PM / Hibernate: fix the number of pages used for hibernate/thaw buffering
[linux-flexiantxendom0-3.2.10.git] / kernel / auditsc.c
index 3ece960..af1de0f 100644 (file)
 
 #include <linux/init.h>
 #include <asm/types.h>
-#include <asm/atomic.h>
+#include <linux/atomic.h>
 #include <linux/fs.h>
 #include <linux/namei.h>
 #include <linux/mm.h>
-#include <linux/module.h>
+#include <linux/export.h>
+#include <linux/slab.h>
 #include <linux/mount.h>
 #include <linux/socket.h>
 #include <linux/mqueue.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"
 
+/* flags stating the success for a syscall */
+#define AUDITSC_INVALID 0
+#define AUDITSC_SUCCESS 1
+#define AUDITSC_FAILURE 2
+
 /* AUDIT_NAMES is the number of slots we reserve in the audit_context
- * for saving names from getname(). */
-#define AUDIT_NAMES    20
+ * for saving names from getname().  If we get more names we will allocate
+ * a name dynamically and also add those to the list anchored by names_list. */
+#define AUDIT_NAMES    5
 
 /* Indicates that audit should log the full pathname. */
 #define AUDIT_NAME_FULL -1
@@ -100,9 +107,8 @@ struct audit_cap_data {
  *
  * Further, in fs/namei.c:path_lookup() we store the inode and device. */
 struct audit_names {
+       struct list_head list;          /* audit_context->names_list */
        const char      *name;
-       int             name_len;       /* number of name's characters to log */
-       unsigned        name_put;       /* call __putname() for this name */
        unsigned long   ino;
        dev_t           dev;
        umode_t         mode;
@@ -112,6 +118,14 @@ struct audit_names {
        u32             osid;
        struct audit_cap_data fcap;
        unsigned int    fcap_ver;
+       int             name_len;       /* number of name's characters to log */
+       bool            name_put;       /* call __putname() for this name */
+       /*
+        * This was an allocated audit_names and not from the array of
+        * names allocated in the task audit context.  Thus this name
+        * should be freed on syscall exit
+        */
+       bool            should_free;
 };
 
 struct audit_aux_data {
@@ -124,21 +138,6 @@ struct audit_aux_data {
 /* Number of target pids per aux struct. */
 #define AUDIT_AUX_PIDS 16
 
-struct audit_aux_data_mq_open {
-       struct audit_aux_data   d;
-       int                     oflag;
-       mode_t                  mode;
-       struct mq_attr          attr;
-};
-
-struct audit_aux_data_mq_sendrecv {
-       struct audit_aux_data   d;
-       mqd_t                   mqdes;
-       size_t                  msg_len;
-       unsigned int            msg_prio;
-       struct timespec         abs_timeout;
-};
-
 struct audit_aux_data_execve {
        struct audit_aux_data   d;
        int argc;
@@ -146,11 +145,6 @@ struct audit_aux_data_execve {
        struct mm_struct *mm;
 };
 
-struct audit_aux_data_fd_pair {
-       struct  audit_aux_data d;
-       int     fd[2];
-};
-
 struct audit_aux_data_pids {
        struct audit_aux_data   d;
        pid_t                   target_pid[AUDIT_AUX_PIDS];
@@ -185,16 +179,25 @@ struct audit_tree_refs {
 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 */
        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 */
-       int                 name_count;
-       struct audit_names  names[AUDIT_NAMES];
+       u64                 prio;
+       int                 return_valid; /* return code is valid */
+       /*
+        * The names_list is the list of all audit_names collected during this
+        * syscall.  The first AUDIT_NAMES entries in the names_list will
+        * actually be from the preallocated_names array for performance
+        * reasons.  Except during allocation they should never be referenced
+        * through the preallocated_names array and should only be found/used
+        * by running the names_list.
+        */
+       struct audit_names  preallocated_names[AUDIT_NAMES];
+       int                 name_count; /* total records in names_list */
+       struct list_head    names_list; /* anchor for struct audit_names->list */
        char *              filterkey;  /* key for rule that triggered record */
        struct path         pwd;
        struct audit_context *previous; /* For nested syscalls */
@@ -217,6 +220,7 @@ struct audit_context {
        char                target_comm[TASK_COMM_LEN];
 
        struct audit_tree_refs *trees, *first_trees;
+       struct list_head killed_trees;
        int tree_count;
 
        int type;
@@ -228,12 +232,12 @@ struct audit_context {
                struct {
                        uid_t                   uid;
                        gid_t                   gid;
-                       mode_t                  mode;
+                       umode_t                 mode;
                        u32                     osid;
                        int                     has_perm;
                        uid_t                   perm_uid;
                        gid_t                   perm_gid;
-                       mode_t                  perm_mode;
+                       umode_t                 perm_mode;
                        unsigned long           qbytes;
                } ipc;
                struct {
@@ -244,7 +248,27 @@ struct audit_context {
                        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;
+                       umode_t                 mode;
+                       struct mq_attr          attr;
+               } mq_open;
+               struct {
+                       pid_t                   pid;
+                       struct audit_cap_data   cap;
+               } capset;
+               struct {
+                       int                     fd;
+                       int                     flags;
+               } mmap;
        };
+       int fds[2];
 
 #if AUDIT_DEBUG
        int                 put_count;
@@ -252,7 +276,6 @@ struct audit_context {
 #endif
 };
 
-#define ACC_MODE(x) ("\004\002\006\006"[(x)&O_ACCMODE])
 static inline int open_arg(int flags, int mask)
 {
        int n = ACC_MODE(flags);
@@ -304,21 +327,21 @@ static int audit_match_perm(struct audit_context *ctx, int mask)
        }
 }
 
-static int audit_match_filetype(struct audit_context *ctx, int which)
+static int audit_match_filetype(struct audit_context *ctx, int val)
 {
-       unsigned index = which & ~S_IFMT;
-       mode_t mode = which & S_IFMT;
+       struct audit_names *n;
+       umode_t mode = (umode_t)val;
 
        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;
+       list_for_each_entry(n, &ctx->names_list, list) {
+               if ((n->ino != -1) &&
+                   ((n->mode & S_IFMT) == mode))
+                       return 1;
+       }
+
+       return 0;
 }
 
 /*
@@ -332,6 +355,14 @@ static int audit_match_filetype(struct audit_context *ctx, int which)
  */
 
 #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;
@@ -432,21 +463,158 @@ static int match_tree_refs(struct audit_context *ctx, struct audit_tree *tree)
        return 0;
 }
 
