blob: c8583e351d27841456032ca77adb47c57ef9bdef [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.
// This file declares a HttpTransactionFactory implementation that can be
// layered on top of another HttpTransactionFactory to add HTTP caching. The
// caching logic follows RFC 7234 (any exceptions are called out in the code).
// The HttpCache takes a disk_cache::Backend as a parameter, and uses that for
// the cache storage.
// See HttpTransactionFactory and HttpTransaction for more details.
#include <list>
#include <map>
#include <memory>
#include <string>
#include <unordered_map>
#include "base/files/file_path.h"
#include "base/macros.h"
#include "base/memory/weak_ptr.h"
#include "base/threading/thread_checker.h"
#include "base/time/clock.h"
#include "base/time/time.h"
#include "net/base/cache_type.h"
#include "net/base/completion_once_callback.h"
#include "net/base/load_states.h"
#include "net/base/net_export.h"
#include "net/base/request_priority.h"
#include "net/http/http_network_session.h"
#include "net/http/http_transaction_factory.h"
class GURL;
namespace base {
namespace trace_event {
class ProcessMemoryDump;
} // namespace base
namespace disk_cache {
class Backend;
class Entry;
} // namespace disk_cache
namespace net {
class HttpNetworkSession;
class HttpResponseInfo;
class IOBuffer;
class NetLog;
class ViewCacheHelper;
struct HttpRequestInfo;
class NET_EXPORT HttpCache : public HttpTransactionFactory {
// The cache mode of operation.
enum Mode {
// Normal mode just behaves like a standard web cache.
// Disables reads and writes from the cache.
// Equivalent to setting LOAD_DISABLE_CACHE on every request.
// A BackendFactory creates a backend object to be used by the HttpCache.
class NET_EXPORT BackendFactory {
virtual ~BackendFactory() {}
// The actual method to build the backend. Returns a net error code. If
// ERR_IO_PENDING is returned, the |callback| will be notified when the
// operation completes, and |backend| must remain valid until the
// notification arrives.
// The implementation must not access the factory object after invoking the
// |callback| because the object can be deleted from within the callback.
virtual int CreateBackend(NetLog* net_log,
std::unique_ptr<disk_cache::Backend>* backend,
CompletionOnceCallback callback) = 0;
// A default backend factory for the common use cases.
class NET_EXPORT DefaultBackend : public BackendFactory {
// |path| is the destination for any files used by the backend. If
// |max_bytes| is zero, a default value will be calculated automatically.
DefaultBackend(CacheType type,
BackendType backend_type,
const base::FilePath& path,
int max_bytes);
~DefaultBackend() override;
// Returns a factory for an in-memory cache.
static std::unique_ptr<BackendFactory> InMemory(int max_bytes);
// BackendFactory implementation.
int CreateBackend(NetLog* net_log,
std::unique_ptr<disk_cache::Backend>* backend,
CompletionOnceCallback callback) override;
CacheType type_;
BackendType backend_type_;
const base::FilePath path_;
int max_bytes_;
// Whether a transaction can join parallel writing or not is a function of the
// transaction as well as the current writers (if present). This enum
// captures that decision as well as when a Writers object is first created.
// This is also used to log metrics so should be consistent with the values in
// enums.xml and should only be appended to.
enum ParallelWritingPattern {
// Used as the default value till the transaction is in initial headers
// phase.
// The transaction creates a writers object. This is only logged for
// transactions that did not fail to join existing writers earlier.
// The transaction joins existing writers.
// The transaction cannot join existing writers since either itself or
// existing writers instance is serving a range request.
// The transaction cannot join existing writers since either itself or
// existing writers instance is serving a non GET request.
// The transaction cannot join existing writers since it does not have cache
// write privileges.
// Writers does not exist and the transaction does not need to create one
// since it is going to read from the cache.
// On adding a value here, make sure to add in enums.xml as well.
// The number of minutes after a resource is prefetched that it can be used
// again without validation.
static const int kPrefetchReuseMins = 5;
// The disk cache is initialized lazily (by CreateTransaction) in this case.
// Provide an existing HttpNetworkSession, the cache can construct a
// network layer with a shared HttpNetworkSession in order for multiple
// network layers to share information (e.g. authentication data). The
// HttpCache takes ownership of the |backend_factory|.
