kernel/auditsc.c

   1 /* auditsc.c -- System-call auditing support
   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         long                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         int                 arch;
 139
 140 #if AUDIT_DEBUG
 141         int                 put_count;
 142         int                 ino_count;
 143 #endif
 144 };
 145
 146                                 /* Public API */
 147 /* There are three lists of rules -- one to search at task creation
 148  * time, one to search at syscall entry time, and another to search at
 149  * syscall exit time. */
 150 static LIST_HEAD(audit_tsklist);
 151 static LIST_HEAD(audit_entlist);
 152 static LIST_HEAD(audit_extlist);
 153
 154 struct audit_entry {
 155         struct list_head  list;
 156         struct rcu_head   rcu;
 157         struct audit_rule rule;
 158 };
 159
 160 /* Check to see if two rules are identical.  It is called from
 161  * audit_del_rule during AUDIT_DEL. */
 162 static int audit_compare_rule(struct audit_rule *a, struct audit_rule *b)
 163 {
 164         int i;
 165
 166         if (a->flags != b->flags)
 167                 return 1;
 168
 169         if (a->action != b->action)
 170                 return 1;
 171
 172         if (a->field_count != b->field_count)
 173                 return 1;
 174
 175         for (i = 0; i < a->field_count; i++) {
 176                 if (a->fields[i] != b->fields[i]
 177                     || a->values[i] != b->values[i])
 178                         return 1;
 179         }
 180
 181         for (i = 0; i < AUDIT_BITMASK_SIZE; i++)
 182                 if (a->mask[i] != b->mask[i])
 183                         return 1;
 184
 185         return 0;
 186 }
 187
 188 /* Note that audit_add_rule and audit_del_rule are called via
 189  * audit_receive() in audit.c, and are protected by
 190  * audit_netlink_sem. */
 191 static inline int audit_add_rule(struct audit_entry *entry,
 192                                  struct list_head *list)
 193 {
 194         if (entry->rule.flags & AUDIT_PREPEND) {
 195                 entry->rule.flags &= ~AUDIT_PREPEND;
 196                 list_add_rcu(&entry->list, list);
 197         } else {
 198                 list_add_tail_rcu(&entry->list, list);
 199         }
 200         return 0;
 201 }
 202
 203 static void audit_free_rule(struct rcu_head *head)
 204 {
 205         struct audit_entry *e = container_of(head, struct audit_entry, rcu);
 206         kfree(e);
 207 }
 208
 209 /* Note that audit_add_rule and audit_del_rule are called via
 210  * audit_receive() in audit.c, and are protected by
 211  * audit_netlink_sem. */
 212 static inline int audit_del_rule(struct audit_rule *rule,
 213                                  struct list_head *list)
 214 {
 215         struct audit_entry  *e;
 216
 217         /* Do not use the _rcu iterator here, since this is the only
 218          * deletion routine. */
 219         list_for_each_entry(e, list, list) {
 220                 if (!audit_compare_rule(rule, &e->rule)) {
 221                         list_del_rcu(&e->list);
 222                         call_rcu(&e->rcu, audit_free_rule);
 223                         return 0;
 224                 }
 225         }
 226         return -EFAULT;         /* No matching rule */
 227 }
 228
 229 #ifdef CONFIG_NET
 230 /* Copy rule from user-space to kernel-space.  Called during
 231  * AUDIT_ADD. */
 232 static int audit_copy_rule(struct audit_rule *d, struct audit_rule *s)
 233 {
 234         int i;
 235
 236         if (s->action != AUDIT_NEVER
 237             && s->action != AUDIT_POSSIBLE
 238             && s->action != AUDIT_ALWAYS)
 239                 return -1;
 240         if (s->field_count < 0 || s->field_count > AUDIT_MAX_FIELDS)
 241                 return -1;
 242
 243         d->flags        = s->flags;
 244         d->action       = s->action;
 245         d->field_count  = s->field_count;
 246         for (i = 0; i < d->field_count; i++) {
 247                 d->fields[i] = s->fields[i];
 248                 d->values[i] = s->values[i];
 249         }
 250         for (i = 0; i < AUDIT_BITMASK_SIZE; i++) d->mask[i] = s->mask[i];
 251         return 0;
 252 }
 253
 254 int audit_receive_filter(int type, int pid, int uid, int seq, void *data,
 255                                                         uid_t loginuid)
 256 {
 257         u32                flags;
