tunnels: fix netns vs proto registration ordering
[safe/jmp/linux-2.6] / kernel / auditsc.c
index ba08788..fc0f928 100644 (file)
@@ -3,7 +3,7 @@
  *
  * Copyright 2003-2004 Red Hat Inc., Durham, North Carolina.
  * Copyright 2005 Hewlett-Packard Development Company, L.P.
- * Copyright (C) 2005 IBM Corporation
+ * Copyright (C) 2005, 2006 IBM Corporation
  * All Rights Reserved.
  *
  * This program is free software; you can redistribute it and/or modify
@@ -29,6 +29,9 @@
  * this file -- see entry.S) is based on a GPL'd patch written by
  * okir@suse.de and Copyright 2003 SuSE Linux AG.
  *
+ * POSIX message queue support added by George Wilson <ltcgcw@us.ibm.com>,
+ * 2006.
+ *
  * The support of additional filter rules compares (>, <, >=, <=) was
  * added by Dustin Kirkland <dustin.kirkland@us.ibm.com>, 2005.
  *
 #include <linux/init.h>
 #include <asm/types.h>
 #include <asm/atomic.h>
-#include <asm/types.h>
 #include <linux/fs.h>
 #include <linux/namei.h>
 #include <linux/mm.h>
 #include <linux/module.h>
 #include <linux/mount.h>
 #include <linux/socket.h>
+#include <linux/mqueue.h>
 #include <linux/audit.h>
 #include <linux/personality.h>
 #include <linux/time.h>
 #include <asm/unistd.h>
 #include <linux/security.h>
 #include <linux/list.h>
+#include <linux/tty.h>
+#include <linux/binfmts.h>
+#include <linux/highmem.h>
+#include <linux/syscalls.h>
+#include <linux/inotify.h>
+#include <linux/capability.h>
+#include <linux/fs_struct.h>
 
 #include "audit.h"
 
-extern struct list_head audit_filter_list[];
-
-/* No syscall auditing will take place unless audit_enabled != 0. */
-extern int audit_enabled;
-
 /* AUDIT_NAMES is the number of slots we reserve in the audit_context
  * for saving names from getname(). */
 #define AUDIT_NAMES    20
 
-/* AUDIT_NAMES_RESERVED is the number of slots we reserve in the
- * audit_context from being used for nameless inodes from
- * path_lookup. */
-#define AUDIT_NAMES_RESERVED 7
+/* Indicates that audit should log the full pathname. */
+#define AUDIT_NAME_FULL -1
+
+/* no execve audit message should be longer than this (userspace limits) */
+#define MAX_EXECVE_AUDIT_LEN 7500
+
+/* number of audit rules */
+int audit_n_rules;
+
+/* determines whether we collect data for signals sent */
+int audit_signals;
+
+struct audit_cap_data {
+       kernel_cap_t            permitted;
+       kernel_cap_t            inheritable;
+       union {
+               unsigned int    fE;             /* effective bit of a file capability */
+               kernel_cap_t    effective;      /* effective set of a process */
+       };
+};
 
 /* When fs/namei.c:getname() is called, we store the pointer in name and
  * we don't let putname() free it (instead we free all of the saved
@@ -81,14 +102,17 @@ extern int audit_enabled;
  * Further, in fs/namei.c:path_lookup() we store the inode and device. */
 struct audit_names {
        const char      *name;
+       int             name_len;       /* number of name's characters to log */
+       unsigned        name_put;       /* call __putname() for this name */
        unsigned long   ino;
-       unsigned long   pino;
        dev_t           dev;
        umode_t         mode;
        uid_t           uid;
        gid_t           gid;
        dev_t           rdev;
-       char            *ctx;
+       u32             osid;
+       struct audit_cap_data fcap;
+       unsigned int    fcap_ver;
 };
 
 struct audit_aux_data {
@@ -98,75 +122,333 @@ struct audit_aux_data {
 
 #define AUDIT_AUX_IPCPERM      0
 
-struct audit_aux_data_ipcctl {
+/* Number of target pids per aux struct. */
+#define AUDIT_AUX_PIDS 16
+
+struct audit_aux_data_execve {
        struct audit_aux_data   d;
-       struct ipc_perm         p;
-       unsigned long           qbytes;
-       uid_t                   uid;
-       gid_t                   gid;
-       mode_t                  mode;
-       char                    *ctx;
+       int argc;
+       int envc;
+       struct mm_struct *mm;
 };
 
-struct audit_aux_data_socketcall {
+struct audit_aux_data_pids {
        struct audit_aux_data   d;
-       int                     nargs;
-       unsigned long           args[0];
+       pid_t                   target_pid[AUDIT_AUX_PIDS];
+       uid_t                   target_auid[AUDIT_AUX_PIDS];
+       uid_t                   target_uid[AUDIT_AUX_PIDS];
+       unsigned int            target_sessionid[AUDIT_AUX_PIDS];
+       u32                     target_sid[AUDIT_AUX_PIDS];
+       char                    target_comm[AUDIT_AUX_PIDS][TASK_COMM_LEN];
+       int                     pid_count;
 };
 
-struct audit_aux_data_sockaddr {
+struct audit_aux_data_bprm_fcaps {
        struct audit_aux_data   d;
-       int                     len;
-       char                    a[0];
+       struct audit_cap_data   fcap;
+       unsigned int            fcap_ver;
+       struct audit_cap_data   old_pcap;
+       struct audit_cap_data   new_pcap;
 };
 
-struct audit_aux_data_path {
+struct audit_aux_data_capset {
        struct audit_aux_data   d;
-       struct dentry           *dentry;
-       struct vfsmount         *mnt;
+       pid_t                   pid;
+       struct audit_cap_data   cap;
+};
+
+struct audit_tree_refs {
+       struct audit_tree_refs *next;
+       struct audit_chunk *c[31];
 };
 
 /* The per-task audit context. */
 struct audit_context {
+       int                 dummy;      /* must be the first element */
        int                 in_syscall; /* 1 if task is in a syscall */
-       enum audit_state    state;
+       enum audit_state    state, current_state;
        unsigned int        serial;     /* serial number for record */
-       struct timespec     ctime;      /* time of syscall entry */
-       uid_t               loginuid;   /* login uid (identity) */
        int                 major;      /* syscall number */
+       struct timespec     ctime;      /* time of syscall entry */
        unsigned long       argv[4];    /* syscall arguments */
-       int                 return_valid; /* return code is valid */
        long                return_code;/* syscall return code */
-       int                 auditable;  /* 1 if record should be written */
+       u64                 prio;
+       int                 return_valid; /* return code is valid */
        int                 name_count;
        struct audit_names  names[AUDIT_NAMES];
-       struct dentry *     pwd;
-       struct vfsmount *   pwdmnt;
+       char *              filterkey;  /* key for rule that triggered record */
+       struct path         pwd;
        struct audit_context *previous; /* For nested syscalls */
        struct audit_aux_data *aux;
-
+       struct audit_aux_data *aux_pids;
+       struct sockaddr_storage *sockaddr;
+       size_t sockaddr_len;
                                /* Save things to print about task_struct */
-       pid_t               pid;
+       pid_t               pid, ppid;
        uid_t               uid, euid, suid, fsuid;
        gid_t               gid, egid, sgid, fsgid;
        unsigned long       personality;
        int                 arch;
 
+       pid_t               target_pid;
+       uid_t               target_auid;
+       uid_t               target_uid;
+       unsigned int        target_sessionid;
+       u32                 target_sid;
+       char                target_comm[TASK_COMM_LEN];
+
+       struct audit_tree_refs *trees, *first_trees;
+       struct list_head killed_trees;
+       int tree_count;
+
+       int type;
+       union {
+               struct {
+                       int nargs;
+                       long args[6];
+               } socketcall;
+               struct {
+                       uid_t                   uid;
+                       gid_t                   gid;
+                       mode_t                  mode;
+                       u32                     osid;
+                       int                     has_perm;
+                       uid_t                   perm_uid;
+                       gid_t                   perm_gid;
+                       mode_t                  perm_mode;
+                       unsigned long           qbytes;
+               } ipc;
+               struct {
+                       mqd_t                   mqdes;
+                       struct mq_attr          mqstat;
+               } mq_getsetattr;
+               struct {
+                       mqd_t                   mqdes;
+                       int                     sigev_signo;
+               } mq_notify;
+               struct {
+                       mqd_t                   mqdes;
+                       size_t                  msg_len;
+                       unsigned int            msg_prio;
+                       struct timespec         abs_timeout;
+               } mq_sendrecv;
+               struct {
+                       int                     oflag;
+                       mode_t                  mode;
+                       struct mq_attr          attr;
+               } mq_open;
+               struct {
+                       pid_t                   pid;
+                       struct audit_cap_data   cap;
+               } capset;
+       };
+       int fds[2];
+
 #if AUDIT_DEBUG
        int                 put_count;
        int                 ino_count;
 #endif
 };
 
+static inline int open_arg(int flags, int mask)
+{
+       int n = ACC_MODE(flags);
+       if (flags & (O_TRUNC | O_CREAT))
+               n |= AUDIT_PERM_WRITE;
+       return n & mask;
+}
 
+static int audit_match_perm(struct audit_context *ctx, int mask)
+{
+       unsigned n;
+       if (unlikely(!ctx))
+               return 0;
+       n = ctx->major;
+
+       switch (audit_classify_syscall(ctx->arch, n)) {
+       case 0: /* native */
+               if ((mask & AUDIT_PERM_WRITE) &&
+                    audit_match_class(AUDIT_CLASS_WRITE, n))
+                       return 1;
+               if ((mask & AUDIT_PERM_READ) &&
+                    audit_match_class(AUDIT_CLASS_READ, n))
+                       return 1;
+               if ((mask & AUDIT_PERM_ATTR) &&
+                    audit_match_class(AUDIT_CLASS_CHATTR, n))
+                       return 1;
+               return 0;
+       case 1: /* 32bit on biarch */
+               if ((mask & AUDIT_PERM_WRITE) &&
+                    audit_match_class(AUDIT_CLASS_WRITE_32, n))
+                       return 1;
+               if ((mask & AUDIT_PERM_READ) &&
+                    audit_match_class(AUDIT_CLASS_READ_32, n))
+                       return 1;
+               if ((mask & AUDIT_PERM_ATTR) &&
+                    audit_match_class(AUDIT_CLASS_CHATTR_32, n))
+                       return 1;
+               return 0;
+       case 2: /* open */
+               return mask & ACC_MODE(ctx->argv[1]);
+       case 3: /* openat */
+               return mask & ACC_MODE(ctx->argv[2]);
+       case 4: /* socketcall */
+               return ((mask & AUDIT_PERM_WRITE) && ctx->argv[0] == SYS_BIND);
+       case 5: /* execve */
+               return mask & AUDIT_PERM_EXEC;
+       default:
+               return 0;
+       }
+}
+
+static int audit_match_filetype(struct audit_context *ctx, int which)
+{
+       unsigned index = which & ~S_IFMT;
+       mode_t mode = which & S_IFMT;
+
+       if (unlikely(!ctx))
+               return 0;
+
+       if (index >= ctx->name_count)
+               return 0;
+       if (ctx->names[index].ino == -1)
+               return 0;
+       if ((ctx->names[index].mode ^ mode) & S_IFMT)
+               return 0;
+       return 1;
+}
+
+/*
+ * We keep a linked list of fixed-sized (31 pointer) arrays of audit_chunk *;
+ * ->first_trees points to its beginning, ->trees - to the current end of data.
+ * ->tree_count is the number of free entries in array pointed to by ->trees.
+ * Original condition is (NULL, NULL, 0); as soon as it grows we never revert to NULL,
+ * "empty" becomes (p, p, 31) afterwards.  We don't shrink the list (and seriously,
+ * it's going to remain 1-element for almost any setup) until we free context itself.
+ * References in it _are_ dropped - at the same time we free/drop aux stuff.
+ */
+
+#ifdef CONFIG_AUDIT_TREE
+static void audit_set_auditable(struct audit_context *ctx)
+{
+       if (!ctx->prio) {
+               ctx->prio = 1;
+               ctx->current_state = AUDIT_RECORD_CONTEXT;
+       }
+}
+
+static int put_tree_ref(struct audit_context *ctx, struct audit_chunk *chunk)
+{
+       struct audit_tree_refs *p = ctx->trees;
+       int left = ctx->tree_count;
+       if (likely(left)) {
+               p->c[--left] = chunk;
+               ctx->tree_count = left;
+               return 1;
+       }
+       if (!p)
+               return 0;
+       p = p->next;
+       if (p) {
+               p->c[30] = chunk;
+               ctx->trees = p;
+               ctx->tree_count = 30;
+               return 1;
+       }
+       return 0;
+}
+
+static int grow_tree_refs(struct audit_context *ctx)
+{
+       struct audit_tree_refs *p = ctx->trees;
+       ctx->trees = kzalloc(sizeof(struct audit_tree_refs), GFP_KERNEL);
+       if (!ctx->trees) {
+               ctx->trees = p;
+               return 0;
+       }
+       if (p)
+               p->next = ctx->trees;
+       else
+               ctx->first_trees = ctx->trees;
+       ctx->tree_count = 31;
+       return 1;
+}
+#endif
+
+static void unroll_tree_refs(struct audit_context *ctx,
+                     struct audit_tree_refs *p, int count)
+{
+#ifdef CONFIG_AUDIT_TREE
+       struct audit_tree_refs *q;
+       int n;
+       if (!p) {
+               /* we started with empty chain */
+               p = ctx->first_trees;
+               count = 31;
+               /* if the very first allocation has failed, nothing to do */
+               if (!p)
+                       return;
+       }
+       n = count;
+       for (q = p; q != ctx->trees; q = q->next, n = 31) {
+               while (n--) {
+                       audit_put_chunk(q->c[n]);
+                       q->c[n] = NULL;
+               }
+       }
+       while (n-- > ctx->tree_count) {
+               audit_put_chunk(q->c[n]);
+               q->c[n] = NULL;
+       }
+       ctx->trees = p;
+       ctx->tree_count = count;
+#endif
+}
+
+static void free_tree_refs(struct audit_context *ctx)
+{
+       struct audit_tree_refs *p, *q;
+       for (p = ctx->first_trees; p; p = q) {
+               q = p->next;
+               kfree(p);
+       }
+}
+
+static int match_tree_refs(struct audit_context *ctx, struct audit_tree *tree)
+{
+#ifdef CONFIG_AUDIT_TREE
+       struct audit_tree_refs *p;
+       int n;
+       if (!tree)
+               return 0;
+       /* full ones */
+       for (p = ctx->first_trees; p != ctx->trees; p = p->next) {
+               for (n = 0; n < 31; n++)
+                       if (audit_tree_match(p->c[n], tree))
+                               return 1;
+       }
+       /* partial */
+       if (p) {
+               for (n = ctx->tree_count; n < 31; n++)
+                       if (audit_tree_match(p->c[n], tree))
+                               return 1;
+       }
+#endif
+       return 0;
+}
+
+/* Determine if any context name data matches a rule's watch data */
 /* Compare a task_struct with an audit_rule.  Return 1 on match, 0
  * otherwise. */
 static int audit_filter_rules(struct task_struct *tsk,
                              struct audit_krule *rule,
                              struct audit_context *ctx,
+                             struct audit_names *name,
                              enum audit_state *state)
 {
-       int i, j;
+       const struct cred *cred = get_task_cred(tsk);
+       int i, j, need_sid = 1;
+       u32 sid;
 
