Re: [f2fs-dev] [RFC PATCH v5] f2fs: support data compression

From: Jaegeuk Kim
Date: Wed Jan 15 2020 - 16:38:31 EST


On 01/15, Chao Yu wrote:
> On 2020/1/15 6:48, Jaegeuk Kim wrote:
> > On 01/14, Chao Yu wrote:
> >> On 2020/1/14 0:11, Jaegeuk Kim wrote:
> >>> On 01/13, Chao Yu wrote:
> >>>> On 2020/1/12 2:02, Jaegeuk Kim wrote:
> >>>>> On 01/11, Chao Yu wrote:
> >>>>>> On 2020/1/11 7:52, Jaegeuk Kim wrote:
> >>>>>>> On 01/06, Jaegeuk Kim wrote:
> >>>>>>>> On 01/06, Chao Yu wrote:
> >>>>>>>>> On 2020/1/3 14:50, Chao Yu wrote:
> >>>>>>>>>> This works to me. Could you run fsstress tests on compressed root directory?
> >>>>>>>>>> It seems still there are some bugs.
> >>>>>>>>>
> >>>>>>>>> Jaegeuk,
> >>>>>>>>>
> >>>>>>>>> Did you mean running por_fsstress testcase?
> >>>>>>>>>
> >>>>>>>>> Now, at least I didn't hit any problem for normal fsstress case.
> >>>>>>>>
> >>>>>>>> Yup. por_fsstress
> >>>>>>>
> >>>>>>> Please check https://github.com/jaegeuk/f2fs/commits/g-dev-test.
> >>>>>>> I've fixed
> >>>>>>> - truncation offset
> >>>>>>> - i_compressed_blocks and its lock coverage
> >>>>>>> - error handling
> >>>>>>> - etc
> >>>>>>
> >>>>>> I changed as below, and por_fsstress stops panic the system.
> >>>>>>
> >>>>>> Could you merge all these fixes into original patch?
> >>>>>
> >>>>> Yup, let m roll up some early patches first once test results become good.
> >>>>
> >>>> I didn't encounter issue any more, how about por_fsstress test result in your
> >>>> enviornment? If there is, please share the call stack with me.
> >>>
> >>> Sure, will do, once I hit an issue. BTW, I'm hitting another unreacheable nat
> >>> entry issue during por_stress without compression. :(
> >>
> >> Did you enable any features during por_fsstress test?
> >>
> >> I only hit below warning during por_fsstress test on image w/o compression.
> >>
> >> ------------[ cut here ]------------
> >> WARNING: CPU: 10 PID: 33483 at fs/fs-writeback.c:1448 __writeback_single_inode+0x28c/0x340
> >> Call Trace:
> >> writeback_single_inode+0xad/0x120
> >> sync_inode_metadata+0x3d/0x60
> >> f2fs_sync_inode_meta+0x90/0xe0 [f2fs]
> >> block_operations+0x17c/0x360 [f2fs]
> >> f2fs_write_checkpoint+0x101/0xff0 [f2fs]
> >> f2fs_sync_fs+0xa8/0x130 [f2fs]
> >> f2fs_do_sync_file+0x19c/0x880 [f2fs]
> >> do_fsync+0x38/0x60
> >> __x64_sys_fsync+0x10/0x20
> >> do_syscall_64+0x5f/0x220
> >> entry_SYSCALL_64_after_hwframe+0x44/0xa9
> >
> > Does gc_mutex patch fix this?
>
> No, gc_mutex patch fixes another problem.
>
> BTW, it looks like a bug of VFS.

One fix below and rerun tests now.

---
fs/f2fs/compress.c | 18 ++++++++++--------
1 file changed, 10 insertions(+), 8 deletions(-)

diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
index 2480bff1e115..45a6f20ceb3e 100644
--- a/fs/f2fs/compress.c
+++ b/fs/f2fs/compress.c
@@ -556,10 +556,8 @@ static int f2fs_compressed_blocks(struct compress_ctx *cc)

blkaddr = datablock_addr(dn.inode,
dn.node_page, dn.ofs_in_node + i);
- if (blkaddr != NULL_ADDR) {
+ if (blkaddr != NULL_ADDR)
ret++;
- break;
- }
}
}
fail:
@@ -955,10 +953,14 @@ static int f2fs_write_raw_pages(struct compress_ctx *cc,
enum iostat_type io_type)
{
struct address_space *mapping = cc->inode->i_mapping;
- int i, _submitted, compr_blocks, ret;
- int err = 0;
+ int _submitted, compr_blocks, ret;
+ int i = -1, err = 0;

compr_blocks = f2fs_compressed_blocks(cc);
+ if (compr_blocks < 0) {
+ err = compr_blocks;
+ goto out_err;
+ }

for (i = 0; i < cc->cluster_size; i++) {
if (!cc->rpages[i])
@@ -997,7 +999,7 @@ static int f2fs_write_raw_pages(struct compress_ctx *cc,
out_fail:
/* TODO: revoke partially updated block addresses */
BUG_ON(compr_blocks);
-
+out_err:
for (++i; i < cc->cluster_size; i++) {
if (!cc->rpages[i])
continue;
@@ -1020,9 +1022,9 @@ int f2fs_write_multi_pages(struct compress_ctx *cc,
*submitted = 0;
if (cluster_may_compress(cc)) {
err = f2fs_compress_pages(cc);
- if (err == -EAGAIN)
+ if (err == -EAGAIN) {
goto write;
- else if (err) {
+ } else if (err) {
f2fs_put_rpages_wbc(cc, wbc, true, 1);
goto destroy_out;
}
--
2.24.0.525.g8f36a354ae-goog