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 /* 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                                                         uid_t loginuid)
 255 {
 256         u32                flags;
 257         struct audit_entry *entry;
 258         int                err = 0;
 259
 260         switch (type) {
 261         case AUDIT_LIST:
 262                 /* The *_rcu iterators not needed here because we are
 263                    always called with audit_netlink_sem held. */
 264                 list_for_each_entry(entry, &audit_tsklist, list)
 265                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 266                                          &entry->rule, sizeof(entry->rule));
 267                 list_for_each_entry(entry, &audit_entlist, list)
 268                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 269                                          &entry->rule, sizeof(entry->rule));
 270                 list_for_each_entry(entry, &audit_extlist, list)
 271                         audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
 272                                          &entry->rule, sizeof(entry->rule));
 273                 audit_send_reply(pid, seq, AUDIT_LIST, 1, 1, NULL, 0);
 274                 break;
 275         case AUDIT_ADD:
 276                 if (!(entry = kmalloc(sizeof(*entry), GFP_KERNEL)))
 277                         return -ENOMEM;
 278                 if (audit_copy_rule(&entry->rule, data)) {
 279                         kfree(entry);
 280                         return -EINVAL;
 281                 }
 282                 flags = entry->rule.flags;
 283                 if (!err && (flags & AUDIT_PER_TASK))
 284                         err = audit_add_rule(entry, &audit_tsklist);
 285                 if (!err && (flags & AUDIT_AT_ENTRY))
 286                         err = audit_add_rule(entry, &audit_entlist);
 287                 if (!err && (flags & AUDIT_AT_EXIT))
 288                         err = audit_add_rule(entry, &audit_extlist);
 289                 audit_log(NULL, "auid %u added an audit rule\n", loginuid);
 290                 break;
 291         case AUDIT_DEL:
 292                 flags =((struct audit_rule *)data)->flags;
 293                 if (!err && (flags & AUDIT_PER_TASK))
 294                         err = audit_del_rule(data, &audit_tsklist);
 295                 if (!err && (flags & AUDIT_AT_ENTRY))
 296                         err = audit_del_rule(data, &audit_entlist);
 297                 if (!err && (flags & AUDIT_AT_EXIT))
 298                         err = audit_del_rule(data, &audit_extlist);
 299                 audit_log(NULL, "auid %u removed an audit rule\n", loginuid);
 300                 break;
 301         default:
 302                 return -EINVAL;
 303         }
 304
 305         return err;
 306 }
 307
 308 /* Compare a task_struct with an audit_rule.  Return 1 on match, 0
 309  * otherwise. */
 310 static int audit_filter_rules(struct task_struct *tsk,
 311                               struct audit_rule *rule,
 312                               struct audit_context *ctx,
 313                               enum audit_state *state)
 314 {
 315         int i, j;
 316
 317         for (i = 0; i < rule->field_count; i++) {
 318                 u32 field  = rule->fields[i] & ~AUDIT_NEGATE;
 319                 u32 value  = rule->values[i];
 320                 int result = 0;
 321
 322                 switch (field) {
 323                 case AUDIT_PID:
 324                         result = (tsk->pid == value);
 325                         break;
 326                 case AUDIT_UID:
 327                         result = (tsk->uid == value);
 328                         break;
 329                 case AUDIT_EUID:
 330                         result = (tsk->euid == value);
 331                         break;
 332                 case AUDIT_SUID:
 333                         result = (tsk->suid == value);
 334                         break;
 335                 case AUDIT_FSUID:
 336                         result = (tsk->fsuid == value);
 337                         break;
 338                 case AUDIT_GID:
 339                         result = (tsk->gid == value);
 340                         break;
 341                 case AUDIT_EGID:
 342                         result = (tsk->egid == value);
 343                         break;
 344                 case AUDIT_SGID:
 345                         result = (tsk->sgid == value);
 346                         break;