        for (i = 0; i < rule->field_count; i++) {
                struct audit_field *f = &rule->fields[i];
@@ -176,35 +458,42 @@ static int audit_filter_rules(struct task_struct *tsk,
                case AUDIT_PID:
                        result = audit_comparator(tsk->pid, f->op, f->val);
                        break;
+               case AUDIT_PPID:
+                       if (ctx) {
+                               if (!ctx->ppid)
+                                       ctx->ppid = sys_getppid();
+                               result = audit_comparator(ctx->ppid, f->op, f->val);
+                       }
+                       break;
                case AUDIT_UID:
-                       result = audit_comparator(tsk->uid, f->op, f->val);
+                       result = audit_comparator(cred->uid, f->op, f->val);
                        break;
                case AUDIT_EUID:
-                       result = audit_comparator(tsk->euid, f->op, f->val);
+                       result = audit_comparator(cred->euid, f->op, f->val);
                        break;
                case AUDIT_SUID:
-                       result = audit_comparator(tsk->suid, f->op, f->val);
+                       result = audit_comparator(cred->suid, f->op, f->val);
                        break;
                case AUDIT_FSUID:
-                       result = audit_comparator(tsk->fsuid, f->op, f->val);
+                       result = audit_comparator(cred->fsuid, f->op, f->val);
                        break;
                case AUDIT_GID:
-                       result = audit_comparator(tsk->gid, f->op, f->val);
+                       result = audit_comparator(cred->gid, f->op, f->val);
                        break;
                case AUDIT_EGID:
-                       result = audit_comparator(tsk->egid, f->op, f->val);
+                       result = audit_comparator(cred->egid, f->op, f->val);
                        break;
                case AUDIT_SGID:
-                       result = audit_comparator(tsk->sgid, f->op, f->val);
+                       result = audit_comparator(cred->sgid, f->op, f->val);
                        break;
                case AUDIT_FSGID:
-                       result = audit_comparator(tsk->fsgid, f->op, f->val);
+                       result = audit_comparator(cred->fsgid, f->op, f->val);
                        break;
                case AUDIT_PERS:
                        result = audit_comparator(tsk->personality, f->op, f->val);
                        break;
                case AUDIT_ARCH:
-                       if (ctx)
+                       if (ctx)
                                result = audit_comparator(ctx->arch, f->op, f->val);
                        break;
 
@@ -221,7 +510,10 @@ static int audit_filter_rules(struct task_struct *tsk,
                        }
                        break;
                case AUDIT_DEVMAJOR:
-                       if (ctx) {
+                       if (name)
+                               result = audit_comparator(MAJOR(name->dev),
+                                                         f->op, f->val);
+                       else if (ctx) {
                                for (j = 0; j < ctx->name_count; j++) {
                                        if (audit_comparator(MAJOR(ctx->names[j].dev),  f->op, f->val)) {
                                                ++result;
@@ -231,7 +523,10 @@ static int audit_filter_rules(struct task_struct *tsk,
                        }
                        break;
                case AUDIT_DEVMINOR:
-                       if (ctx) {
+                       if (name)
+                               result = audit_comparator(MINOR(name->dev),
+                                                         f->op, f->val);
+                       else if (ctx) {
                                for (j = 0; j < ctx->name_count; j++) {
                                        if (audit_comparator(MINOR(ctx->names[j].dev), f->op, f->val)) {
                                                ++result;
@@ -241,20 +536,84 @@ static int audit_filter_rules(struct task_struct *tsk,
                        }
                        break;
                case AUDIT_INODE:
-                       if (ctx) {
+                       if (name)
+                               result = (name->ino == f->val);
+                       else if (ctx) {
                                for (j = 0; j < ctx->name_count; j++) {
-                                       if (audit_comparator(ctx->names[j].ino, f->op, f->val) ||
-                                           audit_comparator(ctx->names[j].pino, f->op, f->val)) {
+                                       if (audit_comparator(ctx->names[j].ino, f->op, f->val)) {
                                                ++result;
                                                break;
                                        }
                                }
                        }
                        break;
+               case AUDIT_WATCH:
+                       if (name && audit_watch_inode(rule->watch) != (unsigned long)-1)
+                               result = (name->dev == audit_watch_dev(rule->watch) &&
+                                         name->ino == audit_watch_inode(rule->watch));
+                       break;
+               case AUDIT_DIR:
+                       if (ctx)
+                               result = match_tree_refs(ctx, rule->tree);
+                       break;
                case AUDIT_LOGINUID:
                        result = 0;
                        if (ctx)
-                               result = audit_comparator(ctx->loginuid, f->op, f->val);
+                               result = audit_comparator(tsk->loginuid, f->op, f->val);
+                       break;
+               case AUDIT_SUBJ_USER:
+               case AUDIT_SUBJ_ROLE:
+               case AUDIT_SUBJ_TYPE:
+               case AUDIT_SUBJ_SEN:
+               case AUDIT_SUBJ_CLR:
+                       /* NOTE: this may return negative values indicating
+                          a temporary error.  We simply treat this as a
+                          match for now to avoid losing information that
+                          may be wanted.   An error message will also be
+                          logged upon error */
+                       if (f->lsm_rule) {
+                               if (need_sid) {
+                                       security_task_getsecid(tsk, &sid);
+                                       need_sid = 0;
+                               }
+                               result = security_audit_rule_match(sid, f->type,
+                                                                 f->op,
+                                                                 f->lsm_rule,
+                                                                 ctx);
+                       }
+                       break;
+               case AUDIT_OBJ_USER:
+               case AUDIT_OBJ_ROLE:
+               case AUDIT_OBJ_TYPE:
+               case AUDIT_OBJ_LEV_LOW:
+               case AUDIT_OBJ_LEV_HIGH:
+                       /* The above note for AUDIT_SUBJ_USER...AUDIT_SUBJ_CLR
+                          also applies here */
+                       if (f->lsm_rule) {
+                               /* Find files that match */
+                               if (name) {
+                                       result = security_audit_rule_match(
+                                                  name->osid, f->type, f->op,
+                                                  f->lsm_rule, ctx);
+                               } else if (ctx) {
+                                       for (j = 0; j < ctx->name_count; j++) {
+                                               if (security_audit_rule_match(
+                                                     ctx->names[j].osid,
+                                                     f->type, f->op,
+                                                     f->lsm_rule, ctx)) {
+                                                       ++result;
+                                                       break;
+                                               }
+                                       }
+                               }
+                               /* Find ipc objects that match */
+                               if (!ctx || ctx->type != AUDIT_IPC)
+                                       break;
+                               if (security_audit_rule_match(ctx->ipc.osid,
+                                                             f->type, f->op,
+                                                             f->lsm_rule, ctx))
+                                       ++result;
+                       }
                        break;
                case AUDIT_ARG0:
                case AUDIT_ARG1:
@@ -263,16 +622,38 @@ static int audit_filter_rules(struct task_struct *tsk,
                        if (ctx)
                                result = audit_comparator(ctx->argv[f->type-AUDIT_ARG0], f->op, f->val);
                        break;
+               case AUDIT_FILTERKEY:
+                       /* ignore this field for filtering */
+                       result = 1;
+                       break;
+               case AUDIT_PERM:
+                       result = audit_match_perm(ctx, f->val);
+                       break;
+               case AUDIT_FILETYPE:
+                       result = audit_match_filetype(ctx, f->val);
+                       break;
                }
 
-               if (!result)
+               if (!result) {
+                       put_cred(cred);
                        return 0;
+               }
+       }
+
+       if (ctx) {
+               if (rule->prio <= ctx->prio)
+                       return 0;
+               if (rule->filterkey) {
+                       kfree(ctx->filterkey);
+                       ctx->filterkey = kstrdup(rule->filterkey, GFP_ATOMIC);
+               }
+               ctx->prio = rule->prio;
        }
        switch (rule->action) {
        case AUDIT_NEVER:    *state = AUDIT_DISABLED;       break;
-       case AUDIT_POSSIBLE: *state = AUDIT_BUILD_CONTEXT;  break;
        case AUDIT_ALWAYS:   *state = AUDIT_RECORD_CONTEXT; break;
        }
+       put_cred(cred);
        return 1;
 }
 
@@ -280,14 +661,16 @@ static int audit_filter_rules(struct task_struct *tsk,
  * completely disabled for this task.  Since we only have the task
  * structure at this point, we can only check uid and gid.
  */
-static enum audit_state audit_filter_task(struct task_struct *tsk)
+static enum audit_state audit_filter_task(struct task_struct *tsk, char **key)
 {
        struct audit_entry *e;
        enum audit_state   state;
 
        rcu_read_lock();
        list_for_each_entry_rcu(e, &audit_filter_list[AUDIT_FILTER_TASK], list) {
-               if (audit_filter_rules(tsk, &e->rule, NULL, &state)) {
+               if (audit_filter_rules(tsk, &e->rule, NULL, NULL, &state)) {
+                       if (state == AUDIT_RECORD_CONTEXT)
+                               *key = kstrdup(e->rule.filterkey, GFP_ATOMIC);
                        rcu_read_unlock();
                        return state;
                }
@@ -317,9 +700,11 @@ static enum audit_state audit_filter_syscall(struct task_struct *tsk,
                int bit  = AUDIT_BIT(ctx->major);
 
                list_for_each_entry_rcu(e, list, list) {
-                       if ((e->rule.mask[word] & bit) == bit
-                                       && audit_filter_rules(tsk, &e->rule, ctx, &state)) {
+                       if ((e->rule.mask[word] & bit) == bit &&
+                           audit_filter_rules(tsk, &e->rule, ctx, NULL,
+                                              &state)) {
                                rcu_read_unlock();
+                               ctx->current_state = state;
                                return state;
                        }
                }
@@ -328,35 +713,76 @@ static enum audit_state audit_filter_syscall(struct task_struct *tsk,
        return AUDIT_BUILD_CONTEXT;
 }
 
