blob: cdfa587f696a6640d1ac824c03e0281fbaf41901 [file] [log] [blame]
// Copyright (c) 2012 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include <stddef.h>
#include <stdint.h>
#include <cstddef>
#include <list>
#include <map>
#include <memory>
#include <set>
#include <string>
#include <utility>
#include <vector>
#include "base/macros.h"
#include "base/memory/ref_counted.h"
#include "base/memory/weak_ptr.h"
#include "base/optional.h"
#include "base/time/time.h"
#include "base/timer/timer.h"
#include "net/base/address_list.h"
#include "net/base/completion_once_callback.h"
#include "net/base/load_states.h"
#include "net/base/load_timing_info.h"
#include "net/base/net_errors.h"
#include "net/base/net_export.h"
#include "net/base/network_change_notifier.h"
#include "net/base/priority_queue.h"
#include "net/base/proxy_server.h"
#include "net/base/request_priority.h"
#include "net/log/net_log_with_source.h"
#include "net/socket/client_socket_handle.h"
#include "net/socket/client_socket_pool.h"
#include "net/socket/connect_job.h"
#include "net/socket/connection_attempts.h"
#include "net/socket/socket_tag.h"
#include "net/socket/ssl_client_socket.h"
#include "net/socket/stream_socket.h"
namespace base {
namespace trace_event {
class ProcessMemoryDump;
} // namespace base
namespace net {
struct CommonConnectJobParams;
struct NetLogSource;
struct NetworkTrafficAnnotationTag;
// TransportClientSocketPool establishes network connections through using
// ConnectJobs, and maintains a list of idle persistent sockets available for
// reuse. It restricts the number of sockets open at a time, both globally, and
// for each unique GroupId, which rougly corresponds to origin and privacy mode
// setting. TransportClientSocketPools is designed to work with HTTP reuse
// semantics, handling each request serially, before reusable sockets are
// returned to the socket pool.
// In order to manage connection limits on a per-Proxy basis, separate
// TransportClientSocketPools are created for each proxy, and another for
// connections that have no proxy.
// TransportClientSocketPool is an internal class that implements almost all
// the functionality from ClientSocketPool.
class NET_EXPORT_PRIVATE TransportClientSocketPool
: public ClientSocketPool,
public NetworkChangeNotifier::IPAddressObserver,
public SSLClientContext::Observer {
using Flags = uint32_t;
// Used to specify specific behavior for the ClientSocketPool.
enum Flag {
NORMAL = 0, // Normal behavior.
NO_IDLE_SOCKETS = 0x1, // Do not return an idle socket. Create a new one.
class NET_EXPORT_PRIVATE Request {
// If |proxy_auth_callback| is null, proxy auth challenges will
// result in an error.
ClientSocketHandle* handle,
CompletionOnceCallback callback,
const ProxyAuthCallback& proxy_auth_callback,
RequestPriority priority,
const SocketTag& socket_tag,
RespectLimits respect_limits,
Flags flags,
scoped_refptr<SocketParams> socket_params,
const base::Optional<NetworkTrafficAnnotationTag>& proxy_annotation_tag,
const NetLogWithSource& net_log);
ClientSocketHandle* handle() const { return handle_; }
CompletionOnceCallback release_callback() { return std::move(callback_); }
const ProxyAuthCallback& proxy_auth_callback() const {
return proxy_auth_callback_;
RequestPriority priority() const { return priority_; }
void set_priority(RequestPriority priority) { priority_ = priority; }
RespectLimits respect_limits() const { return respect_limits_; }
Flags flags() const { return flags_; }
SocketParams* socket_params() const { return socket_params_.get(); }
const base::Optional<NetworkTrafficAnnotationTag>& proxy_annotation_tag()
const {
return proxy_annotation_tag_;
const NetLogWithSource& net_log() const { return net_log_; }
const SocketTag& socket_tag() const { return socket_tag_; }
ConnectJob* job() const { return job_; }
// Associates a ConnectJob with the request. Must be called on a request
// that does not already have a job.
void AssignJob(ConnectJob* job);
// Unassigns the request's |job_| and returns it. Must be called on a
// request with a job.
