sched: guest CPU accounting: add guest-CPU /proc/<pid>/stat fields
[safe/jmp/linux-2.6] / kernel / auditsc.c
index 25d890e..04f3ffb 100644 (file)
 
 extern struct list_head audit_filter_list[];
 
-/* No syscall auditing will take place unless audit_enabled != 0. */
-extern int audit_enabled;
-
 /* AUDIT_NAMES is the number of slots we reserve in the audit_context
  * for saving names from getname(). */
 #define AUDIT_NAMES    20
 
-/* AUDIT_NAMES_RESERVED is the number of slots we reserve in the
- * audit_context from being used for nameless inodes from
- * path_lookup. */
-#define AUDIT_NAMES_RESERVED 7
-
 /* Indicates that audit should log the full pathname. */
 #define AUDIT_NAME_FULL -1
 
@@ -161,7 +153,7 @@ struct audit_aux_data_execve {
        struct audit_aux_data   d;
        int argc;
        int envc;
-       char mem[0];
+       struct mm_struct *mm;
 };
 
 struct audit_aux_data_socketcall {
@@ -181,12 +173,6 @@ struct audit_aux_data_fd_pair {
        int     fd[2];
 };
 
-struct audit_aux_data_path {
-       struct audit_aux_data   d;
-       struct dentry           *dentry;
-       struct vfsmount         *mnt;
-};
-
 struct audit_aux_data_pids {
        struct audit_aux_data   d;
        pid_t                   target_pid[AUDIT_AUX_PIDS];
@@ -662,12 +648,6 @@ static inline void audit_free_aux(struct audit_context *context)
        struct audit_aux_data *aux;
 
        while ((aux = context->aux)) {
-               if (aux->type == AUDIT_AVC_PATH) {
-                       struct audit_aux_data_path *axi = (void *)aux;
-                       dput(axi->dentry);
-                       mntput(axi->mnt);
-               }
-
                context->aux = aux->next;
                kfree(aux);
        }
@@ -839,6 +819,57 @@ static int audit_log_pid_context(struct audit_context *context, pid_t pid,
        return rc;
 }
 
+static void audit_log_execve_info(struct audit_buffer *ab,
+               struct audit_aux_data_execve *axi)
+{
+       int i;
+       long len, ret;
+       const char __user *p;
+       char *buf;
+
+       if (axi->mm != current->mm)
+               return; /* execve failed, no additional info */
+
+       p = (const char __user *)axi->mm->arg_start;
+
+       for (i = 0; i < axi->argc; i++, p += len) {
+               len = strnlen_user(p, MAX_ARG_STRLEN);
+               /*
+                * We just created this mm, if we can't find the strings
+                * we just copied into it something is _very_ wrong. Similar
+                * for strings that are too long, we should not have created
+                * any.
+                */
+               if (!len || len > MAX_ARG_STRLEN) {
+                       WARN_ON(1);
+                       send_sig(SIGKILL, current, 0);
+               }
+
+               buf = kmalloc(len, GFP_KERNEL);
+               if (!buf) {
+                       audit_panic("out of memory for argv string\n");
+                       break;
+               }
+
+               ret = copy_from_user(buf, p, len);
+               /*
+                * There is no reason for this copy to be short. We just
+                * copied them here, and the mm hasn't been exposed to user-
+                * space yet.
+                */
+               if (ret) {
+                       WARN_ON(1);
+                       send_sig(SIGKILL, current, 0);
+               }
+
+               audit_log_format(ab, "a%d=", i);
+               audit_log_untrustedstring(ab, buf);
+               audit_log_format(ab, "\n");
+
+               kfree(buf);
+       }
+}
+
 static void audit_log_exit(struct audit_context *context, struct task_struct *tsk)
 {
        int i, call_panic = 0;
@@ -954,7 +985,7 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
                case AUDIT_IPC: {
                        struct audit_aux_data_ipcctl *axi = (void *)aux;
                        audit_log_format(ab, 
-                                "ouid=%u ogid=%u mode=%x",
+                                "ouid=%u ogid=%u mode=%#o",
                                 axi->uid, axi->gid, axi->mode);
                        if (axi->osid != 0) {
                                char *ctx = NULL;
@@ -973,19 +1004,13 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
                case AUDIT_IPC_SET_PERM: {
                        struct audit_aux_data_ipcctl *axi = (void *)aux;
                        audit_log_format(ab,
-                               "qbytes=%lx ouid=%u ogid=%u mode=%x",
+                               "qbytes=%lx ouid=%u ogid=%u mode=%#o",
                                axi->qbytes, axi->uid, axi->gid, axi->mode);
                        break; }
 
