module: don't unlink the module until we've removed all exposure.

rustyrussell · rustyrussell · commit 944a1fa01266 · 2013-04-17T13:23:02.000+09:30
Otherwise we get a race between unload and reload of the same module: the new module doesn't see the old one in the list, but then fails because it can't register over the still-extant entries in sysfs: [ 103.981925] ------------[ cut here ]------------ [ 103.986902] WARNING: at fs/sysfs/dir.c:536 sysfs_add_one+0xab/0xd0() [ 103.993606] Hardware name: CrownBay Platform [ 103.998075] sysfs: cannot create duplicate filename '/module/pch_gbe' [ 104.004784] Modules linked in: pch_gbe(+) [last unloaded: pch_gbe] [ 104.011362] Pid: 3021, comm: modprobe Tainted: G W 3.9.0-rc5+ hardkernel#5 [ 104.018662] Call Trace: [ 104.021286] [<c103599d>] warn_slowpath_common+0x6d/0xa0 [ 104.026933] [<c1168c8b>] ? sysfs_add_one+0xab/0xd0 [ 104.031986] [<c1168c8b>] ? sysfs_add_one+0xab/0xd0 [ 104.037000] [<c1035a4e>] warn_slowpath_fmt+0x2e/0x30 [ 104.042188] [<c1168c8b>] sysfs_add_one+0xab/0xd0 [ 104.046982] [<c1168dbe>] create_dir+0x5e/0xa0 [ 104.051633] [<c1168e78>] sysfs_create_dir+0x78/0xd0 [ 104.056774] [<c1262bc3>] kobject_add_internal+0x83/0x1f0 [ 104.062351] [<c126daf6>] ? kvasprintf+0x46/0x60 [ 104.067231] [<c1262ebd>] kobject_add_varg+0x2d/0x50 [ 104.072450] [<c1262f07>] kobject_init_and_add+0x27/0x30 [ 104.078075] [<c1089240>] mod_sysfs_setup+0x80/0x540 [ 104.083207] [<c1260851>] ? module_bug_finalize+0x51/0xc0 [ 104.088720] [<c108ab29>] load_module+0x1429/0x18b0 We can teardown sysfs first, then to be sure, put the state in MODULE_STATE_UNFORMED so it's ignored while we deconstruct it. Reported-by: Veaceslav Falico <vfalico@redhat.com> Tested-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
diff --git a/kernel/module.c b/kernel/module.c
@@ -1862,12 +1862,12 @@ static void free_module(struct module *mod)
 {
 	trace_module_free(mod);
 
-	/* Delete from various lists */
-	mutex_lock(&module_mutex);
-	stop_machine(__unlink_module, mod, NULL);
-	mutex_unlock(&module_mutex);
 	mod_sysfs_teardown(mod);
 
+	/* We leave it in list to prevent duplicate loads, but make sure
+	 * that noone uses it while it's being deconstructed. */
+	mod->state = MODULE_STATE_UNFORMED;
+
 	/* Remove dynamic debug info */
 	ddebug_remove_module(mod->name);
 
@@ -1880,6 +1880,11 @@ static void free_module(struct module *mod)
 	/* Free any allocated parameters. */
 	destroy_params(mod->kp, mod->num_kp);
 
+	/* Now we can delete it from the lists */
+	mutex_lock(&module_mutex);
+	stop_machine(__unlink_module, mod, NULL);
+	mutex_unlock(&module_mutex);
+
 	/* This may be NULL, but that's OK */
 	unset_module_init_ro_nx(mod);
 	module_free(mod, mod->module_init);