ConnectJob* ReleaseJob();
ClientSocketHandle* const handle_;
CompletionOnceCallback callback_;
const ProxyAuthCallback proxy_auth_callback_;
RequestPriority priority_;
const RespectLimits respect_limits_;
const Flags flags_;
const scoped_refptr<SocketParams> socket_params_;
const base::Optional<NetworkTrafficAnnotationTag> proxy_annotation_tag_;
const NetLogWithSource net_log_;
const SocketTag socket_tag_;
ConnectJob* job_;
class ConnectJobFactory {
ConnectJobFactory() {}
virtual ~ConnectJobFactory() {}
virtual std::unique_ptr<ConnectJob> NewConnectJob(
ClientSocketPool::GroupId group_id,
scoped_refptr<ClientSocketPool::SocketParams> socket_params,
const base::Optional<NetworkTrafficAnnotationTag>& proxy_annotation_tag,
RequestPriority request_priority,
SocketTag socket_tag,
ConnectJob::Delegate* delegate) const = 0;
int max_sockets,
int max_sockets_per_group,
base::TimeDelta unused_idle_socket_timeout,
const ProxyServer& proxy_server,
bool is_for_websockets,
const CommonConnectJobParams* common_connect_job_params);
// Creates a socket pool with an alternative ConnectJobFactory, for use in
// testing.
// |connect_backup_jobs_enabled| can be set to false to disable backup connect
// jobs (Which are normally enabled).
static std::unique_ptr<TransportClientSocketPool> CreateForTesting(
int max_sockets,
int max_sockets_per_group,
base::TimeDelta unused_idle_socket_timeout,
base::TimeDelta used_idle_socket_timeout,
const ProxyServer& proxy_server,
std::unique_ptr<ConnectJobFactory> connect_job_factory,
SSLClientContext* ssl_client_context,
bool connect_backup_jobs_enabled);
~TransportClientSocketPool() override;
// See LowerLayeredPool::IsStalled for documentation on this function.
bool IsStalled() const override;
// See LowerLayeredPool for documentation on these functions. It is expected
// in the destructor that no higher layer pools remain.
void AddHigherLayeredPool(HigherLayeredPool* higher_pool) override;
void RemoveHigherLayeredPool(HigherLayeredPool* higher_pool) override;
// ClientSocketPool implementation:
int RequestSocket(
const GroupId& group_id,
scoped_refptr<SocketParams> params,
const base::Optional<NetworkTrafficAnnotationTag>& proxy_annotation_tag,
RequestPriority priority,
const SocketTag& socket_tag,
RespectLimits respect_limits,
ClientSocketHandle* handle,
CompletionOnceCallback callback,
const ProxyAuthCallback& proxy_auth_callback,
const NetLogWithSource& net_log) override;
void RequestSockets(
const GroupId& group_id,
scoped_refptr<SocketParams> params,
const base::Optional<NetworkTrafficAnnotationTag>& proxy_annotation_tag,
int num_sockets,
const NetLogWithSource& net_log) override;
void SetPriority(const GroupId& group_id,
ClientSocketHandle* handle,
RequestPriority priority) override;
void CancelRequest(const GroupId& group_id,
ClientSocketHandle* handle,
bool cancel_connect_job) override;
void ReleaseSocket(const GroupId& group_id,
std::unique_ptr<StreamSocket> socket,
int64_t group_generation) override;
void FlushWithError(int error) override;
void CloseIdleSockets() override;
void CloseIdleSocketsInGroup(const GroupId& group_id) override;
int IdleSocketCount() const override;
size_t IdleSocketCountInGroup(const GroupId& group_id) const override;
LoadState GetLoadState(const GroupId& group_id,
const ClientSocketHandle* handle) const override;
base::Value GetInfoAsValue(const std::string& name,
const std::string& type) const override;
void DumpMemoryStats(
base::trace_event::ProcessMemoryDump* pmd,
const std::string& parent_dump_absolute_name) const override;
bool RequestInGroupWithHandleHasJobForTesting(
const GroupId& group_id,
const ClientSocketHandle* handle) const {
return group_map_.find(group_id)->second->RequestWithHandleHasJobForTesting(
size_t NumNeverAssignedConnectJobsInGroupForTesting(
const GroupId& group_id) const {
return NumNeverAssignedConnectJobsInGroup(group_id);
size_t NumUnassignedConnectJobsInGroupForTesting(
const GroupId& group_id) const {
return NumUnassignedConnectJobsInGroup(group_id);
size_t NumConnectJobsInGroupForTesting(const GroupId& group_id) const {
return NumConnectJobsInGroup(group_id);
int NumActiveSocketsInGroupForTesting(const GroupId& group_id) const {
return NumActiveSocketsInGroup(group_id);
bool HasGroupForTesting(const GroupId& group_id) const {
return HasGroup(group_id);
static bool connect_backup_jobs_enabled();
static bool set_connect_backup_jobs_enabled(bool enabled);
// NetworkChangeNotifier::IPAddressObserver methods:
void OnIPAddressChanged() override;
// SSLClientContext::Observer methods.
