1 // Copyright (C) 2019 The Android Open Source Project
2 //
3 // Licensed under the Apache License, Version 2.0 (the "License");
4 // you may not use this file except in compliance with the License.
5 // You may obtain a copy of the License at
6 //
7 //      http://www.apache.org/licenses/LICENSE-2.0
8 //
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS,
11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 // See the License for the specific language governing permissions and
13 // limitations under the License.
14 
15 #pragma once
16 
17 #include <stdint.h>
18 #include <unistd.h>
19 
20 #include <chrono>
21 #include <map>
22 #include <memory>
23 #include <optional>
24 #include <ostream>
25 #include <string>
26 #include <string_view>
27 #include <vector>
28 
29 #include <android-base/unique_fd.h>
30 #include <android/snapshot/snapshot.pb.h>
31 #include <fs_mgr_dm_linear.h>
32 #include <libdm/dm.h>
33 #include <libfiemap/image_manager.h>
34 #include <liblp/builder.h>
35 #include <liblp/liblp.h>
36 #include <libsnapshot/auto_device.h>
37 #include <libsnapshot/cow_writer.h>
38 #include <libsnapshot/return.h>
39 #include <snapuserd/snapuserd_client.h>
40 #include <update_engine/update_metadata.pb.h>
41 
42 #ifndef FRIEND_TEST
43 #define FRIEND_TEST(test_set_name, individual_test) \
44     friend class test_set_name##_##individual_test##_Test
45 #define DEFINED_FRIEND_TEST
46 #endif
47 
48 namespace aidl::android::hardware::boot {
49 enum class MergeStatus;
50 }
51 
52 namespace android {
53 
54 namespace fiemap {
55 class IImageManager;
56 }  // namespace fiemap
57 
58 namespace fs_mgr {
59 struct CreateLogicalPartitionParams;
60 class IPartitionOpener;
61 }  // namespace fs_mgr
62 
63 // Forward declare IBootControl types since we cannot include only the headers
64 // with Soong. Note: keep the enum width in sync.
65 
66 namespace snapshot {
67 
68 struct AutoDeleteCowImage;
69 struct AutoDeleteSnapshot;
70 struct AutoDeviceList;
71 struct PartitionCowCreator;
72 class ISnapshotMergeStats;
73 class SnapshotMergeStats;
74 class SnapshotStatus;
75 
76 using std::chrono::duration_cast;
77 using namespace std::chrono_literals;
78 
79 static constexpr const std::string_view kCowGroupName = "cow";
80 static constexpr char kVirtualAbCompressionProp[] = "ro.virtual_ab.compression.enabled";
81 
82 bool OptimizeSourceCopyOperation(const chromeos_update_engine::InstallOperation& operation,
83                                  chromeos_update_engine::InstallOperation* optimized);
84 
85 enum class CreateResult : unsigned int {
86     ERROR,
87     CREATED,
88     NOT_CREATED,
89 };
90 
91 class ISnapshotManager {
92   public:
93     // Dependency injection for testing.
94     class IDeviceInfo {
95       public:
96         using IImageManager = android::fiemap::IImageManager;
97         using MergeStatus = aidl::android::hardware::boot::MergeStatus;
98 
~IDeviceInfo()99         virtual ~IDeviceInfo() {}
100         virtual std::string GetMetadataDir() const = 0;
101         virtual std::string GetSlotSuffix() const = 0;
102         virtual std::string GetOtherSlotSuffix() const = 0;
103         virtual std::string GetSuperDevice(uint32_t slot) const = 0;
104         virtual const android::fs_mgr::IPartitionOpener& GetPartitionOpener() const = 0;
105         virtual bool IsOverlayfsSetup() const = 0;
106         virtual bool SetBootControlMergeStatus(MergeStatus status) = 0;
107         virtual bool SetSlotAsUnbootable(unsigned int slot) = 0;
108         virtual bool IsRecovery() const = 0;
IsTestDevice()109         virtual bool IsTestDevice() const { return false; }
110         virtual bool IsFirstStageInit() const = 0;
111         virtual std::unique_ptr<IImageManager> OpenImageManager() const = 0;
112         virtual android::dm::IDeviceMapper& GetDeviceMapper() = 0;
113 
114         // Helper method for implementing OpenImageManager.
115         std::unique_ptr<IImageManager> OpenImageManager(const std::string& gsid_dir) const;
116     };
117     virtual ~ISnapshotManager() = default;
118 
119     // Begin an update. This must be called before creating any snapshots. It
120     // will fail if GetUpdateState() != None.
121     virtual bool BeginUpdate() = 0;
122 
123     // Cancel an update; any snapshots will be deleted. This is allowed if the
124     // state == Initiated, None, or Unverified (before rebooting to the new
125     // slot).
126     virtual bool CancelUpdate() = 0;
127 
128     // Mark snapshot writes as having completed. After this, new snapshots cannot
129     // be created, and the device must either cancel the OTA (either before
130     // rebooting or after rolling back), or merge the OTA.