// The HttpCache must be destroyed before the HttpNetworkSession.
// If |is_main_cache| is true, configures the cache to track
// information about servers supporting QUIC.
// TODO(zhongyi): remove |is_main_cache| when we get rid of cache split.
HttpCache(HttpNetworkSession* session,
std::unique_ptr<BackendFactory> backend_factory,
bool is_main_cache);
// Initialize the cache from its component parts. |network_layer| and
// |backend_factory| will be destroyed when the HttpCache is.
HttpCache(std::unique_ptr<HttpTransactionFactory> network_layer,
std::unique_ptr<BackendFactory> backend_factory,
bool is_main_cache);
~HttpCache() override;
HttpTransactionFactory* network_layer() { return network_layer_.get(); }
// Retrieves the cache backend for this HttpCache instance. If the backend
// is not initialized yet, this method will initialize it. The return value is
// a network error code, and it could be ERR_IO_PENDING, in which case the
// |callback| will be notified when the operation completes. The pointer that
// receives the |backend| must remain valid until the operation completes.
int GetBackend(disk_cache::Backend** backend,
CompletionOnceCallback callback);
// Returns the current backend (can be NULL).
disk_cache::Backend* GetCurrentBackend() const;
// Given a header data blob, convert it to a response info object.
static bool ParseResponseInfo(const char* data, int len,
HttpResponseInfo* response_info,
bool* response_truncated);
// Writes |buf_len| bytes of metadata stored in |buf| to the cache entry
// referenced by |url|, as long as the entry's |expected_response_time| has
// not changed. This method returns without blocking, and the operation will
// be performed asynchronously without any completion notification.
// Takes ownership of |buf|.
virtual void WriteMetadata(const GURL& url,
RequestPriority priority,
base::Time expected_response_time,
IOBuffer* buf,
int buf_len);
// Get/Set the cache's mode.
void set_mode(Mode value) { mode_ = value; }
Mode mode() { return mode_; }
// Get/Set the cache's clock. These are public only for testing.
void SetClockForTesting(base::Clock* clock) { clock_ = clock; }
base::Clock* clock() const { return clock_; }
// Close currently active sockets so that fresh page loads will not use any
// recycled connections. For sockets currently in use, they may not close
// immediately, but they will not be reusable. This is for debugging.
void CloseAllConnections();
// Close all idle connections. Will close all sockets not in active use.
void CloseIdleConnections();
// Called whenever an external cache in the system reuses the resource
// referred to by |url| and |http_method|.
void OnExternalCacheHit(const GURL& url, const std::string& http_method);
// Causes all transactions created after this point to simulate lock timeout
// and effectively bypass the cache lock whenever there is lock contention.
void SimulateCacheLockTimeoutForTesting() { bypass_lock_for_test_ = true; }
// Causes all transactions created after this point to simulate lock timeout
// and effectively bypass the cache lock whenever there is lock contention
// after the transaction has completed its headers phase.
void SimulateCacheLockTimeoutAfterHeadersForTesting() {
bypass_lock_after_headers_for_test_ = true;
// Causes all transactions created after this point to generate a failure
// when attempting to conditionalize a network request.
void FailConditionalizationForTest() {
fail_conditionalization_for_test_ = true;
// HttpTransactionFactory implementation:
int CreateTransaction(RequestPriority priority,
std::unique_ptr<HttpTransaction>* trans) override;
HttpCache* GetCache() override;
HttpNetworkSession* GetSession() override;
base::WeakPtr<HttpCache> GetWeakPtr() { return weak_factory_.GetWeakPtr(); }
// Resets the network layer to allow for tests that probe
// network changes (e.g. host unreachable). The old network layer is
// returned to allow for filter patterns that only intercept
// some creation requests. Note ownership exchange.
std::unique_ptr<HttpTransactionFactory> new_network_layer);
// Dumps memory allocation stats. |parent_dump_absolute_name| is the name
// used by the parent MemoryAllocatorDump in the memory dump hierarchy.
void DumpMemoryStats(base::trace_event::ProcessMemoryDump* pmd,
const std::string& parent_absolute_name) const;
// Types --------------------------------------------------------------------
// Disk cache entry data indices.
enum {
kResponseInfoIndex = 0,
// Must remain at the end of the enum.
class MetadataWriter;
class QuicServerInfoFactoryAdaptor;
class Transaction;
class WorkItem;
class Writers;
friend class WritersTest;
friend class TestHttpCacheTransaction;
friend class TestHttpCache;
friend class Transaction;
friend class ViewCacheHelper;
struct PendingOp; // Info for an entry under construction.
