)
 468 {
 469         struct sk_buff  *skb;
 470         unsigned int qlen;
 471
 472         down(&audit_netlink_sem);
 473
 474         for (qlen = skb_queue_len(&sk->sk_receive_queue); qlen; qlen--) {
 475                 skb = skb_dequeue(&sk->sk_receive_queue);
 476                 audit_receive_skb(skb);
 477                 kfree_skb(skb);
 478         }
 479         up(&audit_netlink_sem);
 480 }
 481
 482 /* Move data from tmp buffer into an skb.  This is an extra copy, and
 483  * that is unfortunate.  However, the copy will only occur when a record
 484  * is being written to user space, which is already a high-overhead
 485  * operation.  (Elimination of the copy is possible, for example, by
 486  * writing directly into a pre-allocated skb, at the cost of wasting
 487  * memory. */
 488 static void audit_log_move(struct audit_buffer *ab)
 489 {
 490         struct sk_buff  *skb;
 491         char            *start;
 492         int             extra = ab->nlh ? 0 : NLMSG_SPACE(0);
 493
 494         /* possible resubmission */
 495         if (ab->len == 0)
 496                 return;
 497
 498         skb = skb_peek_tail(&ab->sklist);
 499         if (!skb || skb_tailroom(skb) <= ab->len + extra) {
 500                 skb = alloc_skb(2 * ab->len + extra, GFP_ATOMIC);
 501                 if (!skb) {
 502                         ab->len = 0; /* Lose information in ab->tmp */
 503                         audit_log_lost("out of memory in audit_log_move");
 504                         return;
 505                 }
 506                 __skb_queue_tail(&ab->sklist, skb);
 507                 if (!ab->nlh)
 508                         ab->nlh = (struct nlmsghdr *)skb_put(skb,
 509                                                              NLMSG_SPACE(0));
 510         }
 511         start = skb_put(skb, ab->len);
 512         memcpy(start, ab->tmp, ab->len);
 513         ab->len = 0;
 514 }
 515
 516 /* Iterate over the skbuff in the audit_buffer, sending their contents
 517  * to user space. */
 518 static inline int audit_log_drain(struct audit_buffer *ab)
 519 {
 520         struct sk_buff *skb;
 521
 522         while ((skb = skb_dequeue(&ab->sklist))) {
 523                 int retval = 0;
 524
 525                 if (audit_pid) {
 526                         if (ab->nlh) {
 527                                 ab->nlh->nlmsg_len   = ab->total;
 528                                 ab->nlh->nlmsg_type  = ab->type;
 529                                 ab->nlh->nlmsg_flags = 0;
 530                                 ab->nlh->nlmsg_seq   = 0;
 531                                 ab->nlh->nlmsg_pid   = ab->pid;
 532                         }
 533                         skb_get(skb); /* because netlink_* frees */
 534                         retval = netlink_unicast(audit_sock, skb, audit_pid,
 535                                                  MSG_DONTWAIT);
 536                 }
 537                 if (retval == -EAGAIN &&
 538                     (atomic_read(&audit_backlog)) < audit_backlog_limit) {
 539                         skb_queue_head(&ab->sklist, skb);
 540                         audit_log_end_irq(ab);
 541                         return 1;
 542                 }
 543                 if (retval < 0) {
 544                         if (retval == -ECONNREFUSED) {
 545                                 printk(KERN_ERR
 546                                        "audit: *NO* daemon at audit_pid=%d\n",
 547                                        audit_pid);
 548                                 audit_pid = 0;
 549                         } else
 550                                 audit_log_lost("netlink socket too busy");
