| // Copyright 2024 The Chromium Authors |
| // Use of this source code is governed by a BSD-style license that can be |
| // found in the LICENSE file. |
| |
| #ifndef CONTENT_BROWSER_AGENT_CLUSTER_KEY_H_ |
| #define CONTENT_BROWSER_AGENT_CLUSTER_KEY_H_ |
| |
| #include <optional> |
| #include <variant> |
| |
| #include "content/browser/security/coop/cross_origin_isolation_mode.h" |
| #include "content/common/content_export.h" |
| #include "url/gurl.h" |
| #include "url/origin.h" |
| |
| namespace content { |
| |
| // AgentClusterKey represents the implementation in the browser process of the |
| // AgentClusterKey concept of the HTML spec: |
| // https://html.spec.whatwg.org/multipage/webappapis.html#agent-cluster-key |
| // |
| // SiteInstances have an AgentClusterKey in their SiteInfo, which represents the |
| // AgentClusterKey of the execution contexts hosted by the SiteInstance. In a |
| // BrowsingInstance, all regular web execution contexts needing the same |
| // AgentClusterKey are hosted in the same SiteInstance. There can be exceptions |
| // for non-regular web contexts, such as Guest Views, as they may require to be |
| // hosted in a separate SiteInstance for security reasons. |
| // |
| // The AgentClusterKey is computed upon navigation, or when launching a worker. |
| // It is then passed to RenderFrameHostManager to determine which SiteInstance |
| // is appropriate to host the execution context. |
| // TODO(crbug.com/342365078): Currently, AgentClusterKey is only computed when a |
| // document has a Document-Isolation-Policy. Compute it on all navigations. Once |
| // this is properly done, use the AgentClusterKey to replace the site URL in |
| // SiteInfo, as it will only duplicate the information in AgentClusterKey. |
| class CONTENT_EXPORT AgentClusterKey { |
| public: |
| // Cross-origin isolated agent clusters have an additional isolation key. |
| struct CONTENT_EXPORT CrossOriginIsolationKey { |
| CrossOriginIsolationKey( |
| const url::Origin& common_coi_origin, |
| CrossOriginIsolationMode cross_origin_isolation_mode); |
| CrossOriginIsolationKey(const CrossOriginIsolationKey& other); |
| ~CrossOriginIsolationKey(); |
| bool operator==(const CrossOriginIsolationKey& b) const; |
| // The origin of the document which triggered cross-origin isolation. This |
| // might be different from the origin returned by AgentClusterKey::GetOrigin |
| // when cross-origin isolation was enabled by COOP + COEP. It should always |
| // match when cross-origin isolation was enabled by |
| // Document-Isolation-Policy. |
| url::Origin common_coi_origin; |
| |
| // Whether cross-origin isolation is effective or logical. Effective |
| // cross-origin isolation grants access to extra web APIs. Some platforms |
| // might not have the process model needed to support cross-origin |
| // isolation. In this case, the web-visible isolation restrictions apply, |
| // but do not lead to access to extra APIs. This is logical cross-origin |
| // isolation. |
| CrossOriginIsolationMode cross_origin_isolation_mode; |
| }; |
| |
| // Tracks the state of an Origin-Agent-Cluster request for a document. |
| // The Origin-Agent-Cluster header can be used to request either an |
| // origin-keyed agent cluster (1?) or a site-keyed one (0?). In the absence of |
| // an OAC header, agent clusters will be either site-keyed or origin-keyed by |
| // default, depending on whether features::kOriginKeyedProcessesByDefault is |
| // enabled. |
| enum class OACStatus { |
| kOriginKeyedByHeader, |
| kSiteKeyedByHeader, |
| kOriginKeyedByDefault, |
| kSiteKeyedByDefault |
| }; |
| |
| // The following functions are used to create appropriate AgentClusterKeys for |
| // a navigation. The |oac_status| parameter passed to the AgentClusterKey is |
| // the OAC status computed for the navigation. Note that it corresponds to OAC |