void OnSSLConfigChanged(bool is_cert_database_change) override;
void OnSSLConfigForServerChanged(const HostPortPair& server) override;
class ConnectJobFactoryImpl;
// Entry for a persistent socket which became idle at time |start_time|.
struct IdleSocket {
IdleSocket() : socket(nullptr) {}
// An idle socket can't be used if it is disconnected or has been used
// before and has received data unexpectedly (hence no longer idle). The
// unread data would be mistaken for the beginning of the next response if
// we were to use the socket for a new request.
// Note that a socket that has never been used before (like a preconnected
// socket) may be used even with unread data. This may be, e.g., a SPDY
// SETTINGS frame.
bool IsUsable() const;
StreamSocket* socket;
base::TimeTicks start_time;
using RequestQueue = PriorityQueue<std::unique_ptr<Request>>;
// A Group is allocated per GroupId when there are idle sockets, unbound
// request, or bound requests. Otherwise, the Group object is removed from the
// map.
// A request is "bound" to a ConnectJob when an unbound ConnectJob encounters
// a proxy HTTP auth challenge, and the auth challenge is presented to that
// request. Once a request and ConnectJob are bound together:
// * All auth challenges the ConnectJob sees will be sent to that request.
// * Cancelling the request will cancel the ConnectJob.
// * The final result of the ConnectJob, and any returned socket, will only be
// sent to that bound request, though if the returned socket is returned to
// the socket pool, it can then be used to service any request.
// "assigned" jobs are unbound ConnectJobs that have a corresponding Request.
// If there are 5 Jobs and 10 Requests, the 5 highest priority requests are
// each assigned a Job. If there are 10 Jobs and 5 Requests, the first 5 Jobs
// are each assigned to a request. Assignment is determined by order in their
// corresponding arrays. The assignment concept is used to deal with
// reprioritizing Jobs, and computing a Request's LoadState.
// |active_socket_count| tracks the number of sockets held by clients.
// SanityCheck() will always be true, except during the invocation of a
// method. So all public methods expect the Group to pass SanityCheck() when
// invoked.
class NET_EXPORT_PRIVATE Group : public ConnectJob::Delegate {
using JobList = std::list<std::unique_ptr<ConnectJob>>;
struct BoundRequest {
BoundRequest(std::unique_ptr<ConnectJob> connect_job,
std::unique_ptr<Request> request,
int64_t generation);
BoundRequest(BoundRequest&& other);
BoundRequest& operator=(BoundRequest&& other);
std::unique_ptr<ConnectJob> connect_job;
std::unique_ptr<Request> request;
// Generation of |connect_job|. If it doesn't match the current
// generation, ConnectJob will be destroyed, and a new one created on
// completion.
int64_t generation;
// It's not safe to fail a request in a |CancelAllRequestsWithError| call
// while it's waiting on user input, as the request may have raw pointers
// to objects owned by |connect_job| that it could racily write to after
// |connect_job| is destroyed. Instead, just track an error in that case,
// and fail the request once the ConnectJob completes.
int pending_error;
Group(const GroupId& group_id,
TransportClientSocketPool* client_socket_pool);
~Group() override;
// ConnectJob::Delegate methods:
void OnConnectJobComplete(int result, ConnectJob* job) override;
void OnNeedsProxyAuth(const HttpResponseInfo& response,
HttpAuthController* auth_controller,
base::OnceClosure restart_with_auth_callback,
ConnectJob* job) override;
bool IsEmpty() const {
return active_socket_count_ == 0 && idle_sockets_.empty() &&
jobs_.empty() && unbound_requests_.empty() &&
bool HasAvailableSocketSlot(int max_sockets_per_group) const {
return NumActiveSocketSlots() < max_sockets_per_group;
int NumActiveSocketSlots() const {
return active_socket_count_ + static_cast<int>(jobs_.size()) +
static_cast<int>(idle_sockets_.size()) +
// Returns true if the group could make use of an additional socket slot, if
// it were given one.
bool CanUseAdditionalSocketSlot(int max_sockets_per_group) const {
return HasAvailableSocketSlot(max_sockets_per_group) &&
unbound_requests_.size() > jobs_.size();
// Returns the priority of the top of the unbound request queue
// (which may be less than the maximum priority over the entire
// queue, due to how we prioritize requests with |respect_limits|
// DISABLED over others).