+static int audit_compare_id(uid_t uid1,
+                           struct audit_names *name,
+                           unsigned long name_offset,
+                           struct audit_field *f,
+                           struct audit_context *ctx)
+{
+       struct audit_names *n;
+       unsigned long addr;
+       uid_t uid2;
+       int rc;
+
+       BUILD_BUG_ON(sizeof(uid_t) != sizeof(gid_t));
+
+       if (name) {
+               addr = (unsigned long)name;
+               addr += name_offset;
+
+               uid2 = *(uid_t *)addr;
+               rc = audit_comparator(uid1, f->op, uid2);
+               if (rc)
+                       return rc;
+       }
+
+       if (ctx) {
+               list_for_each_entry(n, &ctx->names_list, list) {
+                       addr = (unsigned long)n;
+                       addr += name_offset;
+
+                       uid2 = *(uid_t *)addr;
+
+                       rc = audit_comparator(uid1, f->op, uid2);
+                       if (rc)
+                               return rc;
+               }
+       }
+       return 0;
+}
+
+static int audit_field_compare(struct task_struct *tsk,
+                              const struct cred *cred,
+                              struct audit_field *f,
+                              struct audit_context *ctx,
+                              struct audit_names *name)
+{
+       switch (f->val) {
+       /* process to file object comparisons */
+       case AUDIT_COMPARE_UID_TO_OBJ_UID:
+               return audit_compare_id(cred->uid,
+                                       name, offsetof(struct audit_names, uid),
+                                       f, ctx);
+       case AUDIT_COMPARE_GID_TO_OBJ_GID:
+               return audit_compare_id(cred->gid,
+                                       name, offsetof(struct audit_names, gid),
+                                       f, ctx);
+       case AUDIT_COMPARE_EUID_TO_OBJ_UID:
+               return audit_compare_id(cred->euid,
+                                       name, offsetof(struct audit_names, uid),
+                                       f, ctx);
+       case AUDIT_COMPARE_EGID_TO_OBJ_GID:
+               return audit_compare_id(cred->egid,
+                                       name, offsetof(struct audit_names, gid),
+                                       f, ctx);
+       case AUDIT_COMPARE_AUID_TO_OBJ_UID:
+               return audit_compare_id(tsk->loginuid,
+                                       name, offsetof(struct audit_names, uid),
+                                       f, ctx);
+       case AUDIT_COMPARE_SUID_TO_OBJ_UID:
+               return audit_compare_id(cred->suid,
+                                       name, offsetof(struct audit_names, uid),
+                                       f, ctx);
+       case AUDIT_COMPARE_SGID_TO_OBJ_GID:
+               return audit_compare_id(cred->sgid,
+                                       name, offsetof(struct audit_names, gid),
+                                       f, ctx);
+       case AUDIT_COMPARE_FSUID_TO_OBJ_UID:
+               return audit_compare_id(cred->fsuid,
+                                       name, offsetof(struct audit_names, uid),
+                                       f, ctx);
+       case AUDIT_COMPARE_FSGID_TO_OBJ_GID:
+               return audit_compare_id(cred->fsgid,
+                                       name, offsetof(struct audit_names, gid),
+                                       f, ctx);
+       /* uid comparisons */
+       case AUDIT_COMPARE_UID_TO_AUID:
+               return audit_comparator(cred->uid, f->op, tsk->loginuid);
+       case AUDIT_COMPARE_UID_TO_EUID:
+               return audit_comparator(cred->uid, f->op, cred->euid);
+       case AUDIT_COMPARE_UID_TO_SUID:
+               return audit_comparator(cred->uid, f->op, cred->suid);
+       case AUDIT_COMPARE_UID_TO_FSUID:
+               return audit_comparator(cred->uid, f->op, cred->fsuid);
+       /* auid comparisons */
+       case AUDIT_COMPARE_AUID_TO_EUID:
+               return audit_comparator(tsk->loginuid, f->op, cred->euid);
+       case AUDIT_COMPARE_AUID_TO_SUID:
+               return audit_comparator(tsk->loginuid, f->op, cred->suid);
+       case AUDIT_COMPARE_AUID_TO_FSUID:
+               return audit_comparator(tsk->loginuid, f->op, cred->fsuid);
+       /* euid comparisons */
+       case AUDIT_COMPARE_EUID_TO_SUID:
+               return audit_comparator(cred->euid, f->op, cred->suid);
+       case AUDIT_COMPARE_EUID_TO_FSUID:
+               return audit_comparator(cred->euid, f->op, cred->fsuid);
+       /* suid comparisons */
+       case AUDIT_COMPARE_SUID_TO_FSUID:
+               return audit_comparator(cred->suid, f->op, cred->fsuid);
+       /* gid comparisons */
+       case AUDIT_COMPARE_GID_TO_EGID:
+               return audit_comparator(cred->gid, f->op, cred->egid);
+       case AUDIT_COMPARE_GID_TO_SGID:
+               return audit_comparator(cred->gid, f->op, cred->sgid);
+       case AUDIT_COMPARE_GID_TO_FSGID:
+               return audit_comparator(cred->gid, f->op, cred->fsgid);
+       /* egid comparisons */
+       case AUDIT_COMPARE_EGID_TO_SGID:
+               return audit_comparator(cred->egid, f->op, cred->sgid);
+       case AUDIT_COMPARE_EGID_TO_FSGID:
+               return audit_comparator(cred->egid, f->op, cred->fsgid);
+       /* sgid comparison */
+       case AUDIT_COMPARE_SGID_TO_FSGID:
+               return audit_comparator(cred->sgid, f->op, cred->fsgid);
+       default:
+               WARN(1, "Missing AUDIT_COMPARE define.  Report as a bug\n");
+               return 0;
+       }
+       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. */
+ * otherwise.
+ *
+ * If task_creation is true, this is an explicit indication that we are
+ * filtering a task rule at task creation time.  This and tsk == current are
+ * the only situations where tsk->cred may be accessed without an rcu read lock.
+ */
 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)
+                             enum audit_state *state,
+                             bool task_creation)
 {
-       const struct cred *cred = get_task_cred(tsk);
-       int i, j, need_sid = 1;
+       const struct cred *cred;
+       int i, need_sid = 1;
        u32 sid;
 
+       cred = rcu_dereference_check(tsk->cred, tsk == current || task_creation);
+
        for (i = 0; i < rule->field_count; i++) {
                struct audit_field *f = &rule->fields[i];
+               struct audit_names *n;
                int result = 0;
 
