[LU-19791] the cached writes are dropped silently if the quota is over limit

• Type: Bug
• Resolution: Unresolved
• Priority: Major
• Fix Version/s: Lustre 2.18.0
• Affects Version/s: Lustre 2.17.0, Lustre 2.16.1
• Severity: 3
• Assignee: Sergey Cheremencev
• Reporter: Hongchao Zhang

Description

The cached writes on clients will fail with -EDQUOT and dropped silently
if the current UID/GID/PRJID is over limit.

the reproducer is as following,

# ./llmount.sh
# mkdir /mnt/lustre/tdir
# lfs project -p 1000 -d -s /mnt/lustre/tdir/
# lctl set_param osd-ldiskfs.*.quota_slave.root_prj_enable=1
# lctl conf_param  lustre.quota.ost=ugp
# lfs setquota -p 1000 -B 300M /mnt/lustre

write data to exceed the quota limit
# dd if=/dev/zero of=/mnt/lustre/tdir/tfile1 bs=1M
dd: error writing '/mnt/lustre/tdir/tfile1': Disk quota exceeded
317+0 records in
316+0 records out

write data on another client
# # dd if=/dev/urandom of=/mnt/lustre2/tdir/tfile2 bs=1M count=11+0 records in
1+0 records out
1048576 bytes (1.0 MB, 1.0 MiB) copied, 0.0152292 s, 68.9 MB/s

sync and drop page cache
# sync && sync && sync
# echo 3 >/proc/sys/vm/drop_caches 

the file written above is empty
# cat /mnt/lustre/tdir/tfile2 >/tmp/tfile_copy
# ls -al /tmp/tfile_copy 
-rw-r--r-- 1 root root 0 Jan 16 03:10 /tmp/tfile_copy

the possible reason for it is the "OBD_BRW_SYNC" is set even if the write is not sync

int osc_io_submit(const struct lu_env *env, struct cl_io *io,
                  const struct cl_io_slice *ios, enum cl_req_type crt,
                  struct cl_2queue *queue)
{
        ,,,
        osc_page_submit(env, opg, crt, brw_flags);
        ...
}

void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
                     enum cl_req_type crt, int brw_flags)
{                                       
        ...
        oap->oap_brw_flags = OBD_BRW_SYNC | brw_flags;
        ...
}