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 static int      audit_backlog_wait_time = 60 * HZ;
  83 static int      audit_backlog_wait_overflow = 0;
  84
  85 /* The identity of the user shutting down the audit system. */
  86 uid_t           audit_sig_uid = -1;
  87 pid_t           audit_sig_pid = -1;
  88
  89 /* Records can be lost in several ways:
  90    0) [suppressed in audit_alloc]
  91    1) out of memory in audit_log_start [kmalloc of struct audit_buffer]
  92    2) out of memory in audit_log_move [alloc_skb]
  93    3) suppressed due to audit_rate_limit
  94    4) suppressed due to audit_backlog_limit
  95 */
  96 static atomic_t    audit_lost = ATOMIC_INIT(0);
  97
  98 /* The netlink socket. */
  99 static struct sock *audit_sock;
 100
 101 /* The audit_freelist is a list of pre-allocated audit buffers (if more
 102  * than AUDIT_MAXFREE are in use, the audit buffer is freed instead of
 103  * being placed on the freelist). */
 104 static DEFINE_SPINLOCK(audit_freelist_lock);
 105 static int         audit_freelist_count = 0;
 106 static LIST_HEAD(audit_freelist);
 107
 108 static struct sk_buff_head audit_skb_queue;
 109 static struct task_struct *kauditd_task;
 110 static DECLARE_WAIT_QUEUE_HEAD(kauditd_wait);
 111 static DECLARE_WAIT_QUEUE_HEAD(audit_backlog_wait);
 112
 113 /* The netlink socket is only to be read by 1 CPU, which lets us assume
 114  * that list additions and deletions never happen simultaneously in
 115  * auditsc.c */
 116 DECLARE_MUTEX(audit_netlink_sem);
 117
 118 /* AUDIT_BUFSIZ is the size of the temporary buffer used for formatting
 119  * audit records.  Since printk uses a 1024 byte buffer, this buffer
 120  * should be at least that large. */
 121 #define AUDIT_BUFSIZ 1024
 122
 123 /* AUDIT_MAXFREE is the number of empty audit_buffers we keep on the
 124  * audit_freelist.  Doing so eliminates many kmalloc/kfree calls. */
 125 #define AUDIT_MAXFREE  (2*NR_CPUS)
 126
 127 /* The audit_buffer is used when formatting an audit record.  The caller
 128  * locks briefly to get the record off the freelist or to allocate the
 129  * buffer, and locks briefly to send the buffer to the netlink layer or
 130  * to place it on a transmit queue.  Multiple audit_buffers can be in
 131  * use simultaneously. */
 132 struct audit_buffer {
 133         struct list_head     list;
 134         struct sk_buff       *skb;      /* formatted skb ready to send */
 135         struct audit_context *ctx;      /* NULL or associated context */
 136         gfp_t                gfp_mask;
 137 };
 138
 139 static void audit_set_pid(struct audit_buffer *ab, pid_t pid)
 140 {
 141         struct nlmsghdr *nlh = (struct nlmsghdr *)ab->skb->data;
 142         nlh->nlmsg_pid = pid;
 143 }
 144
 145 static void audit_panic(const char *message)
 146 {
 147         switch (audit_failure)
 148         {
 149         case AUDIT_FAIL_SILENT:
 150                 break;
 151         case AUDIT_FAIL_PRINTK:
 152                 printk(KERN_ERR "audit: %s\n", message);
 153                 break;
 154         case AUDIT_FAIL_PANIC:
 155                 panic("audit: %s\n", message);
 156                 break;
 157         }
 158 }
 159
 160 static inline int audit_rate_check(void)
 161 {
 162         static unsigned long    last_check = 0;
 163         static int              messages   = 0;
 164         static DEFINE_SPINLOCK(lock);
 165         unsigned long           flags;
 166         unsigned long           now;
 167         unsigned long           elapsed;
 168         int                     retval     = 0;
 169
 170         if (!audit_rate_limit) return 1;
 171
 172         spin_lock_irqsave(&lock, flags);
 173         if (++messages < audit_rate_limit) {
 174                 retval = 1;
 175         } else {
 176                 now     = jiffies;
 177                 elapsed = now - last_check;
 178                 if (elapsed > HZ) {
 179                         last_check = now;
 180                         messages   = 0;
 181                         retval     = 1;
 182                 }
 183         }
 184         spin_unlock_irqrestore(&lock, flags);
 185
 186         return retval;
 187 }
 188
 189 /* Emit at least 1 message per second, even if audit_rate_check is
 190  * throttling. */