131     // Before calling this function, all snapshots must be mapped.
132     // If |wipe| is set to true, wipe is scheduled after reboot, and snapshots
133     // may need to be merged before wiping.
134     virtual bool FinishedSnapshotWrites(bool wipe) = 0;
135 
136     // Set feature flags on an ISnapshotMergeStats object.
137     virtual void SetMergeStatsFeatures(ISnapshotMergeStats* stats) = 0;
138 
139     // Update an ISnapshotMergeStats object with statistics about COW usage.
140     // This should be called before the merge begins as otherwise snapshots
141     // may be deleted.
142     virtual void UpdateCowStats(ISnapshotMergeStats* stats) = 0;
143 
144     // Initiate a merge on all snapshot devices. This should only be used after an
145     // update has been marked successful after booting.
146     virtual bool InitiateMerge() = 0;
147 
148     // Perform any necessary post-boot actions. This should be run soon after
149     // /data is mounted.
150     //
151     // If a merge is in progress, this function will block until the merge is
152     // completed.
153     //    - Callback is called periodically during the merge. If callback()
154     //      returns false during the merge, ProcessUpdateState() will pause
155     //      and returns Merging.
156     // If a merge or update was cancelled, this will clean up any
157     // update artifacts and return.
158     //
159     // Note that after calling this, GetUpdateState() may still return that a
160     // merge is in progress:
161     //   MergeFailed indicates that a fatal error occurred. WaitForMerge() may
162     //   called any number of times again to attempt to make more progress, but
163     //   we do not expect it to succeed if a catastrophic error occurred.
164     //
165     //   MergeNeedsReboot indicates that the merge has completed, but cleanup
166     //   failed. This can happen if for some reason resources were not closed
167     //   properly. In this case another reboot is needed before we can take
168     //   another OTA. However, WaitForMerge() can be called again without
169     //   rebooting, to attempt to finish cleanup anyway.
170     //
171     //   MergeCompleted indicates that the update has fully completed.
172     //   GetUpdateState will return None, and a new update can begin.
173     //
174     // The optional callback allows the caller to periodically check the
175     // progress with GetUpdateState().
176     virtual UpdateState ProcessUpdateState(const std::function<bool()>& callback = {},
177                                            const std::function<bool()>& before_cancel = {}) = 0;
178 
179     // If ProcessUpdateState() returned MergeFailed, this returns the appropriate
180     // code. Otherwise, MergeFailureCode::Ok is returned.
181     virtual MergeFailureCode ReadMergeFailureCode() = 0;
182 
183     // If an update is in progress, return the source build fingerprint.
184     virtual std::string ReadSourceBuildFingerprint() = 0;
185 
186     // Find the status of the current update, if any.
187     //
188     // |progress| depends on the returned status:
189     //   Merging: Value in the range [0, 100]
190     //   MergeCompleted: 100
191     //   Other: 0
192     virtual UpdateState GetUpdateState(double* progress = nullptr) = 0;
193 
194     // Returns true if compression is enabled for the current update. This always returns false if
195     // UpdateState is None, or no snapshots have been created.
196     virtual bool UpdateUsesCompression() = 0;
197 
198     // Returns true if userspace snapshots is enabled for the current update.
199     virtual bool UpdateUsesUserSnapshots() = 0;
200 
201     // Create necessary COW device / files for OTA clients. New logical partitions will be added to
202     // group "cow" in target_metadata. Regions of partitions of current_metadata will be
203     // "write-protected" and snapshotted.
204     virtual Return CreateUpdateSnapshots(
205             const chromeos_update_engine::DeltaArchiveManifest& manifest) = 0;
206 
207     // Map a snapshotted partition for OTA clients to write to. Write-protected regions are
208     // determined previously in CreateSnapshots.
209     //
210     // |snapshot_path| must not be nullptr.
211     //
212     // This method will return false if ro.virtual_ab.compression.enabled is true.
213     virtual bool MapUpdateSnapshot(const android::fs_mgr::CreateLogicalPartitionParams& params,
214                                    std::string* snapshot_path) = 0;
215 
216     // Create an ICowWriter to build a snapshot against a target partition. The partition name
217     // must be suffixed. If a source partition exists, it must be specified as well. The source
218     // partition will only be used if raw bytes are needed. The source partition should be an
219     // absolute path to the device, not a partition name.
220     virtual std::unique_ptr<ICowWriter> OpenSnapshotWriter(
221             const android::fs_mgr::CreateLogicalPartitionParams& params,
222             std::optional<uint64_t> label = {}) = 0;
223 
224     // Unmap a snapshot device or CowWriter that was previously opened with MapUpdateSnapshot,
225     // OpenSnapshotWriter. All outstanding open descriptors, writers, or
226     // readers must be deleted before this is called.