 347                 case AUDIT_FSGID:
 348                         result = (tsk->fsgid == value);
 349                         break;
 350                 case AUDIT_PERS:
 351                         result = (tsk->personality == value);
 352                         break;
 353                 case AUDIT_ARCH:
 354                         if (ctx)
 355                                 result = (ctx->arch == value);
 356                         break;
 357
 358                 case AUDIT_EXIT:
 359                         if (ctx && ctx->return_valid)
 360                                 result = (ctx->return_code == value);
 361                         break;
 362                 case AUDIT_SUCCESS:
 363                         if (ctx && ctx->return_valid)
 364                                 result = (ctx->return_valid == AUDITSC_SUCCESS);
 365                         break;
 366                 case AUDIT_DEVMAJOR:
 367                         if (ctx) {
 368                                 for (j = 0; j < ctx->name_count; j++) {
 369                                         if (MAJOR(ctx->names[j].dev)==value) {
 370                                                 ++result;
 371                                                 break;
 372                                         }
 373                                 }
 374                         }
 375                         break;
 376                 case AUDIT_DEVMINOR:
 377                         if (ctx) {
 378                                 for (j = 0; j < ctx->name_count; j++) {
 379                                         if (MINOR(ctx->names[j].dev)==value) {
 380                                                 ++result;
 381                                                 break;
 382                                         }
 383                                 }
 384                         }
 385                         break;
 386                 case AUDIT_INODE:
 387                         if (ctx) {
 388                                 for (j = 0; j < ctx->name_count; j++) {
 389                                         if (ctx->names[j].ino == value) {
 390                                                 ++result;
 391                                                 break;
 392                                         }
 393                                 }
 394                         }
 395                         break;
 396                 case AUDIT_LOGINUID:
 397                         result = 0;
 398                         if (ctx)
 399                                 result = (ctx->loginuid == value);
 400                         break;
 401                 case AUDIT_ARG0:
 402                 case AUDIT_ARG1:
 403                 case AUDIT_ARG2:
 404                 case AUDIT_ARG3:
 405                         if (ctx)
 406                                 result = (ctx->argv[field-AUDIT_ARG0]==value);
 407                         break;
 408                 }
 409
 410                 if (rule->fields[i] & AUDIT_NEGATE)
 411                         result = !result;
 412                 if (!result)
 413                         return 0;
 414         }
 415         switch (rule->action) {
 416         case AUDIT_NEVER:    *state = AUDIT_DISABLED;       break;
 417         case AUDIT_POSSIBLE: *state = AUDIT_BUILD_CONTEXT;  break;
 418         case AUDIT_ALWAYS:   *state = AUDIT_RECORD_CONTEXT; break;
 419         }
 420         return 1;
 421 }
 422
 423 /* At process creation time, we can determine if system-call auditing is
 424  * completely disabled for this task.  Since we only have the task
 425  * structure at this point, we can only check uid and gid.
 426  */
 427 static enum audit_state audit_filter_task(struct task_struct *tsk)
 428 {
 429         struct audit_entry *e;
 430         enum audit_state   state;
 431
 432         rcu_read_lock();
 433         list_for_each_entry_rcu(e, &audit_tsklist, list) {
 434                 if (audit_filter_rules(tsk, &e->rule, NULL, &state)) {
 435                         rcu_read_unlock();
 436                         return state;
 437                 }
 438         }
 439         rcu_read_unlock();
 440         return AUDIT_BUILD_CONTEXT;
 441 }
 442
 443 /* At syscall entry and exit time, this filter is called if the
 444  * audit_state is not low enough that auditing cannot take place, but is
 445  * also not high enough that we already know we have to write and audit
 446  * record (i.e., the state is AUDIT_SETUP_CONTEXT or  AUDIT_BUILD_CONTEXT).