| // backed by actual process isolation, and not logical OAC (ie OAC enabled |
| // only in the renderer process). |
| // |
| // Following the deprecation of document.domain by default (a.k.a. |
| // Origin-Agent-Cluster by default), AgentClusterKeys should be origin keyed |
| // unless the document sends a "Origin-Agent-Cluster: ?0" header. However, |
| // without SiteInstanceGroup, this would lead to extra process creation. So |
| // when computing AgentClusterKeys for all navigations, we make |
| // them site-keyed by default. Eventually, we should make them origin-keyed, |
| // but have SiteInfos with a same-site AgentClusterKey share a |
| // SiteInstanceGroup when kOriginKeyedProcessesByDefault is not enabled. |
| // See crbug.com/40176090. |
| // |
| // When using CreateSiteKeyed, the |oac_status| must be kSiteKeyed*. |
| // |
| // When using CreateOriginKeyed, the |oac_status| does not have to be |
| // kOriginKeyed*. The browser might want to |
| // assign origin-keyed agent clusters in some cases, even when the document |
| // did not request OAC and kOriginKeyedProcessesByDefault is not enabled. This |
| // does not happen in practice currently, but should happen when we convert |
| // the following cases to always create origin-keyed AgentClusterKeys: |
| // - origin-isolated sandboxed data iframes |
| // - legacy kStrictOriginIsolation mode. |
| // |
| // When using CreateWithCrossOriginIsolationKey, the |oac_status| does not |
| // have to be kOriginKeyed*. Cross-origin isolated contexts are always |
| // origin-keyed per spec, regardless of the OAC header. |
| static AgentClusterKey CreateSiteKeyed(const GURL& site_url, |
| const OACStatus& oac_status); |
| static AgentClusterKey CreateOriginKeyed(const url::Origin& origin, |
| const OACStatus& oac_status); |
| |
| static AgentClusterKey CreateWithCrossOriginIsolationKey( |
| const url::Origin& origin, |
| const AgentClusterKey::CrossOriginIsolationKey& isolation_key, |
| const OACStatus& oac_status); |
| |
| // The default constructor will create an AgentClusterKey site-keyed to the |
| // empty URL. |
| // TODO(crbug.com/342366372): Once SiteInstanceGroup has launched for all |
| // SiteInstances, the default constructor should return an origin-keyed |
| // AgentClusterKey with an empty origin. |
| AgentClusterKey(); |
| AgentClusterKey(const AgentClusterKey& other); |
| ~AgentClusterKey(); |
| |
| // Whether the Agent Cluster is keyed using Site URL or Origin. |
| bool IsSiteKeyed() const; |
| bool IsOriginKeyed() const; |
| |
| // The status of the Origin-Agent-Cluster header request for the navigation |
| // that created the AgentClusterKey. |
| // This is mainly used in SiteInfo to distinguish between SiteInfos that |
| // received process isolation for their origin due to an explicit OAC opt-in |
| // via header (kOriginKeyedByHeader) from the SiteInfos that received process |
| // isolation due to features::kOriginKeyedProcessesByDefault |
| // (kOriginKeyedByDefault). The former must be tracked per BrowsingInstance to |
| // maintain a consistent OAC state, while the latter do not need to do so. |
| // Note that this only applies to OAC that is backed by process isolation. OAC |
| // can also be logical, in which case it will only apply in the renderer |
| // process and is not tracked in the AgentClusterKey (and by extension |
| // SiteInfo). Also note that having an |oac_status_| of kOriginKeyedByHeader |
| // or kOriginKeyedByDefault will make the AgentClusterKey origin-keyed, the |
| // reverse is not true. It is possible for the AgentClusterKey to be |
| // origin-keyed and |oac_status_| to be kSiteKeyedByDefault, for example in |
| // the case of a cross-origin isolated document with DocumentIsolationPolicy. |
| const OACStatus& oac_status() const { return oac_status_; } |
| |
| // The site URL or the origin of the AgentClusterKey. Each function should |