227     virtual bool UnmapUpdateSnapshot(const std::string& target_partition_name) = 0;
228 
229     // If this returns true, first-stage mount must call
230     // CreateLogicalAndSnapshotPartitions rather than CreateLogicalPartitions.
231     virtual bool NeedSnapshotsInFirstStageMount() = 0;
232 
233     // Perform first-stage mapping of snapshot targets. This replaces init's
234     // call to CreateLogicalPartitions when snapshots are present.
235     virtual bool CreateLogicalAndSnapshotPartitions(
236             const std::string& super_device, const std::chrono::milliseconds& timeout_ms = {}) = 0;
237 
238     // Map all snapshots. This is analogous to CreateLogicalAndSnapshotPartitions, except it maps
239     // the target slot rather than the current slot. It should only be used immediately after
240     // applying an update, before rebooting to the new slot.
241     virtual bool MapAllSnapshots(const std::chrono::milliseconds& timeout_ms = {}) = 0;
242 
243     // Unmap all snapshots. This should be called to undo MapAllSnapshots().
244     virtual bool UnmapAllSnapshots() = 0;
245 
246     // This method should be called preceding any wipe or flash of metadata or
247     // userdata. It is only valid in recovery or fastbootd, and it ensures that
248     // a merge has been completed.
249     //
250     // When userdata will be wiped or flashed, it is necessary to clean up any
251     // snapshot state. If a merge is in progress, the merge must be finished.
252     // If a snapshot is present but not yet merged, the slot must be marked as
253     // unbootable.
254     //
255     // Returns true on success (or nothing to do), false on failure. The
256     // optional callback fires periodically to query progress via GetUpdateState.
257     virtual bool HandleImminentDataWipe(const std::function<void()>& callback = {}) = 0;
258 
259     // Force a merge to complete in recovery. This is similar to HandleImminentDataWipe
260     // but does not expect a data wipe after.
261     virtual bool FinishMergeInRecovery() = 0;
262 
263     // This method is only allowed in recovery and is used as a helper to
264     // initialize the snapshot devices as a requirement to mount a snapshotted
265     // /system in recovery.
266     // This function returns:
267     // - CreateResult::CREATED if snapshot devices were successfully created;
268     // - CreateResult::NOT_CREATED if it was not necessary to create snapshot
269     // devices;
270     // - CreateResult::ERROR if a fatal error occurred, mounting /system should
271     // be aborted.
272     // This function mounts /metadata when called, and unmounts /metadata upon
273     // return.
274     virtual CreateResult RecoveryCreateSnapshotDevices() = 0;
275 
276     // Same as RecoveryCreateSnapshotDevices(), but does not auto mount/umount
277     // /metadata.
278     virtual CreateResult RecoveryCreateSnapshotDevices(
279             const std::unique_ptr<AutoDevice>& metadata_device) = 0;
280 
281     // Dump debug information.
282     virtual bool Dump(std::ostream& os) = 0;
283 
284     // Ensure metadata directory is mounted in recovery. When the returned
285     // AutoDevice is destroyed, the metadata directory is automatically
286     // unmounted.
287     // Return nullptr if any failure.
288     // In Android mode, Return an AutoDevice that does nothing
289     // In recovery, return an AutoDevice that does nothing if metadata entry
290     // is not found in fstab.
291     // Note: if this function is called the second time before the AutoDevice returned from the
292     // first call is destroyed, the device will be unmounted when any of these AutoDevices is
293     // destroyed. For example:
294     //   auto a = mgr->EnsureMetadataMounted(); // mounts
295     //   auto b = mgr->EnsureMetadataMounted(); // does nothing
296     //   b.reset() // unmounts
297     //   a.reset() // does nothing
298     virtual std::unique_ptr<AutoDevice> EnsureMetadataMounted() = 0;
299 
300     // Return the associated ISnapshotMergeStats instance. Never null.
301     virtual ISnapshotMergeStats* GetSnapshotMergeStatsInstance() = 0;
302 };
303 
304 class SnapshotManager final : public ISnapshotManager {
305     using CreateLogicalPartitionParams = android::fs_mgr::CreateLogicalPartitionParams;
306     using IPartitionOpener = android::fs_mgr::IPartitionOpener;
307     using LpMetadata = android::fs_mgr::LpMetadata;
308     using MetadataBuilder = android::fs_mgr::MetadataBuilder;
309     using DeltaArchiveManifest = chromeos_update_engine::DeltaArchiveManifest;
310     using MergeStatus = aidl::android::hardware::boot::MergeStatus;
311     using FiemapStatus = android::fiemap::FiemapStatus;
312 
313     friend class SnapshotMergeStats;
314 
315   public:
316     ~SnapshotManager();
317 
318     // Return a new SnapshotManager instance, or null on error. The device
319     // pointer is owned for the lifetime of SnapshotManager. If null, a default
320     // instance will be created.
321     static std::unique_ptr<SnapshotManager> New(IDeviceInfo* device = nullptr);
322 
323     // This is similar to New(), except designed specifically for first-stage
324     // init or recovery.