-/* This should be called with task_lock() held. */
+/* At syscall exit time, this filter is called if any audit_names[] have been
+ * collected during syscall processing.  We only check rules in sublists at hash
+ * buckets applicable to the inode numbers in audit_names[].
+ * Regarding audit_state, same rules apply as for audit_filter_syscall().
+ */
+void audit_filter_inodes(struct task_struct *tsk, struct audit_context *ctx)
+{
+       int i;
+       struct audit_entry *e;
+       enum audit_state state;
+
+       if (audit_pid && tsk->tgid == audit_pid)
+               return;
+
+       rcu_read_lock();
+       for (i = 0; i < ctx->name_count; i++) {
+               int word = AUDIT_WORD(ctx->major);
+               int bit  = AUDIT_BIT(ctx->major);
+               struct audit_names *n = &ctx->names[i];
+               int h = audit_hash_ino((u32)n->ino);
+               struct list_head *list = &audit_inode_hash[h];
+
+               if (list_empty(list))
+                       continue;
+
+               list_for_each_entry_rcu(e, list, list) {
+                       if ((e->rule.mask[word] & bit) == bit &&
+                           audit_filter_rules(tsk, &e->rule, ctx, n, &state)) {
+                               rcu_read_unlock();
+                               ctx->current_state = state;
+                               return;
+                       }
+               }
+       }
+       rcu_read_unlock();
+}
+
 static inline struct audit_context *audit_get_context(struct task_struct *tsk,
                                                      int return_valid,
-                                                     int return_code)
+                                                     long return_code)
 {
        struct audit_context *context = tsk->audit_context;
 
        if (likely(!context))
                return NULL;
        context->return_valid = return_valid;
-       context->return_code  = return_code;
 
-       if (context->in_syscall && !context->auditable) {
-               enum audit_state state;
-               state = audit_filter_syscall(tsk, context, &audit_filter_list[AUDIT_FILTER_EXIT]);
-               if (state == AUDIT_RECORD_CONTEXT)
-                       context->auditable = 1;
+       /*
+        * we need to fix up the return code in the audit logs if the actual
+        * return codes are later going to be fixed up by the arch specific
+        * signal handlers
+        *
+        * This is actually a test for:
+        * (rc == ERESTARTSYS ) || (rc == ERESTARTNOINTR) ||
+        * (rc == ERESTARTNOHAND) || (rc == ERESTART_RESTARTBLOCK)
+        *
+        * but is faster than a bunch of ||
+        */
+       if (unlikely(return_code <= -ERESTARTSYS) &&
+           (return_code >= -ERESTART_RESTARTBLOCK) &&
+           (return_code != -ENOIOCTLCMD))
+               context->return_code = -EINTR;
+       else
+               context->return_code  = return_code;
+
+       if (context->in_syscall && !context->dummy) {
+               audit_filter_syscall(tsk, context, &audit_filter_list[AUDIT_FILTER_EXIT]);
+               audit_filter_inodes(tsk, context);
        }
 
-       context->pid = tsk->pid;
-       context->uid = tsk->uid;
-       context->gid = tsk->gid;
-       context->euid = tsk->euid;
-       context->suid = tsk->suid;
-       context->fsuid = tsk->fsuid;
-       context->egid = tsk->egid;
-       context->sgid = tsk->sgid;
-       context->fsgid = tsk->fsgid;
-       context->personality = tsk->personality;
        tsk->audit_context = NULL;
        return context;
 }
@@ -366,8 +792,7 @@ static inline void audit_free_names(struct audit_context *context)
        int i;
 
 #if AUDIT_DEBUG == 2
-       if (context->auditable
-           ||context->put_count + context->ino_count != context->name_count) {
+       if (context->put_count + context->ino_count != context->name_count) {
                printk(KERN_ERR "%s:%d(:%d): major=%d in_syscall=%d"
                       " name_count=%d put_count=%d"
                       " ino_count=%d [NOT freeing]\n",
@@ -390,19 +815,13 @@ static inline void audit_free_names(struct audit_context *context)
 #endif
 
        for (i = 0; i < context->name_count; i++) {
-               char *p = context->names[i].ctx;
-               context->names[i].ctx = NULL;
-               kfree(p);
-               if (context->names[i].name)
+               if (context->names[i].name && context->names[i].name_put)
                        __putname(context->names[i].name);
        }
        context->name_count = 0;
-       if (context->pwd)
-               dput(context->pwd);
-       if (context->pwdmnt)
-               mntput(context->pwdmnt);
-       context->pwd = NULL;
-       context->pwdmnt = NULL;
+       path_put(&context->pwd);
+       context->pwd.dentry = NULL;
+       context->pwd.mnt = NULL;
 }
 
 static inline void audit_free_aux(struct audit_context *context)
@@ -410,30 +829,21 @@ static inline void audit_free_aux(struct audit_context *context)
        struct audit_aux_data *aux;
 
        while ((aux = context->aux)) {
-               if (aux->type == AUDIT_AVC_PATH) {
-                       struct audit_aux_data_path *axi = (void *)aux;
-                       dput(axi->dentry);
-                       mntput(axi->mnt);
-               }
-               if ( aux->type == AUDIT_IPC ) {
-                       struct audit_aux_data_ipcctl *axi = (void *)aux;
-                       if (axi->ctx)
-                               kfree(axi->ctx);
-               }
-
                context->aux = aux->next;
                kfree(aux);
        }
+       while ((aux = context->aux_pids)) {
+               context->aux_pids = aux->next;
+               kfree(aux);
+       }
 }
 
 static inline void audit_zero_context(struct audit_context *context,
                                      enum audit_state state)
 {
-       uid_t loginuid = context->loginuid;
-
        memset(context, 0, sizeof(*context));
        context->state      = state;
-       context->loginuid   = loginuid;
+       context->prio = state == AUDIT_RECORD_CONTEXT ? ~0ULL : 0;
 }
 
 static inline struct audit_context *audit_alloc_context(enum audit_state state)
@@ -443,6 +853,7 @@ static inline struct audit_context *audit_alloc_context(enum audit_state state)
        if (!(context = kmalloc(sizeof(*context), GFP_KERNEL)))
                return NULL;
        audit_zero_context(context, state);
+       INIT_LIST_HEAD(&context->killed_trees);
        return context;
 }
 
@@ -459,23 +870,21 @@ int audit_alloc(struct task_struct *tsk)
 {
        struct audit_context *context;
        enum audit_state     state;
+       char *key = NULL;
 
-       if (likely(!audit_enabled))
+       if (likely(!audit_ever_enabled))
                return 0; /* Return if not auditing. */
 
-       state = audit_filter_task(tsk);
+       state = audit_filter_task(tsk, &key);
        if (likely(state == AUDIT_DISABLED))
                return 0;
 
        if (!(context = audit_alloc_context(state))) {
+               kfree(key);
                audit_log_lost("out of memory in audit_alloc");
                return -ENOMEM;
        }
-
-                               /* Preserve login uid */
-       context->loginuid = -1;
-       if (current->audit_context)
-               context->loginuid = current->audit_context->loginuid;
+       context->filterkey = key;
 
        tsk->audit_context  = context;
        set_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
@@ -497,7 +906,11 @@ static inline void audit_free_context(struct audit_context *context)
                               context->name_count, count);
                }
                audit_free_names(context);
+               unroll_tree_refs(context, NULL, 0);
+               free_tree_refs(context);
                audit_free_aux(context);
+               kfree(context->filterkey);
+               kfree(context->sockaddr);
                kfree(context);
                context  = previous;
        } while (context);
@@ -505,76 +918,422 @@ static inline void audit_free_context(struct audit_context *context)
                printk(KERN_ERR "audit: freed %d contexts\n", count);
 }
 
-static void audit_log_task_context(struct audit_buffer *ab, gfp_t gfp_mask)
+void audit_log_task_context(struct audit_buffer *ab)
 {
        char *ctx = NULL;
-       ssize_t len = 0;
+       unsigned len;
+       int error;
+       u32 sid;
+
+       security_task_getsecid(current, &sid);
+       if (!sid)
+               return;
 
-       len = security_getprocattr(current, "current", NULL, 0);
-       if (len < 0) {
-               if (len != -EINVAL)
+       error = security_secid_to_secctx(sid, &ctx, &len);
+       if (error) {
+               if (error != -EINVAL)
                        goto error_path;
                return;
        }
 
-       ctx = kmalloc(len, gfp_mask);
-       if (!ctx)
-               goto error_path;
-
-       len = security_getprocattr(current, "current", ctx, len);
-       if (len < 0 )
-               goto error_path;
-
        audit_log_format(ab, " subj=%s", ctx);
+       security_release_secctx(ctx, len);
        return;
 
 error_path:
-       if (ctx)
-               kfree(ctx);
        audit_panic("error in audit_log_task_context");
        return;
 }
 
-static void audit_log_task_info(struct audit_buffer *ab, gfp_t gfp_mask)
+EXPORT_SYMBOL(audit_log_task_context);
+
+static void audit_log_task_info(struct audit_buffer *ab, struct task_struct *tsk)
 {
-       char name[sizeof(current->comm)];
-       struct mm_struct *mm = current->mm;
+       char name[sizeof(tsk->comm)];
+       struct mm_struct *mm = tsk->mm;
        struct vm_area_struct *vma;
 
-       get_task_comm(name, current);
+       /* tsk == current */
+
+       get_task_comm(name, tsk);
        audit_log_format(ab, " comm=");
        audit_log_untrustedstring(ab, name);
 
-       if (!mm)
-               return;
+       if (mm) {
+               down_read(&mm->mmap_sem);
+               vma = mm->mmap;
+               while (vma) {
+                       if ((vma->vm_flags & VM_EXECUTABLE) &&
+                           vma->vm_file) {
+                               audit_log_d_path(ab, "exe=",
+                                                &vma->vm_file->f_path);
+                               break;
+                       }
+                       vma = vma->vm_next;
+               }
+               up_read(&mm->mmap_sem);
+       }
+       audit_log_task_context(ab);
+}
+
+static int audit_log_pid_context(struct audit_context *context, pid_t pid,
+                                uid_t auid, uid_t uid, unsigned int sessionid,
+                                u32 sid, char *comm)
+{
+       struct audit_buffer *ab;
+       char *ctx = NULL;
+       u32 len;
+       int rc = 0;
+
+       ab = audit_log_start(context, GFP_KERNEL, AUDIT_OBJ_PID);
+       if (!ab)
+               return rc;
+
+       audit_log_format(ab, "opid=%d oauid=%d ouid=%d oses=%d", pid, auid,
+                        uid, sessionid);
+       if (security_secid_to_secctx(sid, &ctx, &len)) {
+               audit_log_format(ab, " obj=(none)");
+               rc = 1;
+       } else {
+               audit_log_format(ab, " obj=%s", ctx);
+               security_release_secctx(ctx, len);
+       }
+       audit_log_format(ab, " ocomm=");
+       audit_log_untrustedstring(ab, comm);
+       audit_log_end(ab);
+
+       return rc;
+}
+
+/*
+ * to_send and len_sent accounting are very loose estimates.  We aren't
+ * really worried about a hard cap to MAX_EXECVE_AUDIT_LEN so much as being
+ * within about 500 bytes (next page boundry)
+ *
+ * why snprintf?  an int is up to 12 digits long.  if we just assumed when
+ * logging that a[%d]= was going to be 16 characters long we would be wasting
+ * space in every audit message.  In one 7500 byte message we can log up to
+ * about 1000 min size arguments.  That comes down to about 50% waste of space
+ * if we didn't do the snprintf to find out how long arg_num_len was.
+ */
+static int audit_log_single_execve_arg(struct audit_context *context,
+                                       struct audit_buffer **ab,
+                                       int arg_num,
+                                       size_t *len_sent,
+                                       const char __user *p,
+                                       char *buf)
+{
+       char arg_num_len_buf[12];
+       const char __user *tmp_p = p;
+       /* how many digits are in arg_num? 5 is the length of ' a=""' */
+       size_t arg_num_len = snprintf(arg_num_len_buf, 12, "%d", arg_num) + 5;
+       size_t len, len_left, to_send;
+       size_t max_execve_audit_len = MAX_EXECVE_AUDIT_LEN;
+       unsigned int i, has_cntl = 0, too_long = 0;
+       int ret;
+
+       /* strnlen_user includes the null we don't want to send */
+       len_left = len = strnlen_user(p, MAX_ARG_STRLEN) - 1;
 
