Linux: Fix zfs_prune panics v2 (openzfs#17121)

It turns out that approach taken in the original version of the patch
was wrong. So now, we're taking approach in-line with how kernel
actually does it - when sb is being torn down, access to it
is serialized via sb->s_umount rwsem, only when that lock is taken
is it okay to work with s_flags - and the other mistake I was doing
was trying to make SB_ACTIVE work, but apparently the kernel checks
the negative variant - not SB_DYING and not SB_BORN.

Kernels pre-6.6 don't have SB_DYING, but check if sb is hashed
instead.

Signed-off-by: Pavel Snajdr <[email protected]>
Reviewed-by: Tony Hutter <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Pavel Snajdr <[email protected]>

Author: snajpa

config/kernel-sb-dying.m4

@@ -0,0 +1,19 @@
+dnl #
+dnl # SB_DYING exists since Linux 6.6
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_SRC_SB_DYING], [
+	ZFS_LINUX_TEST_SRC([sb_dying], [
+		#include <linux/fs.h>
+	],[
+		(void) SB_DYING;
+	])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_SB_DYING], [
+	AC_MSG_CHECKING([whether SB_DYING is defined])
+	ZFS_LINUX_TEST_RESULT([sb_dying], [
+		AC_MSG_RESULT(yes)
+	],[
+		AC_MSG_RESULT(no)
+	])
+])

config/kernel.m4

@@ -103,6 +103,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_SRC], [
 	ZFS_AC_KERNEL_SRC_SECURITY_INODE
 	ZFS_AC_KERNEL_SRC_FST_MOUNT
 	ZFS_AC_KERNEL_SRC_BDI
+	ZFS_AC_KERNEL_SRC_SB_DYING
 	ZFS_AC_KERNEL_SRC_SET_NLINK
 	ZFS_AC_KERNEL_SRC_SGET
 	ZFS_AC_KERNEL_SRC_LSEEK_EXECUTE
@@ -255,6 +256,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_RESULT], [
 	ZFS_AC_KERNEL_SECURITY_INODE
 	ZFS_AC_KERNEL_FST_MOUNT
 	ZFS_AC_KERNEL_BDI
+	ZFS_AC_KERNEL_SB_DYING
 	ZFS_AC_KERNEL_SET_NLINK
 	ZFS_AC_KERNEL_SGET
 	ZFS_AC_KERNEL_LSEEK_EXECUTE

module/os/linux/zfs/zpl_super.c

@@ -381,17 +381,25 @@ zpl_prune_sb(uint64_t nr_to_scan, void *arg)
 	int objects = 0;
 
 	/*
-	 * deactivate_locked_super calls shrinker_free and only then
-	 * sops->kill_sb cb, resulting in UAF on umount when trying to reach
-	 * for the shrinker functions in zpl_prune_sb of in-umount dataset.
-	 * Increment if s_active is not zero, but don't prune if it is -
-	 * umount could be underway.
+	 * Ensure the superblock is not in the process of being torn down.
 	 */
-	if (atomic_inc_not_zero(&sb->s_active)) {
-		(void) -zfs_prune(sb, nr_to_scan, &objects);
-		atomic_dec(&sb->s_active);
+#ifdef HAVE_SB_DYING
+	if (down_read_trylock(&sb->s_umount)) {
+		if (!(sb->s_flags & SB_DYING) && sb->s_root &&
+		    (sb->s_flags & SB_BORN)) {
+			(void) zfs_prune(sb, nr_to_scan, &objects);
+		}
+		up_read(&sb->s_umount);
 	}
-
+#else
+	if (down_read_trylock(&sb->s_umount)) {
+		if (!hlist_unhashed(&sb->s_instances) &&
+		    sb->s_root && (sb->s_flags & SB_BORN)) {
+			(void) zfs_prune(sb, nr_to_scan, &objects);
+		}
+		up_read(&sb->s_umount);
+	}
+#endif
 }
 
 const struct super_operations zpl_super_operations = {