 258         struct audit_entry *entry;
 259         int                err = 0;
 260
 261         switch (type) {
 262         case AUDIT_LIST:
 263                 /* The *_rcu iterators not needed here because we are
 264                    always called with audit_netlink_sem held. */
 265                 list_for_each_entry(entry, &audit_tsklist, list)
 266                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 267                                          &entry->rule, sizeof(entry->rule));
 268                 list_for_each_entry(entry, &audit_entlist, list)
 269                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 270                                          &entry->rule, sizeof(entry->rule));
 271                 list_for_each_entry(entry, &audit_extlist, list)
 272                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 273                                          &entry->rule, sizeof(entry->rule));
 274                 audit_send_reply(pid, seq, AUDIT_LIST, 1, 1, NULL, 0);
 275                 break;
 276         case AUDIT_ADD:
 277                 if (!(entry = kmalloc(sizeof(*entry), GFP_KERNEL)))
 278                         return -ENOMEM;
 279                 if (audit_copy_rule(&entry->rule, data)) {
 280                         kfree(entry);
 281                         return -EINVAL;
 282                 }
 283                 flags = entry->rule.flags;
 284                 if (!err && (flags & AUDIT_PER_TASK))
 285                         err = audit_add_rule(entry, &audit_tsklist);
 286                 if (!err && (flags & AUDIT_AT_ENTRY))
 287                         err = audit_add_rule(entry, &audit_entlist);
 288                 if (!err && (flags & AUDIT_AT_EXIT))
 289                         err = audit_add_rule(entry, &audit_extlist);
 290                 audit_log(NULL, "auid %u added an audit rule\n", loginuid);
 291                 break;
 292         case AUDIT_DEL:
 293                 flags =((struct audit_rule *)data)->flags;
 294                 if (!err && (flags & AUDIT_PER_TASK))
 295                         err = audit_del_rule(data, &audit_tsklist);
 296                 if (!err && (flags & AUDIT_AT_ENTRY))
 297                         err = audit_del_rule(data, &audit_entlist);
 298                 if (!err && (flags & AUDIT_AT_EXIT))
 299                         err = audit_del_rule(data, &audit_extlist);
 300                 audit_log(NULL, "auid %u removed an audit rule\n", loginuid);
 301                 break;
 302         default:
 303                 return -EINVAL;
 304         }
 305
 306         return err;
 307 }
 308 #endif
 309
 310 /* Compare a task_struct with an audit_rule.  Return 1 on match, 0
 311  * otherwise. */
 312 static int audit_filter_rules(struct task_struct *tsk,
 313                               struct audit_rule *rule,
 314                               struct audit_context *ctx,
 315                               enum audit_state *state)
 316 {
 317         int i, j;
 318
 319         for (i = 0; i < rule->field_count; i++) {
 320                 u32 field  = rule->fields[i] & ~AUDIT_NEGATE;
 321                 u32 value  = rule->values[i];
 322                 int result = 0;
 323
 324                 switch (field) {
 325                 case AUDIT_PID:
 326                         result = (tsk->pid == value);
 327                         break;
 328                 case AUDIT_UID:
 329                         result = (tsk->uid == value);
 330                         break;
 331                 case AUDIT_EUID:
 332                         result = (tsk->euid == value);
 333                         break;
 334                 case AUDIT_SUID:
 335                         result = (tsk->suid == value);
 336                         break;
 337                 case AUDIT_FSUID:
 338                         result = (tsk->fsuid == value);
 339                         break;
 340                 case AUDIT_GID:
 341                         result = (tsk->gid == value);
 342                         break;
 343                 case AUDIT_EGID:
 344                         result = (tsk->egid == value);
 345                         break;
 346                 case AUDIT_SGID:
 347                         result = (tsk->sgid == value);
 348                         break;
 349                 case AUDIT_FSGID:
 350                         result = (tsk->fsgid == value);