 551                 }
 552                 if (!audit_pid) { /* No daemon */
 553                         int offset = ab->nlh ? NLMSG_SPACE(0) : 0;
 554                         int len    = skb->len - offset;
 555                         skb->data[offset + len] = '\0';
 556                         printk(KERN_ERR "%s\n", skb->data + offset);
 557                 }
 558                 kfree_skb(skb);
 559                 ab->nlh = NULL;
 560         }
 561         return 0;
 562 }
 563
 564 /* Initialize audit support at boot time. */
 565 static int __init audit_init(void)
 566 {
 567         printk(KERN_INFO "audit: initializing netlink socket (%s)\n",
 568                audit_default ? "enabled" : "disabled");
 569         audit_sock = netlink_kernel_create(NETLINK_AUDIT, audit_receive);
 570         if (!audit_sock)
 571                 audit_panic("cannot initialize netlink socket");
 572
 573         audit_initialized = 1;
 574         audit_enabled = audit_default;
 575         audit_log(NULL, "initialized");
 576         return 0;
 577 }
 578
 579 #else
 580 /* Without CONFIG_NET, we have no skbuffs.  For now, print what we have
 581  * in the buffer. */
 582 static void audit_log_move(struct audit_buffer *ab)
 583 {
 584         printk(KERN_ERR "%*.*s\n", ab->len, ab->len, ab->tmp);
 585         ab->len = 0;
 586 }
 587
 588 static inline int audit_log_drain(struct audit_buffer *ab)
 589 {
 590         return 0;
 591 }
 592
 593 /* Initialize audit support at boot time. */
 594 int __init audit_init(void)
 595 {
 596         printk(KERN_INFO "audit: initializing WITHOUT netlink support\n");
 597         audit_sock = NULL;
 598         audit_pid  = 0;
 599
 600         audit_initialized = 1;
 601         audit_enabled = audit_default;
 602         audit_log(NULL, "initialized");
 603         return 0;
 604 }
 605 #endif
 606
 607 __initcall(audit_init);
 608
 609 /* Process kernel command-line parameter at boot time.  audit=0 or audit=1. */
 610 static int __init audit_enable(char *str)
 611 {
 612         audit_default = !!simple_strtol(str, NULL, 0);
 613         printk(KERN_INFO "audit: %s%s\n",
 614                audit_default ? "enabled" : "disabled",
 615                audit_initialized ? "" : " (after initialization)");
 616         if (audit_initialized)
 617                 audit_enabled = audit_default;
 618         return 0;
 619 }
 620
 621 __setup("audit=", audit_enable);
 622
 623 static void audit_buffer_free(struct audit_buffer *ab)
 624 {
 625         unsigned long flags;
 626
 627         atomic_dec(&audit_backlog);
 628         spin_lock_irqsave(&audit_freelist_lock, flags);
 629         if (++audit_freelist_count > AUDIT_MAXFREE)
 630                 kfree(ab);
 631         else
 632                 list_add(&ab->list, &audit_freelist);
 633         spin_unlock_irqrestore(&audit_freelist_lock, flags);
 634 }
 635
 636 static struct audit_buffer * audit_buffer_alloc(int gfp_mask)
 637 {
 638         unsigned long flags;
 639         struct audit_buffer *ab = NULL;
 640
 641         spin_lock_irqsave(&audit_freelist_lock, flags);
 642         if (!list_empty(&audit_freelist)) {
 643                 ab = list_entry(audit_freelist.next,
 644                                 struct audit_buffer, list);
 645                 list_del(&ab->list);
 646                 --audit_freelist_count;
 647         }
 648         spin_unlock_irqrestore(&audit_freelist_lock, flags);
 649
 650         if (!ab) {
 651                 ab = kmalloc(sizeof(*ab), GFP_ATOMIC);
 652                 if (!ab)
 653                         goto out;
 654         }
 655         atomic_inc(&audit_backlog);
 656 out:
 657         return ab;
 658 }
 659
 660 /* Obtain an audit buffer.  This routine does locking to obtain the
 661  * audit buffer, but then no locking is required for calls to
 662  * audit_log_*format.  If the tsk is a task that is currently in a
 663  * syscall, then the syscall is marked as auditable and an audit record
 664  * will be written at syscall exit.  If there is no associated task, tsk
 665  * should be NULL. */
 666 struct audit_buffer *audit_log_start(struct audit_context *ctx)
 667 {
 668         struct audit_buffer     *ab     = NULL;
 669         struct timespec         t;
 670         unsigned int            serial;
 671
 672         if (!audit_initialized)
 673                 return NULL;
 674
 675         if (audit_backlog_limit
 676             && atomic_read(&audit_backlog) > audit_backlog_limit) {
 677                 if (audit_rate_check())
 678                         printk(KERN_WARNING
 679                                "audit: audit_backlog=%d > "
 680                                "audit_backlog_limit=%d\n",
 681                                atomic_read(&audit_backlog),
 682                                audit_backlog_limit);
 683                 audit_log_lost("backlog limit exceeded");
 684                 return NULL;
 685         }
 686
 687         ab = audit_buffer_alloc(GFP_ATOMIC);
 688         if (!ab) {
 689                 audit_log_lost("out of memory in audit_log_start");
 690                 return NULL;
 691         }
 692
 693         skb_queue_head_init(&ab->sklist);
 694
 695         ab->ctx   = ctx;
 696         ab->len   = 0;
 697         ab->nlh   = NULL;
 698         ab->total = 0;
 699         ab->type  = AUDIT_KERNEL;
 700         ab->pid   = 0;
 701
 702 #ifdef CONFIG_AUDITSYSCALL
 703         if (ab->ctx)
 704                 audit_get_stamp(ab->ctx, &t, &serial);
 705         else
 706 #endif
 707         {
 708                 t = CURRENT_TIME;
 709                 serial = 0;
 710         }
 711         audit_log_format(ab, "audit(%lu.%03lu:%u): ",
 712                          t.tv_sec, t.tv_nsec/1000000, serial);
 713         return ab;
 714 }
 715
 716
 717 /* Format an audit message into the audit buffer.  If there isn't enough
 718  * room in the audit buffer, more room will be allocated and vsnprint
 719  * will be called a second time.  Currently, we assume that a printk
 720  * can't format message larger than 1024 bytes, so we don't either. */
 721 static void audit_log_vformat(struct audit_buffer *ab, const char *fmt,
 722                               va_list args)
 723 {
 724         int len, avail;
 725
 726         if (!ab)
 727                 return;
 728
 729         avail = sizeof(ab->tmp) - ab->len;
 730         if (avail <= 0) {
 731                 audit_log_move(ab);
 732                 avail = sizeof(ab->tmp) - ab->len;
 733         }
 734         len   = vsnprintf(ab->tmp + ab->len, avail, fmt, args);
 735         if (len >= avail) {
 736                 /* The printk buffer is 1024 bytes long, so if we get
 737                  * here and AUDIT_BUFSIZ is at least 1024, then we can
 738                  * log everything that printk could have logged. */
 739                 audit_log_move(ab);
 740                 avail = sizeof(ab->tmp) - ab->len;
 741                 len   = vsnprintf(ab->tmp + ab->len, avail, fmt, args);
 742         }
 743         ab->len   += (len < avail) ? len : avail;
 744         ab->total += (len < avail) ? len : avail;
 745 }
 746
 747 /* Format a message into the audit buffer.  All the work is done in
 748  * audit_log_vformat. */
 749 void audit_log_format(struct audit_buffer *ab, const char *fmt, ...)