                switch (f->type) {
@@ -505,12 +673,14 @@ static int audit_filter_rules(struct task_struct *tsk,
                        }
                        break;
                case AUDIT_DEVMAJOR:
-                       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)) {
+                       if (name) {
+                               if (audit_comparator(MAJOR(name->dev), f->op, f->val) ||
+                                   audit_comparator(MAJOR(name->rdev), f->op, f->val))
+                                       ++result;
+                       } else if (ctx) {
+                               list_for_each_entry(n, &ctx->names_list, list) {
+                                       if (audit_comparator(MAJOR(n->dev), f->op, f->val) ||
+                                           audit_comparator(MAJOR(n->rdev), f->op, f->val)) {
                                                ++result;
                                                break;
                                        }
@@ -518,12 +688,14 @@ static int audit_filter_rules(struct task_struct *tsk,
                        }
                        break;
                case AUDIT_DEVMINOR:
-                       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)) {
+                       if (name) {
+                               if (audit_comparator(MINOR(name->dev), f->op, f->val) ||
+                                   audit_comparator(MINOR(name->rdev), f->op, f->val))
+                                       ++result;
+                       } else if (ctx) {
+                               list_for_each_entry(n, &ctx->names_list, list) {
+                                       if (audit_comparator(MINOR(n->dev), f->op, f->val) ||
+                                           audit_comparator(MINOR(n->rdev), f->op, f->val)) {
                                                ++result;
                                                break;
                                        }
@@ -534,8 +706,32 @@ static int audit_filter_rules(struct task_struct *tsk,
                        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)) {
+                               list_for_each_entry(n, &ctx->names_list, list) {
+                                       if (audit_comparator(n->ino, f->op, f->val)) {
+                                               ++result;
+                                               break;
+                                       }
+                               }
+                       }
+                       break;
+               case AUDIT_OBJ_UID:
+                       if (name) {
+                               result = audit_comparator(name->uid, f->op, f->val);
+                       } else if (ctx) {
+                               list_for_each_entry(n, &ctx->names_list, list) {
+                                       if (audit_comparator(n->uid, f->op, f->val)) {
+                                               ++result;
+                                               break;
+                                       }
+                               }
+                       }
+                       break;
+               case AUDIT_OBJ_GID:
+                       if (name) {
+                               result = audit_comparator(name->gid, f->op, f->val);
+                       } else if (ctx) {
+                               list_for_each_entry(n, &ctx->names_list, list) {
+                                       if (audit_comparator(n->gid, f->op, f->val)) {
                                                ++result;
                                                break;
                                        }
@@ -543,9 +739,8 @@ static int audit_filter_rules(struct task_struct *tsk,
                        }
                        break;
                case AUDIT_WATCH:
-                       if (name && rule->watch->ino != (unsigned long)-1)
-                               result = (name->dev == rule->watch->dev &&
-                                         name->ino == rule->watch->ino);
+                       if (name)
+                               result = audit_watch_compare(rule->watch, name->ino, name->dev);
                        break;
                case AUDIT_DIR:
                        if (ctx)
@@ -591,11 +786,10 @@ static int audit_filter_rules(struct task_struct *tsk,
                                                   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)) {
+                                       list_for_each_entry(n, &ctx->names_list, list) {
+                                               if (security_audit_rule_match(n->osid, f->type,
+                                                                             f->op, f->lsm_rule,
+                                                                             ctx)) {
                                                        ++result;
                                                        break;
                                                }
@@ -627,20 +821,27 @@ static int audit_filter_rules(struct task_struct *tsk,
                case AUDIT_FILETYPE:
                        result = audit_match_filetype(ctx, f->val);
                        break;
+               case AUDIT_FIELD_COMPARE:
+                       result = audit_field_compare(tsk, cred, f, ctx, name);
+                       break;
                }
+               if (!result)
+                       return 0;
+       }
 
-               if (!result) {
-                       put_cred(cred);
+       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;
        }
-       if (rule->filterkey && ctx)
-               ctx->filterkey = kstrdup(rule->filterkey, GFP_ATOMIC);
        switch (rule->action) {
        case AUDIT_NEVER:    *state = AUDIT_DISABLED;       break;
        case AUDIT_ALWAYS:   *state = AUDIT_RECORD_CONTEXT; break;
        }
-       put_cred(cred);
        return 1;
 }
 
@@ -648,14 +849,17 @@ 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, NULL, &state)) {
+               if (audit_filter_rules(tsk, &e->rule, NULL, NULL,
+                                      &state, true)) {
+                       if (state == AUDIT_RECORD_CONTEXT)
+                               *key = kstrdup(e->rule.filterkey, GFP_ATOMIC);
                        rcu_read_unlock();
                        return state;
                }
@@ -687,8 +891,9 @@ static enum audit_state audit_filter_syscall(struct task_struct *tsk,
                list_for_each_entry_rcu(e, list, list) {
                        if ((e->rule.mask[word] & bit) == bit &&
                            audit_filter_rules(tsk, &e->rule, ctx, NULL,
-                                              &state)) {
+                                              &state, false)) {
                                rcu_read_unlock();
+                               ctx->current_state = state;
                                return state;
                        }
                }
@@ -697,56 +902,64 @@ static enum audit_state audit_filter_syscall(struct task_struct *tsk,
        return AUDIT_BUILD_CONTEXT;
 }
 
-/* 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().
+/*
+ * Given an audit_name check the inode hash table to see if they match.
+ * Called holding the rcu read lock to protect the use of audit_inode_hash
  */
-enum audit_state audit_filter_inodes(struct task_struct *tsk,
-                                    struct audit_context *ctx)
-{
-       int i;
+static int audit_filter_inode_name(struct task_struct *tsk,
+                                  struct audit_names *n,
+                                  struct audit_context *ctx) {
+       int word, bit;
+       int h = audit_hash_ino((u32)n->ino);
+       struct list_head *list = &audit_inode_hash[h];
        struct audit_entry *e;
        enum audit_state state;
 
-       if (audit_pid && tsk->tgid == audit_pid)
-               return AUDIT_DISABLED;
-
-       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];
+       word = AUDIT_WORD(ctx->major);
+       bit  = AUDIT_BIT(ctx->major);
 
-               if (list_empty(list))
-                       continue;
+       if (list_empty(list))
+               return 0;
 
-               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();
-                               return state;
-                       }
+       list_for_each_entry_rcu(e, list, list) {
+               if ((e->rule.mask[word] & bit) == bit &&
+                   audit_filter_rules(tsk, &e->rule, ctx, n, &state, false)) {
+                       ctx->current_state = state;
+                       return 1;
                }
        }
-       rcu_read_unlock();
-       return AUDIT_BUILD_CONTEXT;
+
+       return 0;
 }
 
-void audit_set_auditable(struct audit_context *ctx)
+/* 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)
 {
-       ctx->auditable = 1;
+       struct audit_names *n;
+
+       if (audit_pid && tsk->tgid == audit_pid)
+               return;
+
+       rcu_read_lock();
+
+       list_for_each_entry(n, &ctx->names_list, list) {
+               if (audit_filter_inode_name(tsk, n, ctx))
+                       break;
+       }
+       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))
+       if (!context)
                return NULL;
        context->return_valid = return_valid;
 
@@ -768,34 +981,21 @@ static inline struct audit_context *audit_get_context(struct task_struct *tsk,
        else
                context->return_code  = return_code;
 
-       if (context->in_syscall && !context->dummy && !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;
-                       goto get_context;
-               }
-
-               state = audit_filter_inodes(tsk, context);
-               if (state == AUDIT_RECORD_CONTEXT)
-                       context->auditable = 1;
-
+       if (context->in_syscall && !context->dummy) {
+               audit_filter_syscall(tsk, context, &audit_filter_list[AUDIT_FILTER_EXIT]);
+               audit_filter_inodes(tsk, context);
        }
 
-get_context:
-
        tsk->audit_context = NULL;
        return context;
 }
 
 static inline void audit_free_names(struct audit_context *context)
 {
-       int i;
+       struct audit_names *n, *next;
 