 191 void audit_log_lost(const char *message)
 192 {
 193         static unsigned long    last_msg = 0;
 194         static DEFINE_SPINLOCK(lock);
 195         unsigned long           flags;
 196         unsigned long           now;
 197         int                     print;
 198
 199         atomic_inc(&audit_lost);
 200
 201         print = (audit_failure == AUDIT_FAIL_PANIC || !audit_rate_limit);
 202
 203         if (!print) {
 204                 spin_lock_irqsave(&lock, flags);
 205                 now = jiffies;
 206                 if (now - last_msg > HZ) {
 207                         print = 1;
 208                         last_msg = now;
 209                 }
 210                 spin_unlock_irqrestore(&lock, flags);
 211         }
 212
 213         if (print) {
 214                 printk(KERN_WARNING
 215                        "audit: audit_lost=%d audit_rate_limit=%d audit_backlog_limit=%d\n",
 216                        atomic_read(&audit_lost),
 217                        audit_rate_limit,
 218                        audit_backlog_limit);
 219                 audit_panic(message);
 220         }
 221
 222 }
 223
 224 static int audit_set_rate_limit(int limit, uid_t loginuid)
 225 {
 226         int old          = audit_rate_limit;
 227         audit_rate_limit = limit;
 228         audit_log(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE,
 229                         "audit_rate_limit=%d old=%d by auid=%u",
 230                         audit_rate_limit, old, loginuid);
 231         return old;
 232 }
 233
 234 static int audit_set_backlog_limit(int limit, uid_t loginuid)
 235 {
 236         int old          = audit_backlog_limit;
 237         audit_backlog_limit = limit;
 238         audit_log(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE,
 239                         "audit_backlog_limit=%d old=%d by auid=%u",
 240                         audit_backlog_limit, old, loginuid);
 241         return old;
 242 }
 243
 244 static int audit_set_enabled(int state, uid_t loginuid)
 245 {
 246         int old          = audit_enabled;
 247         if (state != 0 && state != 1)
 248                 return -EINVAL;
 249         audit_enabled = state;
 250         audit_log(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE,
 251                         "audit_enabled=%d old=%d by auid=%u",
 252                         audit_enabled, old, loginuid);
 253         return old;
 254 }
 255
 256 static int audit_set_failure(int state, uid_t loginuid)
 257 {
 258         int old          = audit_failure;
 259         if (state != AUDIT_FAIL_SILENT
 260             && state != AUDIT_FAIL_PRINTK
 261             && state != AUDIT_FAIL_PANIC)
 262                 return -EINVAL;
 263         audit_failure = state;
 264         audit_log(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE,
 265                         "audit_failure=%d old=%d by auid=%u",
 266                         audit_failure, old, loginuid);
 267         return old;
 268 }
 269
 270 static int kauditd_thread(void *dummy)
 271 {
 272         struct sk_buff *skb;
 273
 274         while (1) {
 275                 skb = skb_dequeue(&audit_skb_queue);
 276                 wake_up(&audit_backlog_wait);
 277                 if (skb) {
 278                         if (audit_pid) {
 279                                 int err = netlink_unicast(audit_sock, skb, audit_pid, 0);
 280                                 if (err < 0) {
 281                                         BUG_ON(err != -ECONNREFUSED); /* Shoudn't happen */
 282                                         printk(KERN_ERR "audit: *NO* daemon at audit_pid=%d\n", audit_pid);
 283                                         audit_pid = 0;
 284                                 }
 285                         } else {
 286                                 printk(KERN_NOTICE "%s\n", skb->data + NLMSG_SPACE(0));
 287                                 kfree_skb(skb);
 288                         }
 289                 } else {
 290                         DECLARE_WAITQUEUE(wait, current);
 291                         set_current_state(TASK_INTERRUPTIBLE);
 292                         add_wait_queue(&kauditd_wait, &wait);
 293
 294                         if (!skb_queue_len(&audit_skb_queue)) {
 295                                 try_to_freeze();