325     static std::unique_ptr<SnapshotManager> NewForFirstStageMount(IDeviceInfo* device = nullptr);
326 
327     // Helper function for first-stage init to check whether a SnapshotManager
328     // might be needed to perform first-stage mounts.
329     static bool IsSnapshotManagerNeeded();
330 
331     // Helper function for second stage init to restorecon on the rollback indicator.
332     static std::string GetGlobalRollbackIndicatorPath();
333 
334     // Populate |snapuserd_argv| with the necessary arguments to restart snapuserd
335     // after loading selinux policy.
336     bool PrepareSnapuserdArgsForSelinux(std::vector<std::string>* snapuserd_argv);
337 
338     // Detach dm-user devices from the first stage snapuserd. Load
339     // new dm-user tables after loading selinux policy.
340     bool DetachFirstStageSnapuserdForSelinux();
341 
342     // Perform the transition from the selinux stage of snapuserd into the
343     // second-stage of snapuserd. This process involves re-creating the dm-user
344     // table entries for each device, so that they connect to the new daemon.
345     // Once all new tables have been activated, we ask the first-stage daemon
346     // to cleanly exit.
347     bool PerformSecondStageInitTransition();
348 
349     // ISnapshotManager overrides.
350     bool BeginUpdate() override;
351     bool CancelUpdate() override;
352     bool FinishedSnapshotWrites(bool wipe) override;
353     void UpdateCowStats(ISnapshotMergeStats* stats) override;
354     MergeFailureCode ReadMergeFailureCode() override;
355     bool InitiateMerge() override;
356     UpdateState ProcessUpdateState(const std::function<bool()>& callback = {},
357                                    const std::function<bool()>& before_cancel = {}) override;
358     UpdateState GetUpdateState(double* progress = nullptr) override;
359     bool UpdateUsesCompression() override;
360     bool UpdateUsesUserSnapshots() override;
361     Return CreateUpdateSnapshots(const DeltaArchiveManifest& manifest) override;
362     bool MapUpdateSnapshot(const CreateLogicalPartitionParams& params,
363                            std::string* snapshot_path) override;
364     std::unique_ptr<ICowWriter> OpenSnapshotWriter(
365             const android::fs_mgr::CreateLogicalPartitionParams& params,
366             std::optional<uint64_t> label) override;
367     bool UnmapUpdateSnapshot(const std::string& target_partition_name) override;
368     bool NeedSnapshotsInFirstStageMount() override;
369     bool CreateLogicalAndSnapshotPartitions(
370             const std::string& super_device,
371             const std::chrono::milliseconds& timeout_ms = {}) override;
372     bool HandleImminentDataWipe(const std::function<void()>& callback = {}) override;
373     bool FinishMergeInRecovery() override;
374     CreateResult RecoveryCreateSnapshotDevices() override;
375     CreateResult RecoveryCreateSnapshotDevices(
376             const std::unique_ptr<AutoDevice>& metadata_device) override;
377     bool Dump(std::ostream& os) override;
378     std::unique_ptr<AutoDevice> EnsureMetadataMounted() override;
379     ISnapshotMergeStats* GetSnapshotMergeStatsInstance() override;
380     bool MapAllSnapshots(const std::chrono::milliseconds& timeout_ms = {}) override;
381     bool UnmapAllSnapshots() override;
382     std::string ReadSourceBuildFingerprint() override;
383     void SetMergeStatsFeatures(ISnapshotMergeStats* stats) override;
384 
385     // We can't use WaitForFile during first-stage init, because ueventd is not
386     // running and therefore will not automatically create symlinks. Instead,
387     // we let init provide us with the correct function to use to ensure
388     // uevents have been processed and symlink/mknod calls completed.
SetUeventRegenCallback(std::function<bool (const std::string &)> callback)389     void SetUeventRegenCallback(std::function<bool(const std::string&)> callback) {
390         uevent_regen_callback_ = callback;
391     }
392 
393     // If true, compression is enabled for this update. This is used by
394     // first-stage to decide whether to launch snapuserd.
395     bool IsSnapuserdRequired();
396 
397     // This is primarily used to device reboot. If OTA update is in progress,
398     // init will avoid killing processes
399     bool IsUserspaceSnapshotUpdateInProgress();
400 
401     enum class SnapshotDriver {
402         DM_SNAPSHOT,
403         DM_USER,
404     };
405 
406     // Add new public entries above this line.