 #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",
@@ -803,10 +1003,9 @@ static inline void audit_free_names(struct audit_context *context)
                       context->serial, context->major, context->in_syscall,
                       context->name_count, context->put_count,
                       context->ino_count);
-               for (i = 0; i < context->name_count; i++) {
+               list_for_each_entry(n, &context->names_list, list) {
                        printk(KERN_ERR "names[%d] = %p = %s\n", i,
-                              context->names[i].name,
-                              context->names[i].name ?: "(null)");
+                              n->name, n->name ?: "(null)");
                }
                dump_stack();
                return;
@@ -817,9 +1016,12 @@ static inline void audit_free_names(struct audit_context *context)
        context->ino_count  = 0;
 #endif
 
-       for (i = 0; i < context->name_count; i++) {
-               if (context->names[i].name && context->names[i].name_put)
-                       __putname(context->names[i].name);
+       list_for_each_entry_safe(n, next, &context->names_list, list) {
+               list_del(&n->list);
+               if (n->name && n->name_put)
+                       __putname(n->name);
+               if (n->should_free)
+                       kfree(n);
        }
        context->name_count = 0;
        path_put(&context->pwd);
@@ -846,6 +1048,7 @@ static inline void audit_zero_context(struct audit_context *context,
 {
        memset(context, 0, sizeof(*context));
        context->state      = state;
+       context->prio = state == AUDIT_RECORD_CONTEXT ? ~0ULL : 0;
 }
 
 static inline struct audit_context *audit_alloc_context(enum audit_state state)
@@ -855,6 +1058,8 @@ 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);
+       INIT_LIST_HEAD(&context->names_list);
        return context;
 }
 
@@ -871,18 +1076,21 @@ int audit_alloc(struct task_struct *tsk)
 {
        struct audit_context *context;
        enum audit_state     state;
+       char *key = NULL;
 
        if (likely(!audit_ever_enabled))
                return 0; /* Return if not auditing. */
 
-       state = audit_filter_task(tsk);
-       if (likely(state == AUDIT_DISABLED))
+       state = audit_filter_task(tsk, &key);
+       if (state == AUDIT_DISABLED)
                return 0;
 
        if (!(context = audit_alloc_context(state))) {
+               kfree(key);
                audit_log_lost("out of memory in audit_alloc");
                return -ENOMEM;
        }
+       context->filterkey = key;
 
        tsk->audit_context  = context;
        set_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
@@ -963,7 +1171,7 @@ static void audit_log_task_info(struct audit_buffer *ab, struct task_struct *tsk
                while (vma) {
                        if ((vma->vm_flags & VM_EXECUTABLE) &&
                            vma->vm_file) {
-                               audit_log_d_path(ab, "exe=",
+                               audit_log_d_path(ab, " exe=",
                                                 &vma->vm_file->f_path);
                                break;
                        }
@@ -1006,7 +1214,7 @@ static int audit_log_pid_context(struct audit_context *context, pid_t pid,
 /*
  * 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)
+ * within about 500 bytes (next page boundary)
  *
  * 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
@@ -1023,8 +1231,8 @@ static int audit_log_single_execve_arg(struct audit_context *context,
 {
        char arg_num_len_buf[12];
        const char __user *tmp_p = p;
-       /* how many digits are in arg_num? 3 is the length of a=\n */
-       size_t arg_num_len = snprintf(arg_num_len_buf, 12, "%d", arg_num) + 3;
+       /* 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;
@@ -1109,7 +1317,7 @@ static int audit_log_single_execve_arg(struct audit_context *context,
                 * so we can be sure nothing was lost.
                 */
                if ((i == 0) && (too_long))
-                       audit_log_format(*ab, "a%d_len=%zu ", arg_num,
+                       audit_log_format(*ab, " a%d_len=%zu", arg_num,
                                         has_cntl ? 2*len : len);
 
                /*
@@ -1129,15 +1337,14 @@ static int audit_log_single_execve_arg(struct audit_context *context,
                buf[to_send] = '\0';
 
                /* actually log it */
-               audit_log_format(*ab, "a%d", arg_num);
+               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_format(*ab, "\"%s\"", buf);
-               audit_log_format(*ab, "\n");
+                       audit_log_string(*ab, buf);
 
                p += to_send;
                len_left -= to_send;
@@ -1155,8 +1362,8 @@ 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;
+       int i, len;
+       size_t len_sent = 0;
        const char __user *p;
        char *buf;
 
@@ -1165,7 +1372,7 @@ static void audit_log_execve_info(struct audit_context *context,
 
        p = (const char __user *)axi->mm->arg_start;
 
-       audit_log_format(*ab, "argc=%d ", axi->argc);
+       audit_log_format(*ab, "argc=%d", axi->argc);
 
        /*
         * we need some kernel buffer to hold the userspace args.  Just
@@ -1238,7 +1445,7 @@ static void show_special(struct audit_context *context, int *call_panic)
        case AUDIT_IPC: {
                u32 osid = context->ipc.osid;
 
-               audit_log_format(ab, "ouid=%u ogid=%u mode=%#o",
+               audit_log_format(ab, "ouid=%u ogid=%u mode=%#ho",
                         context->ipc.uid, context->ipc.gid, context->ipc.mode);
                if (osid) {
                        char *ctx = NULL;
@@ -1256,7 +1463,7 @@ static void show_special(struct audit_context *context, int *call_panic)
                        ab = audit_log_start(context, GFP_KERNEL,
                                             AUDIT_IPC_SET_PERM);
                        audit_log_format(ab,
-                               "qbytes=%lx ouid=%u ogid=%u mode=%#o",
+                               "qbytes=%lx ouid=%u ogid=%u mode=%#ho",
                                context->ipc.qbytes,
                                context->ipc.perm_uid,
                                context->ipc.perm_gid,
@@ -1265,6 +1472,26 @@ static void show_special(struct audit_context *context, int *call_panic)
                                return;
                }
                break; }
+       case AUDIT_MQ_OPEN: {
+               audit_log_format(ab,
+                       "oflag=0x%x mode=%#ho 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,
@@ -1279,7 +1506,79 @@ static void show_special(struct audit_context *context, int *call_panic)
                        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; }
+       case AUDIT_MMAP: {
+               audit_log_format(ab, "fd=%d flags=0x%x", context->mmap.fd,
+                                context->mmap.flags);
+               break; }
+       }
+       audit_log_end(ab);
+}
+
+static void audit_log_name(struct audit_context *context, struct audit_names *n,
+                          int record_num, int *call_panic)
+{
+       struct audit_buffer *ab;
+       ab = audit_log_start(context, GFP_KERNEL, AUDIT_PATH);
+       if (!ab)
+               return; /* audit_panic has been called */
+
+       audit_log_format(ab, "item=%d", record_num);
+
+       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=%#ho"
+                                " 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);
 }
 
