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           dev;
  93         umode_t         mode;
  94         uid_t           uid;
  95         gid_t           gid;
  96         dev_t           rdev;
  97 };
  98
  99 struct audit_aux_data {
 100         struct audit_aux_data   *next;
 101         int                     type;
 102 };
 103
 104 #define AUDIT_AUX_IPCPERM       0
 105
 106 struct audit_aux_data_ipcctl {
 107         struct audit_aux_data   d;
 108         struct ipc_perm         p;
 109         unsigned long           qbytes;
 110         uid_t                   uid;
 111         gid_t                   gid;
 112         mode_t                  mode;
 113 };
 114
 115
 116 /* The per-task audit context. */
 117 struct audit_context {
 118         int                 in_syscall; /* 1 if task is in a syscall */
 119         enum audit_state    state;
 120         unsigned int        serial;     /* serial number for record */
 121         struct timespec     ctime;      /* time of syscall entry */
 122         uid_t               loginuid;   /* login uid (identity) */
 123         int                 major;      /* syscall number */
 124         unsigned long       argv[4];    /* syscall arguments */
 125         int                 return_valid; /* return code is valid */
 126         int                 return_code;/* syscall return code */
 127         int                 auditable;  /* 1 if record should be written */
 128         int                 name_count;
 129         struct audit_names  names[AUDIT_NAMES];
 130         struct audit_context *previous; /* For nested syscalls */
 131         struct audit_aux_data *aux;
 132
 133                                 /* Save things to print about task_struct */
 134         pid_t               pid;
 135         uid_t               uid, euid, suid, fsuid;
 136         gid_t               gid, egid, sgid, fsgid;
 137         unsigned long       personality;
 138
 139 #if AUDIT_DEBUG
 140         int                 put_count;
 141         int                 ino_count;
 142 #endif
 143 };
 144
 145                                 /* Public API */
 146 /* There are three lists of rules -- one to search at task creation
 147  * time, one to search at syscall entry time, and another to search at
 148  * syscall exit time. */
 149 static LIST_HEAD(audit_tsklist);
 150 static LIST_HEAD(audit_entlist);
 151 static LIST_HEAD(audit_extlist);
 152
 153 struct audit_entry {
 154         struct list_head  list;
 155         struct rcu_head   rcu;
 156         struct audit_rule rule;
 157 };
 158
 159 /* Check to see if two rules are identical.  It is called from
 160  * audit_del_rule during AUDIT_DEL. */
 161 static int audit_compare_rule(struct audit_rule *a, struct audit_rule *b)
 162 {
 163         int i;
 164
 165         if (a->flags != b->flags)
 166                 return 1;
 167
 168         if (a->action != b->action)
 169                 return 1;
 170
 171         if (a->field_count != b->field_count)
 172                 return 1;
 173
 174         for (i = 0; i < a->field_count; i++) {
 175                 if (a->fields[i] != b->fields[i]
 176                     || a->values[i] != b->values[i])
 177                         return 1;
 178         }
 179
 180         for (i = 0; i < AUDIT_BITMASK_SIZE; i++)
 181                 if (a->mask[i] != b->mask[i])
 182                         return 1;
 183
 184         return 0;
 185 }
 186
 187 /* Note that audit_add_rule and audit_del_rule are called via
 188  * audit_receive() in audit.c, and are protected by
 189  * audit_netlink_sem. */
 190 static inline int audit_add_rule(struct audit_entry *entry,
 191                                  struct list_head *list)
 192 {
 193         if (entry->rule.flags & AUDIT_PREPEND) {
 194                 entry->rule.flags &= ~AUDIT_PREPEND;
 195                 list_add_rcu(&entry->list, list);
 196         } else {
 197                 list_add_tail_rcu(&entry->list, list);
 198         }
 199         return 0;
 200 }
 201
 202 static void audit_free_rule(struct rcu_head *head)
 203 {
 204         struct audit_entry *e = container_of(head, struct audit_entry, rcu);
 205         kfree(e);
 206 }
 207
 208 /* Note that audit_add_rule and audit_del_rule are called via
 209  * audit_receive() in audit.c, and are protected by
 210  * audit_netlink_sem. */
 211 static inline int audit_del_rule(struct audit_rule *rule,
 212                                  struct list_head *list)
 213 {
 214         struct audit_entry  *e;
 215
 216         /* Do not use the _rcu iterator here, since this is the only
 217          * deletion routine. */
 218         list_for_each_entry(e, list, list) {
 219                 if (!audit_compare_rule(rule, &e->rule)) {
 220                         list_del_rcu(&e->list);
 221                         call_rcu(&e->rcu, audit_free_rule);
