On Fri, 19 Jun 2009 03:36:42 +0900 (JST), Ryusuke Konishi <[email protected]>
wrote:
> Hi,
> On Fri, 12 Jun 2009 13:23:44 +0900 (JST), Ryusuke Konishi wrote:
> > On Tue, 9 Jun 2009 13:04:04 -0600, "Dunphy, Bill" wrote:
> > >
> > > Ryusuke,
> > >
> > > I've simplified my test approach by simply copying a single 10 byte file
> > > repeatedly. When doing so, I get consistent NILFS log output at the
> > > time of hang. Examination of /proc/sysreq-trigger also shows the copy
> > > process to be in the same state as before (at sys_write).
> >
> > Thank you for sending me the trace log. I will look into the problem
> > later.
> >
> > Ryusuke Konishi
>
> In another thread, I've sent a patch that may cause the hang problem.
> Here, I attach the out-of-tree version of the patch.
s/cause/fix/
Ryusuke
> Does this make a difference in your problem ?
>
> Thanks,
> Ryusuke Konishi
> --
>
> diff --git a/fs/dat.c b/fs/dat.c
> index 036b05d..b05d9d4 100644
> --- a/fs/dat.c
> +++ b/fs/dat.c
> @@ -149,15 +149,6 @@ void nilfs_dat_commit_start(struct inode *dat, struct
> nilfs_palloc_req *req,
> entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
> req->pr_entry_bh, kaddr);
> entry->de_start = cpu_to_le64(nilfs_mdt_cno(dat));
> - if (entry->de_blocknr != cpu_to_le64(0) ||
> - entry->de_end != cpu_to_le64(NILFS_CNO_MAX)) {
> - printk(KERN_CRIT
> - "%s: vbn = %llu, start = %llu, end = %llu, pbn = %llu\n",
> - __func__, (unsigned long long)req->pr_entry_nr,
> - (unsigned long long)le64_to_cpu(entry->de_start),
> - (unsigned long long)le64_to_cpu(entry->de_end),
> - (unsigned long long)le64_to_cpu(entry->de_blocknr));
> - }
> entry->de_blocknr = cpu_to_le64(blocknr);
> kunmap_atomic(kaddr, KM_USER0);
>
> diff --git a/fs/segment.c b/fs/segment.c
> index 3cbb3fe..225d5ae 100644
> --- a/fs/segment.c
> +++ b/fs/segment.c
> @@ -2009,26 +2009,13 @@ static int nilfs_segctor_write(struct nilfs_sc_info
> *sci,
> err = nilfs_segbuf_write(segbuf, &wi);
>
> res = nilfs_segbuf_wait(segbuf, &wi);
> - err = unlikely(err) ? : res;
> + err = unlikely(err) ? err : res;
> if (unlikely(err))
> return err;
> }
> return 0;
> }
>
> -static int nilfs_page_has_uncleared_buffer(struct page *page)
> -{
> - struct buffer_head *head, *bh;
> -
> - head = bh = page_buffers(page);
> - do {
> - if (buffer_dirty(bh) && !list_empty(&bh->b_assoc_buffers))
> - return 1;
> - bh = bh->b_this_page;
> - } while (bh != head);
> - return 0;
> -}
> -
> static void __nilfs_end_page_io(struct page *page, int err)
> {
> if (!err) {
> @@ -2056,12 +2043,11 @@ static void nilfs_end_page_io(struct page *page, int
> err)
> if (!page)
> return;
>
> - if (buffer_nilfs_node(page_buffers(page)) &&
> - nilfs_page_has_uncleared_buffer(page))
> - /* For b-tree node pages, this function may be called twice
> - or more because they might be split in a segment.
> - This check assures that cleanup has been done for all
> - buffers in a split btnode page. */
> + if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page))
> + /*
> + * For b-tree node pages, this function may be called twice
> + * or more because they might be split in a segment.
> + */
> return;
>
> __nilfs_end_page_io(page, err);
> @@ -2125,7 +2111,7 @@ static void nilfs_segctor_abort_write(struct
> nilfs_sc_info *sci,
> }
> if (bh->b_page != fs_page) {
> nilfs_end_page_io(fs_page, err);
> - if (unlikely(fs_page == failed_page))
> + if (fs_page && fs_page == failed_page)
> goto done;
> fs_page = bh->b_page;
> }
> _______________________________________________
> users mailing list
> [email protected]
> https://www.nilfs.org/mailman/listinfo/users
_______________________________________________
users mailing list
[email protected]
https://www.nilfs.org/mailman/listinfo/users