 351                         break;
 352                 case AUDIT_PERS:
 353                         result = (tsk->personality == value);
 354                         break;
 355                 case AUDIT_ARCH:
 356                         if (ctx)
 357                                 result = (ctx->arch == value);
 358                         break;
 359
 360                 case AUDIT_EXIT:
 361                         if (ctx && ctx->return_valid)
 362                                 result = (ctx->return_code == value);
 363                         break;
 364                 case AUDIT_SUCCESS:
 365                         if (ctx && ctx->return_valid)
 366                                 result = (ctx->return_valid == AUDITSC_SUCCESS);
 367                         break;
 368                 case AUDIT_DEVMAJOR:
 369                         if (ctx) {
 370                                 for (j = 0; j < ctx->name_count; j++) {
 371                                         if (MAJOR(ctx->names[j].dev)==value) {
 372                                                 ++result;
 373                                                 break;
 374                                         }
 375                                 }
 376                         }
 377                         break;
 378                 case AUDIT_DEVMINOR:
 379                         if (ctx) {
 380                                 for (j = 0; j < ctx->name_count; j++) {
 381                                         if (MINOR(ctx->names[j].dev)==value) {
 382                                                 ++result;
 383                                                 break;
 384                                         }
 385                                 }
 386                         }
 387                         break;
 388                 case AUDIT_INODE:
 389                         if (ctx) {
 390                                 for (j = 0; j < ctx->name_count; j++) {
 391                                         if (ctx->names[j].ino == value) {
 392                                                 ++result;
 393                                                 break;
 394                                         }
 395                                 }
 396                         }
 397                         break;
 398                 case AUDIT_LOGINUID:
 399                         result = 0;
 400                         if (ctx)
 401                                 result = (ctx->loginuid == value);
 402                         break;
 403                 case AUDIT_ARG0:
 404                 case AUDIT_ARG1:
 405                 case AUDIT_ARG2:
 406                 case AUDIT_ARG3:
 407                         if (ctx)
 408                                 result = (ctx->argv[field-AUDIT_ARG0]==value);
 409                         break;
 410                 }
 411
 412                 if (rule->fields[i] & AUDIT_NEGATE)
 413                         result = !result;
 414                 if (!result)
 415                         return 0;
 416         }
 417         switch (rule->action) {
 418         case AUDIT_NEVER:    *state = AUDIT_DISABLED;       break;
 419         case AUDIT_POSSIBLE: *state = AUDIT_BUILD_CONTEXT;  break;
 420         case AUDIT_ALWAYS:   *state = AUDIT_RECORD_CONTEXT; break;
 421         }
 422         return 1;
 423 }
 424
 425 /* At process creation time, we can determine if system-call auditing is
 426  * completely disabled for this task.  Since we only have the task
 427  * structure at this point, we can only check uid and gid.
 428  */
 429 static enum audit_state audit_filter_task(struct task_struct *tsk)
 430 {
 431         struct audit_entry *e;
 432         enum audit_state   state;
 433
 434         rcu_read_lock();
 435         list_for_each_entry_rcu(e, &audit_tsklist, list) {
 436                 if (audit_filter_rules(tsk, &e->rule, NULL, &state)) {
 437                         rcu_read_unlock();
 438                         return state;
 439                 }
 440         }
 441         rcu_read_unlock();
 442         return AUDIT_BUILD_CONTEXT;
 443 }
 444
 445 /* At syscall entry and exit time, this filter is called if the
 446  * audit_state is not low enough that auditing cannot take place, but is
 447  * also not high enough that we already know we have to write and audit
 448  * record (i.e., the state is AUDIT_SETUP_CONTEXT or  AUDIT_BUILD_CONTEXT).
