Commit 555dbf1a authored by Trond Myklebust's avatar Trond Myklebust Committed by Chuck Lever
Browse files

nfsd: Replace use of rwsem with errseq_t 



The nfsd_file nf_rwsem is currently being used to separate file write
and commit instances to ensure that we catch errors and apply them to
the correct write/commit.
We can improve scalability at the expense of a little accuracy (some
extra false positives) by replacing the nf_rwsem with more careful
use of the errseq_t mechanism to track errors across the different
operations.

Signed-off-by: default avatarTrond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: default avatarChuck Lever <chuck.lever@oracle.com>
[ cel: rebased on zero-verifier fix ]
parent f11ad7aa

fs/nfsd/filecache.c

+0 −1
Original line number Original line Diff line number Diff line
@@ -189,7 +189,6 @@ nfsd_file_alloc(struct inode *inode, unsigned int may, unsigned int hashval, 
				__set_bit(NFSD_FILE_BREAK_READ, &nf->nf_flags);

				__set_bit(NFSD_FILE_BREAK_READ, &nf->nf_flags);

		}

		}

		nf->nf_mark = NULL;

		nf->nf_mark = NULL;

		init_rwsem(&nf->nf_rwsem);

		trace_nfsd_file_alloc(nf);

		trace_nfsd_file_alloc(nf);

	}

	}

	return nf;

	return nf;

fs/nfsd/filecache.h

+0 −1
Original line number Original line Diff line number Diff line
@@ -46,7 +46,6 @@ struct nfsd_file { 
	refcount_t		nf_ref;

	refcount_t		nf_ref;

	unsigned char		nf_may;

	unsigned char		nf_may;

	struct nfsd_file_mark	*nf_mark;

	struct nfsd_file_mark	*nf_mark;

	struct rw_semaphore	nf_rwsem;

};

};





int nfsd_file_cache_init(void);

int nfsd_file_cache_init(void);

fs/nfsd/nfs4proc.c

+9 −7
Original line number Original line Diff line number Diff line
@@ -1510,6 +1510,9 @@ static void nfsd4_init_copy_res(struct nfsd4_copy *copy, bool sync) 




static ssize_t _nfsd_copy_file_range(struct nfsd4_copy *copy)

static ssize_t _nfsd_copy_file_range(struct nfsd4_copy *copy)

{

{

	struct file *dst = copy->nf_dst->nf_file;

	struct file *src = copy->nf_src->nf_file;

	errseq_t since;

	ssize_t bytes_copied = 0;

	ssize_t bytes_copied = 0;

	u64 bytes_total = copy->cp_count;

	u64 bytes_total = copy->cp_count;

	u64 src_pos = copy->cp_src_pos;

	u64 src_pos = copy->cp_src_pos;
@@ -1522,8 +1525,7 @@ static ssize_t _nfsd_copy_file_range(struct nfsd4_copy *copy) 
	do {

	do {

		if (kthread_should_stop())

		if (kthread_should_stop())

			break;

			break;

		bytes_copied = nfsd_copy_file_range(copy->nf_src->nf_file,

		bytes_copied = nfsd_copy_file_range(src, src_pos, dst, dst_pos,

				src_pos, copy->nf_dst->nf_file, dst_pos,

						    bytes_total);

						    bytes_total);

		if (bytes_copied <= 0)

		if (bytes_copied <= 0)

			break;

			break;
@@ -1534,11 +1536,11 @@ static ssize_t _nfsd_copy_file_range(struct nfsd4_copy *copy) 
	} while (bytes_total > 0 && !copy->cp_synchronous);

	} while (bytes_total > 0 && !copy->cp_synchronous);