RequestPriority TopPendingPriority() const {
// NOTE: FirstMax().value()->priority() is not the same as
// FirstMax().priority()!
return unbound_requests_.FirstMax().value()->priority();
// Set a timer to create a backup job if it takes too long to
// create one and if a timer isn't already running.
void StartBackupJobTimer(const GroupId& group_id);
bool BackupJobTimerIsRunning() const;
// If there's a ConnectJob that's never been assigned to Request,
// decrements |never_assigned_job_count_| and returns true.
// Otherwise, returns false.
bool TryToUseNeverAssignedConnectJob();
void AddJob(std::unique_ptr<ConnectJob> job, bool is_preconnect);
// Remove |job| from this group, which must already own |job|. Returns the
// removed ConnectJob.
std::unique_ptr<ConnectJob> RemoveUnboundJob(ConnectJob* job);
void RemoveAllUnboundJobs();
bool has_unbound_requests() const { return !unbound_requests_.empty(); }
size_t unbound_request_count() const { return unbound_requests_.size(); }
size_t ConnectJobCount() const;
// Returns the connect job correspding to |handle|. In particular, if
// |handle| is bound to a ConnectJob, returns that job. If |handle| is
// "assigned" a ConnectJob, return that job. Otherwise, returns nullptr.
ConnectJob* GetConnectJobForHandle(const ClientSocketHandle* handle) const;
// Inserts the request into the queue based on priority
// order. Older requests are prioritized over requests of equal
// priority.
void InsertUnboundRequest(std::unique_ptr<Request> request);
// Gets (but does not remove) the next unbound request. Returns
// NULL if there are no unbound requests.
const Request* GetNextUnboundRequest() const;
// Gets and removes the next unbound request. Returns NULL if
// there are no unbound requests.
std::unique_ptr<Request> PopNextUnboundRequest();
// Finds the unbound request for |handle| and removes it. Returns
// the removed unbound request, or NULL if there was none.
std::unique_ptr<Request> FindAndRemoveUnboundRequest(
ClientSocketHandle* handle);
// Sets a pending error for all bound requests. Bound requests may be in the
// middle of a callback, so can't be failed at arbitrary points in time.
void SetPendingErrorForAllBoundRequests(int pending_error);
// Attempts to bind the highest priority unbound request to |connect_job|,
// and returns the bound request. If the request has previously been bound
// to |connect_job|, returns the previously bound request. If there are no
// requests, or the highest priority request doesn't have a proxy auth
// callback, returns nullptr.
const Request* BindRequestToConnectJob(ConnectJob* connect_job);
// Finds the request, if any, bound to |connect_job|, and returns the
// BoundRequest or base::nullopt if there was none.
base::Optional<BoundRequest> FindAndRemoveBoundRequestForConnectJob(
ConnectJob* connect_job);
// Finds the bound request, if any, corresponding to |client_socket_handle|
// and returns it. Destroys the ConnectJob bound to the request, if there
// was one.
std::unique_ptr<Request> FindAndRemoveBoundRequest(
ClientSocketHandle* client_socket_handle);
// Change the priority of the request named by |*handle|. |*handle|
// must refer to a request currently present in the group. If |priority|
// is the same as the current priority of the request, this is a no-op.
void SetPriority(ClientSocketHandle* handle, RequestPriority priority);
void IncrementActiveSocketCount() { active_socket_count_++; }
void DecrementActiveSocketCount() { active_socket_count_--; }
void IncrementGeneration() { generation_++; }
// Whether the request in |unbound_requests_| with a given handle has a job.
bool RequestWithHandleHasJobForTesting(
const ClientSocketHandle* handle) const;
const GroupId& group_id() { return group_id_; }
size_t unassigned_job_count() const { return unassigned_jobs_.size(); }
const JobList& jobs() const { return jobs_; }
const std::list<IdleSocket>& idle_sockets() const { return idle_sockets_; }
int active_socket_count() const { return active_socket_count_; }
std::list<IdleSocket>* mutable_idle_sockets() { return &idle_sockets_; }
size_t never_assigned_job_count() const {
return never_assigned_job_count_;
int64_t generation() const { return generation_; }
// Returns the iterator's unbound request after removing it from
// the queue. Expects the Group to pass SanityCheck() when called.
std::unique_ptr<Request> RemoveUnboundRequest(
const RequestQueue::Pointer& pointer);
// Finds the Request which is associated with the given ConnectJob.