        /*
-        * this is brittle; all callers that pass GFP_ATOMIC will have
-        * NULL current->mm and we won't get here.
+        * We just created this mm, if we can't find the strings
+        * we just copied into it something is _very_ wrong. Similar
+        * for strings that are too long, we should not have created
+        * any.
         */
-       down_read(&mm->mmap_sem);
-       vma = mm->mmap;
-       while (vma) {
-               if ((vma->vm_flags & VM_EXECUTABLE) &&
-                   vma->vm_file) {
-                       audit_log_d_path(ab, "exe=",
-                                        vma->vm_file->f_dentry,
-                                        vma->vm_file->f_vfsmnt);
+       if (unlikely((len == -1) || len > MAX_ARG_STRLEN - 1)) {
+               WARN_ON(1);
+               send_sig(SIGKILL, current, 0);
+               return -1;
+       }
+
+       /* walk the whole argument looking for non-ascii chars */
+       do {
+               if (len_left > MAX_EXECVE_AUDIT_LEN)
+                       to_send = MAX_EXECVE_AUDIT_LEN;
+               else
+                       to_send = len_left;
+               ret = copy_from_user(buf, tmp_p, to_send);
+               /*
+                * There is no reason for this copy to be short. We just
+                * copied them here, and the mm hasn't been exposed to user-
+                * space yet.
+                */
+               if (ret) {
+                       WARN_ON(1);
+                       send_sig(SIGKILL, current, 0);
+                       return -1;
+               }
+               buf[to_send] = '\0';
+               has_cntl = audit_string_contains_control(buf, to_send);
+               if (has_cntl) {
+                       /*
+                        * hex messages get logged as 2 bytes, so we can only
+                        * send half as much in each message
+                        */
+                       max_execve_audit_len = MAX_EXECVE_AUDIT_LEN / 2;
                        break;
                }
-               vma = vma->vm_next;
+               len_left -= to_send;
+               tmp_p += to_send;
+       } while (len_left > 0);
+
+       len_left = len;
+
+       if (len > max_execve_audit_len)
+               too_long = 1;
+
+       /* rewalk the argument actually logging the message */
+       for (i = 0; len_left > 0; i++) {
+               int room_left;
+
+               if (len_left > max_execve_audit_len)
+                       to_send = max_execve_audit_len;
+               else
+                       to_send = len_left;
+
+               /* do we have space left to send this argument in this ab? */
+               room_left = MAX_EXECVE_AUDIT_LEN - arg_num_len - *len_sent;
+               if (has_cntl)
+                       room_left -= (to_send * 2);
+               else
+                       room_left -= to_send;
+               if (room_left < 0) {
+                       *len_sent = 0;
+                       audit_log_end(*ab);
+                       *ab = audit_log_start(context, GFP_KERNEL, AUDIT_EXECVE);
+                       if (!*ab)
+                               return 0;
+               }
+
+               /*
+                * first record needs to say how long the original string was
+                * so we can be sure nothing was lost.
+                */
+               if ((i == 0) && (too_long))
+                       audit_log_format(*ab, " a%d_len=%zu", arg_num,
+                                        has_cntl ? 2*len : len);
+
+               /*
+                * normally arguments are small enough to fit and we already
+                * filled buf above when we checked for control characters
+                * so don't bother with another copy_from_user
+                */
+               if (len >= max_execve_audit_len)
+                       ret = copy_from_user(buf, p, to_send);
+               else
+                       ret = 0;
+               if (ret) {
+                       WARN_ON(1);
+                       send_sig(SIGKILL, current, 0);
+                       return -1;
+               }
+               buf[to_send] = '\0';
+
+               /* actually log it */
+               audit_log_format(*ab, " a%d", arg_num);
+               if (too_long)
+                       audit_log_format(*ab, "[%d]", i);
+               audit_log_format(*ab, "=");
+               if (has_cntl)
+                       audit_log_n_hex(*ab, buf, to_send);
+               else
+                       audit_log_string(*ab, buf);
+
+               p += to_send;
+               len_left -= to_send;
+               *len_sent += arg_num_len;
+               if (has_cntl)
+                       *len_sent += to_send * 2;
+               else
+                       *len_sent += to_send;
+       }
+       /* include the null we didn't log */
+       return len + 1;
+}
+
+static void audit_log_execve_info(struct audit_context *context,
+                                 struct audit_buffer **ab,
+                                 struct audit_aux_data_execve *axi)
+{
+       int i;
+       size_t len, len_sent = 0;
+       const char __user *p;
+       char *buf;
+
+       if (axi->mm != current->mm)
+               return; /* execve failed, no additional info */
+
+       p = (const char __user *)axi->mm->arg_start;
+
+       audit_log_format(*ab, "argc=%d", axi->argc);
+
+       /*
+        * we need some kernel buffer to hold the userspace args.  Just
+        * allocate one big one rather than allocating one of the right size
+        * for every single argument inside audit_log_single_execve_arg()
+        * should be <8k allocation so should be pretty safe.
+        */
+       buf = kmalloc(MAX_EXECVE_AUDIT_LEN + 1, GFP_KERNEL);
+       if (!buf) {
+               audit_panic("out of memory for argv string\n");
+               return;
+       }
+
+       for (i = 0; i < axi->argc; i++) {
+               len = audit_log_single_execve_arg(context, ab, i,
+                                                 &len_sent, p, buf);
+               if (len <= 0)
+                       break;
+               p += len;
+       }
+       kfree(buf);
+}
+
+static void audit_log_cap(struct audit_buffer *ab, char *prefix, kernel_cap_t *cap)
+{
+       int i;
+
+       audit_log_format(ab, " %s=", prefix);
+       CAP_FOR_EACH_U32(i) {
+               audit_log_format(ab, "%08x", cap->cap[(_KERNEL_CAPABILITY_U32S-1) - i]);
+       }
+}
+
+static void audit_log_fcaps(struct audit_buffer *ab, struct audit_names *name)
+{
+       kernel_cap_t *perm = &name->fcap.permitted;
+       kernel_cap_t *inh = &name->fcap.inheritable;
+       int log = 0;
+
+       if (!cap_isclear(*perm)) {
+               audit_log_cap(ab, "cap_fp", perm);
+               log = 1;
        }
-       up_read(&mm->mmap_sem);
-       audit_log_task_context(ab, gfp_mask);
+       if (!cap_isclear(*inh)) {
+               audit_log_cap(ab, "cap_fi", inh);
+               log = 1;
+       }
+
+       if (log)
+               audit_log_format(ab, " cap_fe=%d cap_fver=%x", name->fcap.fE, name->fcap_ver);
 }
 
-static void audit_log_exit(struct audit_context *context, gfp_t gfp_mask)
+static void show_special(struct audit_context *context, int *call_panic)
 {
+       struct audit_buffer *ab;
        int i;
+
+       ab = audit_log_start(context, GFP_KERNEL, context->type);
+       if (!ab)
+               return;
+
+       switch (context->type) {
+       case AUDIT_SOCKETCALL: {
+               int nargs = context->socketcall.nargs;
+               audit_log_format(ab, "nargs=%d", nargs);
+               for (i = 0; i < nargs; i++)
+                       audit_log_format(ab, " a%d=%lx", i,
+                               context->socketcall.args[i]);
+               break; }
+       case AUDIT_IPC: {
+               u32 osid = context->ipc.osid;
+
+               audit_log_format(ab, "ouid=%u ogid=%u mode=%#o",
+                        context->ipc.uid, context->ipc.gid, context->ipc.mode);
+               if (osid) {
+                       char *ctx = NULL;
+                       u32 len;
+                       if (security_secid_to_secctx(osid, &ctx, &len)) {
+                               audit_log_format(ab, " osid=%u", osid);
+                               *call_panic = 1;
+                       } else {
+                               audit_log_format(ab, " obj=%s", ctx);
+                               security_release_secctx(ctx, len);
+                       }
+               }
+               if (context->ipc.has_perm) {
+                       audit_log_end(ab);
+                       ab = audit_log_start(context, GFP_KERNEL,
+                                            AUDIT_IPC_SET_PERM);
+                       audit_log_format(ab,
+                               "qbytes=%lx ouid=%u ogid=%u mode=%#o",
+                               context->ipc.qbytes,
+                               context->ipc.perm_uid,
+                               context->ipc.perm_gid,
+                               context->ipc.perm_mode);
+                       if (!ab)
+                               return;
+               }
+               break; }
+       case AUDIT_MQ_OPEN: {
+               audit_log_format(ab,
+                       "oflag=0x%x mode=%#o mq_flags=0x%lx mq_maxmsg=%ld "
+                       "mq_msgsize=%ld mq_curmsgs=%ld",
+                       context->mq_open.oflag, context->mq_open.mode,
+                       context->mq_open.attr.mq_flags,
+                       context->mq_open.attr.mq_maxmsg,
+                       context->mq_open.attr.mq_msgsize,
+                       context->mq_open.attr.mq_curmsgs);
+               break; }
+       case AUDIT_MQ_SENDRECV: {
+               audit_log_format(ab,
+                       "mqdes=%d msg_len=%zd msg_prio=%u "
+                       "abs_timeout_sec=%ld abs_timeout_nsec=%ld",
+                       context->mq_sendrecv.mqdes,
+                       context->mq_sendrecv.msg_len,
+                       context->mq_sendrecv.msg_prio,
+                       context->mq_sendrecv.abs_timeout.tv_sec,
+                       context->mq_sendrecv.abs_timeout.tv_nsec);
+               break; }
+       case AUDIT_MQ_NOTIFY: {
+               audit_log_format(ab, "mqdes=%d sigev_signo=%d",
+                               context->mq_notify.mqdes,
+                               context->mq_notify.sigev_signo);
+               break; }
+       case AUDIT_MQ_GETSETATTR: {
+               struct mq_attr *attr = &context->mq_getsetattr.mqstat;
+               audit_log_format(ab,
+                       "mqdes=%d mq_flags=0x%lx mq_maxmsg=%ld mq_msgsize=%ld "
+                       "mq_curmsgs=%ld ",
+                       context->mq_getsetattr.mqdes,
+                       attr->mq_flags, attr->mq_maxmsg,
+                       attr->mq_msgsize, attr->mq_curmsgs);
+               break; }
+       case AUDIT_CAPSET: {
+               audit_log_format(ab, "pid=%d", context->capset.pid);
+               audit_log_cap(ab, "cap_pi", &context->capset.cap.inheritable);
+               audit_log_cap(ab, "cap_pp", &context->capset.cap.permitted);
+               audit_log_cap(ab, "cap_pe", &context->capset.cap.effective);
+               break; }
+       }
+       audit_log_end(ab);
+}
+
+static void audit_log_exit(struct audit_context *context, struct task_struct *tsk)
+{
+       const struct cred *cred;
+       int i, call_panic = 0;
        struct audit_buffer *ab;
        struct audit_aux_data *aux;
+       const char *tty;
+
+       /* tsk == current */
+       context->pid = tsk->pid;
+       if (!context->ppid)
+               context->ppid = sys_getppid();
+       cred = current_cred();
+       context->uid   = cred->uid;
+       context->gid   = cred->gid;
+       context->euid  = cred->euid;
+       context->suid  = cred->suid;
+       context->fsuid = cred->fsuid;
+       context->egid  = cred->egid;
+       context->sgid  = cred->sgid;
+       context->fsgid = cred->fsgid;
+       context->personality = tsk->personality;
 