407 
408   private:
409     FRIEND_TEST(SnapshotTest, CleanFirstStageMount);
410     FRIEND_TEST(SnapshotTest, CreateSnapshot);
411     FRIEND_TEST(SnapshotTest, FirstStageMountAfterRollback);
412     FRIEND_TEST(SnapshotTest, FirstStageMountAndMerge);
413     FRIEND_TEST(SnapshotTest, FlagCheck);
414     FRIEND_TEST(SnapshotTest, FlashSuperDuringMerge);
415     FRIEND_TEST(SnapshotTest, FlashSuperDuringUpdate);
416     FRIEND_TEST(SnapshotTest, MapPartialSnapshot);
417     FRIEND_TEST(SnapshotTest, MapSnapshot);
418     FRIEND_TEST(SnapshotTest, Merge);
419     FRIEND_TEST(SnapshotTest, MergeFailureCode);
420     FRIEND_TEST(SnapshotTest, NoMergeBeforeReboot);
421     FRIEND_TEST(SnapshotTest, UpdateBootControlHal);
422     FRIEND_TEST(SnapshotTest, BootSnapshotWithoutSlotSwitch);
423     FRIEND_TEST(SnapshotUpdateTest, AddPartition);
424     FRIEND_TEST(SnapshotUpdateTest, ConsistencyCheckResume);
425     FRIEND_TEST(SnapshotUpdateTest, DaemonTransition);
426     FRIEND_TEST(SnapshotUpdateTest, DataWipeAfterRollback);
427     FRIEND_TEST(SnapshotUpdateTest, DataWipeRollbackInRecovery);
428     FRIEND_TEST(SnapshotUpdateTest, DataWipeWithStaleSnapshots);
429     FRIEND_TEST(SnapshotUpdateTest, FlagCheck);
430     FRIEND_TEST(SnapshotUpdateTest, FullUpdateFlow);
431     FRIEND_TEST(SnapshotUpdateTest, MergeCannotRemoveCow);
432     FRIEND_TEST(SnapshotUpdateTest, MergeInRecovery);
433     FRIEND_TEST(SnapshotUpdateTest, QueryStatusError);
434     FRIEND_TEST(SnapshotUpdateTest, SnapshotStatusFileWithoutCow);
435     FRIEND_TEST(SnapshotUpdateTest, SpaceSwapUpdate);
436     FRIEND_TEST(SnapshotUpdateTest, MapAllSnapshotsWithoutSlotSwitch);
437     friend class SnapshotTest;
438     friend class SnapshotUpdateTest;
439     friend class FlashAfterUpdateTest;
440     friend class LockTestConsumer;
441     friend class SnapshotFuzzEnv;
442     friend class MapSnapshots;
443     friend struct AutoDeleteCowImage;
444     friend struct AutoDeleteSnapshot;
445     friend struct PartitionCowCreator;
446 
447     using DmTargetSnapshot = android::dm::DmTargetSnapshot;
448     using IImageManager = android::fiemap::IImageManager;
449     using TargetInfo = android::dm::DeviceMapper::TargetInfo;
450 
451     explicit SnapshotManager(IDeviceInfo* info);
452 
453     // This is created lazily since it can connect via binder.
454     bool EnsureImageManager();
455 
456     // Ensure we're connected to snapuserd.
457     bool EnsureSnapuserdConnected(std::chrono::milliseconds timeout_ms = 10s);
458 
459     // Helpers for first-stage init.
device()460     const std::unique_ptr<IDeviceInfo>& device() const { return device_; }
461 
462     // Helper functions for tests.
image_manager()463     IImageManager* image_manager() const { return images_.get(); }
set_use_first_stage_snapuserd(bool value)464     void set_use_first_stage_snapuserd(bool value) { use_first_stage_snapuserd_ = value; }
465 
466     // Since libsnapshot is included into multiple processes, we flock() our
467     // files for simple synchronization. LockedFile is a helper to assist with
468     // this. It also serves as a proof-of-lock for some functions.
469     class LockedFile final {
470       public:
LockedFile(const std::string & path,android::base::unique_fd && fd,int lock_mode)471         LockedFile(const std::string& path, android::base::unique_fd&& fd, int lock_mode)
472             : path_(path), fd_(std::move(fd)), lock_mode_(lock_mode) {}
473         ~LockedFile();
lock_mode()474         int lock_mode() const { return lock_mode_; }
475 
476       private:
477         std::string path_;
478         android::base::unique_fd fd_;
479         int lock_mode_;
480     };
481     static std::unique_ptr<LockedFile> OpenFile(const std::string& file, int lock_flags);
482 
483     SnapshotDriver GetSnapshotDriver(LockedFile* lock);
484 
485     // Create a new snapshot record. This creates the backing COW store and
486     // persists information needed to map the device. The device can be mapped
487     // with MapSnapshot().
488     //
489     // |status|.device_size should be the size of the base_device that will be passed
490     // via MapDevice(). |status|.snapshot_size should be the number of bytes in the
491     // base device, starting from 0, that will be snapshotted. |status|.cow_file_size
492     // should be the amount of space that will be allocated to store snapshot
493     // deltas.
494     //
495     // If |status|.snapshot_size < |status|.device_size, then the device will always
496     // be mapped with two table entries: a dm-snapshot range covering
497     // snapshot_size, and a dm-linear range covering the remainder.