 449  */
 450 static enum audit_state audit_filter_syscall(struct task_struct *tsk,
 451                                              struct audit_context *ctx,
 452                                              struct list_head *list)
 453 {
 454         struct audit_entry *e;
 455         enum audit_state   state;
 456         int                word = AUDIT_WORD(ctx->major);
 457         int                bit  = AUDIT_BIT(ctx->major);
 458
 459         rcu_read_lock();
 460         list_for_each_entry_rcu(e, list, list) {
 461                 if ((e->rule.mask[word] & bit) == bit
 462                     && audit_filter_rules(tsk, &e->rule, ctx, &state)) {
 463                         rcu_read_unlock();
 464                         return state;
 465                 }
 466         }
 467         rcu_read_unlock();
 468         return AUDIT_BUILD_CONTEXT;
 469 }
 470
 471 /* This should be called with task_lock() held. */
 472 static inline struct audit_context *audit_get_context(struct task_struct *tsk,
 473                                                       int return_valid,
 474                                                       int return_code)
 475 {
 476         struct audit_context *context = tsk->audit_context;
 477
 478         if (likely(!context))
 479                 return NULL;
 480         context->return_valid = return_valid;
 481         context->return_code  = return_code;
 482
 483         if (context->in_syscall && !context->auditable) {
 484                 enum audit_state state;
 485                 state = audit_filter_syscall(tsk, context, &audit_extlist);
 486                 if (state == AUDIT_RECORD_CONTEXT)
 487                         context->auditable = 1;
 488         }
 489
 490         context->pid = tsk->pid;
 491         context->uid = tsk->uid;
 492         context->gid = tsk->gid;
 493         context->euid = tsk->euid;
 494         context->suid = tsk->suid;
 495         context->fsuid = tsk->fsuid;
 496         context->egid = tsk->egid;
 497         context->sgid = tsk->sgid;
 498         context->fsgid = tsk->fsgid;
 499         context->personality = tsk->personality;
 500         tsk->audit_context = NULL;
 501         return context;
 502 }
 503
 504 static inline void audit_free_names(struct audit_context *context)
 505 {
 506         int i;
 507
 508 #if AUDIT_DEBUG == 2
 509         if (context->auditable
 510             ||context->put_count + context->ino_count != context->name_count) {
 511                 printk(KERN_ERR "audit.c:%d(:%d): major=%d in_syscall=%d"
 512                        " name_count=%d put_count=%d"
 513                        " ino_count=%d [NOT freeing]\n",
 514                        __LINE__,
 515                        context->serial, context->major, context->in_syscall,
 516                        context->name_count, context->put_count,
 517                        context->ino_count);
 518                 for (i = 0; i < context->name_count; i++)
 519                         printk(KERN_ERR "names[%d] = %p = %s\n", i,
 520                                context->names[i].name,
 521                                context->names[i].name);
 522                 dump_stack();
 523                 return;
 524         }
 525 #endif
 526 #if AUDIT_DEBUG
 527         context->put_count  = 0;
 528         context->ino_count  = 0;
 529 #endif
 530
 531         for (i = 0; i < context->name_count; i++)
 532                 if (context->names[i].name)
 533                         __putname(context->names[i].name);
 534         context->name_count = 0;
 535 }
 536
 537 static inline void audit_free_aux(struct audit_context *context)
 538 {
 539         struct audit_aux_data *aux;
 540
 541         while ((aux = context->aux)) {
 542                 context->aux = aux->next;
 543                 kfree(aux);
 544         }
 545 }
 546
 547 static inline void audit_zero_context(struct audit_context *context,
 548                                       enum audit_state state)
 549 {
 550         uid_t loginuid = context->loginuid;
 551
 552         memset(context, 0, sizeof(*context));
 553         context->state      = state;
 554         context->loginuid   = loginuid;
 555 }
 556
 557 static inline struct audit_context *audit_alloc_context(enum audit_state state)
 558 {
 559         struct audit_context *context;
 560
 561         if (!(context = kmalloc(sizeof(*context), GFP_KERNEL)))
 562                 return NULL;
 563         audit_zero_context(context, state);
 564         return context;
 565 }
 566
 567 /* Filter on the task information and allocate a per-task audit context
 568  * if necessary.  Doing so turns on system call auditing for the
 569  * specified task.  This is called from copy_process, so no lock is
 570  * needed. */
 571 int audit_alloc(struct task_struct *tsk)
 572 {
 573         struct audit_context *context;
 574         enum audit_state     state;
 575
 576         if (likely(!audit_enabled))
 577                 return 0; /* Return if not auditing. */
