kernel/auditsc.c

   1 /* auditsc.c -- System-call auditing support -*- linux-c -*-
   2  * Handles all system-call specific auditing features.
   3  *
   4  * Copyright 2003-2004 Red Hat Inc., Durham, North Carolina.
   5  * All Rights Reserved.
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, write to the Free Software
  19  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  * Written by Rickard E. (Rik) Faith <faith@redhat.com>
  22  *
  23  * Many of the ideas implemented here are from Stephen C. Tweedie,
  24  * especially the idea of avoiding a copy by using getname.
  25  *
  26  * The method for actual interception of syscall entry and exit (not in
  27  * this file -- see entry.S) is based on a GPL'd patch written by
  28  * okir@suse.de and Copyright 2003 SuSE Linux AG.
  29  *
  30  */
  31
  32 #include <linux/init.h>
  33 #include <asm/atomic.h>
  34 #include <asm/types.h>
  35 #include <linux/mm.h>
  36 #include <linux/module.h>
  37
  38 #include <linux/audit.h>
  39 #include <linux/personality.h>
  40 #include <linux/time.h>
  41 #include <asm/unistd.h>
  42
  43 /* 0 = no checking
  44    1 = put_count checking
  45    2 = verbose put_count checking
  46 */
  47 #define AUDIT_DEBUG 0
  48
  49 /* No syscall auditing will take place unless audit_enabled != 0. */
  50 extern int audit_enabled;
  51
  52 /* AUDIT_NAMES is the number of slots we reserve in the audit_context
  53  * for saving names from getname(). */
  54 #define AUDIT_NAMES    20
  55
  56 /* AUDIT_NAMES_RESERVED is the number of slots we reserve in the
  57  * audit_context from being used for nameless inodes from
  58  * path_lookup. */
  59 #define AUDIT_NAMES_RESERVED 7
  60
  61 /* At task start time, the audit_state is set in the audit_context using
  62    a per-task filter.  At syscall entry, the audit_state is augmented by
  63    the syscall filter. */
  64 enum audit_state {
  65         AUDIT_DISABLED,         /* Do not create per-task audit_context.
  66                                  * No syscall-specific audit records can
  67                                  * be generated. */
  68         AUDIT_SETUP_CONTEXT,    /* Create the per-task audit_context,
  69                                  * but don't necessarily fill it in at
  70                                  * syscall entry time (i.e., filter
  71                                  * instead). */
  72         AUDIT_BUILD_CONTEXT,    /* Create the per-task audit_context,
  73                                  * and always fill it in at syscall
  74                                  * entry time.  This makes a full
  75                                  * syscall record available if some
  76                                  * other part of the kernel decides it
  77                                  * should be recorded. */
  78         AUDIT_RECORD_CONTEXT    /* Create the per-task audit_context,
  79                                  * always fill it in at syscall entry
  80                                  * time, and always write out the audit
  81                                  * record at syscall exit time.  */
  82 };
  83
  84 /* When fs/namei.c:getname() is called, we store the pointer in name and
  85  * we don't let putname() free it (instead we free all of the saved
  86  * pointers at syscall exit time).