// Returns nullptr if none is found. Expects the Group to pass SanityCheck()
// when called.
RequestQueue::Pointer FindUnboundRequestWithJob(
const ConnectJob* job) const;
// Finds the Request in |unbound_requests_| which is the first request
// without a job. Returns a null pointer if all requests have jobs. Does not
// expect the Group to pass SanityCheck() when called, but does expect all
// jobs to either be assigned to a request or in |unassigned_jobs_|. Expects
// that no requests with jobs come after any requests without a job.
RequestQueue::Pointer GetFirstRequestWithoutJob() const;
// Tries to assign an unassigned |job| to a request. If no requests need a
// job, |job| is added to |unassigned_jobs_|.
// When called, does not expect the Group to pass SanityCheck(), but does
// expect it to have passed SanityCheck() before the given ConnectJob was
// either created or had the request it was assigned to removed.
void TryToAssignUnassignedJob(ConnectJob* job);
// Tries to assign a job to the given request. If any unassigned jobs are
// available, the first unassigned job is assigned to the request.
// Otherwise, if the request is ahead of the last request with a job, the
// job is stolen from the last request with a job.
// When called, does not expect the Group to pass SanityCheck(), but does
// expect that:
// - the request associated with |request_pointer| must not have
// an assigned ConnectJob,
// - the first min( jobs_.size(), unbound_requests_.size() - 1 ) Requests
// other than the given request must have ConnectJobs, i.e. the group
// must have passed SanityCheck() before the passed in Request was either
// added or had its job unassigned.
void TryToAssignJobToRequest(RequestQueue::Pointer request_pointer);
// Transfers the associated ConnectJob from one Request to another. Expects
// the source request to have a job, and the destination request to not have
// a job. Does not expect the Group to pass SanityCheck() when called.
void TransferJobBetweenRequests(Request* source, Request* dest);
// Called when the backup socket timer fires.
void OnBackupJobTimerFired(const GroupId& group_id);
// Checks that:
// - |unassigned_jobs_| is empty iff there are at least as many requests
// as jobs.
// - Exactly the first |jobs_.size() - unassigned_jobs_.size()| requests
// have ConnectJobs.
// - No requests are assigned a ConnectJob in |unassigned_jobs_|.
// - No requests are assigned a ConnectJob not in |jobs_|.
// - No two requests are assigned the same ConnectJob.
// - All entries in |unassigned_jobs_| are also in |jobs_|.
// - There are no duplicate entries in |unassigned_jobs_|.
void SanityCheck() const;
const GroupId group_id_;
TransportClientSocketPool* const client_socket_pool_;
// Total number of ConnectJobs that have never been assigned to a Request.
// Since jobs use late binding to requests, which ConnectJobs have or have
// not been assigned to a request are not tracked. This is incremented on
// preconnect and decremented when a preconnect is assigned, or when there
// are fewer than |never_assigned_job_count_| ConnectJobs. Not incremented
// when a request is cancelled.
size_t never_assigned_job_count_;
std::list<IdleSocket> idle_sockets_;
JobList jobs_; // For bookkeeping purposes, there is a copy of the raw
// pointer of each element of |jobs_| stored either in
// |unassigned_jobs_|, or as the associated |job_| of an
// element of |unbound_requests_|.
std::list<ConnectJob*> unassigned_jobs_;
RequestQueue unbound_requests_;
int active_socket_count_; // number of active sockets used by clients
// A timer for when to start the backup job.
base::OneShotTimer backup_job_timer_;
// List of Requests bound to ConnectJobs currently undergoing proxy auth.
