Skip to contentSkip to conversation

xfs: properly order xfsstats proc/sysfs init and teardown

PATCH Applied Eric Sandeen 3 participants 3 messages Nov 14 – Nov 18 XFS filesystem

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xfs: properly order xfsstats proc/sysfs init and teardown

The xfs statistics are accessible both through procfs and through sysctls, so they need to be fully set up before exposure via those interfaces, and those interfaces both need to be torn down before the underlying percpu structures are freed. Today we get that wrong on both ends - we expose them through xfs_init_procfs before allocating the percpu structures, and we free those percpu structures before tearing down sysctls and procfs.

Fix init_xfs_fs by calling alloc_percpu(xfsstats) prior to xfs_init_procfs(), and fix the error unwind path to reflect this, then also move free_percpu() to after tearing down both procfs and sysctls in exit_xfs_fs.

Fixes 80529c45ab66 ("xfs: pass xfsstats structures to handlers and macros")
---
 fs/xfs/xfs_super.c | 22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)
fs/xfs/xfs_super.c +11 -11
index 5a06132aa384..769c622cff54 100644
@@ -2668,9 +2668,15 @@ init_xfs_fs(void)
 	if (error)
 		goto out_destroy_wq;
 
+	xfsstats.xs_stats = alloc_percpu(struct xfsstats);
+	if (!xfsstats.xs_stats) {
+		error = -ENOMEM;
+		goto out_mru_cache_uninit;
+	}
+
 	error = xfs_init_procfs();
 	if (error)
-		goto out_mru_cache_uninit;
+		goto out_free_stats;
 
 	error = xfs_sysctl_register();
 	if (error)
@@ -2686,16 +2692,10 @@ init_xfs_fs(void)
 
 	xfsstats.xs_kobj.kobject.kset = xfs_kset;
 
-	xfsstats.xs_stats = alloc_percpu(struct xfsstats);
-	if (!xfsstats.xs_stats) {
-		error = -ENOMEM;
-		goto out_kset_unregister;
-	}
-
 	error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
 			       "stats");
 	if (error)
-		goto out_free_stats;
+		goto out_kset_unregister;
 
 	error = xchk_global_stats_setup(xfs_debugfs);
 	if (error)
@@ -2727,8 +2727,6 @@ init_xfs_fs(void)
 	xchk_global_stats_teardown();
  out_remove_stats_kobj:
 	xfs_sysfs_del(&xfsstats.xs_kobj);
- out_free_stats:
-	free_percpu(xfsstats.xs_stats);
  out_kset_unregister:
 	kset_unregister(xfs_kset);
  out_debugfs_unregister:
@@ -2736,6 +2734,8 @@ init_xfs_fs(void)
 	xfs_sysctl_unregister();
  out_cleanup_procfs:
 	xfs_cleanup_procfs();
+ out_free_stats:
+	free_percpu(xfsstats.xs_stats);
  out_mru_cache_uninit:
 	xfs_mru_cache_uninit();
  out_destroy_wq:
@@ -2756,11 +2756,11 @@ exit_xfs_fs(void)
 #endif
 	xchk_global_stats_teardown();
 	xfs_sysfs_del(&xfsstats.xs_kobj);
-	free_percpu(xfsstats.xs_stats);
 	kset_unregister(xfs_kset);
 	debugfs_remove(xfs_debugfs);
 	xfs_sysctl_unregister();
 	xfs_cleanup_procfs();
+	free_percpu(xfsstats.xs_stats);
 	xfs_mru_cache_uninit();
 	xfs_destroy_workqueues();
 	xfs_destroy_caches();
Signature · 1 line
2.56.0
maintainer Applied
#2 lore raw
In reply to Eric Sandeen
Quoted 9 lines · Eric Sandeen

The xfs statistics are accessible both through procfs and through sysctls, so they need to be fully set up before exposure via those interfaces, and those interfaces both need to be torn down before the underlying percpu structures are freed. Today we get that wrong on both ends - we expose them through xfs_init_procfs before allocating the percpu structures, and we free those percpu structures before tearing down sysctls and procfs.

[...]

Applied to for-next, thanks!

[1/1] xfs: properly order xfsstats proc/sysfs init and teardown

      commit: 765db499ad5837365b445b26b6fc81f30f134758

Best regards,

Signature · 1 line
Carlos Maiolino <[email protected]>