  87  *
  88  * Further, in fs/namei.c:path_lookup() we store the inode and device. */
  89 struct audit_names {
  90         const char      *name;
  91         unsigned long   ino;
  92         dev_t           rdev;
  93 };
  94
  95 /* The per-task audit context. */
  96 struct audit_context {
  97         int                 in_syscall; /* 1 if task is in a syscall */
  98         enum audit_state    state;
  99         unsigned int        serial;     /* serial number for record */
 100         struct timespec     ctime;      /* time of syscall entry */
 101         uid_t               loginuid;   /* login uid (identity) */
 102         int                 major;      /* syscall number */
 103         unsigned long       argv[4];    /* syscall arguments */
 104         int                 return_valid; /* return code is valid */
 105         int                 return_code;/* syscall return code */
 106         int                 auditable;  /* 1 if record should be written */
 107         int                 name_count;
 108         struct audit_names  names[AUDIT_NAMES];
 109         struct audit_context *previous; /* For nested syscalls */
 110
 111                                 /* Save things to print about task_struct */
 112         pid_t               pid;
 113         uid_t               uid, euid, suid, fsuid;
 114         gid_t               gid, egid, sgid, fsgid;
 115         unsigned long       personality;
 116
 117 #if AUDIT_DEBUG
 118         int                 put_count;
 119         int                 ino_count;
 120 #endif
 121 };
 122
 123                                 /* Public API */
 124 /* There are three lists of rules -- one to search at task creation
 125  * time, one to search at syscall entry time, and another to search at
 126  * syscall exit time. */
 127 static LIST_HEAD(audit_tsklist);
 128 static LIST_HEAD(audit_entlist);
 129 static LIST_HEAD(audit_extlist);
 130
 131 struct audit_entry {
 132         struct list_head  list;
 133         struct rcu_head   rcu;
 134         struct audit_rule rule;
 135 };
 136
 137 /* Check to see if two rules are identical.  It is called from
 138  * audit_del_rule during AUDIT_DEL. */
 139 static int audit_compare_rule(struct audit_rule *a, struct audit_rule *b)
 140 {
 141         int i;
 142
 143         if (a->flags != b->flags)
 144                 return 1;
 145
 146         if (a->action != b->action)
 147                 return 1;
 148
 149         if (a->field_count != b->field_count)
 150                 return 1;
 151
 152         for (i = 0; i < a->field_count; i++) {
 153                 if (a->fields[i] != b->fields[i]
 154                     || a->values[i] != b->values[i])
 155                         return 1;
 156         }
 157
 158         for (i = 0; i < AUDIT_BITMASK_SIZE; i++)
 159                 if (a->mask[i] != b->mask[i])
 160                         return 1;
 161
 162         return 0;
 163 }
 164
 165 /* Note that audit_add_rule and audit_del_rule are called via
 166  * audit_receive() in audit.c, and are protected by
 167  * audit_netlink_sem. */
 168 static inline int audit_add_rule(struct audit_entry *entry,
 169                                  struct list_head *list)
 170 {
 171         if (entry->rule.flags & AUDIT_PREPEND) {
 172                 entry->rule.flags &= ~AUDIT_PREPEND;
 173                 list_add_rcu(&entry->list, list);
 174         } else {
 175                 list_add_tail_rcu(&entry->list, list);
 176         }
 177         return 0;
 178 }
 179
 180 static void audit_free_rule(struct rcu_head *head)
 181 {
 182         struct audit_entry *e = container_of(head, struct audit_entry, rcu);
 183         kfree(e);
 184 }
 185
 186 /* Note that audit_add_rule and audit_del_rule are called via
 187  * audit_receive() in audit.c, and are protected by
 188  * audit_netlink_sem. */
 189 static inline int audit_del_rule(struct audit_rule *rule,
 190                                  struct list_head *list)
 191 {
 192         struct audit_entry  *e;
 193
 194         /* Do not use the _rcu iterator here, since this is the only
 195          * deletion routine. */
 196         list_for_each_entry(e, list, list) {
 197                 if (!audit_compare_rule(rule, &e->rule)) {
 198                         list_del_rcu(&e->list);
 199                         call_rcu(&e->rcu, audit_free_rule);
 200                         return 0;
 201                 }
 202         }
 203         return -EFAULT;         /* No matching rule */
 204 }
 205
 206 #ifdef CONFIG_NET
 207 /* Copy rule from user-space to kernel-space.  Called during
 208  * AUDIT_ADD. */
 209 static int audit_copy_rule(struct audit_rule *d, struct audit_rule *s)
 210 {
 211         int i;
 212
 213         if (s->action != AUDIT_NEVER
 214             && s->action != AUDIT_POSSIBLE
 215             && s->action != AUDIT_ALWAYS)
 216                 return -1;
 217         if (s->field_count < 0 || s->field_count > AUDIT_MAX_FIELDS)
 218                 return -1;
 219
 220         d->flags        = s->flags;
 221         d->action       = s->action;
 222         d->field_count  = s->field_count;
 223         for (i = 0; i < d->field_count; i++) {
 224                 d->fields[i] = s->fields[i];
