[lustre-devel] [PATCH 1/3] lustre: use generic_error_remove_page()

James Simmons jsimmons at infradead.org
Mon Jun 25 17:26:31 PDT 2018


> >> On Jun 24, 2018, at 8:02 PM, NeilBrown <neilb at suse.com> wrote:
> >> 
> >> 
> >> lustre's internal ll_invalidate_page() is behaviourally identical to
> >> generic_error_remove_page().
> >> In the case of lustre it isn't a memory hardware error that requires
> >> the page being invalidated, it is the loss of a lock, which will like
> >> result in the data changing on the server.
> >> In either case, we don't want the page to be accessed any more, so the
> >> same removal is appropriate.
> >> 
> >> Signed-off-by: NeilBrown <neilb at suse.com>
> >> ---
> >> 
> >> I've replaced
> >> [PATCH 08/24] lustre: use truncate_inode_page in place of truncate_complete_page
> >> with 3 patches, this and the next two.
> >
> > This looks reasonable. Are you running any tests on this?
> 
> Just the sanity tests on a 4-node vcluster.  They haven't caused any
> noticeable regressions.
> 

I also have tested this patch set using a single client node with
a stand alone MGS server, 2 MDS servers with 2 MDTs each, and a
single OSS server with 2 OST.

> >
> > Acked-by: Oleg Drokin <green at linuxhacker.ru>
> 
> Thanks,
> NeilBrown
> 
> >
> >> 
> >> Thanks,
> >> NeilBrown
> >> 
> >> 
> >> drivers/staging/lustre/lustre/llite/llite_internal.h | 17 -----------------
> >> drivers/staging/lustre/lustre/llite/vvp_io.c         |  2 +-
> >> drivers/staging/lustre/lustre/llite/vvp_page.c       |  2 +-
> >> 3 files changed, 2 insertions(+), 19 deletions(-)
> >> 
> >> diff --git a/drivers/staging/lustre/lustre/llite/llite_internal.h b/drivers/staging/lustre/lustre/llite/llite_internal.h
> >> index c08a6e14b6d7..22dcabf6de0f 100644
> >> --- a/drivers/staging/lustre/lustre/llite/llite_internal.h
> >> +++ b/drivers/staging/lustre/lustre/llite/llite_internal.h
> >> @@ -928,23 +928,6 @@ void policy_from_vma(union ldlm_policy_data *policy, struct vm_area_struct *vma,
> >> struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
> >> 			       size_t count);
> >> 
> >> -static inline void ll_invalidate_page(struct page *vmpage)
> >> -{
> >> -	struct address_space *mapping = vmpage->mapping;
> >> -	loff_t offset = vmpage->index << PAGE_SHIFT;
> >> -
> >> -	LASSERT(PageLocked(vmpage));
> >> -	if (!mapping)
> >> -		return;
> >> -
> >> -	/*
> >> -	 * truncate_complete_page() calls
> >> -	 * a_ops->invalidatepage()->cl_page_delete()->vvp_page_delete().
> >> -	 */
> >> -	ll_teardown_mmaps(mapping, offset, offset + PAGE_SIZE);
> >> -	truncate_complete_page(mapping, vmpage);
> >> -}
> >> -
> >> #define    ll_s2sbi(sb)	(s2lsi(sb)->lsi_llsbi)
> >> 
> >> /* don't need an addref as the sb_info should be holding one */
> >> diff --git a/drivers/staging/lustre/lustre/llite/vvp_io.c b/drivers/staging/lustre/lustre/llite/vvp_io.c
> >> index e7a4778e02e4..5a67955974ad 100644
> >> --- a/drivers/staging/lustre/lustre/llite/vvp_io.c
> >> +++ b/drivers/staging/lustre/lustre/llite/vvp_io.c
> >> @@ -1098,7 +1098,7 @@ static int vvp_io_fault_start(const struct lu_env *env,
> >> 	LASSERT(PageLocked(vmpage));
> >> 
> >> 	if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_FAULT_TRUNC_RACE))
> >> -		ll_invalidate_page(vmpage);
> >> +		generic_error_remove_page(vmpage->mapping, vmpage);
> >> 
> >> 	size = i_size_read(inode);
> >> 	/* Though we have already held a cl_lock upon this page, but
> >> diff --git a/drivers/staging/lustre/lustre/llite/vvp_page.c b/drivers/staging/lustre/lustre/llite/vvp_page.c
> >> index 6eb0565ddc22..dcc4d8faa0cd 100644
> >> --- a/drivers/staging/lustre/lustre/llite/vvp_page.c
> >> +++ b/drivers/staging/lustre/lustre/llite/vvp_page.c
> >> @@ -147,7 +147,7 @@ static void vvp_page_discard(const struct lu_env *env,
> >> 	if (vpg->vpg_defer_uptodate && !vpg->vpg_ra_used)
> >> 		ll_ra_stats_inc(vmpage->mapping->host, RA_STAT_DISCARDED);
> >> 
> >> -	ll_invalidate_page(vmpage);
> >> +	generic_error_remove_page(vmpage->mapping, vmpage);
> >> }
> >> 
> >> static void vvp_page_delete(const struct lu_env *env,
> >> -- 
> >> 2.14.0.rc0.dirty
> >> 
> 


More information about the lustre-devel mailing list