RepoMirrors/ceph
1
0
Fork 0
mirror of https://github.com/ceph/ceph synced 2025-04-01 23:02:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

test_librbd_fsx: flush before discard in krbd mode

Browse Source 
We need to sync dirty pages before issuing BLKDISCARD, explained in
a comment.

Fixes: #11448
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
This commit is contained in:
Ilya Dryomov 2015-04-23 19:22:05 +03:00
parent 2e8d476e7a
commit bd050240ce
1 changed files with 26 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
src/test/librbd/fsx.c
View File

@ -658,6 +658,20 @@ krbd_discard(struct rbd_ctx *ctx, uint64_t off, uint64_t len)
uint64_t range[2] = { off, len };
int ret;
/*
* BLKDISCARD doesn't affect dirty pages. This means we can't
* rely on discarded sectors to match good_buf (i.e. contain
* zeros) without a preceding cache flush:
*
* write 0..3M
* discard 1..2M
*
* results in "data data data" rather than "data 0000 data".
*/
ret = __krbd_flush(ctx);
if (ret < 0)
return ret;
/*
* off and len must be 512-byte aligned, otherwise BLKDISCARD
* will fail with -EINVAL. This means that -K (enable krbd
@ -697,11 +711,18 @@ krbd_resize(struct rbd_ctx *ctx, uint64_t size)
assert(size % truncbdy == 0);
/*
* This is essential: when krbd detects a size change, it calls
* revalidate_disk(), which ends up calling invalidate_bdev(),
* which invalidates only clean buffers. The cache flush makes
* it invalidate everything, which is what we need if we are
* shrinking.
* When krbd detects a size change, it calls revalidate_disk(),
* which ends up calling invalidate_bdev(), which invalidates
* clean pages and does nothing about dirty pages beyond the
* new size. The preceding cache flush makes sure those pages
* are invalidated, which is what we need on shrink:
*
* write 0..1M
* resize 0
* resize 2M
* write 1..2M
*
* results in "data data" rather than "0000 data".
*/
ret = __krbd_flush(ctx);
if (ret < 0)