 225                 d->values[i] = s->values[i];
 226         }
 227         for (i = 0; i < AUDIT_BITMASK_SIZE; i++) d->mask[i] = s->mask[i];
 228         return 0;
 229 }
 230
 231 int audit_receive_filter(int type, int pid, int uid, int seq, void *data)
 232 {
 233         u32                flags;
 234         struct audit_entry *entry;
 235         int                err = 0;
 236
 237         switch (type) {
 238         case AUDIT_LIST:
 239                 /* The *_rcu iterators not needed here because we are
 240                    always called with audit_netlink_sem held. */
 241                 list_for_each_entry(entry, &audit_tsklist, list)
 242                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 243                                          &entry->rule, sizeof(entry->rule));
 244                 list_for_each_entry(entry, &audit_entlist, list)
 245                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 246                                          &entry->rule, sizeof(entry->rule));
 247                 list_for_each_entry(entry, &audit_extlist, list)
 248                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 249                                          &entry->rule, sizeof(entry->rule));
 250                 audit_send_reply(pid, seq, AUDIT_LIST, 1, 1, NULL, 0);
 251                 break;
 252         case AUDIT_ADD:
 253                 if (!capable(CAP_SYS_ADMIN))
 254                         return -EPERM;
 255                 if (!(entry = kmalloc(sizeof(*entry), GFP_KERNEL)))
 256                         return -ENOMEM;
 257                 if (audit_copy_rule(&entry->rule, data)) {
 258                         kfree(entry);
 259                         return -EINVAL;
 260                 }
 261                 flags = entry->rule.flags;
 262                 if (!err && (flags & AUDIT_PER_TASK))
 263                         err = audit_add_rule(entry, &audit_tsklist);
 264                 if (!err && (flags & AUDIT_AT_ENTRY))
 265                         err = audit_add_rule(entry, &audit_entlist);
 266                 if (!err && (flags & AUDIT_AT_EXIT))
 267                         err = audit_add_rule(entry, &audit_extlist);
 268                 break;
 269         case AUDIT_DEL:
 270                 flags =((struct audit_rule *)data)->flags;
 271                 if (!err && (flags & AUDIT_PER_TASK))
 272                         err = audit_del_rule(data, &audit_tsklist);
 273                 if (!err && (flags & AUDIT_AT_ENTRY))
 274                         err = audit_del_rule(data, &audit_entlist);
 275                 if (!err && (flags & AUDIT_AT_EXIT))
 276                         err = audit_del_rule(data, &audit_extlist);
 277                 break;
 278         default:
 279                 return -EINVAL;
 280         }
 281
 282         return err;
 283 }
 284 #endif
 285
 286 /* Compare a task_struct with an audit_rule.  Return 1 on match, 0
 287  * otherwise. */
 288 static int audit_filter_rules(struct task_struct *tsk,
 289                               struct audit_rule *rule,
 290                               struct audit_context *ctx,
 291                               enum audit_state *state)
 292 {
 293         int i, j;
 294
 295         for (i = 0; i < rule->field_count; i++) {
 296                 u32 field  = rule->fields[i] & ~AUDIT_NEGATE;
 297                 u32 value  = rule->values[i];
 298                 int result = 0;
 299
 300                 switch (field) {
 301                 case AUDIT_PID:
 302                         result = (tsk->pid == value);
 303                         break;
 304                 case AUDIT_UID:
 305                         result = (tsk->uid == value);
 306                         break;
 307                 case AUDIT_EUID:
 308                         result = (tsk->euid == value);
 309                         break;
 310                 case AUDIT_SUID:
 311                         result = (tsk->suid == value);
 312                         break;
 313                 case AUDIT_FSUID:
 314                         result = (tsk->fsuid == value);
 315                         break;
 316                 case AUDIT_GID:
 317                         result = (tsk->gid == value);
 318                         break;
 319                 case AUDIT_EGID:
 320                         result = (tsk->egid == value);
 321                         break;
 322                 case AUDIT_SGID:
 323                         result = (tsk->sgid == value);
 324                         break;
 325                 case AUDIT_FSGID:
 326                         result = (tsk->fsgid == value);
 327                         break;
 328                 case AUDIT_PERS:
 329                         result = (tsk->personality == value);
 330                         break;
 331
 332                 case AUDIT_EXIT:
 333                         if (ctx && ctx->return_valid)
 334                                 result = (ctx->return_code == value);
 335                         break;
 336                 case AUDIT_SUCCESS:
 337                         if (ctx && ctx->return_valid)
 338                                 result = (ctx->return_code >= 0);