-       ab = audit_log_start(context, gfp_mask, AUDIT_SYSCALL);
+       ab = audit_log_start(context, GFP_KERNEL, AUDIT_SYSCALL);
        if (!ab)
                return;         /* audit_panic has been called */
        audit_log_format(ab, "arch=%x syscall=%d",
@@ -582,142 +1341,219 @@ static void audit_log_exit(struct audit_context *context, gfp_t gfp_mask)
        if (context->personality != PER_LINUX)
                audit_log_format(ab, " per=%lx", context->personality);
        if (context->return_valid)
-               audit_log_format(ab, " success=%s exit=%ld", 
+               audit_log_format(ab, " success=%s exit=%ld",
                                 (context->return_valid==AUDITSC_SUCCESS)?"yes":"no",
                                 context->return_code);
+
+       spin_lock_irq(&tsk->sighand->siglock);
+       if (tsk->signal && tsk->signal->tty && tsk->signal->tty->name)
+               tty = tsk->signal->tty->name;
+       else
+               tty = "(none)";
+       spin_unlock_irq(&tsk->sighand->siglock);
+
        audit_log_format(ab,
                  " a0=%lx a1=%lx a2=%lx a3=%lx items=%d"
-                 " pid=%d auid=%u uid=%u gid=%u"
+                 " ppid=%d pid=%d auid=%u uid=%u gid=%u"
                  " euid=%u suid=%u fsuid=%u"
-                 " egid=%u sgid=%u fsgid=%u",
+                 " egid=%u sgid=%u fsgid=%u tty=%s ses=%u",
                  context->argv[0],
                  context->argv[1],
                  context->argv[2],
                  context->argv[3],
                  context->name_count,
+                 context->ppid,
                  context->pid,
-                 context->loginuid,
+                 tsk->loginuid,
                  context->uid,
                  context->gid,
                  context->euid, context->suid, context->fsuid,
-                 context->egid, context->sgid, context->fsgid);
-       audit_log_task_info(ab, gfp_mask);
+                 context->egid, context->sgid, context->fsgid, tty,
+                 tsk->sessionid);
+
+
+       audit_log_task_info(ab, tsk);
+       audit_log_key(ab, context->filterkey);
        audit_log_end(ab);
 
        for (aux = context->aux; aux; aux = aux->next) {
 
-               ab = audit_log_start(context, gfp_mask, aux->type);
+               ab = audit_log_start(context, GFP_KERNEL, aux->type);
                if (!ab)
                        continue; /* audit_panic has been called */
 
                switch (aux->type) {
-               case AUDIT_IPC: {
-                       struct audit_aux_data_ipcctl *axi = (void *)aux;
-                       audit_log_format(ab, 
-                                        " qbytes=%lx iuid=%u igid=%u mode=%x obj=%s",
-                                        axi->qbytes, axi->uid, axi->gid, axi->mode, axi->ctx);
+
+               case AUDIT_EXECVE: {
+                       struct audit_aux_data_execve *axi = (void *)aux;
+                       audit_log_execve_info(context, &ab, axi);
                        break; }
 
-               case AUDIT_SOCKETCALL: {
-                       int i;
-                       struct audit_aux_data_socketcall *axs = (void *)aux;
-                       audit_log_format(ab, "nargs=%d", axs->nargs);
-                       for (i=0; i<axs->nargs; i++)
-                               audit_log_format(ab, " a%d=%lx", i, axs->args[i]);
+               case AUDIT_BPRM_FCAPS: {
+                       struct audit_aux_data_bprm_fcaps *axs = (void *)aux;
+                       audit_log_format(ab, "fver=%x", axs->fcap_ver);
+                       audit_log_cap(ab, "fp", &axs->fcap.permitted);
+                       audit_log_cap(ab, "fi", &axs->fcap.inheritable);
+                       audit_log_format(ab, " fe=%d", axs->fcap.fE);
+                       audit_log_cap(ab, "old_pp", &axs->old_pcap.permitted);
+                       audit_log_cap(ab, "old_pi", &axs->old_pcap.inheritable);
+                       audit_log_cap(ab, "old_pe", &axs->old_pcap.effective);
+                       audit_log_cap(ab, "new_pp", &axs->new_pcap.permitted);
+                       audit_log_cap(ab, "new_pi", &axs->new_pcap.inheritable);
+                       audit_log_cap(ab, "new_pe", &axs->new_pcap.effective);
                        break; }
 
-               case AUDIT_SOCKADDR: {
-                       struct audit_aux_data_sockaddr *axs = (void *)aux;
+               }
+               audit_log_end(ab);
+       }
 
-                       audit_log_format(ab, "saddr=");
-                       audit_log_hex(ab, axs->a, axs->len);
-                       break; }
+       if (context->type)
+               show_special(context, &call_panic);
 
-               case AUDIT_AVC_PATH: {
-                       struct audit_aux_data_path *axi = (void *)aux;
-                       audit_log_d_path(ab, "path=", axi->dentry, axi->mnt);
-                       break; }
+       if (context->fds[0] >= 0) {
+               ab = audit_log_start(context, GFP_KERNEL, AUDIT_FD_PAIR);
+               if (ab) {
+                       audit_log_format(ab, "fd0=%d fd1=%d",
+                                       context->fds[0], context->fds[1]);
+                       audit_log_end(ab);
+               }
+       }
 
+       if (context->sockaddr_len) {
+               ab = audit_log_start(context, GFP_KERNEL, AUDIT_SOCKADDR);
+               if (ab) {
+                       audit_log_format(ab, "saddr=");
+                       audit_log_n_hex(ab, (void *)context->sockaddr,
+                                       context->sockaddr_len);
+                       audit_log_end(ab);
                }
-               audit_log_end(ab);
        }
 
-       if (context->pwd && context->pwdmnt) {
-               ab = audit_log_start(context, gfp_mask, AUDIT_CWD);
+       for (aux = context->aux_pids; aux; aux = aux->next) {
+               struct audit_aux_data_pids *axs = (void *)aux;
+
+               for (i = 0; i < axs->pid_count; i++)
+                       if (audit_log_pid_context(context, axs->target_pid[i],
+                                                 axs->target_auid[i],
+                                                 axs->target_uid[i],
+                                                 axs->target_sessionid[i],
+                                                 axs->target_sid[i],
+                                                 axs->target_comm[i]))
+                               call_panic = 1;
+       }
+
+       if (context->target_pid &&
+           audit_log_pid_context(context, context->target_pid,
+                                 context->target_auid, context->target_uid,
+                                 context->target_sessionid,
+                                 context->target_sid, context->target_comm))
+                       call_panic = 1;
+
+       if (context->pwd.dentry && context->pwd.mnt) {
+               ab = audit_log_start(context, GFP_KERNEL, AUDIT_CWD);
                if (ab) {
-                       audit_log_d_path(ab, "cwd=", context->pwd, context->pwdmnt);
+                       audit_log_d_path(ab, "cwd=", &context->pwd);
                        audit_log_end(ab);
                }
        }
        for (i = 0; i < context->name_count; i++) {
-               unsigned long ino  = context->names[i].ino;
-               unsigned long pino = context->names[i].pino;
+               struct audit_names *n = &context->names[i];
 
-               ab = audit_log_start(context, gfp_mask, AUDIT_PATH);
+               ab = audit_log_start(context, GFP_KERNEL, AUDIT_PATH);
                if (!ab)
                        continue; /* audit_panic has been called */
 
                audit_log_format(ab, "item=%d", i);
 
-               audit_log_format(ab, " name=");
-               if (context->names[i].name)
-                       audit_log_untrustedstring(ab, context->names[i].name);
-               else
-                       audit_log_format(ab, "(null)");
-
-               if (pino != (unsigned long)-1)
-                       audit_log_format(ab, " parent=%lu",  pino);
-               if (ino != (unsigned long)-1)
-                       audit_log_format(ab, " inode=%lu",  ino);
-               if ((pino != (unsigned long)-1) || (ino != (unsigned long)-1))
-                       audit_log_format(ab, " dev=%02x:%02x mode=%#o" 
-                                        " ouid=%u ogid=%u rdev=%02x:%02x", 
-                                        MAJOR(context->names[i].dev), 
-                                        MINOR(context->names[i].dev), 
-                                        context->names[i].mode, 
-                                        context->names[i].uid, 
-                                        context->names[i].gid, 
-                                        MAJOR(context->names[i].rdev), 
-                                        MINOR(context->names[i].rdev));
-               if (context->names[i].ctx) {
-                       audit_log_format(ab, " obj=%s",
-                                       context->names[i].ctx);
+               if (n->name) {
+                       switch(n->name_len) {
+                       case AUDIT_NAME_FULL:
+                               /* log the full path */
+                               audit_log_format(ab, " name=");
+                               audit_log_untrustedstring(ab, n->name);
+                               break;
+                       case 0:
+                               /* name was specified as a relative path and the
+                                * directory component is the cwd */
+                               audit_log_d_path(ab, "name=", &context->pwd);
+                               break;
+                       default:
+                               /* log the name's directory component */
+                               audit_log_format(ab, " name=");
+                               audit_log_n_untrustedstring(ab, n->name,
+                                                           n->name_len);
+                       }
+               } else
+                       audit_log_format(ab, " name=(null)");
+
+               if (n->ino != (unsigned long)-1) {
+                       audit_log_format(ab, " inode=%lu"
+                                        " dev=%02x:%02x mode=%#o"
+                                        " ouid=%u ogid=%u rdev=%02x:%02x",
+                                        n->ino,
+                                        MAJOR(n->dev),
+                                        MINOR(n->dev),
+                                        n->mode,
+                                        n->uid,
+                                        n->gid,
+                                        MAJOR(n->rdev),
+                                        MINOR(n->rdev));
+               }
+               if (n->osid != 0) {
+                       char *ctx = NULL;
+                       u32 len;
+                       if (security_secid_to_secctx(
+                               n->osid, &ctx, &len)) {
+                               audit_log_format(ab, " osid=%u", n->osid);
+                               call_panic = 2;
+                       } else {
+                               audit_log_format(ab, " obj=%s", ctx);
+                               security_release_secctx(ctx, len);
+                       }
                }
 
+               audit_log_fcaps(ab, n);
+
                audit_log_end(ab);
        }
+
+       /* Send end of event record to help user space know we are finished */
+       ab = audit_log_start(context, GFP_KERNEL, AUDIT_EOE);
+       if (ab)
+               audit_log_end(ab);
+       if (call_panic)
+               audit_panic("error converting sid to string");
 }
 
 /**
  * audit_free - free a per-task audit context
  * @tsk: task whose audit context block to free
  *
- * Called from copy_process and __put_task_struct.
+ * Called from copy_process and do_exit
  */
 void audit_free(struct task_struct *tsk)
 {
        struct audit_context *context;
 
-       task_lock(tsk);
        context = audit_get_context(tsk, 0, 0);
-       task_unlock(tsk);
-
        if (likely(!context))
                return;
 
        /* Check for system calls that do not go through the exit
-        * function (e.g., exit_group), then free context block. 
-        * We use GFP_ATOMIC here because we might be doing this 
+        * function (e.g., exit_group), then free context block.
+        * We use GFP_ATOMIC here because we might be doing this
         * in the context of the idle thread */
-       if (context->in_syscall && context->auditable)
-               audit_log_exit(context, GFP_ATOMIC);
+       /* that can happen only if we are called from do_exit() */
+       if (context->in_syscall && context->current_state == AUDIT_RECORD_CONTEXT)
+               audit_log_exit(context, tsk);
+       if (!list_empty(&context->killed_trees))
+               audit_kill_trees(&context->killed_trees);
 
        audit_free_context(context);
 }
 
 /**
  * audit_syscall_entry - fill in an audit record at syscall entry
- * @tsk: task being audited
  * @arch: architecture type
  * @major: major syscall type (function)
  * @a1: additional syscall register 1
@@ -733,14 +1569,16 @@ void audit_free(struct task_struct *tsk)
  * will only be written if another part of the kernel requests that it
  * be written).
  */
-void audit_syscall_entry(struct task_struct *tsk, int arch, int major,
+void audit_syscall_entry(int arch, int major,
                         unsigned long a1, unsigned long a2,
                         unsigned long a3, unsigned long a4)
 {
+       struct task_struct *tsk = current;
        struct audit_context *context = tsk->audit_context;
        enum audit_state     state;
 
-       BUG_ON(!context);
+       if (unlikely(!context))
+               return;
 