                case AUDIT_EXECVE: {
                        struct audit_aux_data_execve *axi = (void *)aux;
-                       int i;
-                       const char *p;
-                       for (i = 0, p = axi->mem; i < axi->argc; i++) {
-                               audit_log_format(ab, "a%d=", i);
-                               p = audit_log_untrustedstring(ab, p);
-                               audit_log_format(ab, "\n");
-                       }
+                       audit_log_execve_info(ab, axi);
                        break; }
 
                case AUDIT_SOCKETCALL: {
@@ -1003,11 +1028,6 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
                        audit_log_hex(ab, axs->a, axs->len);
                        break; }
 
-               case AUDIT_AVC_PATH: {
-                       struct audit_aux_data_path *axi = (void *)aux;
-                       audit_log_d_path(ab, "path=", axi->dentry, axi->mnt);
-                       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]);
@@ -1343,6 +1363,28 @@ 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",
+                              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");
+               return 1;
+       }
+       context->name_count++;
+#if AUDIT_DEBUG
+       context->ino_count++;
+#endif
+       return 0;
+}
+
 /* Copy inode data into an audit_names. */
 static void audit_copy_inode(struct audit_names *name, const struct inode *inode)
 {
@@ -1380,13 +1422,10 @@ void __audit_inode(const char *name, const struct inode *inode)
        else {
                /* FIXME: how much do we care about inodes that have no
                 * associated name? */
-               if (context->name_count >= AUDIT_NAMES - AUDIT_NAMES_RESERVED)
+               if (audit_inc_name_count(context, inode))
                        return;
-               idx = context->name_count++;
+               idx = context->name_count - 1;
                context->names[idx].name = NULL;
-#if AUDIT_DEBUG
-               ++context->ino_count;
-#endif
        }
        audit_copy_inode(&context->names[idx], inode);
 }
@@ -1410,7 +1449,7 @@ void __audit_inode_child(const char *dname, const struct inode *inode,
 {
        int idx;
        struct audit_context *context = current->audit_context;
-       const char *found_name = NULL;
+       const char *found_parent = NULL, *found_child = NULL;
        int dirlen = 0;
 
        if (!context->in_syscall)
@@ -1418,61 +1457,73 @@ void __audit_inode_child(const char *dname, const struct inode *inode,
 
        /* determine matching parent */
        if (!dname)
-               goto update_context;
-       for (idx = 0; idx < context->name_count; idx++)
-               if (context->names[idx].ino == parent->i_ino) {
-                       const char *name = context->names[idx].name;
+               goto add_names;
 
-                       if (!name)
-                               continue;
+       /* parent is more likely, look for it first */
+       for (idx = 0; idx < context->name_count; idx++) {
+               struct audit_names *n = &context->names[idx];
 
-                       if (audit_compare_dname_path(dname, name, &dirlen) == 0) {
-                               context->names[idx].name_len = dirlen;
-                               found_name = name;
-                               break;
-                       }
+               if (!n->name)
+                       continue;
+
+               if (n->ino == parent->i_ino &&
+                   !audit_compare_dname_path(dname, n->name, &dirlen)) {
+                       n->name_len = dirlen; /* update parent data in place */
+                       found_parent = n->name;
+                       goto add_names;
                }
+       }
 