 296                                 schedule();
 297                         }
 298
 299                         __set_current_state(TASK_RUNNING);
 300                         remove_wait_queue(&kauditd_wait, &wait);
 301                 }
 302         }
 303 }
 304
 305 void audit_send_reply(int pid, int seq, int type, int done, int multi,
 306                       void *payload, int size)
 307 {
 308         struct sk_buff  *skb;
 309         struct nlmsghdr *nlh;
 310         int             len = NLMSG_SPACE(size);
 311         void            *data;
 312         int             flags = multi ? NLM_F_MULTI : 0;
 313         int             t     = done  ? NLMSG_DONE  : type;
 314
 315         skb = alloc_skb(len, GFP_KERNEL);
 316         if (!skb)
 317                 return;
 318
 319         nlh              = NLMSG_PUT(skb, pid, seq, t, size);
 320         nlh->nlmsg_flags = flags;
 321         data             = NLMSG_DATA(nlh);
 322         memcpy(data, payload, size);
 323
 324         /* Ignore failure. It'll only happen if the sender goes away,
 325            because our timeout is set to infinite. */
 326         netlink_unicast(audit_sock, skb, pid, 0);
 327         return;
 328
 329 nlmsg_failure:                  /* Used by NLMSG_PUT */
 330         if (skb)
 331                 kfree_skb(skb);
 332 }
 333
 334 /*
 335  * Check for appropriate CAP_AUDIT_ capabilities on incoming audit
 336  * control messages.
 337  */
 338 static int audit_netlink_ok(kernel_cap_t eff_cap, u16 msg_type)
 339 {
 340         int err = 0;
 341
 342         switch (msg_type) {
 343         case AUDIT_GET:
 344         case AUDIT_LIST:
 345         case AUDIT_SET:
 346         case AUDIT_ADD:
 347         case AUDIT_DEL:
 348         case AUDIT_SIGNAL_INFO:
 349                 if (!cap_raised(eff_cap, CAP_AUDIT_CONTROL))
 350                         err = -EPERM;
 351                 break;
 352         case AUDIT_USER:
 353         case AUDIT_FIRST_USER_MSG...AUDIT_LAST_USER_MSG:
 354                 if (!cap_raised(eff_cap, CAP_AUDIT_WRITE))
 355                         err = -EPERM;
 356                 break;
 357         default:  /* bad msg */
 358                 err = -EINVAL;
 359         }
 360
 361         return err;
 362 }
 363
 364 static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
 365 {
 366         u32                     uid, pid, seq;
 367         void                    *data;
 368         struct audit_status     *status_get, status_set;
 369         int                     err;
 370         struct audit_buffer     *ab;
 371         u16                     msg_type = nlh->nlmsg_type;
 372         uid_t                   loginuid; /* loginuid of sender */
 373         struct audit_sig_info   sig_data;
 374
 375         err = audit_netlink_ok(NETLINK_CB(skb).eff_cap, msg_type);
 376         if (err)
 377                 return err;
 378
 379         /* As soon as there's any sign of userspace auditd, start kauditd to talk to it */
 380         if (!kauditd_task)
 381                 kauditd_task = kthread_run(kauditd_thread, NULL, "kauditd");
 382         if (IS_ERR(kauditd_task)) {
 383                 err = PTR_ERR(kauditd_task);
 384                 kauditd_task = NULL;
 385                 return err;
 386         }
 387
 388         pid  = NETLINK_CREDS(skb)->pid;
 389         uid  = NETLINK_CREDS(skb)->uid;
 390         loginuid = NETLINK_CB(skb).loginuid;
 391         seq  = nlh->nlmsg_seq;
 392         data = NLMSG_DATA(nlh);
 393
 394         switch (msg_type) {
 395         case AUDIT_GET:
 396                 status_set.enabled       = audit_enabled;
 397                 status_set.failure       = audit_failure;
 398                 status_set.pid           = audit_pid;
 399                 status_set.rate_limit    = audit_rate_limit;
 400                 status_set.backlog_limit = audit_backlog_limit;
 401                 status_set.lost          = atomic_read(&audit_lost);
 402                 status_set.backlog       = skb_queue_len(&audit_skb_queue);
 403                 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_GET, 0, 0,
 404                                  &status_set, sizeof(status_set));
 405                 break;
 406         case AUDIT_SET:
 407                 if (nlh->nlmsg_len < sizeof(struct audit_status))
 408                         return -EINVAL;
 409                 status_get   = (struct audit_status *)data;
 410                 if (status_get->mask & AUDIT_STATUS_ENABLED) {
 411                         err = audit_set_enabled(status_get->enabled, loginuid);
 412                         if (err < 0) return err;
 413                 }
 414                 if (status_get->mask & AUDIT_STATUS_FAILURE) {
 415                         err = audit_set_failure(status_get->failure, loginuid);
 416                         if (err < 0) return err;
 417                 }
 418                 if (status_get->mask & AUDIT_STATUS_PID) {
 419                         int old   = audit_pid;
 420                         audit_pid = status_get->pid;
 421                         audit_log(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE,
 422                                 "audit_pid=%d old=%d by auid=%u",
 423                                   audit_pid, old, loginuid);
 424                 }
 425                 if (status_get->mask & AUDIT_STATUS_RATE_LIMIT)
 426                         audit_set_rate_limit(status_get->rate_limit, loginuid);
 427                 if (status_get->mask & AUDIT_STATUS_BACKLOG_LIMIT)
 428                         audit_set_backlog_limit(status_get->backlog_limit,
 429                                                         loginuid);
 430                 break;
 431         case AUDIT_USER:
 432         case AUDIT_FIRST_USER_MSG...AUDIT_LAST_USER_MSG:
 433                 if (!audit_enabled && msg_type != AUDIT_USER_AVC)
 434                         return 0;
 435
 436                 err = audit_filter_user(&NETLINK_CB(skb), msg_type);
 437                 if (err == 1) {
 438                         err = 0;
 439                         ab = audit_log_start(NULL, GFP_KERNEL, msg_type);
 440                         if (ab) {
 441                                 audit_log_format(ab,
 442                                                  "user pid=%d uid=%u auid=%u msg='%.1024s'",
 443                                                  pid, uid, loginuid, (char *)data);
 444                                 audit_set_pid(ab, pid);
 445                                 audit_log_end(ab);
 446                         }
 447                 }
 448                 break;
 449         case AUDIT_ADD:
 450         case AUDIT_DEL:
 451                 if (nlh->nlmsg_len < sizeof(struct audit_rule))
 452                         return -EINVAL;
 453                 /* fallthrough */
 454         case AUDIT_LIST:
 455                 err = audit_receive_filter(nlh->nlmsg_type, NETLINK_CB(skb).pid,
 456                                            uid, seq, data, loginuid);
 457                 break;
 458         case AUDIT_SIGNAL_INFO:
 459                 sig_data.uid = audit_sig_uid;
 460                 sig_data.pid = audit_sig_pid;
 461                 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_SIGNAL_INFO,
 462                                 0, 0, &sig_data, sizeof(sig_data));
 463                 break;
 464         default:
 465                 err = -EINVAL;
 466                 break;
 467         }
 468
 469         return err < 0 ? err : 0;
 470 }
 471
 472 /* Get message from skb (based on rtnetlink_rcv_skb).  Each message is
 473  * processed by audit_receive_msg.  Malformed skbs with wrong length are
 474  * discarded silently.  */
 475 static void audit_receive_skb(struct sk_buff *skb)
 476 {
 477         int             err;
 478         struct nlmsghdr *nlh;
 479         u32             rlen;
 480
 481         while (skb->len >= NLMSG_SPACE(0)) {
 482                 nlh = (struct nlmsghdr *)skb->data;
 483                 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
 484                         return;
 485                 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
 486                 if (rlen > skb->len)
 487                         rlen = skb->len;
 488                 if ((err = audit_receive_msg(skb, nlh))) {
 489                         netlink_ack(skb, nlh, err);
 490                 } else if (nlh->nlmsg_flags & NLM_F_ACK)
 491                         netlink_ack(skb, nlh, 0);
 492                 skb_pull(skb, rlen);