@@ -1290,6 +1589,7 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
        struct audit_buffer *ab;
        struct audit_aux_data *aux;
        const char *tty;
+       struct audit_names *n;
 
        /* tsk == current */
        context->pid = tsk->pid;
@@ -1346,11 +1646,7 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
 
 
        audit_log_task_info(ab, tsk);
-       if (context->filterkey) {
-               audit_log_format(ab, " key=");
-               audit_log_untrustedstring(ab, context->filterkey);
-       } else
-               audit_log_format(ab, " key=(null)");
+       audit_log_key(ab, context->filterkey);
        audit_log_end(ab);
 
        for (aux = context->aux; aux; aux = aux->next) {
@@ -1360,35 +1656,12 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
                        continue; /* audit_panic has been called */
 
                switch (aux->type) {
-               case AUDIT_MQ_OPEN: {
-                       struct audit_aux_data_mq_open *axi = (void *)aux;
-                       audit_log_format(ab,
-                               "oflag=0x%x mode=%#o mq_flags=0x%lx mq_maxmsg=%ld "
-                               "mq_msgsize=%ld mq_curmsgs=%ld",
-                               axi->oflag, axi->mode, axi->attr.mq_flags,
-                               axi->attr.mq_maxmsg, axi->attr.mq_msgsize,
-                               axi->attr.mq_curmsgs);
-                       break; }
-
-               case AUDIT_MQ_SENDRECV: {
-                       struct audit_aux_data_mq_sendrecv *axi = (void *)aux;
-                       audit_log_format(ab,
-                               "mqdes=%d msg_len=%zd msg_prio=%u "
-                               "abs_timeout_sec=%ld abs_timeout_nsec=%ld",
-                               axi->mqdes, axi->msg_len, axi->msg_prio,
-                               axi->abs_timeout.tv_sec, axi->abs_timeout.tv_nsec);
-                       break; }
 
                case AUDIT_EXECVE: {
                        struct audit_aux_data_execve *axi = (void *)aux;
                        audit_log_execve_info(context, &ab, axi);
                        break; }
 
-               case AUDIT_FD_PAIR: {
-                       struct audit_aux_data_fd_pair *axs = (void *)aux;
-                       audit_log_format(ab, "fd0=%d fd1=%d", axs->fd[0], axs->fd[1]);
-                       break; }
-
                case AUDIT_BPRM_FCAPS: {
                        struct audit_aux_data_bprm_fcaps *axs = (void *)aux;
                        audit_log_format(ab, "fver=%x", axs->fcap_ver);
@@ -1403,14 +1676,6 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
                        audit_log_cap(ab, "new_pe", &axs->new_pcap.effective);
                        break; }
 
-               case AUDIT_CAPSET: {
-                       struct audit_aux_data_capset *axs = (void *)aux;
-                       audit_log_format(ab, "pid=%d", axs->pid);
-                       audit_log_cap(ab, "cap_pi", &axs->cap.inheritable);
-                       audit_log_cap(ab, "cap_pp", &axs->cap.permitted);
-                       audit_log_cap(ab, "cap_pe", &axs->cap.effective);
-                       break; }
-
                }
                audit_log_end(ab);
        }
@@ -1418,6 +1683,15 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
        if (context->type)
                show_special(context, &call_panic);
 
+       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) {
@@ -1451,70 +1725,14 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
        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);
+                       audit_log_d_path(ab, " cwd=", &context->pwd);
                        audit_log_end(ab);
                }
        }
-       for (i = 0; i < context->name_count; i++) {
-               struct audit_names *n = &context->names[i];
-
-               ab = audit_log_start(context, GFP_KERNEL, AUDIT_PATH);
-               if (!ab)
-                       continue; /* audit_panic has been called */
-
-               audit_log_format(ab, "item=%d", i);
-
-               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);
-       }
+       i = 0;
+       list_for_each_entry(n, &context->names_list, list)
+               audit_log_name(context, n, i++, &call_panic);
 
        /* Send end of event record to help user space know we are finished */
        ab = audit_log_start(context, GFP_KERNEL, AUDIT_EOE);
@@ -1530,12 +1748,12 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
  *
  * Called from copy_process and do_exit
  */
-void audit_free(struct task_struct *tsk)
+void __audit_free(struct task_struct *tsk)
 {
        struct audit_context *context;
 
        context = audit_get_context(tsk, 0, 0);
-       if (likely(!context))
+       if (!context)
                return;
 
        /* Check for system calls that do not go through the exit
@@ -1543,8 +1761,10 @@ void audit_free(struct task_struct *tsk)
         * We use GFP_ATOMIC here because we might be doing this
         * in the context of the idle thread */
        /* that can happen only if we are called from do_exit() */
-       if (context->in_syscall && context->auditable)
+       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);
 }
@@ -1566,7 +1786,7 @@ 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(int arch, int major,
+void __audit_syscall_entry(int arch, int major,
                         unsigned long a1, unsigned long a2,
                         unsigned long a3, unsigned long a4)
 {
@@ -1574,7 +1794,7 @@ void audit_syscall_entry(int arch, int major,
        struct audit_context *context = tsk->audit_context;
        enum audit_state     state;
 
-       if (unlikely(!context))
+       if (!context)
                return;
 
        /*
@@ -1627,61 +1847,53 @@ void audit_syscall_entry(int arch, int major,
 
        state = context->state;
        context->dummy = !audit_n_rules;
-       if (!context->dummy && (state == AUDIT_SETUP_CONTEXT || state == AUDIT_BUILD_CONTEXT))
+       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))
+       }
+       if (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 || !ctx->auditable)
-               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->auditable = ctx->auditable;
-       p->in_syscall = ctx->in_syscall;
-       p->filterkey = kstrdup(ctx->filterkey, GFP_KERNEL);
-       p->ppid = current->pid;
-}
-
 /**
  * audit_syscall_exit - deallocate audit context after a system call
- * @valid: success/failure flag
- * @return_code: syscall return value
+ * @success: success value of the syscall
+ * @return_code: return value of the syscall
  *
  * Tear down after system call.  If the audit context has been marked as
  * auditable (either because of the AUDIT_RECORD_CONTEXT state from
- * filtering, or because some other part of the kernel write an audit
+ * filtering, or because some other part of the kernel wrote an audit
  * message), then write out the syscall information.  In call cases,
  * free the names stored from getname().
  */
-void audit_syscall_exit(int valid, long return_code)
+void __audit_syscall_exit(int success, long return_code)
 {
        struct task_struct *tsk = current;
        struct audit_context *context;
 
-       context = audit_get_context(tsk, valid, return_code);
+       if (success)
+               success = AUDITSC_SUCCESS;
+       else
+               success = AUDITSC_FAILURE;
 
-       if (likely(!context))
+       context = audit_get_context(tsk, success, return_code);
+       if (!context)
                return;
 
