Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   ChenXiaoSong <chenxiaosong2@huawei.com>
To:     <trond.myklebust@hammerspace.com>, <anna@kernel.org>,
        <smayhew@redhat.com>
CC:     <linux-nfs@vger.kernel.org>, <linux-kernel@vger.kernel.org>,
        <chenxiaosong2@huawei.com>, <liuyongqiang13@huawei.com>,
        <yi.zhang@huawei.com>, <zhangxiaoxu5@huawei.com>
Subject: [PATCH -next,v2 3/3] Revert "nfs: nfs_file_write() should check for writeback errors"
Date:   Fri, 1 Apr 2022 11:44:09 +0800
Message-ID: <20220401034409.256770-4-chenxiaosong2@huawei.com>
In-Reply-To: <20220401034409.256770-1-chenxiaosong2@huawei.com>
References: <20220401034409.256770-1-chenxiaosong2@huawei.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 7BIT
Content-Type:   text/plain; charset=US-ASCII
Precedence: bulk

This reverts commit ce368536dd614452407dc31e2449eb84681a06af.

Second `dd` of the following reproducer will be very very slow.
filemap_sample_wb_err() always return 0 if old wb error have not been consumed.
filemap_check_wb_err() will return the old error even if there is no new
writeback error between filemap_sample_wb_err() and filemap_check_wb_err().
Then nfs_need_check_write() always return true, nfs_wb_all() will be called
every time in nfs_file_write().

Reproducer:
        nfs server            |       nfs client
 -----------------------------|---------------------------------------------
 # No space left on server    |
 fallocate -l 100G /svr/nospc |
                              | mount -t nfs $nfs_server_ip:/ /mnt
                              |
                              | # Expected error: No space left on device
                              | dd if=/dev/zero of=/mnt/file count=1 ibs=1M
                              |
                              | # Release space on mountpoint
                              | rm /mnt/nospc
                              |
                              | # very very slow
                              | dd if=/dev/zero of=/mnt/file count=1 ibs=1M

generic_perform_write() detect wb error by calling filemap_check_errors():
  generic_perform_write
    nfs_write_end
      nfs_wb_all
        filemap_write_and_wait
          filemap_write_and_wait_range
            filemap_check_errors

filemap_fdatawait_range() detect wb error by calling filemap_check_errors():
  filemap_fdatawait_range
    __filemap_fdatawait_range
    filemap_check_errors

generic_write_sync() will also detect wb error by calling nfs_file_fsync():
  generic_write_sync
    vfs_fsync_range
      nfs_file_fsync
        file_write_and_wait_range
          file_check_and_advance_wb_err
            errseq_check_and_advance

When writeback error is detected in nfs_file_write(), we just goto label "out",
will not goto nfs_need_check_write(). So we remove the useless and problematic
checking.

Signed-off-by: ChenXiaoSong <chenxiaosong2@huawei.com>
---
 fs/nfs/file.c | 12 +++---------
 1 file changed, 3 insertions(+), 9 deletions(-)

diff --git a/fs/nfs/file.c b/fs/nfs/file.c
index 353f1f832519..49f1485a30bb 100644
--- a/fs/nfs/file.c
+++ b/fs/nfs/file.c
@@ -597,14 +597,12 @@ static const struct vm_operations_struct nfs_file_vm_ops = {
 	.page_mkwrite = nfs_vm_page_mkwrite,
 };
 
-static int nfs_need_check_write(struct file *filp, struct inode *inode,
-				int error)
+static int nfs_need_check_write(struct file *filp, struct inode *inode)
 {
 	struct nfs_open_context *ctx;
 
 	ctx = nfs_file_open_context(filp);
-	if (nfs_error_is_fatal_on_server(error) ||
-	    nfs_ctx_key_to_expire(ctx, inode))
+	if (nfs_ctx_key_to_expire(ctx, inode))
 		return 1;
 	return 0;
 }
@@ -615,8 +613,6 @@ ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
 	struct inode *inode = file_inode(file);
 	unsigned int mntflags = NFS_SERVER(inode)->flags;
 	ssize_t result, written;
-	errseq_t since;
-	int error;
 
 	result = nfs_key_timeout_notify(file, inode);
 	if (result)
@@ -641,7 +637,6 @@ ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
 
 	nfs_clear_invalid_mapping(file->f_mapping);
 
-	since = filemap_sample_wb_err(file->f_mapping);
 	nfs_start_io_write(inode);
 	result = generic_write_checks(iocb, from);
 	if (result > 0) {
@@ -675,8 +670,7 @@ ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
 		goto out;
 
 	/* Return error values */
-	error = filemap_check_wb_err(file->f_mapping, since);
-	if (nfs_need_check_write(file, inode, error)) {
+	if (nfs_need_check_write(file, inode)) {
 		int err = nfs_wb_all(inode);
 		if (err < 0)
 			result = err;
-- 
2.31.1