Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

cross-origin iframes on by default with depth limits

Nick Sweetinge725edb62ac4
diff --git a/browser_use/browser/profile.py b/browser_use/browser/profile.py
index 06b135497..a52ae8273 100644
--- a/browser_use/browser/profile.py
+++ b/browser_use/browser/profile.py
@@ -631,7 +631,7 @@ class BrowserProfile(BrowserConnectArgs, BrowserLaunchPersistentContextArgs, Bro
description='Window position to use for the browser x,y from the top left when headless=False.',
)
cross_origin_iframes: bool = Field(
- default=False,
+ default=True,
description='Enable cross-origin iframe support (OOPIF/Out-of-Process iframes). When False (default), only same-origin frames are processed to avoid complexity and hanging.',
)
diff --git a/browser_use/browser/watchdogs/popups_watchdog.py b/browser_use/browser/watchdogs/popups_watchdog.py
index f517838a2..e04914e27 100644
--- a/browser_use/browser/watchdogs/popups_watchdog.py
+++ b/browser_use/browser/watchdogs/popups_watchdog.py
@@ -53,15 +53,18 @@ class PopupsWatchdog(BaseWatchdog):
self.logger.debug(f"πŸ”” JavaScript {dialog_type} dialog detected: '{message[:50]}...' - accepting immediately")
# Dispatch the event first so tests can observe it
- event = self.browser_session.event_bus.dispatch(
- DialogOpenedEvent(
- frame_id=frame_id,
- dialog_type=dialog_type,
- message=message,
- url=url,
+ try:
+ event = self.browser_session.event_bus.dispatch(
+ DialogOpenedEvent(
+ frame_id=frame_id,
+ dialog_type=dialog_type,
+ message=message,
+ url=url,
+ )
)
- )
- await event.event_result(raise_if_none=False, raise_if_any=True, timeout=5.0)
+ await event.event_result(raise_if_none=False, raise_if_any=True, timeout=5.0)
+ except Exception as e:
+ self.logger.error(f'Failed to dispatch DialogOpenedEvent callback event: {event} error: {type(e).__name__}: {e}')
# Accept the dialog immediately to unblock the browser
try:
diff --git a/browser_use/dom/serializer/serializer.py b/browser_use/dom/serializer/serializer.py
index 26ef9b62a..8e7af86bd 100644
--- a/browser_use/dom/serializer/serializer.py
+++ b/browser_use/dom/serializer/serializer.py
@@ -117,13 +117,17 @@ class DOMTreeSerializer:
return self._clickable_cache[node.node_id]
- def _create_simplified_tree(self, node: EnhancedDOMTreeNode) -> SimplifiedNode | None:
+ def _create_simplified_tree(self, node: EnhancedDOMTreeNode, depth: int = 0) -> SimplifiedNode | None:
"""Step 1: Create a simplified tree with enhanced element detection."""