// To help with testing.
friend class MockHttpCache;
using TransactionList = std::list<Transaction*>;
using TransactionSet = std::unordered_set<Transaction*>;
typedef std::list<std::unique_ptr<WorkItem>> WorkItemList;
// We implement a basic reader/writer lock for the disk cache entry. If there
// is a writer, then all transactions must wait to read the body. But the
// waiting transactions can start their headers phase in parallel. Headers
// phase is allowed for one transaction at a time so that if it doesn't match
// the existing headers, remaining transactions do not also try to match the
// existing entry in parallel leading to wasted network requests. If the
// headers do not match, this entry will be doomed.
// A transaction goes through these state transitions.
// Write mode transactions eligible for shared writing:
// add_to_entry_queue-> headers_transaction -> writers (first writer)
// add_to_entry_queue-> headers_transaction -> done_headers_queue -> writers
// (subsequent writers)
// add_to_entry_queue-> headers_transaction -> done_headers_queue -> readers
// (transactions not eligible for shared writing - once the data is written to
// the cache by writers)
// Read only transactions:
// add_to_entry_queue-> headers_transaction -> done_headers_queue -> readers
// (once the data is written to the cache by writers)
struct NET_EXPORT_PRIVATE ActiveEntry {
explicit ActiveEntry(disk_cache::Entry* entry);
size_t EstimateMemoryUsage() const;
// Returns true if no transactions are associated with this entry.
bool HasNoTransactions();
// Returns true if no transactions are associated with this entry and
// writers is not present.
bool SafeToDestroy();
bool TransactionInReaders(Transaction* transaction) const;
disk_cache::Entry* disk_entry = nullptr;
// Transactions waiting to be added to entry.
TransactionList add_to_entry_queue;
// Transaction currently in the headers phase, either validating the
// response or getting new headers. This can exist simultaneously with
// writers or readers while validating existing headers.
Transaction* headers_transaction = nullptr;
// Transactions that have completed their headers phase and are waiting
// to read the response body or write the response body.
TransactionList done_headers_queue;
// Transactions currently reading from the network and writing to the cache.
std::unique_ptr<Writers> writers;
// Transactions that can only read from the cache. Only one of writers or
// readers can be non-empty at a time.
TransactionSet readers;
// The following variables are true if OnProcessQueuedTransactions is posted
bool will_process_queued_transactions = false;
// True if entry is doomed.
bool doomed = false;
using ActiveEntriesMap =
std::unordered_map<std::string, std::unique_ptr<ActiveEntry>>;
using PendingOpsMap = std::unordered_map<std::string, PendingOp*>;
using ActiveEntriesSet = std::map<ActiveEntry*, std::unique_ptr<ActiveEntry>>;
using PlaybackCacheMap = std::unordered_map<std::string, int>;
// Methods ------------------------------------------------------------------
// Creates the |backend| object and notifies the |callback| when the operation
// completes. Returns an error code.
int CreateBackend(disk_cache::Backend** backend,
CompletionOnceCallback callback);
// Makes sure that the backend creation is complete before allowing the
// provided transaction to use the object. Returns an error code. |trans|
// will be notified via its IO callback if this method returns ERR_IO_PENDING.