 750 {
 751         va_list args;
 752
 753         if (!ab)
 754                 return;
 755         va_start(args, fmt);
 756         audit_log_vformat(ab, fmt, args);
 757         va_end(args);
 758 }
 759
 760 void audit_log_hex(struct audit_buffer *ab, const unsigned char *buf, size_t len)
 761 {
 762         int i;
 763
 764         for (i=0; i<len; i++)
 765                 audit_log_format(ab, "%02x", buf[i]);
 766 }
 767
 768 void audit_log_untrustedstring(struct audit_buffer *ab, const char *string)
 769 {
 770         const unsigned char *p = string;
 771
 772         while (*p) {
 773                 if (*p == '"' || *p == ' ' || *p < 0x20 || *p > 0x7f) {
 774                         audit_log_hex(ab, string, strlen(string));
 775                         return;
 776                 }
 777                 p++;
 778         }
 779         audit_log_format(ab, "\"%s\"", string);
 780 }
 781
 782
 783 /* This is a helper-function to print the d_path without using a static
 784  * buffer or allocating another buffer in addition to the one in
 785  * audit_buffer. */
 786 void audit_log_d_path(struct audit_buffer *ab, const char *prefix,
 787                       struct dentry *dentry, struct vfsmount *vfsmnt)
 788 {
 789         char *p;
 790         int  len, avail;
 791
 792         if (prefix) audit_log_format(ab, " %s", prefix);
 793
 794         if (ab->len > 128)
 795                 audit_log_move(ab);
 796         avail = sizeof(ab->tmp) - ab->len;
 797         p = d_path(dentry, vfsmnt, ab->tmp + ab->len, avail);
 798         if (IS_ERR(p)) {
 799                 /* FIXME: can we save some information here? */
 800                 audit_log_format(ab, "<toolong>");
 801         } else {
 802                                 /* path isn't at start of buffer */
 803                 len        = (ab->tmp + sizeof(ab->tmp) - 1) - p;
 804                 memmove(ab->tmp + ab->len, p, len);
 805                 ab->len   += len;
 806                 ab->total += len;
 807         }
 808 }
 809
 810 /* Remove queued messages from the audit_txlist and send them to userspace. */
 811 static void audit_tasklet_handler(unsigned long arg)
 812 {
 813         LIST_HEAD(list);
 814         struct audit_buffer *ab;
 815         unsigned long       flags;
 816
 817         spin_lock_irqsave(&audit_txlist_lock, flags);
 818         list_splice_init(&audit_txlist, &list);
 819         spin_unlock_irqrestore(&audit_txlist_lock, flags);
 820
 821         while (!list_empty(&list)) {
 822                 ab = list_entry(list.next, struct audit_buffer, list);
 823                 list_del(&ab->list);
 824                 audit_log_end_fast(ab);
 825         }
 826 }
 827
 828 static DECLARE_TASKLET(audit_tasklet, audit_tasklet_handler, 0);
 829
 830 /* The netlink_* functions cannot be called inside an irq context, so
 831  * the audit buffer is places on a queue and a tasklet is scheduled to
 832  * remove them from the queue outside the irq context.  May be called in
 833  * any context. */
 834 static void audit_log_end_irq(struct audit_buffer *ab)
 835 {
 836         unsigned long flags;
 837
 838         if (!ab)
 839                 return;
 840         spin_lock_irqsave(&audit_txlist_lock, flags);
 841         list_add_tail(&ab->list, &audit_txlist);
 842         spin_unlock_irqrestore(&audit_txlist_lock, flags);
 843
 844         tasklet_schedule(&audit_tasklet);
 845 }
 846
 847 /* Send the message in the audit buffer directly to user space.  May not
 848  * be called in an irq context. */
 849 static void audit_log_end_fast(struct audit_buffer *ab)
 850 {
 851         BUG_ON(in_irq());
 852         if (!ab)
 853                 return;
 854         if (!audit_rate_check()) {
 855                 audit_log_lost("rate limit exceeded");
 856         } else {
 857                 audit_log_move(ab);
 858                 if (audit_log_drain(ab))
 859                         return;
 860         }
 861         audit_buffer_free(ab);
 862 }
 863
 864 /* Send or queue the message in the audit buffer, depending on the
 865  * current context.  (A convenience function that may be called in any
 866  * context.) */
 867 void audit_log_end(struct audit_buffer *ab)
 868 {
 869         if (in_irq())
 870                 audit_log_end_irq(ab);
 871         else
 872                 audit_log_end_fast(ab);
 873 }
 874
 875 /* Log an audit record.  This is a convenience function that calls
 876  * audit_log_start, audit_log_vformat, and audit_log_end.  It may be
 877  * called in any context. */
 878 void audit_log(struct audit_context *ctx, const char *fmt, ...)
 879 {
 880         struct audit_buffer *ab;
 881         va_list args;
 882
 883         ab = audit_log_start(ctx);
 884         if (ab) {
 885                 va_start(args, fmt);
 886                 audit_log_vformat(ab, fmt, args);
 887                 va_end(args);
 888                 audit_log_end(ab);
 889         }
 890 }