 447  */
 448 static enum audit_state audit_filter_syscall(struct task_struct *tsk,
 449                                              struct audit_context *ctx,
 450                                              struct list_head *list)
 451 {
 452         struct audit_entry *e;
 453         enum audit_state   state;
 454         int                word = AUDIT_WORD(ctx->major);
 455         int                bit  = AUDIT_BIT(ctx->major);
 456
 457         rcu_read_lock();
 458         list_for_each_entry_rcu(e, list, list) {
 459                 if ((e->rule.mask[word] & bit) == bit
 460                     && audit_filter_rules(tsk, &e->rule, ctx, &state)) {
 461                         rcu_read_unlock();
 462                         return state;
 463                 }
 464         }
 465         rcu_read_unlock();
 466         return AUDIT_BUILD_CONTEXT;
 467 }
 468
 469 /* This should be called with task_lock() held. */
 470 static inline struct audit_context *audit_get_context(struct task_struct *tsk,
 471                                                       int return_valid,
 472                                                       int return_code)
 473 {
 474         struct audit_context *context = tsk->audit_context;
 475
 476         if (likely(!context))
 477                 return NULL;
 478         context->return_valid = return_valid;
 479         context->return_code  = return_code;
 480
 481         if (context->in_syscall && !context->auditable) {
 482                 enum audit_state state;
 483                 state = audit_filter_syscall(tsk, context, &audit_extlist);
 484                 if (state == AUDIT_RECORD_CONTEXT)
 485                         context->auditable = 1;
 486         }
 487
 488         context->pid = tsk->pid;
 489         context->uid = tsk->uid;
 490         context->gid = tsk->gid;
 491         context->euid = tsk->euid;
 492         context->suid = tsk->suid;
 493         context->fsuid = tsk->fsuid;
 494         context->egid = tsk->egid;
 495         context->sgid = tsk->sgid;
 496         context->fsgid = tsk->fsgid;
 497         context->personality = tsk->personality;
 498         tsk->audit_context = NULL;
 499         return context;
 500 }
 501
 502 static inline void audit_free_names(struct audit_context *context)
 503 {
 504         int i;
 505
 506 #if AUDIT_DEBUG == 2
 507         if (context->auditable
 508             ||context->put_count + context->ino_count != context->name_count) {
 509                 printk(KERN_ERR "audit.c:%d(:%d): major=%d in_syscall=%d"
 510                        " name_count=%d put_count=%d"
 511                        " ino_count=%d [NOT freeing]\n",
 512                        __LINE__,
 513                        context->serial, context->major, context->in_syscall,
 514                        context->name_count, context->put_count,
 515                        context->ino_count);
 516                 for (i = 0; i < context->name_count; i++)
 517                         printk(KERN_ERR "names[%d] = %p = %s\n", i,
 518                                context->names[i].name,
 519                                context->names[i].name);
 520                 dump_stack();
 521                 return;
 522         }
 523 #endif
 524 #if AUDIT_DEBUG
 525         context->put_count  = 0;
 526         context->ino_count  = 0;
 527 #endif
 528
 529         for (i = 0; i < context->name_count; i++)
 530                 if (context->names[i].name)
 531                         __putname(context->names[i].name);
 532         context->name_count = 0;
 533 }
 534
 535 static inline void audit_free_aux(struct audit_context *context)
 536 {
 537         struct audit_aux_data *aux;
 538
 539         while ((aux = context->aux)) {
 540                 context->aux = aux->next;
 541                 kfree(aux);
 542         }
 543 }
 544
 545 static inline void audit_zero_context(struct audit_context *context,
 546                                       enum audit_state state)
 547 {
 548         uid_t loginuid = context->loginuid;
 549
 550         memset(context, 0, sizeof(*context));
 551         context->state      = state;
 552         context->loginuid   = loginuid;
 553 }
 554
 555 static inline struct audit_context *audit_alloc_context(enum audit_state state)