498     //
499     // All sizes are specified in bytes, and the device, snapshot, COW partition and COW file sizes
500     // must be a multiple of the sector size (512 bytes).
501     bool CreateSnapshot(LockedFile* lock, PartitionCowCreator* cow_creator, SnapshotStatus* status);
502 
503     // |name| should be the base partition name (e.g. "system_a"). Create the
504     // backing COW image using the size previously passed to CreateSnapshot().
505     Return CreateCowImage(LockedFile* lock, const std::string& name);
506 
507     // Map a snapshot device that was previously created with CreateSnapshot.
508     // If a merge was previously initiated, the device-mapper table will have a
509     // snapshot-merge target instead of a snapshot target. If the timeout
510     // parameter greater than zero, this function will wait the given amount
511     // of time for |dev_path| to become available, and fail otherwise. If
512     // timeout_ms is 0, then no wait will occur and |dev_path| may not yet
513     // exist on return.
514     bool MapSnapshot(LockedFile* lock, const std::string& name, const std::string& base_device,
515                      const std::string& cow_device, const std::chrono::milliseconds& timeout_ms,
516                      std::string* dev_path);
517 
518     // Create a dm-user device for a given snapshot.
519     bool MapDmUserCow(LockedFile* lock, const std::string& name, const std::string& cow_file,
520                       const std::string& base_device, const std::string& base_path_merge,
521                       const std::chrono::milliseconds& timeout_ms, std::string* path);
522 
523     // Map the source device used for dm-user.
524     bool MapSourceDevice(LockedFile* lock, const std::string& name,
525                          const std::chrono::milliseconds& timeout_ms, std::string* path);
526 
527     // Map a COW image that was previous created with CreateCowImage.
528     std::optional<std::string> MapCowImage(const std::string& name,
529                                            const std::chrono::milliseconds& timeout_ms);
530 
531     // Remove the backing copy-on-write image and snapshot states for the named snapshot. The
532     // caller is responsible for ensuring that the snapshot is unmapped.
533     bool DeleteSnapshot(LockedFile* lock, const std::string& name);
534 
535     // Unmap a snapshot device previously mapped with MapSnapshotDevice().
536     bool UnmapSnapshot(LockedFile* lock, const std::string& name);
537 
538     // Unmap a COW image device previously mapped with MapCowImage().
539     bool UnmapCowImage(const std::string& name);
540 
541     // Unmap a COW and remove it from a MetadataBuilder.
542     void UnmapAndDeleteCowPartition(MetadataBuilder* current_metadata);
543 
544     // Remove invalid snapshots if any
545     void RemoveInvalidSnapshots(LockedFile* lock);
546 
547     // Unmap and remove all known snapshots.
548     bool RemoveAllSnapshots(LockedFile* lock);
549 
550     // Boot device off snapshots without slot switch
551     bool BootFromSnapshotsWithoutSlotSwitch();
552 
553     // Remove kBootSnapshotsWithoutSlotSwitch so that device can boot
554     // without snapshots on the current slot
555     bool PrepareDeviceToBootWithoutSnapshot();
556 
557     // Is the kBootSnapshotsWithoutSlotSwitch present
558     bool IsSnapshotWithoutSlotSwitch();
559 
560     // List the known snapshot names.
561     bool ListSnapshots(LockedFile* lock, std::vector<std::string>* snapshots,
562                        const std::string& suffix = "");
563 
564     // Check for a cancelled or rolled back merge, returning true if such a
565     // condition was detected and handled.
566     bool HandleCancelledUpdate(LockedFile* lock, const std::function<bool()>& before_cancel);
567 
568     // Helper for HandleCancelledUpdate. Assumes booting from new slot.
569     bool AreAllSnapshotsCancelled(LockedFile* lock);
570 
571     // Determine whether partition names in |snapshots| have been flashed and
572     // store result to |out|.
573     // Return true if values are successfully retrieved and false on error
574     // (e.g. super partition metadata cannot be read). When it returns true,
575     // |out| stores true for partitions that have been flashed and false for
576     // partitions that have not been flashed.
577     bool GetSnapshotFlashingStatus(LockedFile* lock, const std::vector<std::string>& snapshots,
578                                    std::map<std::string, bool>* out);
579 
580     // Remove artifacts created by the update process, such as snapshots, and
581     // set the update state to None.
582     bool RemoveAllUpdateState(LockedFile* lock, const std::function<bool()>& prolog = {});
583 
584     // Interact with /metadata/ota.
585     std::unique_ptr<LockedFile> OpenLock(int lock_flags);
586     std::unique_ptr<LockedFile> LockShared();
587     std::unique_ptr<LockedFile> LockExclusive();
588     std::string GetLockPath() const;
589 
590     // Interact with /metadata/ota/state.