 578
 579         state = audit_filter_task(tsk);
 580         if (likely(state == AUDIT_DISABLED))
 581                 return 0;
 582
 583         if (!(context = audit_alloc_context(state))) {
 584                 audit_log_lost("out of memory in audit_alloc");
 585                 return -ENOMEM;
 586         }
 587
 588                                 /* Preserve login uid */
 589         context->loginuid = -1;
 590         if (current->audit_context)
 591                 context->loginuid = current->audit_context->loginuid;
 592
 593         tsk->audit_context  = context;
 594         set_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
 595         return 0;
 596 }
 597
 598 static inline void audit_free_context(struct audit_context *context)
 599 {
 600         struct audit_context *previous;
 601         int                  count = 0;
 602
 603         do {
 604                 previous = context->previous;
 605                 if (previous || (count &&  count < 10)) {
 606                         ++count;
 607                         printk(KERN_ERR "audit(:%d): major=%d name_count=%d:"
 608                                " freeing multiple contexts (%d)\n",
 609                                context->serial, context->major,
 610                                context->name_count, count);
 611                 }
 612                 audit_free_names(context);
 613                 audit_free_aux(context);
 614                 kfree(context);
 615                 context  = previous;
 616         } while (context);
 617         if (count >= 10)
 618                 printk(KERN_ERR "audit: freed %d contexts\n", count);
 619 }
 620
 621 static void audit_log_task_info(struct audit_buffer *ab)
 622 {
 623         char name[sizeof(current->comm)];
 624         struct mm_struct *mm = current->mm;
 625         struct vm_area_struct *vma;
 626
 627         get_task_comm(name, current);
 628         audit_log_format(ab, " comm=%s", name);
 629
 630         if (!mm)
 631                 return;
 632
 633         down_read(&mm->mmap_sem);
 634         vma = mm->mmap;
 635         while (vma) {
 636                 if ((vma->vm_flags & VM_EXECUTABLE) &&
 637                     vma->vm_file) {
 638                         audit_log_d_path(ab, "exe=",
 639                                          vma->vm_file->f_dentry,
 640                                          vma->vm_file->f_vfsmnt);
 641                         break;
 642                 }
 643                 vma = vma->vm_next;
 644         }
 645         up_read(&mm->mmap_sem);
 646 }
 647
 648 static void audit_log_exit(struct audit_context *context)
 649 {
 650         int i;
 651         struct audit_buffer *ab;
 652
 653         ab = audit_log_start(context);
 654         if (!ab)
 655                 return;         /* audit_panic has been called */
 656         audit_log_format(ab, "syscall=%d", context->major);
 657         if (context->personality != PER_LINUX)
 658                 audit_log_format(ab, " per=%lx", context->personality);
 659         audit_log_format(ab, " arch=%x", context->arch);
 660         if (context->return_valid)
 661                 audit_log_format(ab, " success=%s exit=%ld",
 662                                  (context->return_valid==AUDITSC_SUCCESS)?"yes":"no",
 663                                  context->return_code);
 664         audit_log_format(ab,
 665                   " a0=%lx a1=%lx a2=%lx a3=%lx items=%d"
 666                   " pid=%d loginuid=%d uid=%d gid=%d"
 667                   " euid=%d suid=%d fsuid=%d"
 668                   " egid=%d sgid=%d fsgid=%d",
 669                   context->argv[0],
 670                   context->argv[1],
 671                   context->argv[2],
 672                   context->argv[3],
 673                   context->name_count,
 674                   context->pid,
 675                   context->loginuid,
 676                   context->uid,
 677                   context->gid,
 678                   context->euid, context->suid, context->fsuid,
 679                   context->egid, context->sgid, context->fsgid);
 680         audit_log_task_info(ab);
 681         audit_log_end(ab);
 682         while (context->aux) {
 683                 struct audit_aux_data *aux;
 684
 685                 ab = audit_log_start(context);
 686                 if (!ab)
 687                         continue; /* audit_panic has been called */
 688
 689                 aux = context->aux;
 690                 context->aux = aux->next;
 691
 692                 audit_log_format(ab, "auxitem=%d", aux->type);
 693                 switch (aux->type) {
 694                 case AUDIT_AUX_IPCPERM: {
 695                         struct audit_aux_data_ipcctl *axi = (void *)aux;
 696                         audit_log_format(ab,
 697                                          " qbytes=%lx uid=%d gid=%d mode=%x",
 698                                          axi->qbytes, axi->uid, axi->gid, axi->mode);
 699                         }