// The transaction is free to use the backend directly at any time after
// receiving the notification.
int GetBackendForTransaction(Transaction* trans);
// Generates the cache key for this request.
std::string GenerateCacheKey(const HttpRequestInfo*);
// Dooms the entry selected by |key|, if it is currently in the list of active
// entries.
void DoomActiveEntry(const std::string& key);
// Dooms the entry selected by |key|. |trans| will be notified via its IO
// callback if this method returns ERR_IO_PENDING. The entry can be
// currently in use or not. If entry is in use and the invoking transaction
// is associated with this entry and this entry is already doomed, this API
// should not be invoked.
int DoomEntry(const std::string& key, Transaction* trans);
// Dooms the entry selected by |key|. |trans| will be notified via its IO
// callback if this method returns ERR_IO_PENDING. The entry should not
// be currently in use.
int AsyncDoomEntry(const std::string& key, Transaction* trans);
// Dooms the entry associated with a GET for a given |url|.
void DoomMainEntryForUrl(const GURL& url);
// Closes a previously doomed entry.
void FinalizeDoomedEntry(ActiveEntry* entry);
// Returns an entry that is currently in use and not doomed, or NULL.
ActiveEntry* FindActiveEntry(const std::string& key);
// Creates a new ActiveEntry and starts tracking it. |disk_entry| is the disk
// cache entry.
ActiveEntry* ActivateEntry(disk_cache::Entry* disk_entry);
// Deletes an ActiveEntry.
void DeactivateEntry(ActiveEntry* entry);
// Deletes an ActiveEntry using an exhaustive search.
void SlowDeactivateEntry(ActiveEntry* entry);
// Returns the PendingOp for the desired |key|. If an entry is not under
// construction already, a new PendingOp structure is created.
PendingOp* GetPendingOp(const std::string& key);
// Deletes a PendingOp.
void DeletePendingOp(PendingOp* pending_op);
// Opens the disk cache entry associated with |key|, returning an ActiveEntry
// in |*entry|. |trans| will be notified via its IO callback if this method
// returns ERR_IO_PENDING. This should not be called if there already is
// an active entry associated with |key|, e.g. you should call FindActiveEntry
// first.
int OpenEntry(const std::string& key, ActiveEntry** entry,
Transaction* trans);
// Creates the disk cache entry associated with |key|, returning an
// ActiveEntry in |*entry|. |trans| will be notified via its IO callback if
// this method returns ERR_IO_PENDING.
int CreateEntry(const std::string& key, ActiveEntry** entry,
Transaction* trans);
// Destroys an ActiveEntry (active or doomed). Should only be called if
// entry->SafeToDestroy() returns true.
void DestroyEntry(ActiveEntry* entry);
// Adds a transaction to an ActiveEntry. This method returns ERR_IO_PENDING
// and the transaction will be notified about completion via its IO callback.
// In a failure case, the callback will be invoked with ERR_CACHE_RACE.
int AddTransactionToEntry(ActiveEntry* entry, Transaction* transaction);
// Transaction invokes this when its response headers phase is complete
// If the transaction is responsible for writing the response body,
// it becomes the writer and returns OK. In other cases ERR_IO_PENDING is
// returned and the transaction will be notified about completion via its
// IO callback. In a failure case, the callback will be invoked with
int DoneWithResponseHeaders(ActiveEntry* entry,
Transaction* transaction,
bool is_partial);
// Called when the transaction has finished working with this entry.
// |entry_is_complete| is true if the transaction finished reading/writing
// from the entry successfully, else it's false.
void DoneWithEntry(ActiveEntry* entry,
Transaction* transaction,
bool entry_is_complete,
bool is_partial);
// Invoked when writers wants to doom the entry and restart any queued and
// headers transactions.
// Virtual so that it can be extended in tests.
virtual void WritersDoomEntryRestartTransactions(ActiveEntry* entry);
// Invoked when current transactions in writers have completed writing to the
// cache. It may be successful completion of the response or failure as given
// by |success|. Must delete the writers object.
// |entry| is the owner of writers.
// |should_keep_entry| indicates if the entry should be doomed/destroyed.
// Virtual so that it can be extended in tests.
virtual void WritersDoneWritingToEntry(ActiveEntry* entry,
bool success,
bool should_keep_entry,
TransactionSet make_readers);
// Called when the transaction has received a non-matching response to
// validation and it's not the transaction responsible for writing the
// response body.
void DoomEntryValidationNoMatch(ActiveEntry* entry);
// Removes and returns all queued transactions in |entry| in FIFO order. This
// includes transactions that have completed the headers phase and those that
// have not been added to the entry yet in that order. |list| is the output
// argument.
void RemoveAllQueuedTransactions(ActiveEntry* entry, TransactionList* list);
// Processes either writer's failure to write response body or
// headers_transactions's failure to write headers.