 222                         return 0;
 223                 }
 224         }
 225         return -EFAULT;         /* No matching rule */
 226 }
 227
 228 #ifdef CONFIG_NET
 229 /* Copy rule from user-space to kernel-space.  Called during
 230  * AUDIT_ADD. */
 231 static int audit_copy_rule(struct audit_rule *d, struct audit_rule *s)
 232 {
 233         int i;
 234
 235         if (s->action != AUDIT_NEVER
 236             && s->action != AUDIT_POSSIBLE
 237             && s->action != AUDIT_ALWAYS)
 238                 return -1;
 239         if (s->field_count < 0 || s->field_count > AUDIT_MAX_FIELDS)
 240                 return -1;
 241
 242         d->flags        = s->flags;
 243         d->action       = s->action;
 244         d->field_count  = s->field_count;
 245         for (i = 0; i < d->field_count; i++) {
 246                 d->fields[i] = s->fields[i];
 247                 d->values[i] = s->values[i];
 248         }
 249         for (i = 0; i < AUDIT_BITMASK_SIZE; i++) d->mask[i] = s->mask[i];
 250         return 0;
 251 }
 252
 253 int audit_receive_filter(int type, int pid, int uid, int seq, void *data)
 254 {
 255         u32                flags;
 256         struct audit_entry *entry;
 257         int                err = 0;
 258
 259         switch (type) {
 260         case AUDIT_LIST:
 261                 /* The *_rcu iterators not needed here because we are
 262                    always called with audit_netlink_sem held. */
 263                 list_for_each_entry(entry, &audit_tsklist, list)
 264                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 265                                          &entry->rule, sizeof(entry->rule));
 266                 list_for_each_entry(entry, &audit_entlist, list)
 267                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 268                                          &entry->rule, sizeof(entry->rule));
 269                 list_for_each_entry(entry, &audit_extlist, list)
 270                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 271                                          &entry->rule, sizeof(entry->rule));
 272                 audit_send_reply(pid, seq, AUDIT_LIST, 1, 1, NULL, 0);
 273                 break;
 274         case AUDIT_ADD:
 275                 if (!(entry = kmalloc(sizeof(*entry), GFP_KERNEL)))
 276                         return -ENOMEM;
 277                 if (audit_copy_rule(&entry->rule, data)) {
 278                         kfree(entry);
 279                         return -EINVAL;
 280                 }
 281                 flags = entry->rule.flags;
 282                 if (!err && (flags & AUDIT_PER_TASK))
 283                         err = audit_add_rule(entry, &audit_tsklist);
 284                 if (!err && (flags & AUDIT_AT_ENTRY))
 285                         err = audit_add_rule(entry, &audit_entlist);
 286                 if (!err && (flags & AUDIT_AT_EXIT))
 287                         err = audit_add_rule(entry, &audit_extlist);
 288                 break;
 289         case AUDIT_DEL:
 290                 flags =((struct audit_rule *)data)->flags;
 291                 if (!err && (flags & AUDIT_PER_TASK))
 292                         err = audit_del_rule(data, &audit_tsklist);
 293                 if (!err && (flags & AUDIT_AT_ENTRY))
 294                         err = audit_del_rule(data, &audit_entlist);
 295                 if (!err && (flags & AUDIT_AT_EXIT))
 296                         err = audit_del_rule(data, &audit_extlist);
 297                 break;
 298         default:
 299                 return -EINVAL;
 300         }
 301
 302         return err;
 303 }
 304 #endif
 305
 306 /* Compare a task_struct with an audit_rule.  Return 1 on match, 0
 307  * otherwise. */
 308 static int audit_filter_rules(struct task_struct *tsk,
 309                               struct audit_rule *rule,
 310                               struct audit_context *ctx,
 311                               enum audit_state *state)
 312 {
 313         int i, j;
 314
 315         for (i = 0; i < rule->field_count; i++) {
 316                 u32 field  = rule->fields[i] & ~AUDIT_NEGATE;
 317                 u32 value  = rule->values[i];
 318                 int result = 0;
 319
 320                 switch (field) {
 321                 case AUDIT_PID:
 322                         result = (tsk->pid == value);
 323                         break;
 324                 case AUDIT_UID:
 325                         result = (tsk->uid == value);
 326                         break;
 327                 case AUDIT_EUID:
 328                         result = (tsk->euid == value);
 329                         break;
 330                 case AUDIT_SUID:
 331                         result = (tsk->suid == value);
 332                         break;
 333                 case AUDIT_FSUID:
 334                         result = (tsk->fsuid == value);
 335                         break;
 336                 case AUDIT_GID:
 337                         result = (tsk->gid == value);
 338                         break;
 339                 case AUDIT_EGID:
 340                         result = (tsk->egid == value);
 341                         break;
 342                 case AUDIT_SGID:
 343                         result = (tsk->sgid == value);
 344                         break;
 345                 case AUDIT_FSGID:
 346                         result = (tsk->fsgid == value);
 347                         break;
 348                 case AUDIT_PERS:
 349                         result = (tsk->personality == value);
 350                         break;
 351
 352                 case AUDIT_EXIT:
 353                         if (ctx && ctx->return_valid)
 354                                 result = (ctx->return_code == value);
 355                         break;
 356                 case AUDIT_SUCCESS:
 357                         if (ctx && ctx->return_valid)
 358                                 result = (ctx->return_code >= 0);
 359                         break;
 360                 case AUDIT_DEVMAJOR:
 361                         if (ctx) {
 362                                 for (j = 0; j < ctx->name_count; j++) {
 363                                         if (MAJOR(ctx->names[j].dev)==value) {
 364                                                 ++result;
 365                                                 break;
 366                                         }
 367                                 }
 368                         }
 369                         break;
 370                 case AUDIT_DEVMINOR:
 371                         if (ctx) {
 372                                 for (j = 0; j < ctx->name_count; j++) {
 373                                         if (MINOR(ctx->names[j].dev)==value) {
 374                                                 ++result;
 375                                                 break;
 376                                         }
 377                                 }
 378                         }
 379                         break;
 380                 case AUDIT_INODE:
 381                         if (ctx) {
 382                                 for (j = 0; j < ctx->name_count; j++) {
 383                                         if (ctx->names[j].ino == value) {
 384                                                 ++result;
 385                                                 break;
 386                                         }
 387                                 }
 388                         }
 389                         break;
 390                 case AUDIT_LOGINUID:
 391                         result = 0;
 392                         if (ctx)
 393                                 result = (ctx->loginuid == value);
 394                         break;
 395                 case AUDIT_ARG0:
 396                 case AUDIT_ARG1:
 397                 case AUDIT_ARG2:
 398                 case AUDIT_ARG3:
 399                         if (ctx)
 400                                 result = (ctx->argv[field-AUDIT_ARG0]==value);
 401                         break;
 402                 }
 403
 404                 if (rule->fields[i] & AUDIT_NEGATE)
 405                         result = !result;
 406                 if (!result)
 407                         return 0;
 408         }
 409         switch (rule->action) {
 410         case AUDIT_NEVER:    *state = AUDIT_DISABLED;       break;
 411         case AUDIT_POSSIBLE: *state = AUDIT_BUILD_CONTEXT;  break;
 412         case AUDIT_ALWAYS:   *state = AUDIT_RECORD_CONTEXT; break;
 413         }
 414         return 1;
 415 }
 416
 417 /* At process creation time, we can determine if system-call auditing is
 418  * completely disabled for this task.  Since we only have the task
 419  * structure at this point, we can only check uid and gid.
 420  */
 421 static enum audit_state audit_filter_task(struct task_struct *tsk)
 422 {
 423         struct audit_entry *e;
 424         enum audit_state   state;
 425
 426         rcu_read_lock();
 427         list_for_each_entry_rcu(e, &audit_tsklist, list) {
 428                 if (audit_filter_rules(tsk, &e->rule, NULL, &state)) {
 429                         rcu_read_unlock();
 430                         return state;
 431                 }
 432         }
 433         rcu_read_unlock();
 434         return AUDIT_BUILD_CONTEXT;
 435 }
 436
 437 /* At syscall entry and exit time, this filter is called if the
 438  * audit_state is not low enough that auditing cannot take place, but is
 439  * also not high enough that we already know we have to write and audit
 440  * record (i.e., the state is AUDIT_SETUP_CONTEXT or  AUDIT_BUILD_CONTEXT).