 556 {
 557         struct audit_context *context;
 558
 559         if (!(context = kmalloc(sizeof(*context), GFP_KERNEL)))
 560                 return NULL;
 561         audit_zero_context(context, state);
 562         return context;
 563 }
 564
 565 /* Filter on the task information and allocate a per-task audit context
 566  * if necessary.  Doing so turns on system call auditing for the
 567  * specified task.  This is called from copy_process, so no lock is
 568  * needed. */
 569 int audit_alloc(struct task_struct *tsk)
 570 {
 571         struct audit_context *context;
 572         enum audit_state     state;
 573
 574         if (likely(!audit_enabled))
 575                 return 0; /* Return if not auditing. */
 576
 577         state = audit_filter_task(tsk);
 578         if (likely(state == AUDIT_DISABLED))
 579                 return 0;
 580
 581         if (!(context = audit_alloc_context(state))) {
 582                 audit_log_lost("out of memory in audit_alloc");
 583                 return -ENOMEM;
 584         }
 585
 586                                 /* Preserve login uid */
 587         context->loginuid = -1;
 588         if (current->audit_context)
 589                 context->loginuid = current->audit_context->loginuid;
 590
 591         tsk->audit_context  = context;
 592         set_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
 593         return 0;
 594 }
 595
 596 static inline void audit_free_context(struct audit_context *context)
 597 {
 598         struct audit_context *previous;
 599         int                  count = 0;
 600
 601         do {
 602                 previous = context->previous;
 603                 if (previous || (count &&  count < 10)) {
 604                         ++count;
 605                         printk(KERN_ERR "audit(:%d): major=%d name_count=%d:"
 606                                " freeing multiple contexts (%d)\n",
 607                                context->serial, context->major,
 608                                context->name_count, count);
 609                 }
 610                 audit_free_names(context);
 611                 audit_free_aux(context);
 612                 kfree(context);
 613                 context  = previous;
 614         } while (context);
 615         if (count >= 10)
 616                 printk(KERN_ERR "audit: freed %d contexts\n", count);
 617 }
 618
 619 static void audit_log_task_info(struct audit_buffer *ab)
 620 {
 621         char name[sizeof(current->comm)];
 622         struct mm_struct *mm = current->mm;
 623         struct vm_area_struct *vma;
 624
 625         get_task_comm(name, current);
 626         audit_log_format(ab, " comm=%s", name);
 627
 628         if (!mm)
 629                 return;
 630
 631         down_read(&mm->mmap_sem);
 632         vma = mm->mmap;
 633         while (vma) {
 634                 if ((vma->vm_flags & VM_EXECUTABLE) &&
 635                     vma->vm_file) {
 636                         audit_log_d_path(ab, "exe=",
 637                                          vma->vm_file->f_dentry,
 638                                          vma->vm_file->f_vfsmnt);
 639                         break;
 640                 }
 641                 vma = vma->vm_next;
 642         }
 643         up_read(&mm->mmap_sem);
 644 }
 645
 646 static void audit_log_exit(struct audit_context *context)
 647 {
 648         int i;
 649         struct audit_buffer *ab;
 650
 651         ab = audit_log_start(context);
 652         if (!ab)
 653                 return;         /* audit_panic has been called */
 654         audit_log_format(ab, "syscall=%d", context->major);
 655         if (context->personality != PER_LINUX)
 656                 audit_log_format(ab, " per=%lx", context->personality);
 657         audit_log_format(ab, " arch=%x", context->arch);
 658         if (context->return_valid)
 659                 audit_log_format(ab, " success=%s exit=%ld",
 660                                  (context->return_valid==AUDITSC_SUCCESS)?"yes":"no",
 661                                  context->return_code);
 662         audit_log_format(ab,
 663                   " a0=%lx a1=%lx a2=%lx a3=%lx items=%d"
 664                   " pid=%d loginuid=%d uid=%d gid=%d"
