gfs2_seek_data() and gfs2_seek_hole() call gfs2_glock_nq_init(), and independent of the outcome, they call gfs2_glock_dq_uninit(). When the gfs2_glock_nq() call in gfs2_glock_nq_init() fails, gfs2_glock_nq_init() returns with an error code and gh.gh_gl is set NULL. This causes a NULL pointer dereference in gfs2_glock_dq(). Fix that by only calling gfs2_glock_dq_uninit() when gfs2_glock_nq_init() succeeds. Fixes: 3a27411cb4bc ("gfs2: Implement SEEK_HOLE / SEEK_DATA via iomap") Reported-by: syzbot+5a5f492ccae698fd7434@syzkaller.appspotmail.com Closes: https://syzkaller.appspot.com/bug?extid=5a5f492ccae698fd7434 Suggested-by: Andreas Gruenbacher Cc: stable@vger.kernel.org Signed-off-by: William Chuang --- v2: Adopt Andreas Gruenbacher's version of the fix -- skip gfs2_glock_dq_uninit() via `goto out` when gfs2_glock_nq_init() fails, rather than guarding the dequeue with gfs2_holder_initialized(); reword the commit message to match. v1 posted at https://lore.kernel.org/r/20261116180002.44336-1-founder@logarcheon.com fs/gfs2/inode.c | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-) diff --git a/fs/gfs2/inode.c b/fs/gfs2/inode.c index 2c2dc459c037..d4678e4832b3 100644 --- a/fs/gfs2/inode.c +++ b/fs/gfs2/inode.c @@ -2253,9 +2253,11 @@ loff_t gfs2_seek_data(struct file *file, loff_t offset) inode_lock_shared(inode); ret = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &gh); - if (!ret) - ret = iomap_seek_data(inode, offset, &gfs2_iomap_ops); + if (ret) + goto out; + ret = iomap_seek_data(inode, offset, &gfs2_iomap_ops); gfs2_glock_dq_uninit(&gh); +out: inode_unlock_shared(inode); if (ret < 0) @@ -2272,9 +2274,11 @@ loff_t gfs2_seek_hole(struct file *file, loff_t offset) inode_lock_shared(inode); ret = gfs2_glock_nq_init(gl, LM_ST_SHARED, 0, &gh); - if (!ret) - ret = iomap_seek_hole(inode, offset, &gfs2_iomap_ops); + if (ret) + goto out; + ret = iomap_seek_hole(inode, offset, &gfs2_iomap_ops); gfs2_glock_dq_uninit(&gh); +out: inode_unlock_shared(inode); if (ret < 0) -- 2.43.0