 441  */
 442 static enum audit_state audit_filter_syscall(struct task_struct *tsk,
 443                                              struct audit_context *ctx,
 444                                              struct list_head *list)
 445 {
 446         struct audit_entry *e;
 447         enum audit_state   state;
 448         int                word = AUDIT_WORD(ctx->major);
 449         int                bit  = AUDIT_BIT(ctx->major);
 450
 451         rcu_read_lock();
 452         list_for_each_entry_rcu(e, list, list) {
 453                 if ((e->rule.mask[word] & bit) == bit
 454                     && audit_filter_rules(tsk, &e->rule, ctx, &state)) {
 455                         rcu_read_unlock();
 456                         return state;
 457                 }
 458         }
 459         rcu_read_unlock();
 460         return AUDIT_BUILD_CONTEXT;
 461 }
 462
 463 /* This should be called with task_lock() held. */
 464 static inline struct audit_context *audit_get_context(struct task_struct *tsk,
 465                                                       int return_valid,
 466                                                       int return_code)
 467 {
 468         struct audit_context *context = tsk->audit_context;
 469
 470         if (likely(!context))
 471                 return NULL;
 472         context->return_valid = return_valid;
 473         context->return_code  = return_code;
 474
 475         if (context->in_syscall && !context->auditable) {
 476                 enum audit_state state;
 477                 state = audit_filter_syscall(tsk, context, &audit_extlist);
 478                 if (state == AUDIT_RECORD_CONTEXT)
 479                         context->auditable = 1;
 480         }
 481
 482         context->pid = tsk->pid;
 483         context->uid = tsk->uid;
 484         context->gid = tsk->gid;
 485         context->euid = tsk->euid;
 486         context->suid = tsk->suid;
 487         context->fsuid = tsk->fsuid;
 488         context->egid = tsk->egid;
 489         context->sgid = tsk->sgid;
 490         context->fsgid = tsk->fsgid;
 491         context->personality = tsk->personality;
 492         tsk->audit_context = NULL;
 493         return context;
 494 }
 495
 496 static inline void audit_free_names(struct audit_context *context)
 497 {
 498         int i;
 499
 500 #if AUDIT_DEBUG == 2
 501         if (context->auditable
 502             ||context->put_count + context->ino_count != context->name_count) {
 503                 printk(KERN_ERR "audit.c:%d(:%d): major=%d in_syscall=%d"
 504                        " name_count=%d put_count=%d"
 505                        " ino_count=%d [NOT freeing]\n",
 506                        __LINE__,
 507                        context->serial, context->major, context->in_syscall,
 508                        context->name_count, context->put_count,
 509                        context->ino_count);
 510                 for (i = 0; i < context->name_count; i++)
 511                         printk(KERN_ERR "names[%d] = %p = %s\n", i,
 512                                context->names[i].name,
 513                                context->names[i].name);
 514                 dump_stack();
 515                 return;
 516         }
 517 #endif
 518 #if AUDIT_DEBUG
 519         context->put_count  = 0;
 520         context->ino_count  = 0;
 521 #endif
 522
 523         for (i = 0; i < context->name_count; i++)
 524                 if (context->names[i].name)
 525                         __putname(context->names[i].name);
 526         context->name_count = 0;
 527 }
 528
 529 static inline void audit_free_aux(struct audit_context *context)
 530 {
 531         struct audit_aux_data *aux;
 532
 533         while ((aux = context->aux)) {
 534                 context->aux = aux->next;
 535                 kfree(aux);
 536         }
 537 }
 538
 539 static inline void audit_zero_context(struct audit_context *context,
 540                                       enum audit_state state)
 541 {
 542         uid_t loginuid = context->loginuid;
 543
 544         memset(context, 0, sizeof(*context));
 545         context->state      = state;
 546         context->loginuid   = loginuid;
 547 }
 548
 549 static inline struct audit_context *audit_alloc_context(enum audit_state state)
 550 {
 551         struct audit_context *context;
 552
 553         if (!(context = kmalloc(sizeof(*context), GFP_KERNEL)))
 554                 return NULL;
 555         audit_zero_context(context, state);
 556         return context;
 557 }
 558
 559 /* Filter on the task information and allocate a per-task audit context