 493         }
 494 }
 495
 496 /* Receive messages from netlink socket. */
 497 static void audit_receive(struct sock *sk, int length)
 498 {
 499         struct sk_buff  *skb;
 500         unsigned int qlen;
 501
 502         down(&audit_netlink_sem);
 503
 504         for (qlen = skb_queue_len(&sk->sk_receive_queue); qlen; qlen--) {
 505                 skb = skb_dequeue(&sk->sk_receive_queue);
 506                 audit_receive_skb(skb);
 507                 kfree_skb(skb);
 508         }
 509         up(&audit_netlink_sem);
 510 }
 511
 512
 513 /* Initialize audit support at boot time. */
 514 static int __init audit_init(void)
 515 {
 516         printk(KERN_INFO "audit: initializing netlink socket (%s)\n",
 517                audit_default ? "enabled" : "disabled");
 518         audit_sock = netlink_kernel_create(NETLINK_AUDIT, 0, audit_receive,
 519                                            THIS_MODULE);
 520         if (!audit_sock)
 521                 audit_panic("cannot initialize netlink socket");
 522
 523         audit_sock->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
 524         skb_queue_head_init(&audit_skb_queue);
 525         audit_initialized = 1;
 526         audit_enabled = audit_default;
 527         audit_log(NULL, GFP_KERNEL, AUDIT_KERNEL, "initialized");
 528         return 0;
 529 }
 530 __initcall(audit_init);
 531
 532 /* Process kernel command-line parameter at boot time.  audit=0 or audit=1. */
 533 static int __init audit_enable(char *str)
 534 {
 535         audit_default = !!simple_strtol(str, NULL, 0);
 536         printk(KERN_INFO "audit: %s%s\n",
 537                audit_default ? "enabled" : "disabled",
 538                audit_initialized ? "" : " (after initialization)");
 539         if (audit_initialized)
 540                 audit_enabled = audit_default;
 541         return 0;
 542 }
 543
 544 __setup("audit=", audit_enable);
 545
 546 static void audit_buffer_free(struct audit_buffer *ab)
 547 {
 548         unsigned long flags;
 549
 550         if (!ab)
 551                 return;
 552
 553         if (ab->skb)
 554                 kfree_skb(ab->skb);
 555
 556         spin_lock_irqsave(&audit_freelist_lock, flags);
 557         if (++audit_freelist_count > AUDIT_MAXFREE)
 558                 kfree(ab);
 559         else
 560                 list_add(&ab->list, &audit_freelist);
 561         spin_unlock_irqrestore(&audit_freelist_lock, flags);
 562 }
 563
 564 static struct audit_buffer * audit_buffer_alloc(struct audit_context *ctx,
 565                                                 gfp_t gfp_mask, int type)
 566 {
 567         unsigned long flags;
 568         struct audit_buffer *ab = NULL;
 569         struct nlmsghdr *nlh;
 570
 571         spin_lock_irqsave(&audit_freelist_lock, flags);
 572         if (!list_empty(&audit_freelist)) {
 573                 ab = list_entry(audit_freelist.next,
 574                                 struct audit_buffer, list);
 575                 list_del(&ab->list);
 576                 --audit_freelist_count;
 577         }
 578         spin_unlock_irqrestore(&audit_freelist_lock, flags);
 579
 580         if (!ab) {
 581                 ab = kmalloc(sizeof(*ab), gfp_mask);
 582                 if (!ab)
 583                         goto err;
 584         }
 585
 586         ab->skb = alloc_skb(AUDIT_BUFSIZ, gfp_mask);
 587         if (!ab->skb)
 588                 goto err;
 589
 590         ab->ctx = ctx;
 591         ab->gfp_mask = gfp_mask;
 592         nlh = (struct nlmsghdr *)skb_put(ab->skb, NLMSG_SPACE(0));
 593         nlh->nlmsg_type = type;
 594         nlh->nlmsg_flags = 0;
 595         nlh->nlmsg_pid = 0;
 596         nlh->nlmsg_seq = 0;
 597         return ab;
 598 err:
 599         audit_buffer_free(ab);
 600         return NULL;
 601 }
 602
 603 /* Compute a serial number for the audit record.  Audit records are
 604  * written to user-space as soon as they are generated, so a complete
 605  * audit record may be written in several pieces.  The timestamp of the
 606  * record and this serial number are used by the user-space tools to
 607  * determine which pieces belong to the same audit record.  The
 608  * (timestamp,serial) tuple is unique for each syscall and is live from
 609  * syscall entry to syscall exit.