-       if (context->in_syscall && context->auditable)
+       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;
@@ -1698,8 +1910,11 @@ void audit_syscall_exit(int valid, long return_code)
                context->target_sid = 0;
                context->sockaddr_len = 0;
                context->type = 0;
-               kfree(context->filterkey);
-               context->filterkey = NULL;
+               context->fds[0] = -1;
+               if (context->state != AUDIT_RECORD_CONTEXT) {
+                       kfree(context->filterkey);
+                       context->filterkey = NULL;
+               }
                tsk->audit_context = context;
        }
 }
@@ -1711,7 +1926,7 @@ static inline void handle_one(const struct inode *inode)
        struct audit_tree_refs *p;
        struct audit_chunk *chunk;
        int count;
-       if (likely(list_empty(&inode->inotify_watches)))
+       if (likely(hlist_empty(&inode->i_fsnotify_marks)))
                return;
        context = current->audit_context;
        p = context->trees;
@@ -1754,7 +1969,7 @@ retry:
        seq = read_seqbegin(&rename_lock);
        for(;;) {
                struct inode *inode = d->d_inode;
-               if (inode && unlikely(!list_empty(&inode->inotify_watches))) {
+               if (inode && unlikely(!hlist_empty(&inode->i_fsnotify_marks))) {
                        struct audit_chunk *chunk;
                        chunk = audit_tree_lookup(inode);
                        if (chunk) {
@@ -1793,6 +2008,30 @@ retry:
 #endif
 }
 
+static struct audit_names *audit_alloc_name(struct audit_context *context)
+{
+       struct audit_names *aname;
+
+       if (context->name_count < AUDIT_NAMES) {
+               aname = &context->preallocated_names[context->name_count];
+               memset(aname, 0, sizeof(*aname));
+       } else {
+               aname = kzalloc(sizeof(*aname), GFP_NOFS);
+               if (!aname)
+                       return NULL;
+               aname->should_free = true;
+       }
+
+       aname->ino = (unsigned long)-1;
+       list_add_tail(&aname->list, &context->names_list);
+
+       context->name_count++;
+#if AUDIT_DEBUG
+       context->ino_count++;
+#endif
+       return aname;
+}
+
 /**
  * audit_getname - add a name to the list
  * @name: name to add
@@ -1803,9 +2042,7 @@ retry:
 void __audit_getname(const char *name)
 {
        struct audit_context *context = current->audit_context;
-
-       if (IS_ERR(name) || !name)
-               return;
+       struct audit_names *n;
 
        if (!context->in_syscall) {
 #if AUDIT_DEBUG == 2
@@ -1815,20 +2052,17 @@ void __audit_getname(const char *name)
 #endif
                return;
        }
-       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.dentry) {
-               read_lock(&current->fs->lock);
-               context->pwd = current->fs->pwd;
-               path_get(&current->fs->pwd);
-               read_unlock(&current->fs->lock);
-       }
 
+       n = audit_alloc_name(context);
+       if (!n)
+               return;
+
+       n->name = name;
+       n->name_len = AUDIT_NAME_FULL;
+       n->name_put = true;
+
+       if (!context->pwd.dentry)
+               get_fs_pwd(current->fs, &context->pwd);
 }
 
 /* audit_putname - intercept a putname request
@@ -1848,12 +2082,13 @@ void audit_putname(const char *name)
                printk(KERN_ERR "%s:%d(:%d): __putname(%p)\n",
                       __FILE__, __LINE__, context->serial, name);
                if (context->name_count) {
+                       struct audit_names *n;
                        int i;
-                       for (i = 0; i < context->name_count; i++)
+
+                       list_for_each_entry(n, &context->names_list, list)
                                printk(KERN_ERR "name[%d] = %p = %s\n", i,
-                                      context->names[i].name,
-                                      context->names[i].name ?: "(null)");
-               }
+                                      n->name, n->name ?: "(null)");
+                       }
 #endif
                __putname(name);
        }
@@ -1874,39 +2109,11 @@ void audit_putname(const char *name)
 #endif
 }
 
-static int audit_inc_name_count(struct audit_context *context,
-                               const struct inode *inode)
-{
-       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);
-
-               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;
-}
-
-
 static inline int audit_copy_fcaps(struct audit_names *name, const struct dentry *dentry)
 {
        struct cpu_vfs_cap_data caps;
        int rc;
 
-       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;
-
        if (!dentry)
                return 0;
 
@@ -1946,35 +2153,29 @@ static void audit_copy_inode(struct audit_names *name, const struct dentry *dent
  */
 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;
+       struct audit_names *n;
 
        if (!context->in_syscall)
                return;
-       if (context->name_count
-           && context->names[context->name_count-1].name
-           && context->names[context->name_count-1].name == name)
-               idx = context->name_count - 1;
-       else if (context->name_count > 1
-                && context->names[context->name_count-2].name
-                && context->names[context->name_count-2].name == name)
-               idx = context->name_count - 2;
-       else {
-               /* FIXME: how much do we care about inodes that have no
-                * associated name? */
-               if (audit_inc_name_count(context, inode))
-                       return;
-               idx = context->name_count - 1;
-               context->names[idx].name = NULL;
+
+       list_for_each_entry_reverse(n, &context->names_list, list) {
+               if (n->name && (n->name == name))
+                       goto out;
        }
+
+       /* unable to find the name from a previous getname() */
+       n = audit_alloc_name(context);
+       if (!n)
+               return;
+out:
        handle_path(dentry);
-       audit_copy_inode(&context->names[idx], dentry, inode);
+       audit_copy_inode(n, dentry, inode);
 }
 
 /**
  * audit_inode_child - collect inode info for created/removed objects
- * @dname: inode's dentry name
  * @dentry: dentry being audited
  * @parent: inode of dentry parent
  *
@@ -1986,13 +2187,14 @@ void __audit_inode(const char *name, const struct dentry *dentry)
  * must be hooked prior, in order to capture the target inode during
  * unsuccessful attempts.
  */
-void __audit_inode_child(const char *dname, const struct dentry *dentry,
+void __audit_inode_child(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;
+       const char *dname = dentry->d_name.name;
+       struct audit_names *n;
        int dirlen = 0;
 
        if (!context->in_syscall)
@@ -2000,14 +2202,9 @@ void __audit_inode_child(const char *dname, const struct dentry *dentry,
 
        if (inode)
                handle_one(inode);
-       /* determine matching parent */
-       if (!dname)
-               goto add_names;
 
        /* parent is more likely, look for it first */
-       for (idx = 0; idx < context->name_count; idx++) {
-               struct audit_names *n = &context->names[idx];
-
+       list_for_each_entry(n, &context->names_list, list) {
                if (!n->name)
                        continue;
 
@@ -2020,9 +2217,7 @@ void __audit_inode_child(const char *dname, const struct dentry *dentry,
        }
 
        /* no matching parent, look for matching child */
-       for (idx = 0; idx < context->name_count; idx++) {
-               struct audit_names *n = &context->names[idx];
-
+       list_for_each_entry(n, &context->names_list, list) {
                if (!n->name)
                        continue;
 