	/* for a non-zero asynchronous copy do a commit of data */

	/* for a non-zero asynchronous copy do a commit of data */

	if (!copy->cp_synchronous && copy->cp_res.wr_bytes_written > 0) {

	if (!copy->cp_synchronous && copy->cp_res.wr_bytes_written > 0) {

		down_write(&copy->nf_dst->nf_rwsem);

		since = READ_ONCE(dst->f_wb_err);

		status = vfs_fsync_range(copy->nf_dst->nf_file,

		status = vfs_fsync_range(dst, copy->cp_dst_pos,

					 copy->cp_dst_pos,

					 copy->cp_res.wr_bytes_written, 0);

					 copy->cp_res.wr_bytes_written, 0);

		up_write(&copy->nf_dst->nf_rwsem);

		if (!status)

			status = filemap_check_wb_err(dst->f_mapping, since);

		if (!status)

		if (!status)

			copy->committed = true;

			copy->committed = true;

	}

	}

fs/nfsd/vfs.c

+15 −25
Original line number Original line Diff line number Diff line
@@ -522,10 +522,11 @@ __be32 nfsd4_clone_file_range(struct nfsd_file *nf_src, u64 src_pos, 
{

{

	struct file *src = nf_src->nf_file;

	struct file *src = nf_src->nf_file;

	struct file *dst = nf_dst->nf_file;

	struct file *dst = nf_dst->nf_file;

	errseq_t since;

	loff_t cloned;

	loff_t cloned;

	__be32 ret = 0;

	__be32 ret = 0;





	down_write(&nf_dst->nf_rwsem);

	since = READ_ONCE(dst->f_wb_err);

	cloned = vfs_clone_file_range(src, src_pos, dst, dst_pos, count, 0);

	cloned = vfs_clone_file_range(src, src_pos, dst, dst_pos, count, 0);

	if (cloned < 0) {

	if (cloned < 0) {

		ret = nfserrno(cloned);

		ret = nfserrno(cloned);
@@ -539,6 +540,8 @@ __be32 nfsd4_clone_file_range(struct nfsd_file *nf_src, u64 src_pos, 
		loff_t dst_end = count ? dst_pos + count - 1 : LLONG_MAX;

		loff_t dst_end = count ? dst_pos + count - 1 : LLONG_MAX;

		int status = vfs_fsync_range(dst, dst_pos, dst_end, 0);

		int status = vfs_fsync_range(dst, dst_pos, dst_end, 0);





		if (!status)

			status = filemap_check_wb_err(dst->f_mapping, since);

		if (!status)

		if (!status)

			status = commit_inode_metadata(file_inode(src));

			status = commit_inode_metadata(file_inode(src));

		if (status < 0) {

		if (status < 0) {
@@ -548,7 +551,6 @@ __be32 nfsd4_clone_file_range(struct nfsd_file *nf_src, u64 src_pos, 
		}

		}

	}

	}

out_err:

out_err:

	up_write(&nf_dst->nf_rwsem);

	return ret;

	return ret;

}

}
@@ -956,6 +958,7 @@ nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfsd_file *nf, 
	struct super_block	*sb = file_inode(file)->i_sb;

	struct super_block	*sb = file_inode(file)->i_sb;

	struct svc_export	*exp;

	struct svc_export	*exp;

	struct iov_iter		iter;

	struct iov_iter		iter;

	errseq_t		since;

	__be32			nfserr;

	__be32			nfserr;

	int			host_err;

	int			host_err;

	int			use_wgather;

	int			use_wgather;
@@ -993,8 +996,8 @@ nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfsd_file *nf, 
		flags |= RWF_SYNC;

		flags |= RWF_SYNC;





	iov_iter_kvec(&iter, WRITE, vec, vlen, *cnt);

	iov_iter_kvec(&iter, WRITE, vec, vlen, *cnt);

	since = READ_ONCE(file->f_wb_err);

	if (flags & RWF_SYNC) {

	if (flags & RWF_SYNC) {

		down_write(&nf->nf_rwsem);

		if (verf)

		if (verf)

			nfsd_copy_boot_verifier(verf,

			nfsd_copy_boot_verifier(verf,

					net_generic(SVC_NET(rqstp),

					net_generic(SVC_NET(rqstp),
@@ -1003,15 +1006,12 @@ nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfsd_file *nf, 
		if (host_err < 0)

		if (host_err < 0)

			nfsd_reset_boot_verifier(net_generic(SVC_NET(rqstp),

			nfsd_reset_boot_verifier(net_generic(SVC_NET(rqstp),

						 nfsd_net_id));

						 nfsd_net_id));

		up_write(&nf->nf_rwsem);