        /*
         * This happens only on certain architectures that make system
@@ -751,7 +1589,7 @@ void audit_syscall_entry(struct task_struct *tsk, int arch, int major,
         *
         * i386     no
         * x86_64   no
-        * ppc64    yes (see arch/ppc64/kernel/misc.S)
+        * ppc64    yes (see arch/powerpc/platforms/iseries/misc.S)
         *
         * This also happens with vm86 emulation in a non-nested manner
         * (entries without exits), so this case must be caught.
@@ -791,20 +1629,43 @@ void audit_syscall_entry(struct task_struct *tsk, int arch, int major,
        context->argv[3]    = a4;
 
        state = context->state;
-       if (state == AUDIT_SETUP_CONTEXT || state == AUDIT_BUILD_CONTEXT)
+       context->dummy = !audit_n_rules;
+       if (!context->dummy && state == AUDIT_BUILD_CONTEXT) {
+               context->prio = 0;
                state = audit_filter_syscall(tsk, context, &audit_filter_list[AUDIT_FILTER_ENTRY]);
+       }
        if (likely(state == AUDIT_DISABLED))
                return;
 
        context->serial     = 0;
        context->ctime      = CURRENT_TIME;
        context->in_syscall = 1;
-       context->auditable  = !!(state == AUDIT_RECORD_CONTEXT);
+       context->current_state  = state;
+       context->ppid       = 0;
+}
+
+void audit_finish_fork(struct task_struct *child)
+{
+       struct audit_context *ctx = current->audit_context;
+       struct audit_context *p = child->audit_context;
+       if (!p || !ctx)
+               return;
+       if (!ctx->in_syscall || ctx->current_state != AUDIT_RECORD_CONTEXT)
+               return;
+       p->arch = ctx->arch;
+       p->major = ctx->major;
+       memcpy(p->argv, ctx->argv, sizeof(ctx->argv));
+       p->ctime = ctx->ctime;
+       p->dummy = ctx->dummy;
+       p->in_syscall = ctx->in_syscall;
+       p->filterkey = kstrdup(ctx->filterkey, GFP_KERNEL);
+       p->ppid = current->pid;
+       p->prio = ctx->prio;
+       p->current_state = ctx->current_state;
 }
 
 /**
  * audit_syscall_exit - deallocate audit context after a system call
- * @tsk: task being audited
  * @valid: success/failure flag
  * @return_code: syscall return value
  *
@@ -814,25 +1675,24 @@ void audit_syscall_entry(struct task_struct *tsk, int arch, int major,
  * message), then write out the syscall information.  In call cases,
  * free the names stored from getname().
  */
-void audit_syscall_exit(struct task_struct *tsk, int valid, long return_code)
+void audit_syscall_exit(int valid, long return_code)
 {
+       struct task_struct *tsk = current;
        struct audit_context *context;
 
-       get_task_struct(tsk);
-       task_lock(tsk);
        context = audit_get_context(tsk, valid, return_code);
-       task_unlock(tsk);
 
-       /* Not having a context here is ok, since the parent may have
-        * called __put_task_struct. */
        if (likely(!context))
-               goto out;
+               return;
 
-       if (context->in_syscall && context->auditable)
-               audit_log_exit(context, GFP_KERNEL);
+       if (context->in_syscall && context->current_state == AUDIT_RECORD_CONTEXT)
+               audit_log_exit(context, tsk);
 
        context->in_syscall = 0;
-       context->auditable  = 0;
+       context->prio = context->state == AUDIT_RECORD_CONTEXT ? ~0ULL : 0;
+
+       if (!list_empty(&context->killed_trees))
+               audit_kill_trees(&context->killed_trees);
 
        if (context->previous) {
                struct audit_context *new_context = context->previous;
@@ -841,11 +1701,110 @@ void audit_syscall_exit(struct task_struct *tsk, int valid, long return_code)
                tsk->audit_context = new_context;
        } else {
                audit_free_names(context);
+               unroll_tree_refs(context, NULL, 0);
                audit_free_aux(context);
+               context->aux = NULL;
+               context->aux_pids = NULL;
+               context->target_pid = 0;
+               context->target_sid = 0;
+               context->sockaddr_len = 0;
+               context->type = 0;
+               context->fds[0] = -1;
+               if (context->state != AUDIT_RECORD_CONTEXT) {
+                       kfree(context->filterkey);
+                       context->filterkey = NULL;
+               }
                tsk->audit_context = context;
        }
- out:
-       put_task_struct(tsk);
+}
+
+static inline void handle_one(const struct inode *inode)
+{
+#ifdef CONFIG_AUDIT_TREE
+       struct audit_context *context;
+       struct audit_tree_refs *p;
+       struct audit_chunk *chunk;
+       int count;
+       if (likely(list_empty(&inode->inotify_watches)))
+               return;
+       context = current->audit_context;
+       p = context->trees;
+       count = context->tree_count;
+       rcu_read_lock();
+       chunk = audit_tree_lookup(inode);
+       rcu_read_unlock();
+       if (!chunk)
+               return;
+       if (likely(put_tree_ref(context, chunk)))
+               return;
+       if (unlikely(!grow_tree_refs(context))) {
+               printk(KERN_WARNING "out of memory, audit has lost a tree reference\n");
+               audit_set_auditable(context);
+               audit_put_chunk(chunk);
+               unroll_tree_refs(context, p, count);
+               return;
+       }
+       put_tree_ref(context, chunk);
+#endif
+}
+
+static void handle_path(const struct dentry *dentry)
+{
+#ifdef CONFIG_AUDIT_TREE
+       struct audit_context *context;
+       struct audit_tree_refs *p;
+       const struct dentry *d, *parent;
+       struct audit_chunk *drop;
+       unsigned long seq;
+       int count;
+
+       context = current->audit_context;
+       p = context->trees;
+       count = context->tree_count;
+retry:
+       drop = NULL;
+       d = dentry;
+       rcu_read_lock();
+       seq = read_seqbegin(&rename_lock);
+       for(;;) {
+               struct inode *inode = d->d_inode;
+               if (inode && unlikely(!list_empty(&inode->inotify_watches))) {
+                       struct audit_chunk *chunk;
+                       chunk = audit_tree_lookup(inode);
+                       if (chunk) {
+                               if (unlikely(!put_tree_ref(context, chunk))) {
+                                       drop = chunk;
+                                       break;
+                               }
+                       }
+               }
+               parent = d->d_parent;
+               if (parent == d)
+                       break;
+               d = parent;
+       }
+       if (unlikely(read_seqretry(&rename_lock, seq) || drop)) {  /* in this order */
+               rcu_read_unlock();
+               if (!drop) {
+                       /* just a race with rename */
+                       unroll_tree_refs(context, p, count);
+                       goto retry;
+               }
+               audit_put_chunk(drop);
+               if (grow_tree_refs(context)) {
+                       /* OK, got more space */
+                       unroll_tree_refs(context, p, count);
+                       goto retry;
+               }
+               /* too bad */
+               printk(KERN_WARNING
+                       "out of memory, audit has lost a tree reference\n");
+               unroll_tree_refs(context, p, count);
+               audit_set_auditable(context);
+               return;
+       }
+       rcu_read_unlock();
+#endif
 }
 
 /**
@@ -855,11 +1814,11 @@ void audit_syscall_exit(struct task_struct *tsk, int valid, long return_code)
  * Add a name to the list of audit names for this context.
  * Called from fs/namei.c:getname().
  */
-void audit_getname(const char *name)
+void __audit_getname(const char *name)
 {
        struct audit_context *context = current->audit_context;
 
-       if (!context || IS_ERR(name) || !name)
+       if (IS_ERR(name) || !name)
                return;
 
        if (!context->in_syscall) {
@@ -872,15 +1831,18 @@ void audit_getname(const char *name)
        }
        BUG_ON(context->name_count >= AUDIT_NAMES);
        context->names[context->name_count].name = name;
+       context->names[context->name_count].name_len = AUDIT_NAME_FULL;
+       context->names[context->name_count].name_put = 1;
        context->names[context->name_count].ino  = (unsigned long)-1;
+       context->names[context->name_count].osid = 0;
        ++context->name_count;
-       if (!context->pwd) {
+       if (!context->pwd.dentry) {
                read_lock(&current->fs->lock);
-               context->pwd = dget(current->fs->pwd);
-               context->pwdmnt = mntget(current->fs->pwdmnt);
+               context->pwd = current->fs->pwd;
+               path_get(&current->fs->pwd);
                read_unlock(&current->fs->lock);
        }
-               
+
 }
 
 /* audit_putname - intercept a putname request
@@ -926,55 +1888,81 @@ void audit_putname(const char *name)
 #endif
 }
 
-void audit_inode_context(int idx, const struct inode *inode)
+static int audit_inc_name_count(struct audit_context *context,
+                               const struct inode *inode)
 {
-       struct audit_context *context = current->audit_context;
-       const char *suffix = security_inode_xattr_getsuffix();
-       char *ctx = NULL;
-       int len = 0;
+       if (context->name_count >= AUDIT_NAMES) {
+               if (inode)
+                       printk(KERN_DEBUG "name_count maxed, losing inode data: "
+                              "dev=%02x:%02x, inode=%lu\n",
+                              MAJOR(inode->i_sb->s_dev),
+                              MINOR(inode->i_sb->s_dev),
+                              inode->i_ino);
 
-       if (!suffix)
-               goto ret;
+               else
+                       printk(KERN_DEBUG "name_count maxed, losing inode data\n");
+               return 1;
+       }
+       context->name_count++;
+#if AUDIT_DEBUG
+       context->ino_count++;
+#endif
+       return 0;
+}
 
-       len = security_inode_getsecurity(inode, suffix, NULL, 0, 0);
-       if (len == -EOPNOTSUPP)
-               goto ret;
-       if (len < 0) 
-               goto error_path;
 
-       ctx = kmalloc(len, GFP_KERNEL);
-       if (!ctx) 
-               goto error_path;
+static inline int audit_copy_fcaps(struct audit_names *name, const struct dentry *dentry)
+{
+       struct cpu_vfs_cap_data caps;
+       int rc;
 
-       len = security_inode_getsecurity(inode, suffix, ctx, len, 0);
-       if (len < 0)
-               goto error_path;
+       memset(&name->fcap.permitted, 0, sizeof(kernel_cap_t));
+       memset(&name->fcap.inheritable, 0, sizeof(kernel_cap_t));
+       name->fcap.fE = 0;
+       name->fcap_ver = 0;
 
-       kfree(context->names[idx].ctx);
-       context->names[idx].ctx = ctx;
-       goto ret;
+       if (!dentry)
+               return 0;
 
-error_path:
-       if (ctx)
-               kfree(ctx);
-       audit_panic("error in audit_inode_context");
-ret:
-       return;
+       rc = get_vfs_caps_from_disk(dentry, &caps);
+       if (rc)
+               return rc;
+
+       name->fcap.permitted = caps.permitted;
+       name->fcap.inheritable = caps.inheritable;
+       name->fcap.fE = !!(caps.magic_etc & VFS_CAP_FLAGS_EFFECTIVE);
+       name->fcap_ver = (caps.magic_etc & VFS_CAP_REVISION_MASK) >> VFS_CAP_REVISION_SHIFT;
+
+       return 0;
 }
 
 
+/* Copy inode data into an audit_names. */
+static void audit_copy_inode(struct audit_names *name, const struct dentry *dentry,
+                            const struct inode *inode)
+{
+       name->ino   = inode->i_ino;
+       name->dev   = inode->i_sb->s_dev;
+       name->mode  = inode->i_mode;
+       name->uid   = inode->i_uid;
+       name->gid   = inode->i_gid;
+       name->rdev  = inode->i_rdev;
+       security_inode_getsecid(inode, &name->osid);
+       audit_copy_fcaps(name, dentry);
+}
+
 /**
  * audit_inode - store the inode and device from a lookup
  * @name: name being audited
- * @inode: inode being audited
- * @flags: lookup flags (as used in path_lookup())
+ * @dentry: dentry being audited
  *
  * Called from fs/namei.c:path_lookup().
  */
-void __audit_inode(const char *name, const struct inode *inode, unsigned flags)
+void __audit_inode(const char *name, const struct dentry *dentry)
 {
        int idx;
        struct audit_context *context = current->audit_context;
+       const struct inode *inode = dentry->d_inode;
 
        if (!context->in_syscall)
                return;
@@ -989,35 +1977,20 @@ void __audit_inode(const char *name, const struct inode *inode, unsigned flags)
        else {
                /* FIXME: how much do we care about inodes that have no
                 * associated name? */
-               if (context->name_count >= AUDIT_NAMES - AUDIT_NAMES_RESERVED)
+               if (audit_inc_name_count(context, inode))
                        return;
-               idx = context->name_count++;
+               idx = context->name_count - 1;
                context->names[idx].name = NULL;
-#if AUDIT_DEBUG
-               ++context->ino_count;
-#endif
-       }
-       context->names[idx].dev   = inode->i_sb->s_dev;
-       context->names[idx].mode  = inode->i_mode;
-       context->names[idx].uid   = inode->i_uid;
-       context->names[idx].gid   = inode->i_gid;
-       context->names[idx].rdev  = inode->i_rdev;
-       audit_inode_context(idx, inode);
-       if ((flags & LOOKUP_PARENT) && (strcmp(name, "/") != 0) && 
-           (strcmp(name, ".") != 0)) {
-               context->names[idx].ino   = (unsigned long)-1;
-               context->names[idx].pino  = inode->i_ino;
-       } else {
-               context->names[idx].ino   = inode->i_ino;
-               context->names[idx].pino  = (unsigned long)-1;
        }
+       handle_path(dentry);
+       audit_copy_inode(&context->names[idx], dentry, inode);
 }
 