@@ -2040,34 +2235,29 @@ void __audit_inode_child(const char *dname, const struct dentry *dentry,
 
 add_names:
        if (!found_parent) {
-               if (audit_inc_name_count(context, parent))
+               n = audit_alloc_name(context);
+               if (!n)
                        return;
-               idx = context->name_count - 1;
-               context->names[idx].name = NULL;
-               audit_copy_inode(&context->names[idx], NULL, parent);
+               audit_copy_inode(n, NULL, parent);
        }
 
        if (!found_child) {
-               if (audit_inc_name_count(context, inode))
+               n = audit_alloc_name(context);
+               if (!n)
                        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;
+                       n->name = found_parent;
+                       n->name_len = AUDIT_NAME_FULL;
                        /* don't call __putname() */
-                       context->names[idx].name_put = 0;
-               } else {
-                       context->names[idx].name = NULL;
+                       n->name_put = false;
                }
 
                if (inode)
-                       audit_copy_inode(&context->names[idx], NULL, inode);
-               else
-                       context->names[idx].ino = (unsigned long)-1;
+                       audit_copy_inode(n, NULL, inode);
        }
 }
 EXPORT_SYMBOL_GPL(__audit_inode_child);
@@ -2090,7 +2280,10 @@ int auditsc_get_stamp(struct audit_context *ctx,
        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;
 }
 
@@ -2098,19 +2291,28 @@ int auditsc_get_stamp(struct audit_context *ctx,
 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
+ * audit_set_loginuid - set current task's audit_context loginuid
  * @loginuid: loginuid value
  *
  * Returns 0.
  *
  * Called (set) from fs/proc/base.c::proc_loginuid_write().
  */
-int audit_set_loginuid(struct task_struct *task, uid_t loginuid)
+int audit_set_loginuid(uid_t loginuid)
 {
-       unsigned int sessionid = atomic_inc_return(&session_id);
+       struct task_struct *task = current;
        struct audit_context *context = task->audit_context;
+       unsigned int sessionid;
 
+#ifdef CONFIG_AUDIT_LOGINUID_IMMUTABLE
+       if (task->loginuid != -1)
+               return -EPERM;
+#else /* CONFIG_AUDIT_LOGINUID_IMMUTABLE */
+       if (!capable(CAP_AUDIT_CONTROL))
+               return -EPERM;
+#endif  /* CONFIG_AUDIT_LOGINUID_IMMUTABLE */
+
+       sessionid = atomic_inc_return(&session_id);
        if (context && context->in_syscall) {
                struct audit_buffer *ab;
 
@@ -2134,140 +2336,54 @@ int audit_set_loginuid(struct task_struct *task, uid_t loginuid)
  * __audit_mq_open - record audit data for a POSIX MQ open
  * @oflag: open flag
  * @mode: mode bits
- * @u_attr: queue attributes
+ * @attr: queue attributes
  *
- * Returns 0 for success or NULL context or < 0 on error.
  */
-int __audit_mq_open(int oflag, mode_t mode, struct mq_attr __user *u_attr)
+void __audit_mq_open(int oflag, umode_t mode, struct mq_attr *attr)
 {
-       struct audit_aux_data_mq_open *ax;
        struct audit_context *context = current->audit_context;
 
-       if (!audit_enabled)
-               return 0;
-
-       if (likely(!context))
-               return 0;
-
-       ax = kmalloc(sizeof(*ax), GFP_ATOMIC);
-       if (!ax)
-               return -ENOMEM;
-
-       if (u_attr != NULL) {
-               if (copy_from_user(&ax->attr, u_attr, sizeof(ax->attr))) {
-                       kfree(ax);
-                       return -EFAULT;
-               }
-       } else
-               memset(&ax->attr, 0, sizeof(ax->attr));
+       if (attr)
+               memcpy(&context->mq_open.attr, attr, sizeof(struct mq_attr));
+       else
+               memset(&context->mq_open.attr, 0, sizeof(struct mq_attr));
 
-       ax->oflag = oflag;
-       ax->mode = mode;
+       context->mq_open.oflag = oflag;
+       context->mq_open.mode = mode;
 
-       ax->d.type = AUDIT_MQ_OPEN;
-       ax->d.next = context->aux;
-       context->aux = (void *)ax;
-       return 0;
+       context->type = AUDIT_MQ_OPEN;
 }
 
 /**
- * __audit_mq_timedsend - record audit data for a POSIX MQ timed send
+ * __audit_mq_sendrecv - record audit data for a POSIX MQ timed send/receive
  * @mqdes: MQ descriptor
  * @msg_len: Message length
  * @msg_prio: Message priority
- * @u_abs_timeout: Message timeout in absolute time
- *
- * Returns 0 for success or NULL context or < 0 on error.
- */
-int __audit_mq_timedsend(mqd_t mqdes, size_t msg_len, unsigned int msg_prio,
-                       const struct timespec __user *u_abs_timeout)
-{
-       struct audit_aux_data_mq_sendrecv *ax;
-       struct audit_context *context = current->audit_context;
-
-       if (!audit_enabled)
-               return 0;
-
-       if (likely(!context))
-               return 0;
-
-       ax = kmalloc(sizeof(*ax), GFP_ATOMIC);
-       if (!ax)
-               return -ENOMEM;
-
-       if (u_abs_timeout != NULL) {
-               if (copy_from_user(&ax->abs_timeout, u_abs_timeout, sizeof(ax->abs_timeout))) {
-                       kfree(ax);
-                       return -EFAULT;
-               }
-       } else
-               memset(&ax->abs_timeout, 0, sizeof(ax->abs_timeout));
-
-       ax->mqdes = mqdes;
-       ax->msg_len = msg_len;
-       ax->msg_prio = msg_prio;
-
-       ax->d.type = AUDIT_MQ_SENDRECV;
-       ax->d.next = context->aux;
-       context->aux = (void *)ax;
-       return 0;
-}
-
-/**
- * __audit_mq_timedreceive - record audit data for a POSIX MQ timed receive
- * @mqdes: MQ descriptor
- * @msg_len: Message length
- * @u_msg_prio: Message priority
- * @u_abs_timeout: Message timeout in absolute time
+ * @abs_timeout: Message timeout in absolute time
  *
- * Returns 0 for success or NULL context or < 0 on error.
  */
-int __audit_mq_timedreceive(mqd_t mqdes, size_t msg_len,
-                               unsigned int __user *u_msg_prio,
-                               const struct timespec __user *u_abs_timeout)
+void __audit_mq_sendrecv(mqd_t mqdes, size_t msg_len, unsigned int msg_prio,
+                       const struct timespec *abs_timeout)
 {
-       struct audit_aux_data_mq_sendrecv *ax;
        struct audit_context *context = current->audit_context;
+       struct timespec *p = &context->mq_sendrecv.abs_timeout;
 