 665                   " euid=%d suid=%d fsuid=%d"
 666                   " egid=%d sgid=%d fsgid=%d",
 667                   context->argv[0],
 668                   context->argv[1],
 669                   context->argv[2],
 670                   context->argv[3],
 671                   context->name_count,
 672                   context->pid,
 673                   context->loginuid,
 674                   context->uid,
 675                   context->gid,
 676                   context->euid, context->suid, context->fsuid,
 677                   context->egid, context->sgid, context->fsgid);
 678         audit_log_task_info(ab);
 679         audit_log_end(ab);
 680         while (context->aux) {
 681                 struct audit_aux_data *aux;
 682
 683                 ab = audit_log_start(context);
 684                 if (!ab)
 685                         continue; /* audit_panic has been called */
 686
 687                 aux = context->aux;
 688                 context->aux = aux->next;
 689
 690                 audit_log_format(ab, "auxitem=%d", aux->type);
 691                 switch (aux->type) {
 692                 case AUDIT_AUX_IPCPERM: {
 693                         struct audit_aux_data_ipcctl *axi = (void *)aux;
 694                         audit_log_format(ab,
 695                                          " qbytes=%lx uid=%d gid=%d mode=%x",
 696                                          axi->qbytes, axi->uid, axi->gid, axi->mode);
 697                         }
 698                 }
 699                 audit_log_end(ab);
 700                 kfree(aux);
 701         }
 702
 703         for (i = 0; i < context->name_count; i++) {
 704                 ab = audit_log_start(context);
 705                 if (!ab)
 706                         continue; /* audit_panic has been called */
 707                 audit_log_format(ab, "item=%d", i);
 708                 if (context->names[i].name) {
 709                         audit_log_format(ab, " name=");
 710                         audit_log_untrustedstring(ab, context->names[i].name);
 711                 }
 712                 if (context->names[i].ino != (unsigned long)-1)
 713                         audit_log_format(ab, " inode=%lu dev=%02x:%02x mode=%#o"
 714                                              " uid=%d gid=%d rdev=%02x:%02x",
 715                                          context->names[i].ino,
 716                                          MAJOR(context->names[i].dev),
 717                                          MINOR(context->names[i].dev),
 718                                          context->names[i].mode,
 719                                          context->names[i].uid,
 720                                          context->names[i].gid,
 721                                          MAJOR(context->names[i].rdev),
 722                                          MINOR(context->names[i].rdev));
 723                 audit_log_end(ab);
 724         }
 725 }
 726
 727 /* Free a per-task audit context.  Called from copy_process and
 728  * __put_task_struct. */
 729 void audit_free(struct task_struct *tsk)
 730 {
 731         struct audit_context *context;
 732
 733         task_lock(tsk);
 734         context = audit_get_context(tsk, 0, 0);
 735         task_unlock(tsk);
 736
 737         if (likely(!context))
 738                 return;
 739
 740         /* Check for system calls that do not go through the exit
 741          * function (e.g., exit_group), then free context block. */
 742         if (context->in_syscall && context->auditable)
 743                 audit_log_exit(context);
 744
 745         audit_free_context(context);
 746 }
 747
 748 /* Compute a serial number for the audit record.  Audit records are
 749  * written to user-space as soon as they are generated, so a complete
 750  * audit record may be written in several pieces.  The timestamp of the
 751  * record and this serial number are used by the user-space daemon to
 752  * determine which pieces belong to the same audit record.  The
 753  * (timestamp,serial) tuple is unique for each syscall and is live from
 754  * syscall entry to syscall exit.
 755  *
 756  * Atomic values are only guaranteed to be 24-bit, so we count down.