| // only be called when the Agent Cluster is site-keyed or origin-keyed |
| // respectively. The functions will CHECK fail if called in the wrong cases. |
| const GURL& GetSite() const; |
| const url::Origin& GetOrigin() const; |
| |
| // This will return a URL based on the site URL or the origin of the |
| // AgentClusterKey depending on whether the AgentClusterKey is site-keyed or |
| // origin-keyed. Prefer comparing the Site URL or the Origin directly when |
| // possible. |
| GURL GetURL() const; |
| |
| // Returns nullopt if the AgentClusterKey is not cross-origin isolated. |
| // Otherwise, returns the CrossOriginIsolationKey associated to the |
| // AgentClusterKey. |
| const std::optional<AgentClusterKey::CrossOriginIsolationKey>& |
| GetCrossOriginIsolationKey() const; |
| |
| // Returns true if the AgentClusterKey is cross-origin isolated. |
| bool IsCrossOriginIsolated() const; |
| |
| bool operator==(const AgentClusterKey& b) const; |
| |
| // Needed for tie comparisons in SiteInfo. |
| bool operator<(const AgentClusterKey& b) const; |
| |
| private: |
| AgentClusterKey(const std::variant<GURL, url::Origin>& key, |
| const std::optional<AgentClusterKey::CrossOriginIsolationKey>& |
| isolation_key, |
| const OACStatus& oac_status); |
| |
| // The origin or site URL that all execution contexts in the agent cluster |
| // must share. By default, this is a site URL and the agent cluster is |
| // site-keyed. The agent cluster can also be origin-keyed, in which case |
| // execution contexts in the agent cluster must share the same origin, as |
| // opposed to the site URL. |
| // |
| // For example, execution contexts with origin "https://example.com" and |
| // "https://subdomain.example.com" can be placed in the same site-keyed agent |
| // cluster with site URL key "https://example.com". But an execution context |
| // with origin "https://subdomain.example.com" cannot be placed in |
| // origin-keyed agent cluster with origin key "https://example.com" (because |
| // it is not same-origin with the origin key of the agent cluster). |
| // |
| // When used in ProcessLocks, in the case of an unlocked AllowAnySite process, |
| // the key_ will be an empty GURL in non-cross-origin isolated cases. For |
| // cross-origin isolated cases, it will be an empty origin (along with the |
| // appropriate cross-origin isolation key). |
| std::variant<GURL, url::Origin> key_; |
| |
| // This is used by DocumentIsolationPolicy to isolate the document in an agent |
| // cluster with the appropriate cross-origin isolation status. Setting this to |
| // nullopt means that the AgentClusterKey is not cross-origin isolated. |
| // TODO(crbug.com/342365083): Currently the CrossOriginIsolationKey is only |
| // set based on DocumentIsolationPolicy. It should also be set for documents |
| // in a page with COOP and COEP. |
| std::optional<AgentClusterKey::CrossOriginIsolationKey> isolation_key_; |
| |
| // Tracks the status of the OAC header opt-in request for the navigation that |
| // created this AgentClusterKey. |
| // Note: this is not taken into account in |
| // AgentClusterKey::operator== because we want a document |
| // with OAC: 1? to have an AgentClusterKey equal to the AgentClusterKey of a |
| // document that got origin isolation through other means |
| // (features::kOriginKeyedProcessesByDefault, |
| // cross-origin isolation provided the cross-origin isolation status |
| // match...). Origin isolation is taken into account because origin-keyed and |
| // site-keyed AgentClusterKeys are never equivalent. The precise manner in |
| // which it was achieved is not. |
| AgentClusterKey::OACStatus oac_status_ = |
| AgentClusterKey::OACStatus::kSiteKeyedByDefault; |
| }; |
| |
| CONTENT_EXPORT std::ostream& operator<<( |
| std::ostream& out, |
| const AgentClusterKey& agent_cluster_key); |
| |
| } // namespace content |
| |
| #endif // CONTENT_BROWSER_AGENT_CLUSTER_KEY_H_ |