591     UpdateState ReadUpdateState(LockedFile* file);
592     SnapshotUpdateStatus ReadSnapshotUpdateStatus(LockedFile* file);
593     bool WriteUpdateState(LockedFile* file, UpdateState state,
594                           MergeFailureCode failure_code = MergeFailureCode::Ok);
595     bool WriteSnapshotUpdateStatus(LockedFile* file, const SnapshotUpdateStatus& status);
596     std::string GetStateFilePath() const;
597 
598     // Interact with /metadata/ota/merge_state.
599     // This file contains information related to the snapshot merge process.
600     std::string GetMergeStateFilePath() const;
601 
602     // Helpers for merging.
603     MergeFailureCode MergeSecondPhaseSnapshots(LockedFile* lock);
604     MergeFailureCode SwitchSnapshotToMerge(LockedFile* lock, const std::string& name);
605     MergeFailureCode RewriteSnapshotDeviceTable(const std::string& dm_name);
606     bool MarkSnapshotMergeCompleted(LockedFile* snapshot_lock, const std::string& snapshot_name);
607     void AcknowledgeMergeSuccess(LockedFile* lock);
608     void AcknowledgeMergeFailure(MergeFailureCode failure_code);
609     MergePhase DecideMergePhase(const SnapshotStatus& status);
610     std::unique_ptr<LpMetadata> ReadCurrentMetadata();
611 
612     enum class MetadataPartitionState {
613         // Partition does not exist.
614         None,
615         // Partition is flashed.
616         Flashed,
617         // Partition is created by OTA client.
618         Updated,
619     };
620     // Helper function to check the state of a partition as described in metadata.
621     MetadataPartitionState GetMetadataPartitionState(const LpMetadata& metadata,
622                                                      const std::string& name);
623 
624     // Note that these require the name of the device containing the snapshot,
625     // which may be the "inner" device. Use GetsnapshotDeviecName().
626     bool QuerySnapshotStatus(const std::string& dm_name, std::string* target_type,
627                              DmTargetSnapshot::Status* status);
628     bool IsSnapshotDevice(const std::string& dm_name, TargetInfo* target = nullptr);
629 
630     // Internal callback for when merging is complete.
631     bool OnSnapshotMergeComplete(LockedFile* lock, const std::string& name,
632                                  const SnapshotStatus& status);
633     bool CollapseSnapshotDevice(LockedFile* lock, const std::string& name,
634                                 const SnapshotStatus& status);
635 
636     struct [[nodiscard]] MergeResult {
637         explicit MergeResult(UpdateState state,
638                              MergeFailureCode failure_code = MergeFailureCode::Ok)
stateMergeResult639             : state(state), failure_code(failure_code) {}
640         UpdateState state;
641         MergeFailureCode failure_code;
642     };
643 
644     // Only the following UpdateStates are used here:
645     //   UpdateState::Merging
646     //   UpdateState::MergeCompleted
647     //   UpdateState::MergeFailed
648     //   UpdateState::MergeNeedsReboot
649     MergeResult CheckMergeState(const std::function<bool()>& before_cancel);
650     MergeResult CheckMergeState(LockedFile* lock, const std::function<bool()>& before_cancel);
651     MergeResult CheckTargetMergeState(LockedFile* lock, const std::string& name,
652                                       const SnapshotUpdateStatus& update_status);
653 
654     auto UpdateStateToStr(enum UpdateState state);
655     // Get status or table information about a device-mapper node with a single target.
656     enum class TableQuery {
657         Table,
658         Status,
659     };
660     bool GetSingleTarget(const std::string& dm_name, TableQuery query,
661                          android::dm::DeviceMapper::TargetInfo* target);
662 
663     // Interact with status files under /metadata/ota/snapshots.
664     bool WriteSnapshotStatus(LockedFile* lock, const SnapshotStatus& status);
665     bool ReadSnapshotStatus(LockedFile* lock, const std::string& name, SnapshotStatus* status);
666     std::string GetSnapshotStatusFilePath(const std::string& name);
667 
668     std::string GetSnapshotBootIndicatorPath();
669     std::string GetRollbackIndicatorPath();
670     std::string GetForwardMergeIndicatorPath();
671     std::string GetOldPartitionMetadataPath();
672     std::string GetBootSnapshotsWithoutSlotSwitchPath();
673 
674     const LpMetadata* ReadOldPartitionMetadata(LockedFile* lock);
675 
676     bool MapAllPartitions(LockedFile* lock, const std::string& super_device, uint32_t slot,
677                           const std::chrono::milliseconds& timeout_ms);
678 
679     // Reason for calling MapPartitionWithSnapshot.
680     enum class SnapshotContext {
681         // For writing or verification (during update_engine).
682         Update,
683 
684         // For mounting a full readable device.
685         Mount,
686     };
687 
688     struct SnapshotPaths {
689         // Target/base device (eg system_b), always present.
690         std::string target_device;
691 
692         // COW name (eg system_cow). Not present if no COW is needed.
