Reporting: Implement cache.

net/reporting/reporting_cache.h

Index: net/reporting/reporting_cache.h
diff --git a/net/reporting/reporting_cache.h b/net/reporting/reporting_cache.h
new file mode 100644
index 0000000000000000000000000000000000000000..3bc60ad1e0949f5ca16bb6fca7c2274d6561b12d
--- /dev/null
+++ b/net/reporting/reporting_cache.h
@@ -0,0 +1,166 @@
+// Copyright 2017 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.
+
+#ifndef NET_REPORTING_REPORTING_CACHE_H_
+#define NET_REPORTING_REPORTING_CACHE_H_
+
+#include <map>
+#include <memory>
+#include <string>
+#include <unordered_map>
+#include <unordered_set>
+#include <vector>
+
+#include "base/macros.h"
+#include "base/stl_util.h"
+#include "base/time/time.h"
+#include "base/values.h"
+#include "net/base/net_export.h"
+#include "net/reporting/reporting_client.h"
+#include "url/gurl.h"
+#include "url/origin.h"
+
+namespace net {
+
+struct ReportingReport;
+
+// The cache holds undelivered reports and clients (per-origin endpoint
+// configurations) in memory. (It is not responsible for persisting them.)
+//
+// This corresponds roughly to the "Reporting cache" in the spec, except that
+// endpoints and clients are stored in a more structurally-convenient way, and
+// endpoint failures/retry-after are tracked in ReportingEndpointManager.

[Randy Smith (Not in Mondays), 2017/03/27 16:13:25]

Willing to have a README.md file with a quick sketch of the goals and a link to
the spec and design doc?  (This is based on the spec reference above, and a
concern that future readers of the code would have to do unreasonable
archaeology back to this CL to find the spec link.)

[Randy Smith (Not in Mondays), 2017/03/27 16:13:25]

Request: Go into a bit more detail in this comment about the semantics of
pending reports.

[Julia Tuttle, 2017/03/27 16:39:56]

Done.

[Julia Tuttle, 2017/03/27 16:39:56]

Yeah, sure, sounds great.

+class NET_EXPORT ReportingCache {
+ public:
+  ReportingCache();
+
+  ~ReportingCache();
+
+  // Adds a report to the cache.
+  //
+  // All parameters correspond to the desired values for the relevant fields in
+  // ReportingReport.
+  void AddReport(const GURL& url,
+                 const std::string& group,
+                 const std::string& type,
+                 std::unique_ptr<const base::Value> body,
+                 base::TimeTicks queued,
+                 int attempts);
+
+  // Gets all reports in the cache. The returned pointers are valid as long as
+  // either no calls to |RemoveReports| have happened or the reports' |pending|
+  // flag has been set to true using |SetReportsPending|. Does not return

[Randy Smith (Not in Mondays), 2017/03/27 16:13:25]

This continues to bother me; handing out raw pointers with lifetime guarantees
is dangerous, and more dangerous if those lifetime guarantees are complicated. 
If you handed out weak pointers with the same guarantee and DCHECK'd them
everywhere they were supposed to be valid I'd be happier.  Would that change be
a large one?

[Julia Tuttle, 2017/03/27 16:39:56]

It wouldn't work great for upcoming CLs -- I need the reports to be around to
log metrics, so I can't actually remove them and let the weak pointer go
dangling.

Do you want the weak pointer just as a precaution against crashes, or as the
actual lifetime management?

[Randy Smith (Not in Mondays), 2017/03/28 15:30:09]

I wasn't suggesting changing the lifetime behavior, just not vending raw
pointers.
Precaution against use-after-frees; it would turn a use-after-free into a crash.
 