-update_context:
-       idx = context->name_count;
-       if (context->name_count == AUDIT_NAMES) {
-               printk(KERN_DEBUG "name_count maxed and losing %s\n",
-                       found_name ?: "(null)");
-               return;
+       /* no matching parent, look for matching child */
+       for (idx = 0; idx < context->name_count; idx++) {
+               struct audit_names *n = &context->names[idx];
+
+               if (!n->name)
+                       continue;
+
+               /* strcmp() is the more likely scenario */
+               if (!strcmp(dname, n->name) ||
+                    !audit_compare_dname_path(dname, n->name, &dirlen)) {
+                       if (inode)
+                               audit_copy_inode(n, inode);
+                       else
+                               n->ino = (unsigned long)-1;
+                       found_child = n->name;
+                       goto add_names;
+               }
        }
-       context->name_count++;
-#if AUDIT_DEBUG
-       context->ino_count++;
-#endif
-       /* Re-use the name belonging to the slot for a matching parent directory.
-        * All names for this context are relinquished in audit_free_names() */
-       context->names[idx].name = found_name;
-       context->names[idx].name_len = AUDIT_NAME_FULL;
-       context->names[idx].name_put = 0;       /* don't call __putname() */
-
-       if (!inode)
-               context->names[idx].ino = (unsigned long)-1;
-       else
-               audit_copy_inode(&context->names[idx], inode);
-
-       /* A parent was not found in audit_names, so copy the inode data for the
-        * provided parent. */
-       if (!found_name) {
-               idx = context->name_count;
-               if (context->name_count == AUDIT_NAMES) {
-                       printk(KERN_DEBUG
-                               "name_count maxed and losing parent inode data: dev=%02x:%02x, inode=%lu",
-                               MAJOR(parent->i_sb->s_dev),
-                               MINOR(parent->i_sb->s_dev),
-                               parent->i_ino);
+
+add_names:
+       if (!found_parent) {
+               if (audit_inc_name_count(context, parent))
                        return;
-               }
-               context->name_count++;
-#if AUDIT_DEBUG
-               context->ino_count++;
-#endif
+               idx = context->name_count - 1;
+               context->names[idx].name = NULL;
                audit_copy_inode(&context->names[idx], parent);
        }
+
+       if (!found_child) {
+               if (audit_inc_name_count(context, inode))
+                       return;
+               idx = context->name_count - 1;
+
+               /* Re-use the name belonging to the slot for a matching parent
+                * directory. All names for this context are relinquished in
+                * audit_free_names() */
+               if (found_parent) {
+                       context->names[idx].name = found_parent;
+                       context->names[idx].name_len = AUDIT_NAME_FULL;
+                       /* don't call __putname() */
+                       context->names[idx].name_put = 0;
+               } else {
+                       context->names[idx].name = NULL;
+               }
+
+               if (inode)
+                       audit_copy_inode(&context->names[idx], inode);
+               else
+                       context->names[idx].ino = (unsigned long)-1;
+       }
 }
 
 /**
@@ -1798,32 +1849,31 @@ int __audit_ipc_set_perm(unsigned long qbytes, uid_t uid, gid_t gid, mode_t mode
        return 0;
 }
 
+int audit_argv_kb = 32;
+
 int audit_bprm(struct linux_binprm *bprm)
 {
        struct audit_aux_data_execve *ax;
        struct audit_context *context = current->audit_context;
-       unsigned long p, next;
-       void *to;
 
        if (likely(!audit_enabled || !context || context->dummy))
                return 0;
 
-       ax = kmalloc(sizeof(*ax) + PAGE_SIZE * MAX_ARG_PAGES - bprm->p,
-                               GFP_KERNEL);
+       /*
+        * Even though the stack code doesn't limit the arg+env size any more,
+        * the audit code requires that _all_ arguments be logged in a single
+        * netlink skb. Hence cap it :-(
+        */
+       if (bprm->argv_len > (audit_argv_kb << 10))
+               return -E2BIG;
+
+       ax = kmalloc(sizeof(*ax), GFP_KERNEL);
        if (!ax)
                return -ENOMEM;
 
        ax->argc = bprm->argc;
        ax->envc = bprm->envc;
-       for (p = bprm->p, to = ax->mem; p < MAX_ARG_PAGES*PAGE_SIZE; p = next) {
-               struct page *page = bprm->page[p / PAGE_SIZE];
-               void *kaddr = kmap(page);
-               next = (p + PAGE_SIZE) & ~(PAGE_SIZE - 1);
-               memcpy(to, kaddr + (p & (PAGE_SIZE - 1)), next - p);
-               to += next - p;
-               kunmap(page);
-       }
-
+       ax->mm = bprm->mm;
        ax->d.type = AUDIT_EXECVE;
        ax->d.next = context->aux;
        context->aux = (void *)ax;
@@ -1926,36 +1976,6 @@ void __audit_ptrace(struct task_struct *t)
 }
 