 757  *
 758  * NOTE: Another possibility is to store the formatted records off the
 759  * audit context (for those records that have a context), and emit them
 760  * all at syscall exit.  However, this could delay the reporting of
 761  * significant errors until syscall exit (or never, if the system
 762  * halts). */
 763 static inline unsigned int audit_serial(void)
 764 {
 765         static atomic_t serial = ATOMIC_INIT(0xffffff);
 766         unsigned int a, b;
 767
 768         do {
 769                 a = atomic_read(&serial);
 770                 if (atomic_dec_and_test(&serial))
 771                         atomic_set(&serial, 0xffffff);
 772                 b = atomic_read(&serial);
 773         } while (b != a - 1);
 774
 775         return 0xffffff - b;
 776 }
 777
 778 /* Fill in audit context at syscall entry.  This only happens if the
 779  * audit context was created when the task was created and the state or
 780  * filters demand the audit context be built.  If the state from the
 781  * per-task filter or from the per-syscall filter is AUDIT_RECORD_CONTEXT,
 782  * then the record will be written at syscall exit time (otherwise, it
 783  * will only be written if another part of the kernel requests that it
 784  * be written). */
 785 void audit_syscall_entry(struct task_struct *tsk, int arch, int major,
 786                          unsigned long a1, unsigned long a2,
 787                          unsigned long a3, unsigned long a4)
 788 {
 789         struct audit_context *context = tsk->audit_context;
 790         enum audit_state     state;
 791
 792         BUG_ON(!context);
 793
 794         /* This happens only on certain architectures that make system
 795          * calls in kernel_thread via the entry.S interface, instead of
 796          * with direct calls.  (If you are porting to a new
 797          * architecture, hitting this condition can indicate that you
 798          * got the _exit/_leave calls backward in entry.S.)
 799          *
 800          * i386     no
 801          * x86_64   no
 802          * ppc64    yes (see arch/ppc64/kernel/misc.S)
 803          *
 804          * This also happens with vm86 emulation in a non-nested manner
 805          * (entries without exits), so this case must be caught.
 806          */
 807         if (context->in_syscall) {
 808                 struct audit_context *newctx;
 809
 810 #if defined(__NR_vm86) && defined(__NR_vm86old)
 811                 /* vm86 mode should only be entered once */
 812                 if (major == __NR_vm86 || major == __NR_vm86old)
 813                         return;
 814 #endif
 815 #if AUDIT_DEBUG
 816                 printk(KERN_ERR
 817                        "audit(:%d) pid=%d in syscall=%d;"
 818                        " entering syscall=%d\n",
 819                        context->serial, tsk->pid, context->major, major);
 820 #endif
 821                 newctx = audit_alloc_context(context->state);
 822                 if (newctx) {
 823                         newctx->previous   = context;
 824                         context            = newctx;
 825                         tsk->audit_context = newctx;
 826                 } else  {
 827                         /* If we can't alloc a new context, the best we
 828                          * can do is to leak memory (any pending putname
 829                          * will be lost).  The only other alternative is
 830                          * to abandon auditing. */
 831                         audit_zero_context(context, context->state);
 832                 }
 833         }
 834         BUG_ON(context->in_syscall || context->name_count);
 835
 836         if (!audit_enabled)
 837                 return;
 838
 839         context->arch       = arch;
 840         context->major      = major;
 841         context->argv[0]    = a1;
 842         context->argv[1]    = a2;
 843         context->argv[2]    = a3;
 844         context->argv[3]    = a4;
 845
 846         state = context->state;
 847         if (state == AUDIT_SETUP_CONTEXT || state == AUDIT_BUILD_CONTEXT)
 848                 state = audit_filter_syscall(tsk, context, &audit_entlist);
 849         if (likely(state == AUDIT_DISABLED))
 850                 return;
 851
 852         context->serial     = audit_serial();
 853         context->ctime      = CURRENT_TIME;