+ # Prevent infinite recursion by limiting depth to 30 levels (reduced for stress test pages)
+ if depth > 30:
+ return None
+
if node.node_type == NodeType.DOCUMENT_NODE:
# for all cldren including shadow roots
for child in node.children_and_shadow_roots:
- simplified_child = self._create_simplified_tree(child)
+ simplified_child = self._create_simplified_tree(child, depth + 1)
if simplified_child:
return simplified_child
@@ -133,7 +137,7 @@ class DOMTreeSerializer:
# Super simple pass-through for shadow DOM elements
simplified = SimplifiedNode(original_node=node, children=[])
for child in node.children_and_shadow_roots:
- simplified_child = self._create_simplified_tree(child)
+ simplified_child = self._create_simplified_tree(child, depth + 1)
if simplified_child:
simplified.children.append(simplified_child)
return simplified
@@ -144,13 +148,9 @@ class DOMTreeSerializer:
return None
if node.node_name == 'IFRAME':
- if node.content_document:
- simplified = SimplifiedNode(original_node=node, children=[])
- for child in node.content_document.children:
- simplified_child = self._create_simplified_tree(child)
- if simplified_child:
- simplified.children.append(simplified_child)
- return simplified
+ # For stress test pages with deeply nested iframes, skip content entirely
+ # to prevent infinite recursion
+ return SimplifiedNode(original_node=node, children=[])
# Use enhanced scoring for inclusion decision
is_interactive = self._is_interactive_cached(node)
@@ -167,7 +167,7 @@ class DOMTreeSerializer:
# Process children
for child in node.children_and_shadow_roots:
- simplified_child = self._create_simplified_tree(child)
+ simplified_child = self._create_simplified_tree(child, depth + 1)
if simplified_child:
simplified.children.append(simplified_child)
diff --git a/browser_use/dom/service.py b/browser_use/dom/service.py
index 9076ece75..3a33be449 100644
--- a/browser_use/dom/service.py
+++ b/browser_use/dom/service.py
@@ -27,6 +27,10 @@ from browser_use.dom.views import (
if TYPE_CHECKING:
from browser_use.browser.session import BrowserSession
+# Global limits to prevent iframe explosion
+MAX_TOTAL_IFRAMES = 3 # Maximum number of iframe documents to process (very conservative to prevent explosions)
+MAX_IFRAME_DEPTH = 1 # Maximum depth for cross-origin iframe recursion (only 1 level deep)
+
class DomService:
"""
@@ -407,16 +411,22 @@ class DomService:
end = time.time()
cdp_timing = {'cdp_calls_total': end - start}
- # DEBUG: Log snapshot info
+ # DEBUG: Log snapshot info and limit documents to prevent explosion
if snapshot and 'documents' in snapshot:
+ original_doc_count = len(snapshot['documents'])
+ # Limit to MAX_TOTAL_IFRAMES documents to prevent iframe explosion
+ if original_doc_count > MAX_TOTAL_IFRAMES:
+ self.logger.warning(f'⚠️ Limiting processing of {original_doc_count} iframes on page to only first {MAX_TOTAL_IFRAMES} to prevent crashes!')
+ snapshot['documents'] = snapshot['documents'][:MAX_TOTAL_IFRAMES]
+
total_nodes = sum(len(doc.get('nodes', [])) for doc in snapshot['documents'])
self.logger.debug(
- f'πŸ” DEBUG: Snapshot contains {len(snapshot["documents"])} documents with {total_nodes} total nodes'
+ f'πŸ” DEBUG: Snapshot contains {len(snapshot["documents"])} frames with {total_nodes} total nodes'
)
# Log iframe-specific info
for doc_idx, doc in enumerate(snapshot['documents']):
if doc_idx > 0: # Not the main document
- self.logger.debug(f'πŸ” DEBUG: Document {doc_idx} has {len(doc.get("nodes", []))} nodes')
+ self.logger.debug(f'πŸ” DEBUG: Iframe #{doc_idx} {doc.get("frameId", "no-frame-id")} {doc.get("url", "no-url")} has {len(doc.get("nodes", []))} nodes')
return TargetAllTrees(
snapshot=snapshot,
@@ -431,6 +441,7 @@ class DomService:
target_id: TargetID,
initial_html_frames: list[EnhancedDOMTreeNode] | None = None,
initial_total_frame_offset: DOMRect | None = None,
+ iframe_depth: int = 0,
) -> EnhancedDOMTreeNode:
"""Get the DOM tree for a specific target.
@@ -438,6 +449,7 @@ class DomService:
target_id: Target ID of the page to get the DOM tree for.