-       if (!audit_enabled)
-               return 0;
-
-       if (likely(!context))
-               return 0;
-
-       ax = kmalloc(sizeof(*ax), GFP_ATOMIC);
-       if (!ax)
-               return -ENOMEM;
-
-       if (u_msg_prio != NULL) {
-               if (get_user(ax->msg_prio, u_msg_prio)) {
-                       kfree(ax);
-                       return -EFAULT;
-               }
-       } else
-               ax->msg_prio = 0;
-
-       if (u_abs_timeout != NULL) {
-               if (copy_from_user(&ax->abs_timeout, u_abs_timeout, sizeof(ax->abs_timeout))) {
-                       kfree(ax);
-                       return -EFAULT;
-               }
-       } else
-               memset(&ax->abs_timeout, 0, sizeof(ax->abs_timeout));
+       if (abs_timeout)
+               memcpy(p, abs_timeout, sizeof(struct timespec));
+       else
+               memset(p, 0, sizeof(struct timespec));
 
-       ax->mqdes = mqdes;
-       ax->msg_len = msg_len;
+       context->mq_sendrecv.mqdes = mqdes;
+       context->mq_sendrecv.msg_len = msg_len;
+       context->mq_sendrecv.msg_prio = msg_prio;
 
-       ax->d.type = AUDIT_MQ_SENDRECV;
-       ax->d.next = context->aux;
-       context->aux = (void *)ax;
-       return 0;
+       context->type = AUDIT_MQ_SENDRECV;
 }
 
 /**
  * __audit_mq_notify - record audit data for a POSIX MQ notify
  * @mqdes: MQ descriptor
- * @u_notification: Notification event
+ * @notification: Notification event
  *
  */
 
@@ -2323,7 +2439,7 @@ void __audit_ipc_obj(struct kern_ipc_perm *ipcp)
  *
  * Called only after audit_ipc_obj().
  */
-void __audit_ipc_set_perm(unsigned long qbytes, uid_t uid, gid_t gid, mode_t mode)
+void __audit_ipc_set_perm(unsigned long qbytes, uid_t uid, gid_t gid, umode_t mode)
 {
        struct audit_context *context = current->audit_context;
 
@@ -2334,14 +2450,11 @@ void __audit_ipc_set_perm(unsigned long qbytes, uid_t uid, gid_t gid, mode_t mod
        context->ipc.has_perm = 1;
 }
 
-int audit_bprm(struct linux_binprm *bprm)
+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_KERNEL);
        if (!ax)
                return -ENOMEM;
@@ -2362,13 +2475,10 @@ int audit_bprm(struct linux_binprm *bprm)
  * @args: args array
  *
  */
-void audit_socketcall(int nargs, unsigned long *args)
+void __audit_socketcall(int nargs, unsigned long *args)
 {
        struct audit_context *context = current->audit_context;
 
-       if (likely(!context || context->dummy))
-               return;
-
        context->type = AUDIT_SOCKETCALL;
        context->socketcall.nargs = nargs;
        memcpy(context->socketcall.args, args, nargs * sizeof(unsigned long));
@@ -2379,29 +2489,12 @@ void audit_socketcall(int nargs, unsigned long *args)
  * @fd1: the first file descriptor
  * @fd2: the second file descriptor
  *
- * Returns 0 for success or NULL context or < 0 on error.
  */
-int __audit_fd_pair(int fd1, int fd2)
+void __audit_fd_pair(int fd1, int fd2)
 {
        struct audit_context *context = current->audit_context;
-       struct audit_aux_data_fd_pair *ax;
-
-       if (likely(!context)) {
-               return 0;
-       }
-
-       ax = kmalloc(sizeof(*ax), GFP_KERNEL);
-       if (!ax) {
-               return -ENOMEM;
-       }
-
-       ax->fd[0] = fd1;
-       ax->fd[1] = fd2;
-
-       ax->d.type = AUDIT_FD_PAIR;
-       ax->d.next = context->aux;
-       context->aux = (void *)ax;
-       return 0;
+       context->fds[0] = fd1;
+       context->fds[1] = fd2;
 }
 
 /**
@@ -2411,13 +2504,10 @@ int __audit_fd_pair(int fd1, int fd2)
  *
  * Returns 0 for success or NULL context or < 0 on error.
  */
-int audit_sockaddr(int len, void *a)
+int __audit_sockaddr(int len, void *a)
 {
        struct audit_context *context = current->audit_context;
 
-       if (likely(!context || context->dummy))
-               return 0;
-
        if (!context->sockaddr) {
                void *p = kmalloc(sizeof(struct sockaddr_storage), GFP_KERNEL);
                if (!p)
@@ -2560,31 +2650,44 @@ int __audit_log_bprm_fcaps(struct linux_binprm *bprm,
  * Record the aguments userspace sent to sys_capset for later printing by the
  * audit system if applicable
  */
-int __audit_log_capset(pid_t pid,
+void __audit_log_capset(pid_t pid,
                       const struct cred *new, const struct cred *old)
 {
-       struct audit_aux_data_capset *ax;
        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 (likely(!audit_enabled || !context || context->dummy))
-               return 0;
-
-       ax = kmalloc(sizeof(*ax), GFP_KERNEL);
-       if (!ax)
-               return -ENOMEM;
+void __audit_mmap_fd(int fd, int flags)
+{
+       struct audit_context *context = current->audit_context;
+       context->mmap.fd = fd;
+       context->mmap.flags = flags;
+       context->type = AUDIT_MMAP;
+}
 
-       ax->d.type = AUDIT_CAPSET;
-       ax->d.next = context->aux;
-       context->aux = (void *)ax;
+static void audit_log_abend(struct audit_buffer *ab, char *reason, long signr)
+{
+       uid_t auid, uid;
+       gid_t gid;
+       unsigned int sessionid;
 
-       ax->pid = pid;
-       ax->cap.effective   = new->cap_effective;
-       ax->cap.inheritable = new->cap_effective;
-       ax->cap.permitted   = new->cap_permitted;
+       auid = audit_get_loginuid(current);
+       sessionid = audit_get_sessionid(current);
+       current_uid_gid(&uid, &gid);
 
-       return 0;
+       audit_log_format(ab, "auid=%u uid=%u gid=%u ses=%u",
+                        auid, uid, gid, sessionid);
+       audit_log_task_context(ab);
+       audit_log_format(ab, " pid=%d comm=", current->pid);
+       audit_log_untrustedstring(ab, current->comm);
+       audit_log_format(ab, " reason=");
+       audit_log_string(ab, reason);
+       audit_log_format(ab, " sig=%ld", signr);
 }
-
 /**
  * audit_core_dumps - record information about processes that end abnormally
  * @signr: signal value
@@ -2595,10 +2698,6 @@ int __audit_log_capset(pid_t pid,
 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;
@@ -2607,23 +2706,24 @@ void audit_core_dumps(long signr)
                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;
+       audit_log_abend(ab, "memory violation", signr);
+       audit_log_end(ab);
+}
 
-               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);
+void __audit_seccomp(unsigned long syscall)
+{
+       struct audit_buffer *ab;
+
+       ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_ANOM_ABEND);
+       audit_log_abend(ab, "seccomp", SIGKILL);
+       audit_log_format(ab, " syscall=%ld", syscall);
        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;
+}