// The Requests and ConnectJobs in this list do not appear in
// |unbound_requests_| or |jobs_|.
std::vector<BoundRequest> bound_requests_;
// An id for the group. It gets incremented every time we FlushWithError()
// the socket pool, or refresh the group. This is so that when sockets get
// released back to the group, we can make sure that they are discarded
// rather than reused. Destroying a group will reset the generation number,
// but as that only happens once there are no outstanding sockets or
// requests associated with the group, that's harmless.
int64_t generation_;
using GroupMap = std::map<GroupId, Group*>;
struct CallbackResultPair {
CallbackResultPair(CompletionOnceCallback callback_in, int result_in);
CallbackResultPair(CallbackResultPair&& other);
CallbackResultPair& operator=(CallbackResultPair&& other);
CompletionOnceCallback callback;
int result;
using PendingCallbackMap =
std::map<const ClientSocketHandle*, CallbackResultPair>;
int max_sockets,
int max_sockets_per_group,
base::TimeDelta unused_idle_socket_timeout,
base::TimeDelta used_idle_socket_timeout,
const ProxyServer& proxy_server,
std::unique_ptr<ConnectJobFactory> connect_job_factory,
SSLClientContext* ssl_client_context,
bool connect_backup_jobs_enabled);
base::TimeDelta ConnectRetryInterval() const {
// TODO(mbelshe): Make this tuned dynamically based on measured RTT.
// For now, just use the max retry interval.
return base::TimeDelta::FromMilliseconds(kMaxConnectRetryIntervalMs);
// TODO(mmenke): de-inline these.
size_t NumNeverAssignedConnectJobsInGroup(const GroupId& group_id) const {
return group_map_.find(group_id)->second->never_assigned_job_count();
size_t NumUnassignedConnectJobsInGroup(const GroupId& group_id) const {
return group_map_.find(group_id)->second->unassigned_job_count();
size_t NumConnectJobsInGroup(const GroupId& group_id) const {
return group_map_.find(group_id)->second->ConnectJobCount();
int NumActiveSocketsInGroup(const GroupId& group_id) const {
return group_map_.find(group_id)->second->active_socket_count();
bool HasGroup(const GroupId& group_id) const;
// Closes all idle sockets if |force| is true. Else, only closes idle
// sockets that timed out or can't be reused. Made public for testing.
void CleanupIdleSockets(bool force);
// Closes one idle socket. Picks the first one encountered.
// TODO(willchan): Consider a better algorithm for doing this. Perhaps we
// should keep an ordered list of idle sockets, and close them in order.
// Requires maintaining more state. It's not clear if it's worth it since
// I'm not sure if we hit this situation often.
bool CloseOneIdleSocket();
// Checks higher layered pools to see if they can close an idle connection.
bool CloseOneIdleConnectionInHigherLayeredPool();
// Closes all idle sockets in |group| if |force| is true. Else, only closes
// idle sockets in |group| that timed out with respect to |now| or can't be
// reused.
void CleanupIdleSocketsInGroup(bool force,
Group* group,
const base::TimeTicks& now);
Group* GetOrCreateGroup(const GroupId& group_id);
void RemoveGroup(const GroupId& group_id);
void RemoveGroup(GroupMap::iterator it);
// Called when the number of idle sockets changes.
void IncrementIdleCount();
void DecrementIdleCount();
// Scans the group map for groups which have an available socket slot and
// at least one pending request. Returns true if any groups are stalled, and
// if so (and if both |group| and |group_id| are not NULL), fills |group|
// and |group_id| with data of the stalled group having highest priority.
bool FindTopStalledGroup(Group** group, GroupId* group_id) const;
// Removes |job| from |group|, which must already own |job|.
void RemoveConnectJob(ConnectJob* job, Group* group);
// Tries to see if we can handle any more requests for |group|.
void OnAvailableSocketSlot(const GroupId& group_id, Group* group);
// Process a pending socket request for a group.
void ProcessPendingRequest(const GroupId& group_id, Group* group);
// Assigns |socket| to |handle| and updates |group|'s counters appropriately.
void HandOutSocket(std::unique_ptr<StreamSocket> socket,
ClientSocketHandle::SocketReuseType reuse_type,
const LoadTimingInfo::ConnectTiming& connect_timing,
ClientSocketHandle* handle,
base::TimeDelta time_idle,
Group* group,
const NetLogWithSource& net_log);
// Adds |socket| to the list of idle sockets for |group|.
void AddIdleSocket(std::unique_ptr<StreamSocket> socket, Group* group);
// Iterates through |group_map_|, canceling all ConnectJobs and deleting
// groups if they are no longer needed.
void CancelAllConnectJobs();
// Iterates through |group_map_|, posting |error| callbacks for all
// requests, and then deleting groups if they are no longer needed.
void CancelAllRequestsWithError(int error);
// Returns true if we can't create any more sockets due to the total limit.
bool ReachedMaxSocketsLimit() const;
// This is the internal implementation of RequestSocket(). It differs in that
// it does not handle logging into NetLog of the queueing status of
// |request|.
int RequestSocketInternal(const GroupId& group_id, const Request& request);
// Assigns an idle socket for the group to the request.
