There is a memory leak report by kmemleak: unreferenced object 0xffff8881786ff9a0 (size 192): comm "mount", pid 8881, jiffies 4301165942 (age 892.453s) hex dump (first 32 bytes): e0 ef 6f 78 81 88 ff ff 70 95 2a 04 81 88 ff ff ..ox....p.*..... b0 f9 6f 78 81 88 ff ff b0 f9 6f 78 81 88 ff ff ..ox......ox.... backtrace: [<ffffffff8170ea67>] kmalloc_trace+0x27/0xa0 [<ffffffffa0a15465>] init_inodes+0x495/0x2010 [gfs2] [<ffffffffa0a1bc27>] gfs2_fill_super+0x18c7/0x25b0 [gfs2] [<ffffffff818e1626>] get_tree_bdev+0x3e6/0x6e0 [<ffffffffa0a13a34>] gfs2_get_tree+0x44/0x220 [gfs2] [<ffffffff818de91d>] vfs_get_tree+0x7d/0x230 [<ffffffff81958073>] path_mount+0xd63/0x1760 [<ffffffff81958b3a>] do_mount+0xca/0xe0 [<ffffffff81958e1c>] __x64_sys_mount+0x12c/0x1a0 [<ffffffff82f2e485>] do_syscall_64+0x35/0x80 [<ffffffff8300006a>] entry_SYSCALL_64_after_hwframe+0x46/0xb0
The root cause was traced to an error handling path in init_journal() when gfs2_jindex_hold() fails. The GFS2 superblock will hold a list of "gfs2_jdesc", and some of them are not freed in error handling path "fail" when gfs2_jindex_hold() fails. Fix it by freeing the memory of "gfs2_jdesc" allocated in the loop in gfs2_jindex_hold() when an error occurs. Fixes: b3b94faa5fe5 ("[GFS2] The core of GFS2") Signed-off-by: Li Zetao <lizet...@huawei.com> --- fs/gfs2/ops_fstype.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/fs/gfs2/ops_fstype.c b/fs/gfs2/ops_fstype.c index c0cf1d2d0ef5..b55bee96619e 100644 --- a/fs/gfs2/ops_fstype.c +++ b/fs/gfs2/ops_fstype.c @@ -625,6 +625,9 @@ static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh) spin_unlock(&sdp->sd_jindex_spin); } + if (error) + gfs2_jindex_free(sdp); + mutex_unlock(&sdp->sd_jindex_mutex); return error; -- 2.25.1