	} else {

	} else {

		down_read(&nf->nf_rwsem);

		if (verf)

		if (verf)

			nfsd_copy_boot_verifier(verf,

			nfsd_copy_boot_verifier(verf,

					net_generic(SVC_NET(rqstp),

					net_generic(SVC_NET(rqstp),

					nfsd_net_id));

					nfsd_net_id));

		host_err = vfs_iter_write(file, &iter, &pos, flags);

		host_err = vfs_iter_write(file, &iter, &pos, flags);

		up_read(&nf->nf_rwsem);

	}

	}

	if (host_err < 0) {

	if (host_err < 0) {

		nfsd_reset_boot_verifier(net_generic(SVC_NET(rqstp),

		nfsd_reset_boot_verifier(net_generic(SVC_NET(rqstp),
@@ -1021,6 +1021,9 @@ nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfsd_file *nf, 
	*cnt = host_err;

	*cnt = host_err;

	nfsd_stats_io_write_add(exp, *cnt);

	nfsd_stats_io_write_add(exp, *cnt);

	fsnotify_modify(file);

	fsnotify_modify(file);

	host_err = filemap_check_wb_err(file->f_mapping, since);

	if (host_err < 0)

		goto out_nfserr;





	if (stable && use_wgather) {

	if (stable && use_wgather) {

		host_err = wait_for_concurrent_writes(file);

		host_err = wait_for_concurrent_writes(file);
@@ -1101,19 +1104,6 @@ nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset, 
}

}





#ifdef CONFIG_NFSD_V3

#ifdef CONFIG_NFSD_V3

static int

nfsd_filemap_write_and_wait_range(struct nfsd_file *nf, loff_t offset,

				  loff_t end)

{

	struct address_space *mapping = nf->nf_file->f_mapping;

	int ret = filemap_fdatawrite_range(mapping, offset, end);



	if (ret)

		return ret;

	filemap_fdatawait_range_keep_errors(mapping, offset, end);

	return 0;

}



/*

/*

 * Commit all pending writes to stable storage.

 * Commit all pending writes to stable storage.

 *

 *
@@ -1144,25 +1134,25 @@ nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 
	if (err)

	if (err)

		goto out;

		goto out;

	if (EX_ISSYNC(fhp->fh_export)) {

	if (EX_ISSYNC(fhp->fh_export)) {

		int err2 = nfsd_filemap_write_and_wait_range(nf, offset, end);

		errseq_t since = READ_ONCE(nf->nf_file->f_wb_err);

		int err2;





		down_write(&nf->nf_rwsem);

		if (!err2)

		err2 = vfs_fsync_range(nf->nf_file, offset, end, 0);

		err2 = vfs_fsync_range(nf->nf_file, offset, end, 0);

		switch (err2) {

		switch (err2) {

		case 0:

		case 0:

			nfsd_copy_boot_verifier(verf, net_generic(nf->nf_net,

			nfsd_copy_boot_verifier(verf, net_generic(nf->nf_net,

						nfsd_net_id));

						nfsd_net_id));

			err2 = filemap_check_wb_err(nf->nf_file->f_mapping,

						    since);

			break;

			break;

		case -EINVAL:

		case -EINVAL:

			err = nfserr_notsupp;

			err = nfserr_notsupp;

			break;

			break;

		default:

		default:

			err = nfserrno(err2);

			nfsd_reset_boot_verifier(net_generic(nf->nf_net,

			nfsd_reset_boot_verifier(net_generic(nf->nf_net,

						 nfsd_net_id));

						 nfsd_net_id));

		}

		}

		up_write(&nf->nf_rwsem);

		err = nfserrno(err2);

	} else

	} else

		nfsd_copy_boot_verifier(verf, net_generic(nf->nf_net,

		nfsd_copy_boot_verifier(verf, net_generic(nf->nf_net,

					nfsd_net_id));

					nfsd_net_id));