 560  * if necessary.  Doing so turns on system call auditing for the
 561  * specified task.  This is called from copy_process, so no lock is
 562  * needed. */
 563 int audit_alloc(struct task_struct *tsk)
 564 {
 565         struct audit_context *context;
 566         enum audit_state     state;
 567
 568         if (likely(!audit_enabled))
 569                 return 0; /* Return if not auditing. */
 570
 571         state = audit_filter_task(tsk);
 572         if (likely(state == AUDIT_DISABLED))
 573                 return 0;
 574
 575         if (!(context = audit_alloc_context(state))) {
 576                 audit_log_lost("out of memory in audit_alloc");
 577                 return -ENOMEM;
 578         }
 579
 580                                 /* Preserve login uid */
 581         context->loginuid = -1;
 582         if (current->audit_context)
 583                 context->loginuid = current->audit_context->loginuid;
 584
 585         tsk->audit_context  = context;
 586         set_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
 587         return 0;
 588 }
 589
 590 static inline void audit_free_context(struct audit_context *context)
 591 {
 592         struct audit_context *previous;
 593         int                  count = 0;
 594
 595         do {
 596                 previous = context->previous;
 597                 if (previous || (count &&  count < 10)) {
 598                         ++count;
 599                         printk(KERN_ERR "audit(:%d): major=%d name_count=%d:"
 600                                " freeing multiple contexts (%d)\n",
 601                                context->serial, context->major,
 602                                context->name_count, count);
 603                 }
 604                 audit_free_names(context);
 605                 audit_free_aux(context);
 606                 kfree(context);
 607                 context  = previous;
 608         } while (context);
 609         if (count >= 10)
 610                 printk(KERN_ERR "audit: freed %d contexts\n", count);
 611 }
 612
 613 static void audit_log_exit(struct audit_context *context)
 614 {
 615         int i;
 616         struct audit_buffer *ab;
 617
 618         ab = audit_log_start(context);
 619         if (!ab)
 620                 return;         /* audit_panic has been called */
 621         audit_log_format(ab, "syscall=%d", context->major);
 622         if (context->personality != PER_LINUX)
 623                 audit_log_format(ab, " per=%lx", context->personality);
 624         if (context->return_valid)
 625                 audit_log_format(ab, " exit=%d", context->return_code);
 626         audit_log_format(ab,
 627                   " a0=%lx a1=%lx a2=%lx a3=%lx items=%d"
 628                   " pid=%d loginuid=%d uid=%d gid=%d"
 629                   " euid=%d suid=%d fsuid=%d"
 630                   " egid=%d sgid=%d fsgid=%d",
 631                   context->argv[0],
 632                   context->argv[1],
 633                   context->argv[2],
 634                   context->argv[3],
 635                   context->name_count,
 636                   context->pid,
 637                   context->loginuid,
 638                   context->uid,
 639                   context->gid,
 640                   context->euid, context->suid, context->fsuid,
 641                   context->egid, context->sgid, context->fsgid);
 642         audit_log_end(ab);
 643         while (context->aux) {
 644                 struct audit_aux_data *aux;
 645
 646                 ab = audit_log_start(context);
 647                 if (!ab)
 648                         continue; /* audit_panic has been called */
 649
 650                 aux = context->aux;
 651                 context->aux = aux->next;
 652
 653                 audit_log_format(ab, "auxitem=%d", aux->type);
 654                 switch (aux->type) {
 655                 case AUDIT_AUX_IPCPERM: {
 656                         struct audit_aux_data_ipcctl *axi = (void *)aux;
 657                         audit_log_format(ab,
 658                                          " qbytes=%lx uid=%d gid=%d mode=%x",
 659                                          axi->qbytes, axi->uid, axi->gid, axi->mode);
 660                         }
 661                 }
 662                 audit_log_end(ab);
 663                 kfree(aux);
 664         }
 665
 666         for (i = 0; i < context->name_count; i++) {
 667                 ab = audit_log_start(context);
 668                 if (!ab)
 669                         continue; /* audit_panic has been called */
 670                 audit_log_format(ab, "item=%d", i);