 /**
  * audit_inode_child - collect inode info for created/removed objects
  * @dname: inode's dentry name
- * @inode: inode being audited
- * @pino: inode number of dentry parent
+ * @dentry: dentry being audited
+ * @parent: inode of dentry parent
  *
  * For syscalls that create or remove filesystem objects, audit_inode
  * can only collect information for the filesystem object's parent.
@@ -1027,66 +2000,91 @@ void __audit_inode(const char *name, const struct inode *inode, unsigned flags)
  * must be hooked prior, in order to capture the target inode during
  * unsuccessful attempts.
  */
-void __audit_inode_child(const char *dname, const struct inode *inode,
-                        unsigned long pino)
+void __audit_inode_child(const char *dname, const struct dentry *dentry,
+                        const struct inode *parent)
 {
        int idx;
        struct audit_context *context = current->audit_context;
+       const char *found_parent = NULL, *found_child = NULL;
+       const struct inode *inode = dentry->d_inode;
+       int dirlen = 0;
 
        if (!context->in_syscall)
                return;
 
+       if (inode)
+               handle_one(inode);
        /* determine matching parent */
-       if (dname)
-               for (idx = 0; idx < context->name_count; idx++)
-                       if (context->names[idx].pino == pino) {
-                               const char *n;
-                               const char *name = context->names[idx].name;
-                               int dlen = strlen(dname);
-                               int nlen = name ? strlen(name) : 0;
-
-                               if (nlen < dlen)
-                                       continue;
-                               
-                               /* disregard trailing slashes */
-                               n = name + nlen - 1;
-                               while ((*n == '/') && (n > name))
-                                       n--;
-
-                               /* find last path component */
-                               n = n - dlen + 1;
-                               if (n < name)
-                                       continue;
-                               else if (n > name) {
-                                       if (*--n != '/')
-                                               continue;
-                                       else
-                                               n++;
-                               }
+       if (!dname)
+               goto add_names;
 
-                               if (strncmp(n, dname, dlen) == 0)
-                                       goto update_context;
-                       }
+       /* parent is more likely, look for it first */
+       for (idx = 0; idx < context->name_count; idx++) {
+               struct audit_names *n = &context->names[idx];
 
-       /* catch-all in case match not found */
-       idx = context->name_count++;
-       context->names[idx].name  = NULL;
-       context->names[idx].pino  = pino;
-#if AUDIT_DEBUG
-       context->ino_count++;
-#endif
+               if (!n->name)
+                       continue;
 
-update_context:
-       if (inode) {
-               context->names[idx].ino   = inode->i_ino;
-               context->names[idx].dev   = inode->i_sb->s_dev;
-               context->names[idx].mode  = inode->i_mode;
-               context->names[idx].uid   = inode->i_uid;
-               context->names[idx].gid   = inode->i_gid;
-               context->names[idx].rdev  = inode->i_rdev;
-               audit_inode_context(idx, inode);
+               if (n->ino == parent->i_ino &&
+                   !audit_compare_dname_path(dname, n->name, &dirlen)) {
+                       n->name_len = dirlen; /* update parent data in place */
+                       found_parent = n->name;
+                       goto add_names;
+               }
+       }
+
+       /* no matching parent, look for matching child */
+       for (idx = 0; idx < context->name_count; idx++) {
+               struct audit_names *n = &context->names[idx];
+
+               if (!n->name)
+                       continue;
+
+               /* strcmp() is the more likely scenario */
+               if (!strcmp(dname, n->name) ||
+                    !audit_compare_dname_path(dname, n->name, &dirlen)) {
+                       if (inode)
+                               audit_copy_inode(n, NULL, inode);
+                       else
+                               n->ino = (unsigned long)-1;
+                       found_child = n->name;
+                       goto add_names;
+               }
+       }
+
+add_names:
+       if (!found_parent) {
+               if (audit_inc_name_count(context, parent))
+                       return;
+               idx = context->name_count - 1;
+               context->names[idx].name = NULL;
+               audit_copy_inode(&context->names[idx], NULL, parent);
+       }
+
+       if (!found_child) {
+               if (audit_inc_name_count(context, inode))
+                       return;
+               idx = context->name_count - 1;
+
+               /* Re-use the name belonging to the slot for a matching parent
+                * directory. All names for this context are relinquished in
+                * audit_free_names() */
+               if (found_parent) {
+                       context->names[idx].name = found_parent;
+                       context->names[idx].name_len = AUDIT_NAME_FULL;
+                       /* don't call __putname() */
+                       context->names[idx].name_put = 0;
+               } else {
+                       context->names[idx].name = NULL;
+               }
+
+               if (inode)
+                       audit_copy_inode(&context->names[idx], NULL, inode);
+               else
+                       context->names[idx].ino = (unsigned long)-1;
        }
 }
+EXPORT_SYMBOL_GPL(__audit_inode_child);
 
 /**
  * auditsc_get_stamp - get local copies of audit_context values
@@ -1096,17 +2094,26 @@ update_context:
  *
  * Also sets the context as auditable.
  */
-void auditsc_get_stamp(struct audit_context *ctx,
+int auditsc_get_stamp(struct audit_context *ctx,
                       struct timespec *t, unsigned int *serial)
 {
+       if (!ctx->in_syscall)
+               return 0;
        if (!ctx->serial)
                ctx->serial = audit_serial();
        t->tv_sec  = ctx->ctime.tv_sec;
        t->tv_nsec = ctx->ctime.tv_nsec;
        *serial    = ctx->serial;
-       ctx->auditable = 1;
+       if (!ctx->prio) {
+               ctx->prio = 1;
+               ctx->current_state = AUDIT_RECORD_CONTEXT;
+       }
+       return 1;
 }
 
+/* global counter which is incremented every time something logs in */
+static atomic_t session_id = ATOMIC_INIT(0);
+
 /**
  * audit_set_loginuid - set a task's audit_context loginuid
  * @task: task whose audit context is being modified
@@ -1118,124 +2125,197 @@ void auditsc_get_stamp(struct audit_context *ctx,
  */
 int audit_set_loginuid(struct task_struct *task, uid_t loginuid)
 {
-       if (task->audit_context) {
+       unsigned int sessionid = atomic_inc_return(&session_id);
+       struct audit_context *context = task->audit_context;
+
+       if (context && context->in_syscall) {
                struct audit_buffer *ab;
 
                ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_LOGIN);
                if (ab) {
                        audit_log_format(ab, "login pid=%d uid=%u "
-                               "old auid=%u new auid=%u",
-                               task->pid, task->uid, 
-                               task->audit_context->loginuid, loginuid);
+                               "old auid=%u new auid=%u"
+                               " old ses=%u new ses=%u",
+                               task->pid, task_uid(task),
+                               task->loginuid, loginuid,
+                               task->sessionid, sessionid);
                        audit_log_end(ab);
                }
-               task->audit_context->loginuid = loginuid;
        }
+       task->sessionid = sessionid;
+       task->loginuid = loginuid;
        return 0;
 }
 
 /**
- * audit_get_loginuid - get the loginuid for an audit_context
- * @ctx: the audit_context
+ * __audit_mq_open - record audit data for a POSIX MQ open
+ * @oflag: open flag
+ * @mode: mode bits
+ * @attr: queue attributes
  *
- * Returns the context's loginuid or -1 if @ctx is NULL.
  */
-uid_t audit_get_loginuid(struct audit_context *ctx)
+void __audit_mq_open(int oflag, mode_t mode, struct mq_attr *attr)
 {
-       return ctx ? ctx->loginuid : -1;
+       struct audit_context *context = current->audit_context;
+
+       if (attr)
+               memcpy(&context->mq_open.attr, attr, sizeof(struct mq_attr));
+       else
+               memset(&context->mq_open.attr, 0, sizeof(struct mq_attr));
+
+       context->mq_open.oflag = oflag;
+       context->mq_open.mode = mode;
+
+       context->type = AUDIT_MQ_OPEN;
 }
 
-static char *audit_ipc_context(struct kern_ipc_perm *ipcp)
+/**
+ * __audit_mq_sendrecv - record audit data for a POSIX MQ timed send/receive
+ * @mqdes: MQ descriptor
+ * @msg_len: Message length
+ * @msg_prio: Message priority
+ * @abs_timeout: Message timeout in absolute time
+ *
+ */
+void __audit_mq_sendrecv(mqd_t mqdes, size_t msg_len, unsigned int msg_prio,
+                       const struct timespec *abs_timeout)
 {
        struct audit_context *context = current->audit_context;
-       char *ctx = NULL;
-       int len = 0;
+       struct timespec *p = &context->mq_sendrecv.abs_timeout;
 
-       if (likely(!context))
-               return NULL;
+       if (abs_timeout)
+               memcpy(p, abs_timeout, sizeof(struct timespec));
+       else
+               memset(p, 0, sizeof(struct timespec));
 
-       len = security_ipc_getsecurity(ipcp, NULL, 0);
-       if (len == -EOPNOTSUPP)
-               goto ret;
-       if (len < 0)
-               goto error_path;
+       context->mq_sendrecv.mqdes = mqdes;
+       context->mq_sendrecv.msg_len = msg_len;
+       context->mq_sendrecv.msg_prio = msg_prio;
 
-       ctx = kmalloc(len, GFP_ATOMIC);
-       if (!ctx)
-               goto error_path;
+       context->type = AUDIT_MQ_SENDRECV;
+}
 
-       len = security_ipc_getsecurity(ipcp, ctx, len);
-       if (len < 0)
-               goto error_path;
+/**
+ * __audit_mq_notify - record audit data for a POSIX MQ notify
+ * @mqdes: MQ descriptor
+ * @notification: Notification event
+ *
+ */
 
-       return ctx;
+void __audit_mq_notify(mqd_t mqdes, const struct sigevent *notification)
+{
+       struct audit_context *context = current->audit_context;
 
-error_path:
-       kfree(ctx);
-       audit_panic("error in audit_ipc_context");
-ret:
-       return NULL;
+       if (notification)
+               context->mq_notify.sigev_signo = notification->sigev_signo;
+       else
+               context->mq_notify.sigev_signo = 0;
+
+       context->mq_notify.mqdes = mqdes;
+       context->type = AUDIT_MQ_NOTIFY;
+}
+
+/**
+ * __audit_mq_getsetattr - record audit data for a POSIX MQ get/set attribute
+ * @mqdes: MQ descriptor
+ * @mqstat: MQ flags
+ *
+ */
+void __audit_mq_getsetattr(mqd_t mqdes, struct mq_attr *mqstat)
+{
+       struct audit_context *context = current->audit_context;
+       context->mq_getsetattr.mqdes = mqdes;
+       context->mq_getsetattr.mqstat = *mqstat;
+       context->type = AUDIT_MQ_GETSETATTR;
 }
 
 /**
- * audit_ipc_perms - record audit data for ipc
+ * audit_ipc_obj - record audit data for ipc object
+ * @ipcp: ipc permissions
+ *
+ */
+void __audit_ipc_obj(struct kern_ipc_perm *ipcp)
+{
+       struct audit_context *context = current->audit_context;
+       context->ipc.uid = ipcp->uid;
+       context->ipc.gid = ipcp->gid;
+       context->ipc.mode = ipcp->mode;
+       context->ipc.has_perm = 0;
+       security_ipc_getsecid(ipcp, &context->ipc.osid);
+       context->type = AUDIT_IPC;
+}
+
+/**
+ * audit_ipc_set_perm - record audit data for new ipc permissions
  * @qbytes: msgq bytes
  * @uid: msgq user id
  * @gid: msgq group id
  * @mode: msgq mode (permissions)
  *
- * Returns 0 for success or NULL context or < 0 on error.
+ * Called only after audit_ipc_obj().
  */
-int audit_ipc_perms(unsigned long qbytes, uid_t uid, gid_t gid, mode_t mode, struct kern_ipc_perm *ipcp)
+void __audit_ipc_set_perm(unsigned long qbytes, uid_t uid, gid_t gid, mode_t mode)
 {
-       struct audit_aux_data_ipcctl *ax;
        struct audit_context *context = current->audit_context;
 