void ProcessEntryFailure(ActiveEntry* entry);
// Restarts headers_transaction and done_headers_queue transactions.
void RestartHeadersPhaseTransactions(ActiveEntry* entry);
// Restarts the headers_transaction by setting its state. Since the
// headers_transaction is awaiting an asynchronous operation completion,
// it will be restarted when it's IO callback is invoked.
void RestartHeadersTransaction(ActiveEntry* entry);
// Resumes processing the queued transactions of |entry|.
void ProcessQueuedTransactions(ActiveEntry* entry);
// Checks if a transaction can be added to the entry. If yes, it will
// invoke the IO callback of the transaction. This is a helper function for
// OnProcessQueuedTransactions. It will take a transaction from
// add_to_entry_queue and make it a headers_transaction, if one doesn't exist
// already.
void ProcessAddToEntryQueue(ActiveEntry* entry);
// Returns if the transaction can join other transactions for writing to
// the cache simultaneously. It is only supported for non-Read only,
// GET requests which are not range requests.
ParallelWritingPattern CanTransactionJoinExistingWriters(
Transaction* transaction);
// Invoked when a transaction that has already completed the response headers
// phase can resume reading/writing the response body. It will invoke the IO
// callback of the transaction. This is a helper function for
// OnProcessQueuedTransactions.
void ProcessDoneHeadersQueue(ActiveEntry* entry);
// Adds a transaction to writers.
void AddTransactionToWriters(ActiveEntry* entry,
Transaction* transaction,
ParallelWritingPattern parallel_writing_pattern);
// Returns true if this transaction can write headers to the entry.
bool CanTransactionWriteResponseHeaders(ActiveEntry* entry,
Transaction* transaction,
bool is_partial,
bool is_match) const;
// Returns true if a transaction is currently writing the response body.
bool IsWritingInProgress(ActiveEntry* entry) const;
// Returns the LoadState of the provided pending transaction.
LoadState GetLoadStateForPendingTransaction(const Transaction* trans);
// Removes the transaction |trans|, from the pending list of an entry
// (PendingOp, active or doomed entry).
void RemovePendingTransaction(Transaction* trans);
// Removes the transaction |trans|, from the pending list of |entry|.
bool RemovePendingTransactionFromEntry(ActiveEntry* entry,
Transaction* trans);
// Removes the transaction |trans|, from the pending list of |pending_op|.
bool RemovePendingTransactionFromPendingOp(PendingOp* pending_op,
Transaction* trans);
// Events (called via PostTask) ---------------------------------------------
void OnProcessQueuedTransactions(ActiveEntry* entry);
// Callbacks ----------------------------------------------------------------
// Processes BackendCallback notifications.
void OnIOComplete(int result, PendingOp* entry);
// Helper to conditionally delete |pending_op| if the HttpCache object it
// is meant for has been deleted.
// TODO(ajwong): The PendingOp lifetime management is very tricky. It might
// be possible to simplify it using either base::Owned() or base::Passed()
// with the callback.
static void OnPendingOpComplete(const base::WeakPtr<HttpCache>& cache,
PendingOp* pending_op,
int result);
// Processes the backend creation notification.
void OnBackendCreated(int result, PendingOp* pending_op);
// Variables ----------------------------------------------------------------
NetLog* net_log_;
// Used when lazily constructing the disk_cache_.
std::unique_ptr<BackendFactory> backend_factory_;
bool building_backend_;
bool bypass_lock_for_test_;
bool bypass_lock_after_headers_for_test_;
bool fail_conditionalization_for_test_;
Mode mode_;
std::unique_ptr<HttpTransactionFactory> network_layer_;
std::unique_ptr<disk_cache::Backend> disk_cache_;
// The set of active entries indexed by cache key.
ActiveEntriesMap active_entries_;
// The set of doomed entries.
ActiveEntriesSet doomed_entries_;
// The set of entries "under construction".
PendingOpsMap pending_ops_;
std::unique_ptr<PlaybackCacheMap> playback_cache_map_;
// A clock that can be swapped out for testing.
base::Clock* clock_;
base::WeakPtrFactory<HttpCache> weak_factory_;
} // namespace net