 671                 if (context->names[i].name)
 672                         audit_log_format(ab, " name=%s",
 673                                          context->names[i].name);
 674                 if (context->names[i].ino != (unsigned long)-1)
 675                         audit_log_format(ab, " inode=%lu dev=%02x:%02x mode=%#o"
 676                                              " uid=%d gid=%d rdev=%02x:%02x",
 677                                          context->names[i].ino,
 678                                          MAJOR(context->names[i].dev),
 679                                          MINOR(context->names[i].dev),
 680                                          context->names[i].mode,
 681                                          context->names[i].uid,
 682                                          context->names[i].gid,
 683                                          MAJOR(context->names[i].rdev),
 684                                          MINOR(context->names[i].rdev));
 685                 audit_log_end(ab);
 686         }
 687 }
 688
 689 /* Free a per-task audit context.  Called from copy_process and
 690  * __put_task_struct. */
 691 void audit_free(struct task_struct *tsk)
 692 {
 693         struct audit_context *context;
 694
 695         task_lock(tsk);
 696         context = audit_get_context(tsk, 0, 0);
 697         task_unlock(tsk);
 698
 699         if (likely(!context))
 700                 return;
 701
 702         /* Check for system calls that do not go through the exit
 703          * function (e.g., exit_group), then free context block. */
 704         if (context->in_syscall && context->auditable)
 705                 audit_log_exit(context);
 706
 707         audit_free_context(context);
 708 }
 709
 710 /* Compute a serial number for the audit record.  Audit records are
 711  * written to user-space as soon as they are generated, so a complete
 712  * audit record may be written in several pieces.  The timestamp of the
 713  * record and this serial number are used by the user-space daemon to
 714  * determine which pieces belong to the same audit record.  The
 715  * (timestamp,serial) tuple is unique for each syscall and is live from
 716  * syscall entry to syscall exit.
 717  *
 718  * Atomic values are only guaranteed to be 24-bit, so we count down.
 719  *
 720  * NOTE: Another possibility is to store the formatted records off the
 721  * audit context (for those records that have a context), and emit them
 722  * all at syscall exit.  However, this could delay the reporting of
 723  * significant errors until syscall exit (or never, if the system
 724  * halts). */
 725 static inline unsigned int audit_serial(void)
 726 {
 727         static atomic_t serial = ATOMIC_INIT(0xffffff);
 728         unsigned int a, b;
 729
 730         do {
 731                 a = atomic_read(&serial);
 732                 if (atomic_dec_and_test(&serial))
 733                         atomic_set(&serial, 0xffffff);
 734                 b = atomic_read(&serial);
 735         } while (b != a - 1);
 736
 737         return 0xffffff - b;
 738 }
 739
 740 /* Fill in audit context at syscall entry.  This only happens if the
 741  * audit context was created when the task was created and the state or
 742  * filters demand the audit context be built.  If the state from the
 743  * per-task filter or from the per-syscall filter is AUDIT_RECORD_CONTEXT,
 744  * then the record will be written at syscall exit time (otherwise, it
 745  * will only be written if another part of the kernel requests that it
 746  * be written). */
 747 void audit_syscall_entry(struct task_struct *tsk, int major,
 748                          unsigned long a1, unsigned long a2,
 749                          unsigned long a3, unsigned long a4)
 750 {
 751         struct audit_context *context = tsk->audit_context;
 752         enum audit_state     state;
 753
 754         BUG_ON(!context);
 755
 756         /* This happens only on certain architectures that make system
 757          * calls in kernel_thread via the entry.S interface, instead of
 758          * with direct calls.  (If you are porting to a new
 759          * architecture, hitting this condition can indicate that you
 760          * got the _exit/_leave calls backward in entry.S.)
 761          *
 762          * i386     no
 763          * x86_64   no
 764          * ppc64    yes (see arch/ppc64/kernel/misc.S)
 765          *
 766          * This also happens with vm86 emulation in a non-nested manner
 767          * (entries without exits), so this case must be caught.