 700                 }
 701                 audit_log_end(ab);
 702                 kfree(aux);
 703         }
 704
 705         for (i = 0; i < context->name_count; i++) {
 706                 ab = audit_log_start(context);
 707                 if (!ab)
 708                         continue; /* audit_panic has been called */
 709                 audit_log_format(ab, "item=%d", i);
 710                 if (context->names[i].name) {
 711                         audit_log_format(ab, " name=");
 712                         audit_log_untrustedstring(ab, context->names[i].name);
 713                 }
 714                 if (context->names[i].ino != (unsigned long)-1)
 715                         audit_log_format(ab, " inode=%lu dev=%02x:%02x mode=%#o"
 716                                              " uid=%d gid=%d rdev=%02x:%02x",
 717                                          context->names[i].ino,
 718                                          MAJOR(context->names[i].dev),
 719                                          MINOR(context->names[i].dev),
 720                                          context->names[i].mode,
 721                                          context->names[i].uid,
 722                                          context->names[i].gid,
 723                                          MAJOR(context->names[i].rdev),
 724                                          MINOR(context->names[i].rdev));
 725                 audit_log_end(ab);
 726         }
 727 }
 728
 729 /* Free a per-task audit context.  Called from copy_process and
 730  * __put_task_struct. */
 731 void audit_free(struct task_struct *tsk)
 732 {
 733         struct audit_context *context;
 734
 735         task_lock(tsk);
 736         context = audit_get_context(tsk, 0, 0);
 737         task_unlock(tsk);
 738
 739         if (likely(!context))
 740                 return;
 741
 742         /* Check for system calls that do not go through the exit
 743          * function (e.g., exit_group), then free context block. */
 744         if (context->in_syscall && context->auditable)
 745                 audit_log_exit(context);
 746
 747         audit_free_context(context);
 748 }
 749
 750 /* Compute a serial number for the audit record.  Audit records are
 751  * written to user-space as soon as they are generated, so a complete
 752  * audit record may be written in several pieces.  The timestamp of the
 753  * record and this serial number are used by the user-space daemon to
 754  * determine which pieces belong to the same audit record.  The
 755  * (timestamp,serial) tuple is unique for each syscall and is live from
 756  * syscall entry to syscall exit.
 757  *
 758  * Atomic values are only guaranteed to be 24-bit, so we count down.
 759  *
 760  * NOTE: Another possibility is to store the formatted records off the
 761  * audit context (for those records that have a context), and emit them
 762  * all at syscall exit.  However, this could delay the reporting of
 763  * significant errors until syscall exit (or never, if the system
 764  * halts). */
 765 static inline unsigned int audit_serial(void)
 766 {
 767         static atomic_t serial = ATOMIC_INIT(0xffffff);
 768         unsigned int a, b;
 769
 770         do {
 771                 a = atomic_read(&serial);
 772                 if (atomic_dec_and_test(&serial))
 773                         atomic_set(&serial, 0xffffff);
 774                 b = atomic_read(&serial);
 775         } while (b != a - 1);
 776
 777         return 0xffffff - b;
 778 }
 779
 780 /* Fill in audit context at syscall entry.  This only happens if the
 781  * audit context was created when the task was created and the state or
 782  * filters demand the audit context be built.  If the state from the
 783  * per-task filter or from the per-syscall filter is AUDIT_RECORD_CONTEXT,
 784  * then the record will be written at syscall exit time (otherwise, it
 785  * will only be written if another part of the kernel requests that it
 786  * be written). */
 787 void audit_syscall_entry(struct task_struct *tsk, int arch, int major,
 788                          unsigned long a1, unsigned long a2,
 789                          unsigned long a3, unsigned long a4)
 790 {
 791         struct audit_context *context = tsk->audit_context;
 792         enum audit_state     state;
 793
 794         BUG_ON(!context);
 795
 796         /* This happens only on certain architectures that make system
 797          * calls in kernel_thread via the entry.S interface, instead of
 798          * with direct calls.  (If you are porting to a new
 799          * architecture, hitting this condition can indicate that you
 800          * got the _exit/_leave calls backward in entry.S.)
 801          *
 802          * i386     no
 803          * x86_64   no
 804          * ppc64    yes (see arch/ppc64/kernel/misc.S)
 805          *
 806          * This also happens with vm86 emulation in a non-nested manner
 807          * (entries without exits), so this case must be caught.