 /**
- * audit_avc_path - record the granting or denial of permissions
- * @dentry: dentry to record
- * @mnt: mnt to record
- *
- * Returns 0 for success or NULL context or < 0 on error.
- *
- * Called from security/selinux/avc.c::avc_audit()
- */
-int audit_avc_path(struct dentry *dentry, struct vfsmount *mnt)
-{
-       struct audit_aux_data_path *ax;
-       struct audit_context *context = current->audit_context;
-
-       if (likely(!context))
-               return 0;
-
-       ax = kmalloc(sizeof(*ax), GFP_ATOMIC);
-       if (!ax)
-               return -ENOMEM;
-
-       ax->dentry = dget(dentry);
-       ax->mnt = mntget(mnt);
-
-       ax->d.type = AUDIT_AVC_PATH;
-       ax->d.next = context->aux;
-       context->aux = (void *)ax;
-       return 0;
-}
-
-/**
  * audit_signal_info - record signal info for shutting down audit subsystem
  * @sig: signal value
  * @t: task being signaled
@@ -1972,19 +1992,19 @@ int __audit_signal_info(int sig, struct task_struct *t)
        extern uid_t audit_sig_uid;
        extern u32 audit_sig_sid;
 
-       if (audit_pid && t->tgid == audit_pid &&
-           (sig == SIGTERM || sig == SIGHUP || sig == SIGUSR1)) {
-               audit_sig_pid = tsk->pid;
-               if (ctx)
-                       audit_sig_uid = ctx->loginuid;
-               else
-                       audit_sig_uid = tsk->uid;
-               selinux_get_task_sid(tsk, &audit_sig_sid);
+       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;
+                       else
+                               audit_sig_uid = tsk->uid;
+                       selinux_get_task_sid(tsk, &audit_sig_sid);
+               }
+               if (!audit_signals || audit_dummy_context())
+                       return 0;
        }
 
-       if (!audit_signals) /* audit_context checked in wrapper */
-               return 0;
-
        /* optimize the common case by putting first signal recipient directly
         * in audit_context */
        if (!ctx->target_pid) {
@@ -2003,7 +2023,7 @@ int __audit_signal_info(int sig, struct task_struct *t)
                axp->d.next = ctx->aux_pids;
                ctx->aux_pids = (void *)axp;
        }
-       BUG_ON(axp->pid_count > AUDIT_AUX_PIDS);
+       BUG_ON(axp->pid_count >= AUDIT_AUX_PIDS);
 
        axp->target_pid[axp->pid_count] = t->tgid;
        selinux_get_task_sid(t, &axp->target_sid[axp->pid_count]);
@@ -2011,3 +2031,42 @@ int __audit_signal_info(int sig, struct task_struct *t)
 
        return 0;
 }
+
+/**
+ * audit_core_dumps - record information about processes that end abnormally
+ * @signr: signal value
+ *
+ * If a process ends with a core dump, something fishy is going on and we
+ * should record the event for investigation.
+ */
+void audit_core_dumps(long signr)
+{
+       struct audit_buffer *ab;
+       u32 sid;
+
+       if (!audit_enabled)
+               return;
+
+       if (signr == SIGQUIT)   /* don't care for those */
+               return;
+
+       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),
+                       current->uid, current->gid);
+       selinux_get_task_sid(current, &sid);
+       if (sid) {
+               char *ctx = NULL;
+               u32 len;
+
+               if (selinux_sid_to_string(sid, &ctx, &len))
+                       audit_log_format(ab, " ssid=%u", sid);
+               else
+                       audit_log_format(ab, " subj=%s", ctx);
+               kfree(ctx);
+       }
+       audit_log_format(ab, " pid=%d comm=", current->pid);
+       audit_log_untrustedstring(ab, current->comm);
+       audit_log_format(ab, " sig=%ld", signr);
+       audit_log_end(ab);
+}