 854         context->in_syscall = 1;
 855         context->auditable  = !!(state == AUDIT_RECORD_CONTEXT);
 856 }
 857
 858 /* Tear down after system call.  If the audit context has been marked as
 859  * auditable (either because of the AUDIT_RECORD_CONTEXT state from
 860  * filtering, or because some other part of the kernel write an audit
 861  * message), then write out the syscall information.  In call cases,
 862  * free the names stored from getname(). */
 863 void audit_syscall_exit(struct task_struct *tsk, int valid, long return_code)
 864 {
 865         struct audit_context *context;
 866
 867         get_task_struct(tsk);
 868         task_lock(tsk);
 869         context = audit_get_context(tsk, valid, return_code);
 870         task_unlock(tsk);
 871
 872         /* Not having a context here is ok, since the parent may have
 873          * called __put_task_struct. */
 874         if (likely(!context))
 875                 return;
 876
 877         if (context->in_syscall && context->auditable)
 878                 audit_log_exit(context);
 879
 880         context->in_syscall = 0;
 881         context->auditable  = 0;
 882
 883         if (context->previous) {
 884                 struct audit_context *new_context = context->previous;
 885                 context->previous  = NULL;
 886                 audit_free_context(context);
 887                 tsk->audit_context = new_context;
 888         } else {
 889                 audit_free_names(context);
 890                 audit_free_aux(context);
 891                 audit_zero_context(context, context->state);
 892                 tsk->audit_context = context;
 893         }
 894         put_task_struct(tsk);
 895 }
 896
 897 /* Add a name to the list.  Called from fs/namei.c:getname(). */
 898 void audit_getname(const char *name)
 899 {
 900         struct audit_context *context = current->audit_context;
 901
 902         if (!context || IS_ERR(name) || !name)
 903                 return;
 904
 905         if (!context->in_syscall) {
 906 #if AUDIT_DEBUG == 2
 907                 printk(KERN_ERR "%s:%d(:%d): ignoring getname(%p)\n",
 908                        __FILE__, __LINE__, context->serial, name);
 909                 dump_stack();
 910 #endif
 911                 return;
 912         }
 913         BUG_ON(context->name_count >= AUDIT_NAMES);
 914         context->names[context->name_count].name = name;
 915         context->names[context->name_count].ino  = (unsigned long)-1;
 916         ++context->name_count;
 917 }
 918
 919 /* Intercept a putname request.  Called from
 920  * include/linux/fs.h:putname().  If we have stored the name from
 921  * getname in the audit context, then we delay the putname until syscall
 922  * exit. */
 923 void audit_putname(const char *name)
 924 {
 925         struct audit_context *context = current->audit_context;
 926
 927         BUG_ON(!context);
 928         if (!context->in_syscall) {
 929 #if AUDIT_DEBUG == 2
 930                 printk(KERN_ERR "%s:%d(:%d): __putname(%p)\n",
 931                        __FILE__, __LINE__, context->serial, name);
 932                 if (context->name_count) {
 933                         int i;
 934                         for (i = 0; i < context->name_count; i++)
 935                                 printk(KERN_ERR "name[%d] = %p = %s\n", i,
 936                                        context->names[i].name,
 937                                        context->names[i].name);
 938                 }
 939 #endif
 940                 __putname(name);
 941         }
 942 #if AUDIT_DEBUG
 943         else {
 944                 ++context->put_count;
 945                 if (context->put_count > context->name_count) {
 946                         printk(KERN_ERR "%s:%d(:%d): major=%d"
 947                                " in_syscall=%d putname(%p) name_count=%d"
 948                                " put_count=%d\n",
 949                                __FILE__, __LINE__,
 950                                context->serial, context->major,
 951                                context->in_syscall, name, context->name_count,
 952                                context->put_count);
 953                         dump_stack();
 954                 }
 955         }
 956 #endif
 957 }
 958
 959 /* Store the inode and device from a lookup.  Called from