 610  *
 611  * NOTE: Another possibility is to store the formatted records off the
 612  * audit context (for those records that have a context), and emit them
 613  * all at syscall exit.  However, this could delay the reporting of
 614  * significant errors until syscall exit (or never, if the system
 615  * halts). */
 616
 617 unsigned int audit_serial(void)
 618 {
 619         static spinlock_t serial_lock = SPIN_LOCK_UNLOCKED;
 620         static unsigned int serial = 0;
 621
 622         unsigned long flags;
 623         unsigned int ret;
 624
 625         spin_lock_irqsave(&serial_lock, flags);
 626         do {
 627                 ret = ++serial;
 628         } while (unlikely(!ret));
 629         spin_unlock_irqrestore(&serial_lock, flags);
 630
 631         return ret;
 632 }
 633
 634 static inline void audit_get_stamp(struct audit_context *ctx,
 635                                    struct timespec *t, unsigned int *serial)
 636 {
 637         if (ctx)
 638                 auditsc_get_stamp(ctx, t, serial);
 639         else {
 640                 *t = CURRENT_TIME;
 641                 *serial = audit_serial();
 642         }
 643 }
 644
 645 /* Obtain an audit buffer.  This routine does locking to obtain the
 646  * audit buffer, but then no locking is required for calls to
 647  * audit_log_*format.  If the tsk is a task that is currently in a
 648  * syscall, then the syscall is marked as auditable and an audit record
 649  * will be written at syscall exit.  If there is no associated task, tsk
 650  * should be NULL. */
 651
 652 struct audit_buffer *audit_log_start(struct audit_context *ctx, gfp_t gfp_mask,
 653                                      int type)
 654 {
 655         struct audit_buffer     *ab     = NULL;
 656         struct timespec         t;
 657         unsigned int            serial;
 658         int reserve;
 659         unsigned long timeout_start = jiffies;
 660
 661         if (!audit_initialized)
 662                 return NULL;
 663
 664         if (gfp_mask & __GFP_WAIT)
 665                 reserve = 0;
 666         else
 667                 reserve = 5; /* Allow atomic callers to go up to five
 668                                 entries over the normal backlog limit */
 669
 670         while (audit_backlog_limit
 671                && skb_queue_len(&audit_skb_queue) > audit_backlog_limit + reserve) {
 672                 if (gfp_mask & __GFP_WAIT && audit_backlog_wait_time
 673                     && time_before(jiffies, timeout_start + audit_backlog_wait_time)) {
 674
 675                         /* Wait for auditd to drain the queue a little */
 676                         DECLARE_WAITQUEUE(wait, current);
 677                         set_current_state(TASK_INTERRUPTIBLE);
 678                         add_wait_queue(&audit_backlog_wait, &wait);
 679
 680                         if (audit_backlog_limit &&
 681                             skb_queue_len(&audit_skb_queue) > audit_backlog_limit)
 682                                 schedule_timeout(timeout_start + audit_backlog_wait_time - jiffies);
 683
 684                         __set_current_state(TASK_RUNNING);
 685                         remove_wait_queue(&audit_backlog_wait, &wait);
 686                         continue;
 687                 }
 688                 if (audit_rate_check())
 689                         printk(KERN_WARNING
 690                                "audit: audit_backlog=%d > "
 691                                "audit_backlog_limit=%d\n",
 692                                skb_queue_len(&audit_skb_queue),
 693                                audit_backlog_limit);
 694                 audit_log_lost("backlog limit exceeded");
 695                 audit_backlog_wait_time = audit_backlog_wait_overflow;
 696                 wake_up(&audit_backlog_wait);
 697                 return NULL;
 698         }
 699
 700         ab = audit_buffer_alloc(ctx, gfp_mask, type);
 701         if (!ab) {
 702                 audit_log_lost("out of memory in audit_log_start");
 703                 return NULL;
 704         }
 705
 706         audit_get_stamp(ab->ctx, &t, &serial);
 707
 708         audit_log_format(ab, "audit(%lu.%03lu:%u): ",
 709                          t.tv_sec, t.tv_nsec/1000000, serial);
 710         return ab;
 711 }
 712
 713 /**
 714  * audit_expand - expand skb in the audit buffer
 715  * @ab: audit_buffer
 716  *
 717  * Returns 0 (no space) on failed expansion, or available space if
 718  * successful.
