add get_name implementation for exports. (openzfs#16833)

This fixes a serious performance problem with NFS handling of large
directories, as the new get_name code is much more efficient than the
default zfs_readdir. This is actually part of
20232ec in 2.3. But I've taken only
the minimum code to implement get_name, and not the rest of the long
name changes.

Signed-off-by: Charles Hedrick <hedrick@rutgers.edu>
Co-authored-by: Charles L. Hedrick <hedrick@ncommunis.cs.rutgers.edu>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>

include/os/linux/zfs/sys/zfs_vnops_os.h

@@ -44,6 +44,7 @@ extern int zfs_write_simple(znode_t *zp, const void *data, size_t len,
     loff_t pos, size_t *resid);
 extern int zfs_lookup(znode_t *dzp, char *nm, znode_t **zpp, int flags,
     cred_t *cr, int *direntflags, pathname_t *realpnp);
+extern int zfs_get_name(znode_t *dzp, char *name, znode_t *zp);
 extern int zfs_create(znode_t *dzp, char *name, vattr_t *vap, int excl,
     int mode, znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp,
     zidmap_t *mnt_ns);

module/os/linux/zfs/zfs_vnops_os.c

@@ -532,6 +532,48 @@ zfs_lookup(znode_t *zdp, char *nm, znode_t **zpp, int flags, cred_t *cr,
 	return (error);
 }
 
+/*
+ * Perform a linear search in directory for the name of specific inode.
+ * Note we don't pass in the buffer size of name because it's hardcoded to
+ * NAME_MAX+1(256) in Linux.
+ *
+ *	IN:	dzp	- znode of directory to search.
+ *		zp	- znode of the target
+ *
+ *	OUT:	name	- dentry name of the target
+ *
+ *	RETURN:	0 on success, error code on failure.
+ */
+int
+zfs_get_name(znode_t *dzp, char *name, znode_t *zp)
+{
+	zfsvfs_t *zfsvfs = ZTOZSB(dzp);
+	int error = 0;
+
+	if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0)
+		return (error);
+
+	if ((error = zfs_verify_zp(zp)) != 0) {
+		zfs_exit(zfsvfs, FTAG);
+		return (error);
+	}
+
+	/* ctldir should have got their name in zfs_vget */
+	if (dzp->z_is_ctldir || zp->z_is_ctldir) {
+		zfs_exit(zfsvfs, FTAG);
+		return (ENOENT);
+	}
+
+	/* buffer len is hardcoded to 256 in Linux kernel */
+	error = zap_value_search(zfsvfs->z_os, dzp->z_id, zp->z_id,
+	    ZFS_DIRENT_OBJ(-1ULL), name);
+
+	zfs_exit(zfsvfs, FTAG);
+	return (error);
+}
+
+
+
 /*
  * Attempt to create a new entry in a directory.  If the entry
  * already exists, truncate the file if permissible, else return

module/os/linux/zfs/zpl_export.c

@@ -24,6 +24,7 @@
  */
 
 
+#include <sys/file.h>
 #include <sys/zfs_znode.h>
 #include <sys/zfs_vnops.h>
 #include <sys/zfs_ctldir.h>
@@ -109,6 +110,36 @@ zpl_fh_to_dentry(struct super_block *sb, struct fid *fh,
 	return (d_obtain_alias(ip));
 }
 
+/*
+ * In case the filesystem contains name longer than 255, we need to override
+ * the default get_name so we don't get buffer overflow. Unfortunately, since
+ * the buffer size is hardcoded in Linux, we will get ESTALE error in this
+ * case.
+ */
+static int
+zpl_get_name(struct dentry *parent, char *name, struct dentry *child)
+{
+	cred_t *cr = CRED();
+	fstrans_cookie_t cookie;
+	struct inode *dir = parent->d_inode;
+	struct inode *ip = child->d_inode;
+	int error;
+
+	if (!dir || !S_ISDIR(dir->i_mode))
+		return (-ENOTDIR);
+
+	crhold(cr);
+	cookie = spl_fstrans_mark();
+	spl_inode_lock_shared(dir);
+	error = -zfs_get_name(ITOZ(dir), name, ITOZ(ip));
+	spl_inode_unlock_shared(dir);
+	spl_fstrans_unmark(cookie);
+	crfree(cr);
+
+	return (error);
+}
+
+
 static struct dentry *
 zpl_get_parent(struct dentry *child)
 {
@@ -153,6 +184,7 @@ zpl_commit_metadata(struct inode *inode)
 const struct export_operations zpl_export_operations = {
 	.encode_fh		= zpl_encode_fh,
 	.fh_to_dentry		= zpl_fh_to_dentry,
+	.get_name		= zpl_get_name,
 	.get_parent		= zpl_get_parent,
 	.commit_metadata	= zpl_commit_metadata,
 };