 339                         break;
 340                 case AUDIT_DEVMAJOR:
 341                         if (ctx) {
 342                                 for (j = 0; j < ctx->name_count; j++) {
 343                                         if (MAJOR(ctx->names[j].rdev)==value) {
 344                                                 ++result;
 345                                                 break;
 346                                         }
 347                                 }
 348                         }
 349                         break;
 350                 case AUDIT_DEVMINOR:
 351                         if (ctx) {
 352                                 for (j = 0; j < ctx->name_count; j++) {
 353                                         if (MINOR(ctx->names[j].rdev)==value) {
 354                                                 ++result;
 355                                                 break;
 356                                         }
 357                                 }
 358                         }
 359                         break;
 360                 case AUDIT_INODE:
 361                         if (ctx) {
 362                                 for (j = 0; j < ctx->name_count; j++) {
 363                                         if (MINOR(ctx->names[j].ino)==value) {
 364                                                 ++result;
 365                                                 break;
 366                                         }
 367                                 }
 368                         }
 369                         break;
 370                 case AUDIT_LOGINUID:
 371                         result = 0;
 372                         if (ctx)
 373                                 result = (ctx->loginuid == value);
 374                         break;
 375                 case AUDIT_ARG0:
 376                 case AUDIT_ARG1:
 377                 case AUDIT_ARG2:
 378                 case AUDIT_ARG3:
 379                         if (ctx)
 380                                 result = (ctx->argv[field-AUDIT_ARG0]==value);
 381                         break;
 382                 }
 383
 384                 if (rule->fields[i] & AUDIT_NEGATE)
 385                         result = !result;
 386                 if (!result)
 387                         return 0;
 388         }
 389         switch (rule->action) {
 390         case AUDIT_NEVER:    *state = AUDIT_DISABLED;       break;
 391         case AUDIT_POSSIBLE: *state = AUDIT_BUILD_CONTEXT;  break;
 392         case AUDIT_ALWAYS:   *state = AUDIT_RECORD_CONTEXT; break;
 393         }
 394         return 1;
 395 }
 396
 397 /* At process creation time, we can determine if system-call auditing is
 398  * completely disabled for this task.  Since we only have the task
 399  * structure at this point, we can only check uid and gid.
 400  */
 401 static enum audit_state audit_filter_task(struct task_struct *tsk)
 402 {
 403         struct audit_entry *e;
 404         enum audit_state   state;
 405
 406         rcu_read_lock();
 407         list_for_each_entry_rcu(e, &audit_tsklist, list) {
 408                 if (audit_filter_rules(tsk, &e->rule, NULL, &state)) {
 409                         rcu_read_unlock();
 410                         return state;
 411                 }
 412         }
 413         rcu_read_unlock();
 414         return AUDIT_BUILD_CONTEXT;
 415 }
 416
 417 /* At syscall entry and exit time, this filter is called if the
 418  * audit_state is not low enough that auditing cannot take place, but is
 419  * also not high enough that we already know we have to write and audit
 420  * record (i.e., the state is AUDIT_SETUP_CONTEXT or  AUDIT_BUILD_CONTEXT).
 421  */
 422 static enum audit_state audit_filter_syscall(struct task_struct *tsk,
 423                                              struct audit_context *ctx,
 424                                              struct list_head *list)
 425 {
 426         struct audit_entry *e;
 427         enum audit_state   state;
 428         int                word = AUDIT_WORD(ctx->major);
 429         int                bit  = AUDIT_BIT(ctx->major);
 430
 431         rcu_read_lock();
 432         list_for_each_entry_rcu(e, list, list) {
 433                 if ((e->rule.mask[word] & bit) == bit
 434                     && audit_filter_rules(tsk, &e->rule, ctx, &state)) {
 435                         rcu_read_unlock();
 436                         return state;
 437                 }
 438         }
 439         rcu_read_unlock();
 440         return AUDIT_BUILD_CONTEXT;
 441 }
 442
 443 /* This should be called with task_lock() held. */
 444 static inline struct audit_context *audit_get_context(struct task_struct *tsk,
 445                                                       int return_valid,
 446                                                       int return_code)
 447 {
 448         struct audit_context *context = tsk->audit_context;
 449
 450         if (likely(!context))
 451                 return NULL;
 452         context->return_valid = return_valid;
 453         context->return_code  = return_code;
 454
 455         if (context->in_syscall && !context->auditable) {
 456                 enum audit_state state;
