futex: Fix uninterruptible loop due to gate_area
[linux-2.6.git] / kernel / auditsc.c
index 2de95d1..ce4b054 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/slab.h>
 #include <linux/mount.h>
 #include <linux/socket.h>
 #include <linux/mqueue.h>
@@ -64,7 +65,6 @@
 #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>
 
@@ -168,12 +168,12 @@ struct audit_context {
        int                 in_syscall; /* 1 if task is in a syscall */
        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 */
        u64                 prio;
+       int                 return_valid; /* return code is valid */
        int                 name_count;
        struct audit_names  names[AUDIT_NAMES];
        char *              filterkey;  /* key for rule that triggered record */
@@ -198,6 +198,7 @@ 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;
@@ -240,6 +241,10 @@ struct audit_context {
                        pid_t                   pid;
                        struct audit_cap_data   cap;
                } capset;
+               struct {
+                       int                     fd;
+                       int                     flags;
+               } mmap;
        };
        int fds[2];
 
@@ -249,7 +254,6 @@ 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);
@@ -439,17 +443,25 @@ static int match_tree_refs(struct audit_context *ctx, struct audit_tree *tree)
 
 /* 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);
+       const struct cred *cred;
        int i, j, 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];
                int result = 0;
@@ -548,9 +560,8 @@ static int audit_filter_rules(struct task_struct *tsk,
                        }
                        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));
+                       if (name)
+                               result = audit_watch_compare(rule->watch, name->ino, name->dev);
                        break;
                case AUDIT_DIR:
                        if (ctx)
@@ -634,10 +645,8 @@ static int audit_filter_rules(struct task_struct *tsk,
                        break;
                }
 
-               if (!result) {
-                       put_cred(cred);
+               if (!result)
                        return 0;
-               }
        }
 
        if (ctx) {
@@ -653,7 +662,6 @@ static int audit_filter_rules(struct task_struct *tsk,
        case AUDIT_NEVER:    *state = AUDIT_DISABLED;       break;
        case AUDIT_ALWAYS:   *state = AUDIT_RECORD_CONTEXT; break;
        }
-       put_cred(cred);
        return 1;
 }
 
@@ -668,7 +676,8 @@ static enum audit_state audit_filter_task(struct task_struct *tsk, char **key)
 
        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();
@@ -702,7 +711,7 @@ 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;
@@ -740,7 +749,8 @@ void audit_filter_inodes(struct task_struct *tsk, struct audit_context *ctx)
 
                list_for_each_entry_rcu(e, list, list) {
                        if ((e->rule.mask[word] & bit) == bit &&
-                           audit_filter_rules(tsk, &e->rule, ctx, n, &state)) {
+                           audit_filter_rules(tsk, &e->rule, ctx, n,
+                                              &state, false)) {
                                rcu_read_unlock();
                                ctx->current_state = state;
                                return;
@@ -853,6 +863,7 @@ 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;
 }
 
@@ -1007,7 +1018,7 @@ 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
@@ -1305,6 +1316,10 @@ static void show_special(struct audit_context *context, int *call_panic)
                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);
 }
@@ -1545,6 +1560,8 @@ void audit_free(struct task_struct *tsk)
        /* 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);
 }
@@ -1688,6 +1705,9 @@ void audit_syscall_exit(int valid, long return_code)
        context->in_syscall = 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;
                context->previous  = NULL;
@@ -1719,7 +1739,7 @@ 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;
@@ -1762,7 +1782,7 @@ 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) {
@@ -1830,13 +1850,8 @@ void __audit_getname(const char *name)
        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);
-       }
-
+       if (!context->pwd.dentry)
+               get_fs_pwd(current->fs, &context->pwd);
 }
 
 /* audit_putname - intercept a putname request
@@ -1887,7 +1902,7 @@ static int audit_inc_name_count(struct audit_context *context,
 {
        if (context->name_count >= AUDIT_NAMES) {
                if (inode)
-                       printk(KERN_DEBUG "name_count maxed, losing inode data: "
+                       printk(KERN_DEBUG "audit: name_count maxed, losing inode data: "
                               "dev=%02x:%02x, inode=%lu\n",
                               MAJOR(inode->i_sb->s_dev),
                               MINOR(inode->i_sb->s_dev),
@@ -1982,7 +1997,6 @@ void __audit_inode(const char *name, const struct dentry *dentry)
 
 /**
  * audit_inode_child - collect inode info for created/removed objects
- * @dname: inode's dentry name
  * @dentry: dentry being audited
  * @parent: inode of dentry parent
  *
@@ -1994,13 +2008,14 @@ 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;
        int dirlen = 0;
 
        if (!context->in_syscall)
@@ -2008,9 +2023,6 @@ 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++) {
@@ -2479,6 +2491,14 @@ void __audit_log_capset(pid_t pid,
        context->type = AUDIT_CAPSET;
 }
 
+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;
+}
+
 /**
  * audit_core_dumps - record information about processes that end abnormally
  * @signr: signal value
@@ -2521,3 +2541,11 @@ void audit_core_dumps(long signr)
        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;
+}