 808          */
 809         if (context->in_syscall) {
 810                 struct audit_context *newctx;
 811
 812 #if defined(__NR_vm86) && defined(__NR_vm86old)
 813                 /* vm86 mode should only be entered once */
 814                 if (major == __NR_vm86 || major == __NR_vm86old)
 815                         return;
 816 #endif
 817 #if AUDIT_DEBUG
 818                 printk(KERN_ERR
 819                        "audit(:%d) pid=%d in syscall=%d;"
 820                        " entering syscall=%d\n",
 821                        context->serial, tsk->pid, context->major, major);
 822 #endif
 823                 newctx = audit_alloc_context(context->state);
 824                 if (newctx) {
 825                         newctx->previous   = context;
 826                         context            = newctx;
 827                         tsk->audit_context = newctx;
 828                 } else  {
 829                         /* If we can't alloc a new context, the best we
 830                          * can do is to leak memory (any pending putname
 831                          * will be lost).  The only other alternative is
 832                          * to abandon auditing. */
 833                         audit_zero_context(context, context->state);
 834                 }
 835         }
 836         BUG_ON(context->in_syscall || context->name_count);
 837
 838         if (!audit_enabled)
 839                 return;
 840
 841         context->arch       = arch;
 842         context->major      = major;
 843         context->argv[0]    = a1;
 844         context->argv[1]    = a2;
 845         context->argv[2]    = a3;
 846         context->argv[3]    = a4;
 847
 848         state = context->state;
 849         if (state == AUDIT_SETUP_CONTEXT || state == AUDIT_BUILD_CONTEXT)
 850                 state = audit_filter_syscall(tsk, context, &audit_entlist);
 851         if (likely(state == AUDIT_DISABLED))
 852                 return;
 853
 854         context->serial     = audit_serial();
 855         context->ctime      = CURRENT_TIME;
 856         context->in_syscall = 1;
 857         context->auditable  = !!(state == AUDIT_RECORD_CONTEXT);
 858 }
 859
 860 /* Tear down after system call.  If the audit context has been marked as
 861  * auditable (either because of the AUDIT_RECORD_CONTEXT state from
 862  * filtering, or because some other part of the kernel write an audit
 863  * message), then write out the syscall information.  In call cases,
 864  * free the names stored from getname(). */
 865 void audit_syscall_exit(struct task_struct *tsk, int valid, long return_code)
 866 {
 867         struct audit_context *context;
 868
 869         get_task_struct(tsk);
 870         task_lock(tsk);
 871         context = audit_get_context(tsk, valid, return_code);
 872         task_unlock(tsk);
 873
 874         /* Not having a context here is ok, since the parent may have
 875          * called __put_task_struct. */
 876         if (likely(!context))
 877                 return;
 878
 879         if (context->in_syscall && context->auditable)
 880                 audit_log_exit(context);
 881
 882         context->in_syscall = 0;
 883         context->auditable  = 0;
 884
 885         if (context->previous) {
 886                 struct audit_context *new_context = context->previous;
 887                 context->previous  = NULL;
 888                 audit_free_context(context);
 889                 tsk->audit_context = new_context;
 890         } else {
 891                 audit_free_names(context);
 892                 audit_free_aux(context);
 893                 audit_zero_context(context, context->state);
 894                 tsk->audit_context = context;
 895         }
 896         put_task_struct(tsk);
 897 }
 898
 899 /* Add a name to the list.  Called from fs/namei.c:getname(). */
 900 void audit_getname(const char *name)
 901 {
 902         struct audit_context *context = current->audit_context;
 903
 904         if (!context || IS_ERR(name) || !name)
 905                 return;
 906
 907         if (!context->in_syscall) {
 908 #if AUDIT_DEBUG == 2
 909                 printk(KERN_ERR "%s:%d(:%d): ignoring getname(%p)\n",
 910                        __FILE__, __LINE__, context->serial, name);
 911                 dump_stack();
 912 #endif
 913                 return;
 914         }
 915         BUG_ON(context->name_count >= AUDIT_NAMES);
 916         context->names[context->name_count].name = name;
 917         context->names[context->name_count].ino  = (unsigned long)-1;
 918         ++context->name_count;
 919 }
 920
 921 /* Intercept a putname request.  Called from
 922  * include/linux/fs.h:putname().  If we have stored the name from
 923  * getname in the audit context, then we delay the putname until syscall
 924  * exit. */
 925 void audit_putname(const char *name)
 926 {
 927         struct audit_context *context = current->audit_context;
 928
 929         BUG_ON(!context);
 930         if (!context->in_syscall) {
 931 #if AUDIT_DEBUG == 2