 457                 state = audit_filter_syscall(tsk, context, &audit_extlist);
 458                 if (state == AUDIT_RECORD_CONTEXT)
 459                         context->auditable = 1;
 460         }
 461
 462         context->pid = tsk->pid;
 463         context->uid = tsk->uid;
 464         context->gid = tsk->gid;
 465         context->euid = tsk->euid;
 466         context->suid = tsk->suid;
 467         context->fsuid = tsk->fsuid;
 468         context->egid = tsk->egid;
 469         context->sgid = tsk->sgid;
 470         context->fsgid = tsk->fsgid;
 471         context->personality = tsk->personality;
 472         tsk->audit_context = NULL;
 473         return context;
 474 }
 475
 476 static inline void audit_free_names(struct audit_context *context)
 477 {
 478         int i;
 479
 480 #if AUDIT_DEBUG == 2
 481         if (context->auditable
 482             ||context->put_count + context->ino_count != context->name_count) {
 483                 printk(KERN_ERR "audit.c:%d(:%d): major=%d in_syscall=%d"
 484                        " name_count=%d put_count=%d"
 485                        " ino_count=%d [NOT freeing]\n",
 486                        __LINE__,
 487                        context->serial, context->major, context->in_syscall,
 488                        context->name_count, context->put_count,
 489                        context->ino_count);
 490                 for (i = 0; i < context->name_count; i++)
 491                         printk(KERN_ERR "names[%d] = %p = %s\n", i,
 492                                context->names[i].name,
 493                                context->names[i].name);
 494                 dump_stack();
 495                 return;
 496         }
 497 #endif
 498 #if AUDIT_DEBUG
 499         context->put_count  = 0;
 500         context->ino_count  = 0;
 501 #endif
 502
 503         for (i = 0; i < context->name_count; i++)
 504                 if (context->names[i].name)
 505                         __putname(context->names[i].name);
 506         context->name_count = 0;
 507 }
 508
 509 static inline void audit_zero_context(struct audit_context *context,
 510                                       enum audit_state state)
 511 {
 512         uid_t loginuid = context->loginuid;
 513
 514         memset(context, 0, sizeof(*context));
 515         context->state      = state;
 516         context->loginuid   = loginuid;
 517 }
 518
 519 static inline struct audit_context *audit_alloc_context(enum audit_state state)
 520 {
 521         struct audit_context *context;
 522
 523         if (!(context = kmalloc(sizeof(*context), GFP_KERNEL)))
 524                 return NULL;
 525         audit_zero_context(context, state);
 526         return context;
 527 }
 528
 529 /* Filter on the task information and allocate a per-task audit context
 530  * if necessary.  Doing so turns on system call auditing for the
 531  * specified task.  This is called from copy_process, so no lock is
 532  * needed. */
 533 int audit_alloc(struct task_struct *tsk)
 534 {
 535         struct audit_context *context;
 536         enum audit_state     state;
 537
 538         if (likely(!audit_enabled))
 539                 return 0; /* Return if not auditing. */
 540
 541         state = audit_filter_task(tsk);
 542         if (likely(state == AUDIT_DISABLED))
 543                 return 0;
 544
 545         if (!(context = audit_alloc_context(state))) {
 546                 audit_log_lost("out of memory in audit_alloc");
 547                 return -ENOMEM;
 548         }
 549
 550                                 /* Preserve login uid */
 551         context->loginuid = -1;
 552         if (tsk->audit_context)
 553                 context->loginuid = tsk->audit_context->loginuid;
 554
 555         tsk->audit_context  = context;
 556         set_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
 557         return 0;
 558 }
 559
 560 static inline void audit_free_context(struct audit_context *context)
 561 {
 562         struct audit_context *previous;
 563         int                  count = 0;
 564
 565         do {
 566                 previous = context->previous;
 567                 if (previous || (count &&  count < 10)) {
 568                         ++count;
 569                         printk(KERN_ERR "audit(:%d): major=%d name_count=%d:"
 570                                " freeing multiple contexts (%d)\n",
 571                                context->serial, context->major,
 572                                context->name_count, count);
 573                 }
 574                 audit_free_names(context);
 575                 kfree(context);
 576                 context  = previous;
 577         } while (context);
 578         if (count >= 10)
 579                 printk(KERN_ERR "audit: freed %d contexts\n", count);
 580 }
 581
 582 static void audit_log_exit(struct audit_context *context)
 583 {
 584         int i;
 585         struct audit_buffer *ab;
 586
 587         ab = audit_log_start(context);
 588         if (!ab)
 589                 return;         /* audit_panic has been called */
 590         audit_log_format(ab, "syscall=%d", context->major);
 591         if (context->personality != PER_LINUX)
 592                 audit_log_format(ab, " per=%lx", context->personality);
