nfs: handle request add failure properly

When we fail to queue a read page to IO descriptor,
we need to clean it up otherwise it is hanging around
preventing nfs module from being removed.

When we fail to queue a write page to IO descriptor,
we need to clean it up and also save the failure status
to open context. Then at file close, we can try to write
pages back again and drop the page if it fails to writeback
in .launder_page, which will be done in the next patch.

Signed-off-by: Peng Tao <tao.peng@primarydata.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>

5 files changed, 67 insertions, 31 deletions

diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index c7e8b87..74fb122 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -912,6 +912,12 @@ void nfs_file_clear_open_context(struct file *filp)
 	if (ctx) {
 		struct inode *inode = d_inode(ctx->dentry);
 
+		/*
+		 * We fatal error on write before. Try to writeback
+		 * every page again.
+		 */
+		if (ctx->error < 0)
+			invalidate_inode_pages2(inode->i_mapping);
 		filp->private_data = NULL;
 		spin_lock(&inode->i_lock);
 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h
index 313d554..68f773d 100644
--- a/fs/nfs/internal.h
+++ b/fs/nfs/internal.h
@@ -711,3 +711,17 @@ static inline u32 nfs_stateid_hash(nfs4_stateid *stateid)
 	return 0;
 }
 #endif
+
+static inline bool nfs_error_is_fatal(int err)
+{
+	switch (err) {
+	case -ERESTARTSYS:
+	case -EIO:
+	case -ENOSPC:
+	case -EROFS:
+	case -E2BIG:
+		return true;
+	default:
+		return false;
+	}
+}
diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 0fb3552..580207b 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -904,18 +904,9 @@ send_layoutget(struct pnfs_layout_hdr *lo,
 		lseg = nfs4_proc_layoutget(lgp, gfp_flags);
 	} while (lseg == ERR_PTR(-EAGAIN));
 
-	if (IS_ERR(lseg)) {
-		switch (PTR_ERR(lseg)) {
-		case -ERESTARTSYS:
-		case -EIO:
-		case -ENOSPC:
-		case -EROFS:
-		case -E2BIG:
-			break;
-		default:
-			return NULL;
-		}
-	} else
+	if (IS_ERR(lseg) && !nfs_error_is_fatal(PTR_ERR(lseg)))
+		lseg = NULL;
+	else
 		pnfs_layout_clear_fail_bit(lo,
 				pnfs_iomode_to_fail_bit(range->iomode));
 
diff --git a/fs/nfs/read.c b/fs/nfs/read.c
index 0bb5801..eb31e23 100644
--- a/fs/nfs/read.c
+++ b/fs/nfs/read.c
@@ -85,6 +85,23 @@ void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio)
 }
 EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds);
 
+static void nfs_readpage_release(struct nfs_page *req)
+{
+	struct inode *inode = d_inode(req->wb_context->dentry);
+
+	dprintk("NFS: read done (%s/%llu %d@%lld)\n", inode->i_sb->s_id,
+		(unsigned long long)NFS_FILEID(inode), req->wb_bytes,
+		(long long)req_offset(req));
+
+	if (nfs_page_group_sync_on_bit(req, PG_UNLOCKPAGE)) {
+		if (PageUptodate(req->wb_page))
+			nfs_readpage_to_fscache(inode, req->wb_page, 0);
+
+		unlock_page(req->wb_page);
+	}
+	nfs_release_request(req);
+}
+
 int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
 		       struct page *page)
 {
@@ -106,7 +123,10 @@ int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
 
 	nfs_pageio_init_read(&pgio, inode, false,
 			     &nfs_async_read_completion_ops);
-	nfs_pageio_add_request(&pgio, new);
+	if (!nfs_pageio_add_request(&pgio, new)) {
+		nfs_list_remove_request(new);
+		nfs_readpage_release(new);
+	}
 	nfs_pageio_complete(&pgio);
 
 	/* It doesn't make sense to do mirrored reads! */
@@ -118,23 +138,6 @@ int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
 	return pgio.pg_error < 0 ? pgio.pg_error : 0;
 }
 
-static void nfs_readpage_release(struct nfs_page *req)
-{
-	struct inode *inode = d_inode(req->wb_context->dentry);
-
-	dprintk("NFS: read done (%s/%llu %d@%lld)\n", inode->i_sb->s_id,
-		(unsigned long long)NFS_FILEID(inode), req->wb_bytes,
-		(long long)req_offset(req));
-
-	if (nfs_page_group_sync_on_bit(req, PG_UNLOCKPAGE)) {
-		if (PageUptodate(req->wb_page))
-			nfs_readpage_to_fscache(inode, req->wb_page, 0);
-
-		unlock_page(req->wb_page);
-	}
-	nfs_release_request(req);
-}
-
 static void nfs_page_group_set_uptodate(struct nfs_page *req)
 {
 	if (nfs_page_group_sync_on_bit(req, PG_UPTODATE))
@@ -361,6 +364,8 @@ readpage_async_filler(void *data, struct page *page)
 	if (len < PAGE_CACHE_SIZE)
 		zero_user_segment(page, len, PAGE_CACHE_SIZE);
 	if (!nfs_pageio_add_request(desc->pgio, new)) {
+		nfs_list_remove_request(new);
+		nfs_readpage_release(new);
 		error = desc->pgio->pg_error;
 		goto out_unlock;
 	}
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index 7b93164..9dafb08 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -545,6 +545,15 @@ try_again:
 	return head;
 }
 
+static void nfs_write_error_remove_page(struct nfs_page *req)
+{
+	nfs_unlock_request(req);
+	nfs_end_page_writeback(req);
+	nfs_release_request(req);
+	generic_error_remove_page(page_file_mapping(req->wb_page),
+				  req->wb_page);
+}
+
 /*
  * Find an associated nfs write request, and prepare to flush it out
  * May return an error if the user signalled nfs_wait_on_request().
@@ -567,8 +576,19 @@ static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
 
 	ret = 0;
 	if (!nfs_pageio_add_request(pgio, req)) {
-		nfs_redirty_request(req);
 		ret = pgio->pg_error;
+		/*
+		 * Remove the problematic req upon fatal errors,
+		 * while other dirty pages can still be around
+		 * until they get flushed.
+		 */
+		if (nfs_error_is_fatal(ret)) {
+			nfs_context_set_write_error(req->wb_context, ret);
+			nfs_write_error_remove_page(req);
+		} else {
+			nfs_redirty_request(req);
+			ret = -EAGAIN;
+		}
 	} else
 		nfs_add_stats(page_file_mapping(page)->host,
 				NFSIOS_WRITEPAGES, 1);