 768          */
 769         if (context->in_syscall) {
 770                 struct audit_context *newctx;
 771
 772 #if defined(__NR_vm86) && defined(__NR_vm86old)
 773                 /* vm86 mode should only be entered once */
 774                 if (major == __NR_vm86 || major == __NR_vm86old)
 775                         return;
 776 #endif
 777 #if AUDIT_DEBUG
 778                 printk(KERN_ERR
 779                        "audit(:%d) pid=%d in syscall=%d;"
 780                        " entering syscall=%d\n",
 781                        context->serial, tsk->pid, context->major, major);
 782 #endif
 783                 newctx = audit_alloc_context(context->state);
 784                 if (newctx) {
 785                         newctx->previous   = context;
 786                         context            = newctx;
 787                         tsk->audit_context = newctx;
 788                 } else  {
 789                         /* If we can't alloc a new context, the best we
 790                          * can do is to leak memory (any pending putname
 791                          * will be lost).  The only other alternative is
 792                          * to abandon auditing. */
 793                         audit_zero_context(context, context->state);
 794                 }
 795         }
 796         BUG_ON(context->in_syscall || context->name_count);
 797
 798         if (!audit_enabled)
 799                 return;
 800
 801         context->major      = major;
 802         context->argv[0]    = a1;
 803         context->argv[1]    = a2;
 804         context->argv[2]    = a3;
 805         context->argv[3]    = a4;
 806
 807         state = context->state;
 808         if (state == AUDIT_SETUP_CONTEXT || state == AUDIT_BUILD_CONTEXT)
 809                 state = audit_filter_syscall(tsk, context, &audit_entlist);
 810         if (likely(state == AUDIT_DISABLED))
 811                 return;
 812
 813         context->serial     = audit_serial();
 814         context->ctime      = CURRENT_TIME;
 815         context->in_syscall = 1;
 816         context->auditable  = !!(state == AUDIT_RECORD_CONTEXT);
 817 }
 818
 819 /* Tear down after system call.  If the audit context has been marked as
 820  * auditable (either because of the AUDIT_RECORD_CONTEXT state from
 821  * filtering, or because some other part of the kernel write an audit
 822  * message), then write out the syscall information.  In call cases,
 823  * free the names stored from getname(). */
 824 void audit_syscall_exit(struct task_struct *tsk, int return_code)
 825 {
 826         struct audit_context *context;
 827
 828         get_task_struct(tsk);
 829         task_lock(tsk);
 830         context = audit_get_context(tsk, 1, return_code);
 831         task_unlock(tsk);
 832
 833         /* Not having a context here is ok, since the parent may have
 834          * called __put_task_struct. */
 835         if (likely(!context))
 836                 return;
 837
 838         if (context->in_syscall && context->auditable)
 839                 audit_log_exit(context);
 840
 841         context->in_syscall = 0;
 842         context->auditable  = 0;
 843         if (context->previous) {
 844                 struct audit_context *new_context = context->previous;
 845                 context->previous  = NULL;
 846                 audit_free_context(context);
 847                 tsk->audit_context = new_context;
 848         } else {
 849                 audit_free_names(context);
 850                 audit_free_aux(context);
 851                 audit_zero_context(context, context->state);
 852                 tsk->audit_context = context;
 853         }
 854         put_task_struct(tsk);
 855 }
 856
 857 /* Add a name to the list.  Called from fs/namei.c:getname(). */
 858 void audit_getname(const char *name)
 859 {
 860         struct audit_context *context = current->audit_context;
 861
 862         if (!context || IS_ERR(name) || !name)
 863                 return;
 864
 865         if (!context->in_syscall) {
 866 #if AUDIT_DEBUG == 2
 867                 printk(KERN_ERR "%s:%d(:%d): ignoring getname(%p)\n",
 868                        __FILE__, __LINE__, context->serial, name);
 869                 dump_stack();
 870 #endif
 871                 return;
 872         }
 873         BUG_ON(context->name_count >= AUDIT_NAMES);
 874         context->names[context->name_count].name = name;
 875         context->names[context->name_count].ino  = (unsigned long)-1;
 876         ++context->name_count;
 877 }
 878
 879 /* Intercept a putname request.  Called from
 880  * include/linux/fs.h:putname().  If we have stored the name from
 881  * getname in the audit context, then we delay the putname until syscall
 882  * exit. */
 883 void audit_putname(const char *name)
 884 {
 885         struct audit_context *context = current->audit_context;
 886
 887         BUG_ON(!context);
 888         if (!context->in_syscall) {