 593         if (context->return_valid)
 594                 audit_log_format(ab, " exit=%u", context->return_code);
 595         audit_log_format(ab,
 596                   " a0=%lx a1=%lx a2=%lx a3=%lx items=%d"
 597                   " pid=%d loginuid=%d uid=%d gid=%d"
 598                   " euid=%d suid=%d fsuid=%d"
 599                   " egid=%d sgid=%d fsgid=%d",
 600                   context->argv[0],
 601                   context->argv[1],
 602                   context->argv[2],
 603                   context->argv[3],
 604                   context->name_count,
 605                   context->pid,
 606                   context->loginuid,
 607                   context->uid,
 608                   context->gid,
 609                   context->euid, context->suid, context->fsuid,
 610                   context->egid, context->sgid, context->fsgid);
 611         audit_log_end(ab);
 612         for (i = 0; i < context->name_count; i++) {
 613                 ab = audit_log_start(context);
 614                 if (!ab)
 615                         continue; /* audit_panic has been called */
 616                 audit_log_format(ab, "item=%d", i);
 617                 if (context->names[i].name)
 618                         audit_log_format(ab, " name=%s",
 619                                          context->names[i].name);
 620                 if (context->names[i].ino != (unsigned long)-1)
 621                         audit_log_format(ab, " inode=%lu",
 622                                          context->names[i].ino);
 623                 /* FIXME: should use format_dev_t, but ab structure is
 624                  * opaque. */
 625                 if (context->names[i].rdev != -1)
 626                         audit_log_format(ab, " dev=%02x:%02x",
 627                                          MAJOR(context->names[i].rdev),
 628                                          MINOR(context->names[i].rdev));
 629                 audit_log_end(ab);
 630         }
 631 }
 632
 633 /* Free a per-task audit context.  Called from copy_process and
 634  * __put_task_struct. */
 635 void audit_free(struct task_struct *tsk)
 636 {
 637         struct audit_context *context;
 638
 639         task_lock(tsk);
 640         context = audit_get_context(tsk, 0, 0);
 641         task_unlock(tsk);
 642
 643         if (likely(!context))
 644                 return;
 645
 646         /* Check for system calls that do not go through the exit
 647          * function (e.g., exit_group), then free context block. */
 648         if (context->in_syscall && context->auditable)
 649                 audit_log_exit(context);
 650
 651         audit_free_context(context);
 652 }
 653
 654 /* Compute a serial number for the audit record.  Audit records are
 655  * written to user-space as soon as they are generated, so a complete
 656  * audit record may be written in several pieces.  The timestamp of the
 657  * record and this serial number are used by the user-space daemon to
 658  * determine which pieces belong to the same audit record.  The
 659  * (timestamp,serial) tuple is unique for each syscall and is live from
 660  * syscall entry to syscall exit.
 661  *
 662  * Atomic values are only guaranteed to be 24-bit, so we count down.
 663  *
 664  * NOTE: Another possibility is to store the formatted records off the
 665  * audit context (for those records that have a context), and emit them
 666  * all at syscall exit.  However, this could delay the reporting of
 667  * significant errors until syscall exit (or never, if the system
 668  * halts). */
 669 static inline unsigned int audit_serial(void)
 670 {
 671         static atomic_t serial = ATOMIC_INIT(0xffffff);
 672         unsigned int a, b;
 673
 674         do {
 675                 a = atomic_read(&serial);
 676                 if (atomic_dec_and_test(&serial))
 677                         atomic_set(&serial, 0xffffff);
 678                 b = atomic_read(&serial);
 679         } while (b != a - 1);
 680
 681         return 0xffffff - b;
 682 }
 683
 684 /* Fill in audit context at syscall entry.  This only happens if the
 685  * audit context was created when the task was created and the state or
 686  * filters demand the audit context be built.  If the state from the
 687  * per-task filter or from the per-syscall filter is AUDIT_RECORD_CONTEXT,
 688  * then the record will be written at syscall exit time (otherwise, it
 689  * will only be written if another part of the kernel requests that it
 690  * be written). */
 691 void audit_syscall_entry(struct task_struct *tsk, int major,
 692                          unsigned long a1, unsigned long a2,
 693                          unsigned long a3, unsigned long a4)
 694 {
 695         struct audit_context *context = tsk->audit_context;
 696         enum audit_state     state;
 697
 698         BUG_ON(!context);
 699
 700         /* This happens only on certain architectures that make system
 701          * calls in kernel_thread via the entry.S interface, instead of
 702          * with direct calls.  (If you are porting to a new
 703          * architecture, hitting this condition can indicate that you
 704          * got the _exit/_leave calls backward in entry.S.)
