On 22/03/2022 19:05, Andrew Price wrote:
Per fstrim(8) we must round up the minlen argument to the fs block size.
The current calculation doesn't take into account devices that have a
discard granularity and requested minlen less than 1 fs block, so the
value can get shifted away to zero in the translation to fs blocks.
The zero minlen passed to gfs2_rgrp_send_discards() then allows
sb_issue_discard() to be called with nr_sects == 0 which returns -EINVAL
and results in gfs2_rgrp_send_discards() returning -EIO.
Adding test results. xfstests interprets fstrim failures as "FITRIM not
supported" and doesn't run them, so I got these results:
5.17:
Ran: generic/260 generic/288 generic/537
Not run: generic/260 generic/288 generic/537
Passed all 3 tests
5.17 + patch:
Ran: generic/260 generic/288 generic/537
Passed all 3 tests
(Those 3 tests are in the ones in the 'quick' and 'trim' groups)
Andy
Make sure minlen is never < 1 fs block by taking the max of the
requested minlen and the fs block size before comparing to the device's
discard granularity and shifting to fs blocks.
Fixes: 076f0faa764ab ("GFS2: Fix FITRIM argument handling")
Signed-off-by: Andrew Price <anpr...@redhat.com>
---
fs/gfs2/rgrp.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/fs/gfs2/rgrp.c b/fs/gfs2/rgrp.c
index 0fb3c01bc557..a34945cbf32a 100644
--- a/fs/gfs2/rgrp.c
+++ b/fs/gfs2/rgrp.c
@@ -1415,7 +1415,8 @@ int gfs2_fitrim(struct file *filp, void __user *argp)
start = r.start >> bs_shift;
end = start + (r.len >> bs_shift);
- minlen = max_t(u64, r.minlen,
+ minlen = max_t(u64, r.minlen, sdp->sd_sb.sb_bsize);
+ minlen = max_t(u64, minlen,
q->limits.discard_granularity) >> bs_shift;
if (end <= start || minlen > sdp->sd_max_rg_data)
