]> err.no Git - linux-2.6/blobdiff - kernel/auditsc.c
[AUDIT] return EINTR not ERESTART*
[linux-2.6] / kernel / auditsc.c
index 04f3ffb8d9d4adfd03b314db621f43c8c3ab11fe..ce8c957201efbe2e95d731e07a31c1e277ed5618 100644 (file)
@@ -45,7 +45,6 @@
 #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>
@@ -66,6 +65,7 @@
 #include <linux/binfmts.h>
 #include <linux/highmem.h>
 #include <linux/syscalls.h>
+#include <linux/inotify.h>
 
 #include "audit.h"
 
@@ -180,6 +180,11 @@ struct audit_aux_data_pids {
        int                     pid_count;
 };
 
+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 */
@@ -187,7 +192,6 @@ struct audit_context {
        enum audit_state    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 */
        unsigned long       argv[4];    /* syscall arguments */
        int                 return_valid; /* return code is valid */
@@ -212,6 +216,9 @@ struct audit_context {
        pid_t               target_pid;
        u32                 target_sid;
 
+       struct audit_tree_refs *trees, *first_trees;
+       int tree_count;
+
 #if AUDIT_DEBUG
        int                 put_count;
        int                 ino_count;
@@ -266,6 +273,117 @@ static int audit_match_perm(struct audit_context *ctx, int mask)
        }
 }
 
+/*
+ * 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 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. */
@@ -321,7 +439,7 @@ static int audit_filter_rules(struct task_struct *tsk,
                        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;
 
@@ -380,10 +498,14 @@ static int audit_filter_rules(struct task_struct *tsk,
                                result = (name->dev == rule->watch->dev &&
                                          name->ino == rule->watch->ino);
                        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:
@@ -579,7 +701,24 @@ static inline struct audit_context *audit_get_context(struct task_struct *tsk,
        if (likely(!context))
                return NULL;
        context->return_valid = return_valid;
-       context->return_code  = return_code;
+
+       /*
+        * 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 && !context->auditable) {
                enum audit_state state;
@@ -660,11 +799,8 @@ static inline void audit_free_aux(struct audit_context *context)
 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;
 }
 
 static inline struct audit_context *audit_alloc_context(enum audit_state state)
@@ -703,11 +839,6 @@ int audit_alloc(struct task_struct *tsk)
                return -ENOMEM;
        }
 
-                               /* Preserve login uid */
-       context->loginuid = -1;
-       if (current->audit_context)
-               context->loginuid = current->audit_context->loginuid;
-
        tsk->audit_context  = context;
        set_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
        return 0;
@@ -728,6 +859,8 @@ 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);
@@ -899,7 +1032,7 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
        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);
 
@@ -922,7 +1055,7 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
                  context->name_count,
                  context->ppid,
                  context->pid,
-                 context->loginuid,
+                 tsk->loginuid,
                  context->uid,
                  context->gid,
                  context->euid, context->suid, context->fsuid,
@@ -1136,8 +1269,8 @@ void audit_free(struct task_struct *tsk)
                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 */
        /* that can happen only if we are called from do_exit() */
        if (context->in_syscall && context->auditable)
@@ -1271,6 +1404,7 @@ void audit_syscall_exit(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;
@@ -1282,6 +1416,95 @@ void audit_syscall_exit(int valid, long return_code)
        }
 }
 
+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");
+               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");
+               unroll_tree_refs(context, p, count);
+               audit_set_auditable(context);
+               return;
+       }
+       rcu_read_unlock();
+#endif
+}
+
 /**
  * audit_getname - add a name to the list
  * @name: name to add
@@ -1317,7 +1540,7 @@ void __audit_getname(const char *name)
                context->pwdmnt = mntget(current->fs->pwdmnt);
                read_unlock(&current->fs->lock);
        }
-               
+
 }
 
 /* audit_putname - intercept a putname request
@@ -1400,14 +1623,15 @@ static void audit_copy_inode(struct audit_names *name, const struct inode *inode
 /**
  * audit_inode - store the inode and device from a lookup
  * @name: name being audited
- * @inode: inode being audited
+ * @dentry: dentry being audited
  *
  * Called from fs/namei.c:path_lookup().
  */
-void __audit_inode(const char *name, const struct inode *inode)
+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;
@@ -1427,13 +1651,14 @@ void __audit_inode(const char *name, const struct inode *inode)
                idx = context->name_count - 1;
                context->names[idx].name = NULL;
        }
+       handle_path(dentry);
        audit_copy_inode(&context->names[idx], inode);
 }
 
 /**
  * audit_inode_child - collect inode info for created/removed objects
  * @dname: inode's dentry name
- * @inode: inode being audited
+ * @dentry: dentry being audited
  * @parent: inode of dentry parent
  *
  * For syscalls that create or remove filesystem objects, audit_inode
@@ -1444,17 +1669,20 @@ void __audit_inode(const char *name, const struct inode *inode)
  * 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,
+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)
                goto add_names;
@@ -1525,6 +1753,7 @@ add_names:
                        context->names[idx].ino = (unsigned long)-1;
        }
 }
+EXPORT_SYMBOL_GPL(__audit_inode_child);
 
 /**
  * auditsc_get_stamp - get local copies of audit_context values
@@ -1558,38 +1787,22 @@ int audit_set_loginuid(struct task_struct *task, uid_t loginuid)
 {
        struct audit_context *context = task->audit_context;
 
-       if (context) {
-               /* Only log if audit is enabled */
-               if (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,
-                                       context->loginuid, loginuid);
-                               audit_log_end(ab);
-                       }
+       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->loginuid, loginuid);
+                       audit_log_end(ab);
                }
-               context->loginuid = loginuid;
        }
+       task->loginuid = loginuid;
        return 0;
 }
 
-/**
- * audit_get_loginuid - get the loginuid for an audit_context
- * @ctx: the audit_context
- *
- * Returns the context's loginuid or -1 if @ctx is NULL.
- */
-uid_t audit_get_loginuid(struct audit_context *ctx)
-{
-       return ctx ? ctx->loginuid : -1;
-}
-
-EXPORT_SYMBOL(audit_get_loginuid);
-
 /**
  * __audit_mq_open - record audit data for a POSIX MQ open
  * @oflag: open flag
@@ -1995,8 +2208,8 @@ int __audit_signal_info(int sig, struct task_struct *t)
        if (audit_pid && t->tgid == audit_pid) {
                if (sig == SIGTERM || sig == SIGHUP || sig == SIGUSR1) {
                        audit_sig_pid = tsk->pid;
-                       if (ctx)
-                               audit_sig_uid = ctx->loginuid;
+                       if (tsk->loginuid != -1)
+                               audit_sig_uid = tsk->loginuid;
                        else
                                audit_sig_uid = tsk->uid;
                        selinux_get_task_sid(tsk, &audit_sig_sid);
@@ -2052,7 +2265,7 @@ void audit_core_dumps(long signr)
 
        ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_ANOM_ABEND);
        audit_log_format(ab, "auid=%u uid=%u gid=%u",
-                       audit_get_loginuid(current->audit_context),
+                       audit_get_loginuid(current),
                        current->uid, current->gid);
        selinux_get_task_sid(current, &sid);
        if (sid) {