 705          *
 706          * i386     no
 707          * x86_64   no
 708          * ppc64    yes (see arch/ppc64/kernel/misc.S)
 709          *
 710          * This also happens with vm86 emulation in a non-nested manner
 711          * (entries without exits), so this case must be caught.
 712          */
 713         if (context->in_syscall) {
 714                 struct audit_context *newctx;
 715
 716 #if defined(__NR_vm86) && defined(__NR_vm86old)
 717                 /* vm86 mode should only be entered once */
 718                 if (major == __NR_vm86 || major == __NR_vm86old)
 719                         return;
 720 #endif
 721 #if AUDIT_DEBUG
 722                 printk(KERN_ERR
 723                        "audit(:%d) pid=%d in syscall=%d;"
 724                        " entering syscall=%d\n",
 725                        context->serial, tsk->pid, context->major, major);
 726 #endif
 727                 newctx = audit_alloc_context(context->state);
 728                 if (newctx) {
 729                         newctx->previous   = context;
 730                         context            = newctx;
 731                         tsk->audit_context = newctx;
 732                 } else  {
 733                         /* If we can't alloc a new context, the best we
 734                          * can do is to leak memory (any pending putname
 735                          * will be lost).  The only other alternative is
 736                          * to abandon auditing. */
 737                         audit_zero_context(context, context->state);
 738                 }
 739         }
 740         BUG_ON(context->in_syscall || context->name_count);
 741
 742         if (!audit_enabled)
 743                 return;
 744
 745         context->major      = major;
 746         context->argv[0]    = a1;
 747         context->argv[1]    = a2;
 748         context->argv[2]    = a3;
 749         context->argv[3]    = a4;
 750
 751         state = context->state;
 752         if (state == AUDIT_SETUP_CONTEXT || state == AUDIT_BUILD_CONTEXT)
 753                 state = audit_filter_syscall(tsk, context, &audit_entlist);
 754         if (likely(state == AUDIT_DISABLED))
 755                 return;
 756
 757         context->serial     = audit_serial();
 758         context->ctime      = CURRENT_TIME;
 759         context->in_syscall = 1;
 760         context->auditable  = !!(state == AUDIT_RECORD_CONTEXT);
 761 }
 762
 763 /* Tear down after system call.  If the audit context has been marked as
 764  * auditable (either because of the AUDIT_RECORD_CONTEXT state from
 765  * filtering, or because some other part of the kernel write an audit
 766  * message), then write out the syscall information.  In call cases,
 767  * free the names stored from getname(). */
 768 void audit_syscall_exit(struct task_struct *tsk, int return_code)
 769 {
 770         struct audit_context *context;
 771
 772         get_task_struct(tsk);
 773         task_lock(tsk);
 774         context = audit_get_context(tsk, 1, return_code);
 775         task_unlock(tsk);
 776
 777         /* Not having a context here is ok, since the parent may have
 778          * called __put_task_struct. */
 779         if (likely(!context))
 780                 return;
 781
 782         if (context->in_syscall && context->auditable)
 783                 audit_log_exit(context);
 784
 785         context->in_syscall = 0;
 786         context->auditable  = 0;
 787         if (context->previous) {
 788                 struct audit_context *new_context = context->previous;
 789                 context->previous  = NULL;
 790                 audit_free_context(context);
 791                 tsk->audit_context = new_context;
 792         } else {
 793                 audit_free_names(context);
 794                 audit_zero_context(context, context->state);
 795                 tsk->audit_context = context;
 796         }
 797         put_task_struct(tsk);
 798 }
 799
 800 /* Add a name to the list.  Called from fs/namei.c:getname(). */
 801 void audit_getname(const char *name)
 802 {
 803         struct audit_context *context = current->audit_context;
 804
 805         BUG_ON(!context);
 806         if (!context->in_syscall) {
 807 #if AUDIT_DEBUG == 2
 808                 printk(KERN_ERR "%s:%d(:%d): ignoring getname(%p)\n",