 889 #if AUDIT_DEBUG == 2
 890                 printk(KERN_ERR "%s:%d(:%d): __putname(%p)\n",
 891                        __FILE__, __LINE__, context->serial, name);
 892                 if (context->name_count) {
 893                         int i;
 894                         for (i = 0; i < context->name_count; i++)
 895                                 printk(KERN_ERR "name[%d] = %p = %s\n", i,
 896                                        context->names[i].name,
 897                                        context->names[i].name);
 898                 }
 899 #endif
 900                 __putname(name);
 901         }
 902 #if AUDIT_DEBUG
 903         else {
 904                 ++context->put_count;
 905                 if (context->put_count > context->name_count) {
 906                         printk(KERN_ERR "%s:%d(:%d): major=%d"
 907                                " in_syscall=%d putname(%p) name_count=%d"
 908                                " put_count=%d\n",
 909                                __FILE__, __LINE__,
 910                                context->serial, context->major,
 911                                context->in_syscall, name, context->name_count,
 912                                context->put_count);
 913                         dump_stack();
 914                 }
 915         }
 916 #endif
 917 }
 918
 919 /* Store the inode and device from a lookup.  Called from
 920  * fs/namei.c:path_lookup(). */
 921 void audit_inode(const char *name, const struct inode *inode)
 922 {
 923         int idx;
 924         struct audit_context *context = current->audit_context;
 925
 926         if (!context->in_syscall)
 927                 return;
 928         if (context->name_count
 929             && context->names[context->name_count-1].name
 930             && context->names[context->name_count-1].name == name)
 931                 idx = context->name_count - 1;
 932         else if (context->name_count > 1
 933                  && context->names[context->name_count-2].name
 934                  && context->names[context->name_count-2].name == name)
 935                 idx = context->name_count - 2;
 936         else {
 937                 /* FIXME: how much do we care about inodes that have no
 938                  * associated name? */
 939                 if (context->name_count >= AUDIT_NAMES - AUDIT_NAMES_RESERVED)
 940                         return;
 941                 idx = context->name_count++;
 942                 context->names[idx].name = NULL;
 943 #if AUDIT_DEBUG
 944                 ++context->ino_count;
 945 #endif
 946         }
 947         context->names[idx].ino  = inode->i_ino;
 948         context->names[idx].dev  = inode->i_sb->s_dev;
 949         context->names[idx].mode = inode->i_mode;
 950         context->names[idx].uid  = inode->i_uid;
 951         context->names[idx].gid  = inode->i_gid;
 952         context->names[idx].rdev = inode->i_rdev;
 953 }
 954
 955 void audit_get_stamp(struct audit_context *ctx,
 956                      struct timespec *t, int *serial)
 957 {
 958         if (ctx) {
 959                 t->tv_sec  = ctx->ctime.tv_sec;
 960                 t->tv_nsec = ctx->ctime.tv_nsec;
 961                 *serial    = ctx->serial;
 962                 ctx->auditable = 1;
 963         } else {
 964                 *t      = CURRENT_TIME;
 965                 *serial = 0;
 966         }
 967 }
 968
 969 extern int audit_set_type(struct audit_buffer *ab, int type);
 970
 971 int audit_set_loginuid(struct audit_context *ctx, uid_t loginuid)
 972 {
 973         if (ctx) {
 974                 struct audit_buffer *ab;
 975
 976                 ab = audit_log_start(NULL);
 977                 if (ab) {
 978                         audit_log_format(ab, "login pid=%d uid=%u "
 979                                 "old loginuid=%u new loginuid=%u",
 980                                 ctx->pid, ctx->uid, ctx->loginuid, loginuid);
 981                         audit_set_type(ab, AUDIT_LOGIN);
 982                         audit_log_end(ab);
 983                 }
 984                 ctx->loginuid = loginuid;
 985         }
 986         return 0;
 987 }
 988
 989 uid_t audit_get_loginuid(struct audit_context *ctx)
 990 {
 991         return ctx ? ctx->loginuid : -1;
 992 }
 993
 994 int audit_ipc_perms(unsigned long qbytes, uid_t uid, gid_t gid, mode_t mode)
 995 {
 996         struct audit_aux_data_ipcctl *ax;
 997         struct audit_context *context = current->audit_context;
 998
 999         if (likely(!context))
1000                 return 0;
1001
1002         ax = kmalloc(sizeof(*ax), GFP_KERNEL);
1003         if (!ax)
1004                 return -ENOMEM;
1005
1006         ax->qbytes = qbytes;
1007         ax->uid = uid;
1008         ax->gid = gid;
1009         ax->mode = mode;
1010
1011         ax->d.type = AUDIT_AUX_IPCPERM;
1012         ax->d.next = context->aux;
1013         context->aux = (void *)ax;
1014         return 0;
1015 }