initial_html_frames: List of HTML frame nodes encountered so far
initial_total_frame_offset: Accumulated coordinate offset
+ iframe_depth: Current depth of iframe nesting to prevent infinite recursion
"""
trees = await self._get_all_trees(target_id)
@@ -617,33 +629,61 @@ class DomService:
# TODO: hacky way to disable cross origin iframes for now
self.cross_origin_iframes and node['nodeName'].upper() == 'IFRAME' and node.get('contentDocument', None) is None
): # None meaning there is no content
- # Use get_all_frames to find the iframe's target
- frame_id = node.get('frameId', None)
- if frame_id:
- all_frames, _ = await self.browser_session.get_all_frames()
- frame_info = all_frames.get(frame_id)
- iframe_document_target = None
- if frame_info and frame_info.get('frameTargetId'):
- # Get the target info for this iframe
- targets = await self.browser_session.cdp_client.send.Target.getTargets()
- iframe_document_target = next(
- (t for t in targets['targetInfos'] if t['targetId'] == frame_info['frameTargetId']), None
- )
+ # Check iframe depth to prevent infinite recursion
+ if iframe_depth >= MAX_IFRAME_DEPTH:
+ self.logger.debug(f'Skipping iframe at depth {iframe_depth} to prevent infinite recursion (max depth: {MAX_IFRAME_DEPTH})')
else:
- iframe_document_target = None
- # if target actually exists in one of the frames, just recursively build the dom tree for it
- if iframe_document_target:
- self.logger.debug(f'Getting content document for iframe {node.get("frameId", None)}')
- content_document = await self.get_dom_tree(
- target_id=iframe_document_target.get('targetId'),
- # TODO: experiment with this values -> not sure whether the whole cross origin iframe should be ALWAYS included as soon as some part of it is visible or not.
- # Current config: if the cross origin iframe is AT ALL visible, then just include everything inside of it!
- # initial_html_frames=updated_html_frames,
- initial_total_frame_offset=total_frame_offset,
- )
-
- dom_tree_node.content_document = content_document
- dom_tree_node.content_document.parent_node = dom_tree_node
+ # Check if iframe is visible and large enough (>= 200px in both dimensions)
+ should_process_iframe = False
+
+ # First check if the iframe element itself is visible
+ if dom_tree_node.is_visible:
+ # Check iframe dimensions
+ if dom_tree_node.snapshot_node and dom_tree_node.snapshot_node.bounds:
+ bounds = dom_tree_node.snapshot_node.bounds
+ width = bounds.get('width', 0)
+ height = bounds.get('height', 0)
+
+ # Only process if iframe is at least 200px in both dimensions
+ if width >= 200 and height >= 200:
+ should_process_iframe = True
+ self.logger.debug(f'Processing cross-origin iframe: visible=True, width={width}, height={height}')
+ else:
+ self.logger.debug(f'Skipping small cross-origin iframe: width={width}, height={height} (needs >= 200px)')
+ else:
+ self.logger.debug('Skipping cross-origin iframe: no bounds available')
+ else:
+ self.logger.debug('Skipping invisible cross-origin iframe')
+
+ if should_process_iframe:
+ # Use get_all_frames to find the iframe's target
+ frame_id = node.get('frameId', None)
+ if frame_id:
+ all_frames, _ = await self.browser_session.get_all_frames()
+ frame_info = all_frames.get(frame_id)
+ iframe_document_target = None
+ if frame_info and frame_info.get('frameTargetId'):
+ # Get the target info for this iframe
+ targets = await self.browser_session.cdp_client.send.Target.getTargets()
+ iframe_document_target = next(
+ (t for t in targets['targetInfos'] if t['targetId'] == frame_info['frameTargetId']), None
+ )
+ else:
+ iframe_document_target = None
+ # if target actually exists in one of the frames, just recursively build the dom tree for it
+ if iframe_document_target:
+ self.logger.debug(f'Getting content document for iframe {node.get("frameId", None)} at depth {iframe_depth + 1}')
+ content_document = await self.get_dom_tree(
+ target_id=iframe_document_target.get('targetId'),
+ # TODO: experiment with this values -> not sure whether the whole cross origin iframe should be ALWAYS included as soon as some part of it is visible or not.
+ # Current config: if the cross origin iframe is AT ALL visible, then just include everything inside of it!
+ # initial_html_frames=updated_html_frames,
+ initial_total_frame_offset=total_frame_offset,
+ iframe_depth=iframe_depth + 1,
+ )
+
+ dom_tree_node.content_document = content_document
+ dom_tree_node.content_document.parent_node = dom_tree_node
return dom_tree_node