// Returns |true| if an idle socket is available, false otherwise.
bool AssignIdleSocketToRequest(const Request& request, Group* group);
static void LogBoundConnectJobToRequest(
const NetLogSource& connect_job_source,
const Request& request);
// Same as CloseOneIdleSocket() except it won't close an idle socket in
// |group|. If |group| is NULL, it is ignored. Returns true if it closed a
// socket.
bool CloseOneIdleSocketExceptInGroup(const Group* group);
// Checks if there are stalled socket groups that should be notified
// for possible wakeup.
void CheckForStalledSocketGroups();
// Posts a task to call InvokeUserCallback() on the next iteration through the
// current message loop. Inserts |callback| into |pending_callback_map_|,
// keyed by |handle|. Apply |socket_tag| to the socket if socket successfully
// created.
void InvokeUserCallbackLater(ClientSocketHandle* handle,
CompletionOnceCallback callback,
int rv,
const SocketTag& socket_tag);
// These correspond to ConnectJob::Delegate methods, and are invoked by the
// Group a ConnectJob belongs to.
void OnConnectJobComplete(Group* group, int result, ConnectJob* job);
void OnNeedsProxyAuth(Group* group,
const HttpResponseInfo& response,
HttpAuthController* auth_controller,
base::OnceClosure restart_with_auth_callback,
ConnectJob* job);
// Invokes the user callback for |handle|. By the time this task has run,
// it's possible that the request has been cancelled, so |handle| may not
// exist in |pending_callback_map_|. We look up the callback and result code
// in |pending_callback_map_|.
void InvokeUserCallback(ClientSocketHandle* handle);
// Tries to close idle sockets in a higher level socket pool as long as this
// this pool is stalled.
void TryToCloseSocketsInLayeredPools();
// Closes all idle sockets and cancels all unbound ConnectJobs associated with
// |it->second|. Also increments the group's generation number, ensuring any
// currently existing handed out socket will be silently closed when it is
// returned to the socket pool. Bound ConnectJobs will only be destroyed on
// once they complete, as they may be waiting on user input. No request
// (including bound ones) will be failed as a result of this call - instead,
// new ConnectJobs will be created.
// The group may be removed if this leaves the group empty. The caller must
// call CheckForStalledSocketGroups() after all applicable groups have been
// refreshed.
void RefreshGroup(GroupMap::iterator it, const base::TimeTicks& now);
GroupMap group_map_;
// Map of the ClientSocketHandles for which we have a pending Task to invoke a
// callback. This is necessary since, before we invoke said callback, it's
// possible that the request is cancelled.
PendingCallbackMap pending_callback_map_;
// The total number of idle sockets in the system.
int idle_socket_count_;
// Number of connecting sockets across all groups.
int connecting_socket_count_;
// Number of connected sockets we handed out across all groups.
int handed_out_socket_count_;
// The maximum total number of sockets. See ReachedMaxSocketsLimit.
const int max_sockets_;
// The maximum number of sockets kept per group.
const int max_sockets_per_group_;
// The time to wait until closing idle sockets.
const base::TimeDelta unused_idle_socket_timeout_;
const base::TimeDelta used_idle_socket_timeout_;
const ProxyServer proxy_server_;
const std::unique_ptr<ConnectJobFactory> connect_job_factory_;
// TODO(vandebo) Remove when backup jobs move to TransportClientSocketPool
bool connect_backup_jobs_enabled_;
// Pools that create connections through |this|. |this| will try to close
// their idle sockets when it stalls. Must be empty on destruction.
std::set<HigherLayeredPool*> higher_pools_;
SSLClientContext* const ssl_client_context_;
base::WeakPtrFactory<TransportClientSocketPool> weak_factory_{this};
} // namespace net