 809                        __FILE__, __LINE__, context->serial, name);
 810                 dump_stack();
 811 #endif
 812                 return;
 813         }
 814         BUG_ON(context->name_count >= AUDIT_NAMES);
 815         context->names[context->name_count].name = name;
 816         context->names[context->name_count].ino  = (unsigned long)-1;
 817         context->names[context->name_count].rdev = -1;
 818         ++context->name_count;
 819 }
 820
 821 /* Intercept a putname request.  Called from
 822  * include/linux/fs.h:putname().  If we have stored the name from
 823  * getname in the audit context, then we delay the putname until syscall
 824  * exit. */
 825 void audit_putname(const char *name)
 826 {
 827         struct audit_context *context = current->audit_context;
 828
 829         BUG_ON(!context);
 830         if (!context->in_syscall) {
 831 #if AUDIT_DEBUG == 2
 832                 printk(KERN_ERR "%s:%d(:%d): __putname(%p)\n",
 833                        __FILE__, __LINE__, context->serial, name);
 834                 if (context->name_count) {
 835                         int i;
 836                         for (i = 0; i < context->name_count; i++)
 837                                 printk(KERN_ERR "name[%d] = %p = %s\n", i,
 838                                        context->names[i].name,
 839                                        context->names[i].name);
 840                 }
 841 #endif
 842                 __putname(name);
 843         }
 844 #if AUDIT_DEBUG
 845         else {
 846                 ++context->put_count;
 847                 if (context->put_count > context->name_count) {
 848                         printk(KERN_ERR "%s:%d(:%d): major=%d"
 849                                " in_syscall=%d putname(%p) name_count=%d"
 850                                " put_count=%d\n",
 851                                __FILE__, __LINE__,
 852                                context->serial, context->major,
 853                                context->in_syscall, name, context->name_count,
 854                                context->put_count);
 855                         dump_stack();
 856                 }
 857         }
 858 #endif
 859 }
 860 EXPORT_SYMBOL(audit_putname);
 861
 862 /* Store the inode and device from a lookup.  Called from
 863  * fs/namei.c:path_lookup(). */
 864 void audit_inode(const char *name, unsigned long ino, dev_t rdev)
 865 {
 866         int idx;
 867         struct audit_context *context = current->audit_context;
 868
 869         if (!context->in_syscall)
 870                 return;
 871         if (context->name_count
 872             && context->names[context->name_count-1].name
 873             && context->names[context->name_count-1].name == name)
 874                 idx = context->name_count - 1;
 875         else if (context->name_count > 1
 876                  && context->names[context->name_count-2].name
 877                  && context->names[context->name_count-2].name == name)
 878                 idx = context->name_count - 2;
 879         else {
 880                 /* FIXME: how much do we care about inodes that have no
 881                  * associated name? */
 882                 if (context->name_count >= AUDIT_NAMES - AUDIT_NAMES_RESERVED)
 883                         return;
 884                 idx = context->name_count++;
 885                 context->names[idx].name = NULL;
 886 #if AUDIT_DEBUG
 887                 ++context->ino_count;
 888 #endif
 889         }
 890         context->names[idx].ino  = ino;
 891         context->names[idx].rdev = rdev;
 892 }
 893
 894 void audit_get_stamp(struct audit_context *ctx,
 895                      struct timespec *t, int *serial)
 896 {
 897         if (ctx) {
 898                 t->tv_sec  = ctx->ctime.tv_sec;
 899                 t->tv_nsec = ctx->ctime.tv_nsec;
 900                 *serial    = ctx->serial;
 901                 ctx->auditable = 1;
 902         } else {
 903                 *t      = CURRENT_TIME;
 904                 *serial = 0;
 905         }
 906 }
 907
 908 int audit_set_loginuid(struct audit_context *ctx, uid_t loginuid)
 909 {
 910         if (ctx) {
 911                 if (loginuid < 0)
 912                         return -EINVAL;
 913                 ctx->loginuid = loginuid;
 914         }
 915         return 0;
 916 }