gfs2: Fix gfs2_release for non-writers regression

When a file is opened for writing, the vfs code (do_dentry_open)
calls get_write_access for the inode, thus incrementing the inode's write
count. That writer normally then creates a multi-block reservation for
the inode (i_res) that can be re-used by other writers, which speeds up
writes for applications that stupidly loop on open/write/close.
When the writes are all done, the multi-block reservation should be
deleted when the file is closed by the last "writer."

Commit 0ec9b9ea4f83 broke that concept when it moved the call to
gfs2_rs_delete before the check for FMODE_WRITE.  Non-writers have no
business removing the multi-block reservations of writers. In fact, if
someone opens and closes the file for RO while a writer has a
multi-block reservation, the RO closer will delete the reservation
midway through the write, and this results in:

kernel BUG at fs/gfs2/rgrp.c:677! (or thereabouts) which is:
BUG_ON(rs->rs_requested); from function gfs2_rs_deltree.

This patch moves the check back inside the check for FMODE_WRITE.

Fixes: 0ec9b9ea4f83 ("gfs2: Check for active reservation in gfs2_release")
Cc: stable@vger.kernel.org # v5.12+
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>

diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 3e718cfc19a793eb29225eb685cc17b00f60d6c0..8c39a8571b1fa9bf4db49405dfcbdd73184181bb 100644 (file)
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -704,10 +704,11 @@ static int gfs2_release(struct inode *inode, struct file *file)
        kfree(file->private_data);
        file->private_data = NULL;
 
-       if (gfs2_rs_active(&ip->i_res))
-               gfs2_rs_delete(ip, &inode->i_writecount);
-       if (file->f_mode & FMODE_WRITE)
+       if (file->f_mode & FMODE_WRITE) {
+               if (gfs2_rs_active(&ip->i_res))
+                       gfs2_rs_delete(ip, &inode->i_writecount);
                gfs2_qa_put(ip);
+       }
        return 0;
 }