Linux: add zpl_drop_inode

This is an attempt to rectify the situation in zfs_zget() on Linux,
where we might be racing with inode reclaim from the kernel.

The code is ugly and I'm not sure it's perfectly OK to do it this way
anyway.

I am by no means sure the refcount check is right, lets develop this
idea in the open, I'm eager to see what the testbots say here.

Signed-off-by: Pavel Snajdr <snajpa@snajpa.net>

module/os/linux/zfs/zfs_znode_os.c

@@ -1053,7 +1053,6 @@ zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
 
 	*zpp = NULL;
 
-again:
 	zh = zfs_znode_hold_enter(zfsvfs, obj_num);
 
 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
@@ -1076,55 +1075,20 @@ zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
 	if (hdl != NULL) {
 		zp = sa_get_userdata(hdl);
 
-
 		/*
 		 * Since "SA" does immediate eviction we
 		 * should never find a sa handle that doesn't
 		 * know about the znode.
 		 */
 
 		ASSERT3P(zp, !=, NULL);
+		VERIFY3P(igrab(ZTOI(zp)), !=, NULL);
+		*zpp = zp;
 
-		mutex_enter(&zp->z_lock);
-		ASSERT3U(zp->z_id, ==, obj_num);
-		/*
-		 * If zp->z_unlinked is set, the znode is already marked
-		 * for deletion and should not be discovered. Check this
-		 * after checking igrab() due to fsetxattr() & O_TMPFILE.
-		 *
-		 * If igrab() returns NULL the VFS has independently
-		 * determined the inode should be evicted and has
-		 * called iput_final() to start the eviction process.
-		 * The SA handle is still valid but because the VFS
-		 * requires that the eviction succeed we must drop
-		 * our locks and references to allow the eviction to
-		 * complete.  The zfs_zget() may then be retried.
-		 *
-		 * This unlikely case could be optimized by registering
-		 * a sops->drop_inode() callback.  The callback would
-		 * need to detect the active SA hold thereby informing
-		 * the VFS that this inode should not be evicted.
-		 */
-		if (igrab(ZTOI(zp)) == NULL) {
-			if (zp->z_unlinked)
-				err = SET_ERROR(ENOENT);
-			else
-				err = SET_ERROR(EAGAIN);
-		} else {
-			*zpp = zp;
-			err = 0;
-		}
-
-		mutex_exit(&zp->z_lock);
 		sa_buf_rele(db, NULL);
 		zfs_znode_hold_exit(zfsvfs, zh);
 
-		if (err == EAGAIN) {
-			/* inode might need this to finish evict */
-			cond_resched();
-			goto again;
-		}
-		return (err);
+		return 0;
 	}
 
 	/*

module/os/linux/zfs/zpl_super.c

@@ -378,12 +378,30 @@ zpl_prune_sb(uint64_t nr_to_scan, void *arg)
 	(void) -zfs_prune(sb, nr_to_scan, &objects);
 }
 
+static int
+zpl_drop_inode(struct inode *ip)
+{
+	znode_t *zp = ITOZ(ip);
+	dmu_buf_t *db;
+	int error;
+
+	if (!zp->z_unlinked && zp->z_sa_hdl &&
+	    (db = sa_get_db(zp->z_sa_hdl)) &&
+	    dmu_buf_refcount(db))
+		return (1);
+
+	error = generic_drop_inode(ip);
+
+	return (error);
+}
+
 const struct super_operations zpl_super_operations = {
 	.alloc_inode		= zpl_inode_alloc,
 	.destroy_inode		= zpl_inode_destroy,
 	.dirty_inode		= zpl_dirty_inode,
 	.write_inode		= NULL,
 	.evict_inode		= zpl_evict_inode,
+	.drop_inode		= zpl_drop_inode,
 	.put_super		= zpl_put_super,
 	.sync_fs		= zpl_sync_fs,
 	.statfs			= zpl_statfs,