 932                 printk(KERN_ERR "%s:%d(:%d): __putname(%p)\n",
 933                        __FILE__, __LINE__, context->serial, name);
 934                 if (context->name_count) {
 935                         int i;
 936                         for (i = 0; i < context->name_count; i++)
 937                                 printk(KERN_ERR "name[%d] = %p = %s\n", i,
 938                                        context->names[i].name,
 939                                        context->names[i].name);
 940                 }
 941 #endif
 942                 __putname(name);
 943         }
 944 #if AUDIT_DEBUG
 945         else {
 946                 ++context->put_count;
 947                 if (context->put_count > context->name_count) {
 948                         printk(KERN_ERR "%s:%d(:%d): major=%d"
 949                                " in_syscall=%d putname(%p) name_count=%d"
 950                                " put_count=%d\n",
 951                                __FILE__, __LINE__,
 952                                context->serial, context->major,
 953                                context->in_syscall, name, context->name_count,
 954                                context->put_count);
 955                         dump_stack();
 956                 }
 957         }
 958 #endif
 959 }
 960
 961 /* Store the inode and device from a lookup.  Called from
 962  * fs/namei.c:path_lookup(). */
 963 void audit_inode(const char *name, const struct inode *inode)
 964 {
 965         int idx;
 966         struct audit_context *context = current->audit_context;
 967
 968         if (!context->in_syscall)
 969                 return;
 970         if (context->name_count
 971             && context->names[context->name_count-1].name
 972             && context->names[context->name_count-1].name == name)
 973                 idx = context->name_count - 1;
 974         else if (context->name_count > 1
 975                  && context->names[context->name_count-2].name
 976                  && context->names[context->name_count-2].name == name)
 977                 idx = context->name_count - 2;
 978         else {
 979                 /* FIXME: how much do we care about inodes that have no
 980                  * associated name? */
 981                 if (context->name_count >= AUDIT_NAMES - AUDIT_NAMES_RESERVED)
 982                         return;
 983                 idx = context->name_count++;
 984                 context->names[idx].name = NULL;
 985 #if AUDIT_DEBUG
 986                 ++context->ino_count;
 987 #endif
 988         }
 989         context->names[idx].ino  = inode->i_ino;
 990         context->names[idx].dev  = inode->i_sb->s_dev;
 991         context->names[idx].mode = inode->i_mode;
 992         context->names[idx].uid  = inode->i_uid;
 993         context->names[idx].gid  = inode->i_gid;
 994         context->names[idx].rdev = inode->i_rdev;
 995 }
 996
 997 void audit_get_stamp(struct audit_context *ctx,
 998                      struct timespec *t, unsigned int *serial)
 999 {
1000         if (ctx) {
1001                 t->tv_sec  = ctx->ctime.tv_sec;
1002                 t->tv_nsec = ctx->ctime.tv_nsec;
1003                 *serial    = ctx->serial;
1004                 ctx->auditable = 1;
1005         } else {
1006                 *t      = CURRENT_TIME;
1007                 *serial = 0;
1008         }
1009 }
1010
1011 extern int audit_set_type(struct audit_buffer *ab, int type);
1012
1013 int audit_set_loginuid(struct task_struct *task, uid_t loginuid)
1014 {
1015         if (task->audit_context) {
1016                 struct audit_buffer *ab;
1017
1018                 ab = audit_log_start(NULL);
1019                 if (ab) {
1020                         audit_log_format(ab, "login pid=%d uid=%u "
1021                                 "old loginuid=%u new loginuid=%u",
1022                                 task->pid, task->uid,
1023                                 task->audit_context->loginuid, loginuid);
1024                         audit_set_type(ab, AUDIT_LOGIN);
1025                         audit_log_end(ab);
1026                 }
1027                 task->audit_context->loginuid = loginuid;
1028         }
1029         return 0;
1030 }
1031
1032 uid_t audit_get_loginuid(struct audit_context *ctx)
1033 {
1034         return ctx ? ctx->loginuid : -1;
1035 }
1036
1037 int audit_ipc_perms(unsigned long qbytes, uid_t uid, gid_t gid, mode_t mode)
1038 {
1039         struct audit_aux_data_ipcctl *ax;
1040         struct audit_context *context = current->audit_context;
1041
1042         if (likely(!context))
1043                 return 0;
1044
1045         ax = kmalloc(sizeof(*ax), GFP_KERNEL);
1046         if (!ax)
1047                 return -ENOMEM;
1048
1049         ax->qbytes = qbytes;
1050         ax->uid = uid;
1051         ax->gid = gid;
1052         ax->mode = mode;
1053
1054         ax->d.type = AUDIT_AUX_IPCPERM;
1055         ax->d.next = context->aux;
1056         context->aux = (void *)ax;
1057         return 0;
1058 }