Eric Biggers <ebigg...@kernel.org> 于2022年11月2日周三 11:18写道：
>
> On Wed, Nov 02, 2022 at 11:06:17AM +0800, fengnan chang wrote:
> >
> >
> > > 2022年11月2日 10:05，Eric Biggers <ebigg...@kernel.org> 写道：
> > >
> > > On Wed, Jun 08, 2022 at 09:48:52PM +0800, Fengnan Chang wrote:
> > >> When decompressed failed, f2fs_prepare_compress_overwrite will enter
> > >> endless loop, may casue hungtask.
> > >>
> > >> [ 14.088665] F2FS-fs (nvme0n1): lz4 decompress failed, ret:-4155
> > >> [ 14.089851] F2FS-fs (nvme0n1): lz4 decompress failed, ret:-4155
> > >>
> > >> Signed-off-by: Fengnan Chang <fengnanch...@gmail.com>
> > >
> > > This commit message only explains the problem, not the solution. It
> > > should
> > > explain the solution too.
> > >
> > > Also, if it's fixing a bug, it needs 'Fixes' and 'Cc stable' tags.
> >
> > I’ll add more explanation and tags, thanks.
> >
> > >
> > >> diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
> > >> index 24824cd96f36..1764e3859262 100644
> > >> --- a/fs/f2fs/compress.c
> > >> +++ b/fs/f2fs/compress.c
> > >> @@ -1060,7 +1060,7 @@ static int prepare_compress_overwrite(struct
> > >> compress_ctx *cc,
> > >> sector_t last_block_in_bio;
> > >> unsigned fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT;
> > >> pgoff_t start_idx = start_idx_of_cluster(cc);
> > >> - int i, ret;
> > >> + int i, ret, retry_count = 3;
> > >>
> > >> retry:
> > >> ret = f2fs_is_compressed_cluster(cc->inode, start_idx);
> > >> @@ -1120,7 +1120,12 @@ static int prepare_compress_overwrite(struct
> > >> compress_ctx *cc,
> > >> f2fs_put_rpages(cc);
> > >> f2fs_unlock_rpages(cc, i + 1);
> > >> f2fs_destroy_compress_ctx(cc, true);
> > >> - goto retry;
> > >> + retry_count--;
> > >> + if (PageError(page) && !retry_count) {
> > >> + ret = -EIO;
> > >> + goto out;
> > >> + } else
> > >
> > > What is the purpose of using PG_error here?
> >
> > In this version, we set PG_error when compressed failed, so check PG_error
> > here.
> > Maybe we can remove PG_error in later?
> >
>
> Read I/O errors can be detected via PG_uptodate not being set. There
> shouldn't
> be any need for PG_error here.
Yeah, I get it now. Maybe we can remove PG_error in f2fs_verify_cluster too.
diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
index d315c2de136f..13c0bfe45804 100644
--- a/fs/f2fs/compress.c
+++ b/fs/f2fs/compress.c
@@ -1727,10 +1727,9 @@ static void __f2fs_decompress_end_io(struct
decompress_io_ctx *dic, bool failed,
continue;
/* PG_error was set if verity failed. */
- if (failed || PageError(rpage)) {
- ClearPageUptodate(rpage);
+ if (failed) {
/* will re-read again later */
- ClearPageError(rpage);
+ ClearPageUptodate(rpage);
} else {
SetPageUptodate(rpage);
}
@@ -1745,13 +1744,14 @@ static void f2fs_verify_cluster(struct
work_struct *work)
struct decompress_io_ctx *dic =
container_of(work, struct decompress_io_ctx, verity_work);
int i;
+ bool failed = false;
/* Verify the cluster's decompressed pages with fs-verity. */
for (i = 0; i < dic->cluster_size; i++) {
struct page *rpage = dic->rpages[i];
if (rpage && !fsverity_verify_page(rpage))
- SetPageError(rpage);
+ failed = true;
}
__f2fs_decompress_end_io(dic, false, true);
>
> - Eric
_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