 719  */
 720 static inline int audit_expand(struct audit_buffer *ab, int extra)
 721 {
 722         struct sk_buff *skb = ab->skb;
 723         int ret = pskb_expand_head(skb, skb_headroom(skb), extra,
 724                                    ab->gfp_mask);
 725         if (ret < 0) {
 726                 audit_log_lost("out of memory in audit_expand");
 727                 return 0;
 728         }
 729         return skb_tailroom(skb);
 730 }
 731
 732 /* Format an audit message into the audit buffer.  If there isn't enough
 733  * room in the audit buffer, more room will be allocated and vsnprint
 734  * will be called a second time.  Currently, we assume that a printk
 735  * can't format message larger than 1024 bytes, so we don't either. */
 736 static void audit_log_vformat(struct audit_buffer *ab, const char *fmt,
 737                               va_list args)
 738 {
 739         int len, avail;
 740         struct sk_buff *skb;
 741         va_list args2;
 742
 743         if (!ab)
 744                 return;
 745
 746         BUG_ON(!ab->skb);
 747         skb = ab->skb;
 748         avail = skb_tailroom(skb);
 749         if (avail == 0) {
 750                 avail = audit_expand(ab, AUDIT_BUFSIZ);
 751                 if (!avail)
 752                         goto out;
 753         }
 754         va_copy(args2, args);
 755         len = vsnprintf(skb->tail, avail, fmt, args);
 756         if (len >= avail) {
 757                 /* The printk buffer is 1024 bytes long, so if we get
 758                  * here and AUDIT_BUFSIZ is at least 1024, then we can
 759                  * log everything that printk could have logged. */
 760                 avail = audit_expand(ab, max_t(unsigned, AUDIT_BUFSIZ, 1+len-avail));
 761                 if (!avail)
 762                         goto out;
 763                 len = vsnprintf(skb->tail, avail, fmt, args2);
 764         }
 765         if (len > 0)
 766                 skb_put(skb, len);
 767 out:
 768         return;
 769 }
 770
 771 /* Format a message into the audit buffer.  All the work is done in
 772  * audit_log_vformat. */
 773 void audit_log_format(struct audit_buffer *ab, const char *fmt, ...)
 774 {
 775         va_list args;
 776
 777         if (!ab)
 778                 return;
 779         va_start(args, fmt);
 780         audit_log_vformat(ab, fmt, args);
 781         va_end(args);
 782 }
 783
 784 /* This function will take the passed buf and convert it into a string of
 785  * ascii hex digits. The new string is placed onto the skb. */
 786 void audit_log_hex(struct audit_buffer *ab, const unsigned char *buf,
 787                 size_t len)
 788 {
 789         int i, avail, new_len;
 790         unsigned char *ptr;
 791         struct sk_buff *skb;
 792         static const unsigned char *hex = "0123456789ABCDEF";
 793
 794         BUG_ON(!ab->skb);
 795         skb = ab->skb;
 796         avail = skb_tailroom(skb);
 797         new_len = len<<1;
 798         if (new_len >= avail) {
 799                 /* Round the buffer request up to the next multiple */
 800                 new_len = AUDIT_BUFSIZ*(((new_len-avail)/AUDIT_BUFSIZ) + 1);
 801                 avail = audit_expand(ab, new_len);
 802                 if (!avail)
 803                         return;
 804         }
 805
 806         ptr = skb->tail;
 807         for (i=0; i<len; i++) {
 808                 *ptr++ = hex[(buf[i] & 0xF0)>>4]; /* Upper nibble */
 809                 *ptr++ = hex[buf[i] & 0x0F];      /* Lower nibble */
 810         }
 811         *ptr = 0;
 812         skb_put(skb, len << 1); /* new string is twice the old string */
 813 }
 814
 815 /* This code will escape a string that is passed to it if the string
 816  * contains a control character, unprintable character, double quote mark,
 817  * or a space. Unescaped strings will start and end with a double quote mark.