-       if (likely(!context))
+       context->ipc.qbytes = qbytes;
+       context->ipc.perm_uid = uid;
+       context->ipc.perm_gid = gid;
+       context->ipc.perm_mode = mode;
+       context->ipc.has_perm = 1;
+}
+
+int audit_bprm(struct linux_binprm *bprm)
+{
+       struct audit_aux_data_execve *ax;
+       struct audit_context *context = current->audit_context;
+
+       if (likely(!audit_enabled || !context || context->dummy))
                return 0;
 
-       ax = kmalloc(sizeof(*ax), GFP_ATOMIC);
+       ax = kmalloc(sizeof(*ax), GFP_KERNEL);
        if (!ax)
                return -ENOMEM;
 
-       ax->qbytes = qbytes;
-       ax->uid = uid;
-       ax->gid = gid;
-       ax->mode = mode;
-       ax->ctx = audit_ipc_context(ipcp);
-
-       ax->d.type = AUDIT_IPC;
+       ax->argc = bprm->argc;
+       ax->envc = bprm->envc;
+       ax->mm = bprm->mm;
+       ax->d.type = AUDIT_EXECVE;
        ax->d.next = context->aux;
        context->aux = (void *)ax;
        return 0;
 }
 
+
 /**
  * audit_socketcall - record audit data for sys_socketcall
  * @nargs: number of args
  * @args: args array
  *
- * Returns 0 for success or NULL context or < 0 on error.
  */
-int audit_socketcall(int nargs, unsigned long *args)
+void audit_socketcall(int nargs, unsigned long *args)
 {
-       struct audit_aux_data_socketcall *ax;
        struct audit_context *context = current->audit_context;
 
-       if (likely(!context))
-               return 0;
-
-       ax = kmalloc(sizeof(*ax) + nargs * sizeof(unsigned long), GFP_KERNEL);
-       if (!ax)
-               return -ENOMEM;
+       if (likely(!context || context->dummy))
+               return;
 
-       ax->nargs = nargs;
-       memcpy(ax->args, args, nargs * sizeof(unsigned long));
+       context->type = AUDIT_SOCKETCALL;
+       context->socketcall.nargs = nargs;
+       memcpy(context->socketcall.args, args, nargs * sizeof(unsigned long));
+}
 
-       ax->d.type = AUDIT_SOCKETCALL;
-       ax->d.next = context->aux;
-       context->aux = (void *)ax;
-       return 0;
+/**
+ * __audit_fd_pair - record audit data for pipe and socketpair
+ * @fd1: the first file descriptor
+ * @fd2: the second file descriptor
+ *
+ */
+void __audit_fd_pair(int fd1, int fd2)
+{
+       struct audit_context *context = current->audit_context;
+       context->fds[0] = fd1;
+       context->fds[1] = fd2;
 }
 
 /**
@@ -1247,76 +2327,211 @@ int audit_socketcall(int nargs, unsigned long *args)
  */
 int audit_sockaddr(int len, void *a)
 {
-       struct audit_aux_data_sockaddr *ax;
        struct audit_context *context = current->audit_context;
 
-       if (likely(!context))
+       if (likely(!context || context->dummy))
                return 0;
 
-       ax = kmalloc(sizeof(*ax) + len, GFP_KERNEL);
-       if (!ax)
-               return -ENOMEM;
+       if (!context->sockaddr) {
+               void *p = kmalloc(sizeof(struct sockaddr_storage), GFP_KERNEL);
+               if (!p)
+                       return -ENOMEM;
+               context->sockaddr = p;
+       }
 
-       ax->len = len;
-       memcpy(ax->a, a, len);
+       context->sockaddr_len = len;
+       memcpy(context->sockaddr, a, len);
+       return 0;
+}
+
+void __audit_ptrace(struct task_struct *t)
+{
+       struct audit_context *context = current->audit_context;
+
+       context->target_pid = t->pid;
+       context->target_auid = audit_get_loginuid(t);
+       context->target_uid = task_uid(t);
+       context->target_sessionid = audit_get_sessionid(t);
+       security_task_getsecid(t, &context->target_sid);
+       memcpy(context->target_comm, t->comm, TASK_COMM_LEN);
+}
+
+/**
+ * audit_signal_info - record signal info for shutting down audit subsystem
+ * @sig: signal value
+ * @t: task being signaled
+ *
+ * If the audit subsystem is being terminated, record the task (pid)
+ * and uid that is doing that.
+ */
+int __audit_signal_info(int sig, struct task_struct *t)
+{
+       struct audit_aux_data_pids *axp;
+       struct task_struct *tsk = current;
+       struct audit_context *ctx = tsk->audit_context;
+       uid_t uid = current_uid(), t_uid = task_uid(t);
+
+       if (audit_pid && t->tgid == audit_pid) {
+               if (sig == SIGTERM || sig == SIGHUP || sig == SIGUSR1 || sig == SIGUSR2) {
+                       audit_sig_pid = tsk->pid;
+                       if (tsk->loginuid != -1)
+                               audit_sig_uid = tsk->loginuid;
+                       else
+                               audit_sig_uid = uid;
+                       security_task_getsecid(tsk, &audit_sig_sid);
+               }
+               if (!audit_signals || audit_dummy_context())
+                       return 0;
+       }
+
+       /* optimize the common case by putting first signal recipient directly
+        * in audit_context */
+       if (!ctx->target_pid) {
+               ctx->target_pid = t->tgid;
+               ctx->target_auid = audit_get_loginuid(t);
+               ctx->target_uid = t_uid;
+               ctx->target_sessionid = audit_get_sessionid(t);
+               security_task_getsecid(t, &ctx->target_sid);
+               memcpy(ctx->target_comm, t->comm, TASK_COMM_LEN);
+               return 0;
+       }
+
+       axp = (void *)ctx->aux_pids;
+       if (!axp || axp->pid_count == AUDIT_AUX_PIDS) {
+               axp = kzalloc(sizeof(*axp), GFP_ATOMIC);
+               if (!axp)
+                       return -ENOMEM;
+
+               axp->d.type = AUDIT_OBJ_PID;
+               axp->d.next = ctx->aux_pids;
+               ctx->aux_pids = (void *)axp;
+       }
+       BUG_ON(axp->pid_count >= AUDIT_AUX_PIDS);
+
+       axp->target_pid[axp->pid_count] = t->tgid;
+       axp->target_auid[axp->pid_count] = audit_get_loginuid(t);
+       axp->target_uid[axp->pid_count] = t_uid;
+       axp->target_sessionid[axp->pid_count] = audit_get_sessionid(t);
+       security_task_getsecid(t, &axp->target_sid[axp->pid_count]);
+       memcpy(axp->target_comm[axp->pid_count], t->comm, TASK_COMM_LEN);
+       axp->pid_count++;
 
-       ax->d.type = AUDIT_SOCKADDR;
-       ax->d.next = context->aux;
-       context->aux = (void *)ax;
        return 0;
 }
 
 /**
- * audit_avc_path - record the granting or denial of permissions
- * @dentry: dentry to record
- * @mnt: mnt to record
+ * __audit_log_bprm_fcaps - store information about a loading bprm and relevant fcaps
+ * @bprm: pointer to the bprm being processed
+ * @new: the proposed new credentials
+ * @old: the old credentials
  *
- * Returns 0 for success or NULL context or < 0 on error.
+ * Simply check if the proc already has the caps given by the file and if not
+ * store the priv escalation info for later auditing at the end of the syscall
  *
- * Called from security/selinux/avc.c::avc_audit()
+ * -Eric
  */
-int audit_avc_path(struct dentry *dentry, struct vfsmount *mnt)
+int __audit_log_bprm_fcaps(struct linux_binprm *bprm,
+                          const struct cred *new, const struct cred *old)
 {
-       struct audit_aux_data_path *ax;
+       struct audit_aux_data_bprm_fcaps *ax;
        struct audit_context *context = current->audit_context;
+       struct cpu_vfs_cap_data vcaps;
+       struct dentry *dentry;
 
-       if (likely(!context))
-               return 0;
-
-       ax = kmalloc(sizeof(*ax), GFP_ATOMIC);
+       ax = kmalloc(sizeof(*ax), GFP_KERNEL);
        if (!ax)
                return -ENOMEM;
 
-       ax->dentry = dget(dentry);
-       ax->mnt = mntget(mnt);
-
-       ax->d.type = AUDIT_AVC_PATH;
+       ax->d.type = AUDIT_BPRM_FCAPS;
        ax->d.next = context->aux;
        context->aux = (void *)ax;
+
+       dentry = dget(bprm->file->f_dentry);
+       get_vfs_caps_from_disk(dentry, &vcaps);
+       dput(dentry);
+
+       ax->fcap.permitted = vcaps.permitted;
+       ax->fcap.inheritable = vcaps.inheritable;
+       ax->fcap.fE = !!(vcaps.magic_etc & VFS_CAP_FLAGS_EFFECTIVE);
+       ax->fcap_ver = (vcaps.magic_etc & VFS_CAP_REVISION_MASK) >> VFS_CAP_REVISION_SHIFT;
+
+       ax->old_pcap.permitted   = old->cap_permitted;
+       ax->old_pcap.inheritable = old->cap_inheritable;
+       ax->old_pcap.effective   = old->cap_effective;
+
+       ax->new_pcap.permitted   = new->cap_permitted;
+       ax->new_pcap.inheritable = new->cap_inheritable;
+       ax->new_pcap.effective   = new->cap_effective;
        return 0;
 }
 
 /**
- * audit_signal_info - record signal info for shutting down audit subsystem
- * @sig: signal value
- * @t: task being signaled
+ * __audit_log_capset - store information about the arguments to the capset syscall
+ * @pid: target pid of the capset call
+ * @new: the new credentials
+ * @old: the old (current) credentials
  *
- * If the audit subsystem is being terminated, record the task (pid)
- * and uid that is doing that.
+ * Record the aguments userspace sent to sys_capset for later printing by the
+ * audit system if applicable
  */
-void audit_signal_info(int sig, struct task_struct *t)
+void __audit_log_capset(pid_t pid,
+                      const struct cred *new, const struct cred *old)
 {
-       extern pid_t audit_sig_pid;
-       extern uid_t audit_sig_uid;
+       struct audit_context *context = current->audit_context;
+       context->capset.pid = pid;
+       context->capset.cap.effective   = new->cap_effective;
+       context->capset.cap.inheritable = new->cap_effective;
+       context->capset.cap.permitted   = new->cap_permitted;
+       context->type = AUDIT_CAPSET;
+}
 
-       if (unlikely(audit_pid && t->tgid == audit_pid)) {
-               if (sig == SIGTERM || sig == SIGHUP) {
-                       struct audit_context *ctx = current->audit_context;
-                       audit_sig_pid = current->pid;
-                       if (ctx)
-                               audit_sig_uid = ctx->loginuid;
-                       else
-                               audit_sig_uid = current->uid;
+/**
+ * audit_core_dumps - record information about processes that end abnormally
+ * @signr: signal value
+ *
+ * If a process ends with a core dump, something fishy is going on and we
+ * should record the event for investigation.
+ */
+void audit_core_dumps(long signr)
+{
+       struct audit_buffer *ab;
+       u32 sid;
+       uid_t auid = audit_get_loginuid(current), uid;
+       gid_t gid;
+       unsigned int sessionid = audit_get_sessionid(current);
+
+       if (!audit_enabled)
+               return;
+
+       if (signr == SIGQUIT)   /* don't care for those */
+               return;
+
+       ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_ANOM_ABEND);
+       current_uid_gid(&uid, &gid);
+       audit_log_format(ab, "auid=%u uid=%u gid=%u ses=%u",
+                        auid, uid, gid, sessionid);
+       security_task_getsecid(current, &sid);
+       if (sid) {
+               char *ctx = NULL;
+               u32 len;
+
+               if (security_secid_to_secctx(sid, &ctx, &len))
+                       audit_log_format(ab, " ssid=%u", sid);
+               else {
+                       audit_log_format(ab, " subj=%s", ctx);
+                       security_release_secctx(ctx, len);
                }
        }
+       audit_log_format(ab, " pid=%d comm=", current->pid);
+       audit_log_untrustedstring(ab, current->comm);
+       audit_log_format(ab, " sig=%ld", signr);
+       audit_log_end(ab);
+}
+
+struct list_head *audit_killed_trees(void)
+{
+       struct audit_context *ctx = current->audit_context;
+       if (likely(!ctx || !ctx->in_syscall))
+               return NULL;
+       return &ctx->killed_trees;
 }