[LU-8731] lfs df exits with status 0 on failures

• Type: Bug
• Resolution: Fixed
• Priority: Major
• Fix Version/s: Lustre 2.9.0
• Affects Version/s: Lustre 2.9.0
• Labels: test
• Severity: 3
• Assignee: John Hammond
• Reporter: John Hammond

Description

In lfs.c the mntdf() function does not propagate errors from llapi_obd_statfs(). In fact it always returns 0 (except when an invalid pool is specified). This means that 'lfs df' will not fail in some cases that it probably should. For example in the following case

# lfs df /mnt/lustre
UUID                   1K-blocks        Used   Available Use% Mounted on
lustre-MDT0000_UUID : Input/output error
lustre-MDT0001_UUID       125368        1692      114316   1% /mnt/lustre[MDT:1]
error: llapi_obd_statfs: opening '/mnt/lustre': Cannot send after transport endpoint shutdown (108)
lustre-OST0000_UUID       350360       13900      316988   4% /mnt/lustre[OST:0]
lustre-OST0001_UUID       350360       13904      316984   4% /mnt/lustre[OST:1]

filesystem summary:       700720       27804      633972   4% /mnt/lustre
# echo $?
0

Importantly this also means that the lfs_df_check() function from test-framework.sh which is used by client_up() and clients_up() is broken:

lfs_df_check() {
        local clients=${1:-$CLIENTS}

	if [ -z "$clients" ]; then
                $LFS df $MOUNT
        else
                $PDSH $clients "$LFS df $MOUNT" > /dev/null
        fi
}

clients_up() {
        # not every config has many clients
        sleep 1
        lfs_df_check
}

client_up() {
        # usually checked on particular client or locally
        sleep 1
	lfs_df_check $1
}

Issue Links

• is related to: LU-8544 recovery-double-scale test_pairwise_fail: start client on trevis-54vm5 failed
• is related to: LU-7759 umount hanging in modern distros when OST is unavailable