 818  * Strings that are escaped are printed in hex (2 digits per char). */
 819 void audit_log_untrustedstring(struct audit_buffer *ab, const char *string)
 820 {
 821         const unsigned char *p = string;
 822
 823         while (*p) {
 824                 if (*p == '"' || *p < 0x21 || *p > 0x7f) {
 825                         audit_log_hex(ab, string, strlen(string));
 826                         return;
 827                 }
 828                 p++;
 829         }
 830         audit_log_format(ab, "\"%s\"", string);
 831 }
 832
 833 /* This is a helper-function to print the escaped d_path */
 834 void audit_log_d_path(struct audit_buffer *ab, const char *prefix,
 835                       struct dentry *dentry, struct vfsmount *vfsmnt)
 836 {
 837         char *p, *path;
 838
 839         if (prefix)
 840                 audit_log_format(ab, " %s", prefix);
 841
 842         /* We will allow 11 spaces for ' (deleted)' to be appended */
 843         path = kmalloc(PATH_MAX+11, ab->gfp_mask);
 844         if (!path) {
 845                 audit_log_format(ab, "<no memory>");
 846                 return;
 847         }
 848         p = d_path(dentry, vfsmnt, path, PATH_MAX+11);
 849         if (IS_ERR(p)) { /* Should never happen since we send PATH_MAX */
 850                 /* FIXME: can we save some information here? */
 851                 audit_log_format(ab, "<too long>");
 852         } else
 853                 audit_log_untrustedstring(ab, p);
 854         kfree(path);
 855 }
 856
 857 /* The netlink_* functions cannot be called inside an irq context, so
 858  * the audit buffer is places on a queue and a tasklet is scheduled to
 859  * remove them from the queue outside the irq context.  May be called in
 860  * any context. */
 861 void audit_log_end(struct audit_buffer *ab)
 862 {
 863         if (!ab)
 864                 return;
 865         if (!audit_rate_check()) {
 866                 audit_log_lost("rate limit exceeded");
 867         } else {
 868                 if (audit_pid) {
 869                         struct nlmsghdr *nlh = (struct nlmsghdr *)ab->skb->data;
 870                         nlh->nlmsg_len = ab->skb->len - NLMSG_SPACE(0);
 871                         skb_queue_tail(&audit_skb_queue, ab->skb);
 872                         ab->skb = NULL;
 873                         wake_up_interruptible(&kauditd_wait);
 874                 } else {
 875                         printk(KERN_NOTICE "%s\n", ab->skb->data + NLMSG_SPACE(0));
 876                 }
 877         }
 878         audit_buffer_free(ab);
 879 }
 880
 881 /* Log an audit record.  This is a convenience function that calls
 882  * audit_log_start, audit_log_vformat, and audit_log_end.  It may be
 883  * called in any context. */
 884 void audit_log(struct audit_context *ctx, gfp_t gfp_mask, int type,
 885                const char *fmt, ...)
 886 {
 887         struct audit_buffer *ab;
 888         va_list args;
 889
 890         ab = audit_log_start(ctx, gfp_mask, type);
 891         if (ab) {
 892                 va_start(args, fmt);
 893                 audit_log_vformat(ab, fmt, args);
 894                 va_end(args);
 895                 audit_log_end(ab);
 896         }
 897 }