 960  * fs/namei.c:path_lookup(). */
 961 void audit_inode(const char *name, const struct inode *inode)
 962 {
 963         int idx;
 964         struct audit_context *context = current->audit_context;
 965
 966         if (!context->in_syscall)
 967                 return;
 968         if (context->name_count
 969             && context->names[context->name_count-1].name
 970             && context->names[context->name_count-1].name == name)
 971                 idx = context->name_count - 1;
 972         else if (context->name_count > 1
 973                  && context->names[context->name_count-2].name
 974                  && context->names[context->name_count-2].name == name)
 975                 idx = context->name_count - 2;
 976         else {
 977                 /* FIXME: how much do we care about inodes that have no
 978                  * associated name? */
 979                 if (context->name_count >= AUDIT_NAMES - AUDIT_NAMES_RESERVED)
 980                         return;
 981                 idx = context->name_count++;
 982                 context->names[idx].name = NULL;
 983 #if AUDIT_DEBUG
 984                 ++context->ino_count;
 985 #endif
 986         }
 987         context->names[idx].ino  = inode->i_ino;
 988         context->names[idx].dev  = inode->i_sb->s_dev;
 989         context->names[idx].mode = inode->i_mode;
 990         context->names[idx].uid  = inode->i_uid;
 991         context->names[idx].gid  = inode->i_gid;
 992         context->names[idx].rdev = inode->i_rdev;
 993 }
 994
 995 void audit_get_stamp(struct audit_context *ctx,
 996                      struct timespec *t, unsigned int *serial)
 997 {
 998         if (ctx) {
 999                 t->tv_sec  = ctx->ctime.tv_sec;
1000                 t->tv_nsec = ctx->ctime.tv_nsec;
1001                 *serial    = ctx->serial;
1002                 ctx->auditable = 1;
1003         } else {
1004                 *t      = CURRENT_TIME;
1005                 *serial = 0;
1006         }
1007 }
1008
1009 extern int audit_set_type(struct audit_buffer *ab, int type);
1010
1011 int audit_set_loginuid(struct task_struct *task, uid_t loginuid)
1012 {
1013         if (task->audit_context) {
1014                 struct audit_buffer *ab;
1015
1016                 ab = audit_log_start(NULL);
1017                 if (ab) {
1018                         audit_log_format(ab, "login pid=%d uid=%u "
1019                                 "old loginuid=%u new loginuid=%u",
1020                                 task->pid, task->uid,
1021                                 task->audit_context->loginuid, loginuid);
1022                         audit_set_type(ab, AUDIT_LOGIN);
1023                         audit_log_end(ab);
1024                 }
1025                 task->audit_context->loginuid = loginuid;
1026         }
1027         return 0;
1028 }
1029
1030 uid_t audit_get_loginuid(struct audit_context *ctx)
1031 {
1032         return ctx ? ctx->loginuid : -1;
1033 }
1034
1035 int audit_ipc_perms(unsigned long qbytes, uid_t uid, gid_t gid, mode_t mode)
1036 {
1037         struct audit_aux_data_ipcctl *ax;
1038         struct audit_context *context = current->audit_context;
1039
1040         if (likely(!context))
1041                 return 0;
1042
1043         ax = kmalloc(sizeof(*ax), GFP_KERNEL);
1044         if (!ax)
1045                 return -ENOMEM;
1046
1047         ax->qbytes = qbytes;
1048         ax->uid = uid;
1049         ax->gid = gid;
1050         ax->mode = mode;
1051
1052         ax->d.type = AUDIT_AUX_IPCPERM;
1053         ax->d.next = context->aux;
1054         context->aux = (void *)ax;
1055         return 0;
1056 }
1057
1058 void audit_signal_info(int sig, struct task_struct *t)
1059 {
1060         extern pid_t audit_sig_pid;
1061         extern uid_t audit_sig_uid;
1062         extern int audit_pid;
1063
1064         if (unlikely(audit_pid && t->pid == audit_pid)) {
1065                 if (sig == SIGTERM || sig == SIGHUP) {
1066                         struct audit_context *ctx = current->audit_context;
1067                         audit_sig_pid = current->pid;
1068                         if (ctx)
1069                                 audit_sig_uid = ctx->loginuid;
1070                         else
1071                                 audit_sig_uid = current->uid;
1072                 }
1073         }
1074 }
1075