+  // doomed reports (pending reports for which removal has been requested).
+  //
+  // (Clears any existing data in |*reports_out|.)
+  void GetReports(std::vector<const ReportingReport*>* reports_out) const;
+
+  // Marks a set of reports as pending. |reports| must not already be marked as
+  // pending.
+  void SetReportsPending(const std::vector<const ReportingReport*>& reports);
+
+  // Unmarks a set of reports as pending. |reports| must be previously marked as
+  // pending.
+  void ClearReportsPending(const std::vector<const ReportingReport*>& reports);
+
+  // Increments |attempts| on a set of reports.
+  void IncrementReportsAttempts(
+      const std::vector<const ReportingReport*>& reports);
+
+  // Removes a set of reports. Any reports that are pending will not be removed
+  // immediately, but rather marked doomed and removed once they are no longer
+  // pending.
+  void RemoveReports(const std::vector<const ReportingReport*>& reports);
+
+  // Removes all reports. Like |RemoveReports()|, pending reports are doomed
+  // until no longer pending.
+  void RemoveAllReports();
+
+  // Creates or updates a client for a particular origin and a particular
+  // endpoint.
+  //
+  // All parameters correspond to the desired values for the fields in
+  // |Client|.
+  //
+  // |endpoint| must use a cryptographic scheme.
+  void SetClient(const url::Origin& origin,
+                 const GURL& endpoint,
+                 ReportingClient::Subdomains subdomains,
+                 const std::string& group,
+                 base::TimeTicks expires);
+
+  // Gets all of the clients in the cache, regardless of origin or group.
+  //
+  // (Clears any existing data in |*clients_out|.)
+  void GetClients(std::vector<const ReportingClient*>* clients_out) const;
+
+  // Gets all of the clients configured for a particular origin in a particular
+  // group. The returned pointers are only guaranteed to be valid if no calls
+  // have been made to |SetClient| or |RemoveEndpoint| in between.

[Randy Smith (Not in Mondays), 2017/03/27 16:13:25]

As implied above, this type of guarantee bothers me, because it requires all
consumers of this routine to implicitly be aware of all other uses of the cache.
 Based on planned consumer usage, can you make the restriction stricter (i.e.
only valid if used synchronously--do not store), or does that bork a planned
use?  

[Julia Tuttle, 2017/03/27 16:39:56]

Oh, sure, only used synchronously would be great.

+  //
+  // (Clears any existing data in |*clients_out|.)
+  void GetClientsForOriginAndGroup(
+      const url::Origin& origin,
+      const std::string& group,
+      std::vector<const ReportingClient*>* clients_out) const;
+
+  // Removes a set of clients.
+  //
+  // May invalidate ReportingClient pointers returned by |GetClients| or
+  // |GetClientsForOriginAndGroup|.
+  void RemoveClients(const std::vector<const ReportingClient*>& clients);
+
+  // Removes a client for a particular origin and a particular endpoint.
+  void RemoveClientForOriginAndEndpoint(const url::Origin& origin,
+                                        const GURL& endpoint);
+
+  // Removes all clients whose endpoint is |endpoint|.
+  //
+  // May invalidate ReportingClient pointers returned by |GetClients| or
+  // |GetClientsForOriginAndGroup|.
+  void RemoveClientsForEndpoint(const GURL& endpoint);
+
+  // Removes all clients.
+  void RemoveAllClients();
+
+  // Gets the count of reports in the cache, *including* doomed reports.
+  //
+  // Needed to ensure that doomed reports are eventually deleted, since no
+  // method provides a view of *every* report in the cache, just non-doomed
+  // ones.
+  size_t GetFullReportCountForTesting() const { return reports_.size(); }
+
+  bool IsReportPendingForTesting(const ReportingReport* report) const {
+    return base::ContainsKey(pending_reports_, report);
+  }
+
+  bool IsReportDoomedForTesting(const ReportingReport* report) const {
+    return base::ContainsKey(doomed_reports_, report);
+  }
+
+ private:
+  // Owns all clients, keyed by origin, then endpoint URL.
+  // (These would be unordered_map, but neither url::Origin nor GURL has a hash
+  // function implemented.)
+  std::map<url::Origin, std::map<GURL, std::unique_ptr<ReportingClient>>>
+      clients_;
+
+  // Owns all reports, keyed by const raw pointer for easier lookup.
+  std::unordered_map<const ReportingReport*, std::unique_ptr<ReportingReport>>
+      reports_;
+
+  // Reports that have been marked "pending" (in use elsewhere and should not be
+  // deleted until no longer pending).
+  std::unordered_set<const ReportingReport*> pending_reports_;
+
+  // Reports that have been marked "doomed" (would have been deleted, but were
+  // pending when the deletion was requested).
+  std::unordered_set<const ReportingReport*> doomed_reports_;
+
+  DISALLOW_COPY_AND_ASSIGN(ReportingCache);
+};
+
+}  // namespace net
+
+#endif  // NET_REPORTING_REPORTING_CACHE_H_