693         std::string cow_device_name;
694 
695         // dm-snapshot instance. Not present in Update mode for VABC.
696         std::string snapshot_device;
697     };
698 
699     // Helpers for OpenSnapshotWriter.
700     std::unique_ptr<ICowWriter> OpenCompressedSnapshotWriter(LockedFile* lock,
701                                                              const SnapshotStatus& status,
702                                                              const SnapshotPaths& paths,
703                                                              std::optional<uint64_t> label);
704 
705     // Map the base device, COW devices, and snapshot device.
706     bool MapPartitionWithSnapshot(LockedFile* lock, CreateLogicalPartitionParams params,
707                                   SnapshotContext context, SnapshotPaths* paths);
708 
709     // Map the COW devices, including the partition in super and the images.
710     // |params|:
711     //    - |partition_name| should be the name of the top-level partition (e.g. system_b),
712     //            not system_b-cow-img
713     //    - |device_name| and |partition| is ignored
714     //    - |timeout_ms| and the rest is respected
715     // Return the path in |cow_device_path| (e.g. /dev/block/dm-1) and major:minor in
716     // |cow_device_string|
717     bool MapCowDevices(LockedFile* lock, const CreateLogicalPartitionParams& params,
718                        const SnapshotStatus& snapshot_status, AutoDeviceList* created_devices,
719                        std::string* cow_name);
720 
721     // The reverse of MapCowDevices.
722     bool UnmapCowDevices(LockedFile* lock, const std::string& name);
723 
724     // The reverse of MapPartitionWithSnapshot.
725     bool UnmapPartitionWithSnapshot(LockedFile* lock, const std::string& target_partition_name);
726 
727     // Unmap a dm-user device through snapuserd.
728     bool UnmapDmUserDevice(const std::string& dm_user_name);
729 
730     // Unmap a dm-user device for user space snapshots
731     bool UnmapUserspaceSnapshotDevice(LockedFile* lock, const std::string& snapshot_name);
732 
733     // If there isn't a previous update, return true. |needs_merge| is set to false.
734     // If there is a previous update but the device has not boot into it, tries to cancel the
735     //   update and delete any snapshots. Return true if successful. |needs_merge| is set to false.
736     // If there is a previous update and the device has boot into it, do nothing and return true.
737     //   |needs_merge| is set to true.
738     bool TryCancelUpdate(bool* needs_merge);
739 
740     // Helper for CreateUpdateSnapshots.
741     // Creates all underlying images, COW partitions and snapshot files. Does not initialize them.
742     Return CreateUpdateSnapshotsInternal(
743             LockedFile* lock, const DeltaArchiveManifest& manifest,
744             PartitionCowCreator* cow_creator, AutoDeviceList* created_devices,
745             std::map<std::string, SnapshotStatus>* all_snapshot_status);
746 
747     // Initialize snapshots so that they can be mapped later.
748     // Map the COW partition and zero-initialize the header.
749     Return InitializeUpdateSnapshots(
750             LockedFile* lock, uint32_t cow_version, MetadataBuilder* target_metadata,
751             const LpMetadata* exported_target_metadata, const std::string& target_suffix,
752             const std::map<std::string, SnapshotStatus>& all_snapshot_status);
753 
754     // Implementation of UnmapAllSnapshots(), with the lock provided.
755     bool UnmapAllSnapshots(LockedFile* lock);
756 
757     // Unmap all partitions that were mapped by CreateLogicalAndSnapshotPartitions.
758     // This should only be called in recovery.
759     bool UnmapAllPartitionsInRecovery();
760 
761     // Check no snapshot overflows. Note that this returns false negatives if the snapshot
762     // overflows, then is remapped and not written afterwards.
763     bool EnsureNoOverflowSnapshot(LockedFile* lock);
764 
765     enum class Slot { Unknown, Source, Target };
766     friend std::ostream& operator<<(std::ostream& os, SnapshotManager::Slot slot);
767     Slot GetCurrentSlot();
768 
769     // Return the suffix we expect snapshots to have.
770     std::string GetSnapshotSlotSuffix();
771 
772     std::string ReadUpdateSourceSlotSuffix();
773 
774     // Helper for RemoveAllSnapshots.
775     // Check whether |name| should be deleted as a snapshot name.
776     bool ShouldDeleteSnapshot(const std::map<std::string, bool>& flashing_status, Slot current_slot,
777                               const std::string& name);
778 
779     // Create or delete forward merge indicator given |wipe|. Iff wipe is scheduled,
780     // allow forward merge on FDR.
781     bool UpdateForwardMergeIndicator(bool wipe);
782 
783     // Helper for HandleImminentDataWipe.
784     // Call ProcessUpdateState and handle states with special rules before data wipe. Specifically,
785     // if |allow_forward_merge| and allow-forward-merge indicator exists, initiate merge if
786     // necessary.
787     UpdateState ProcessUpdateStateOnDataWipe(bool allow_forward_merge,
788                                              const std::function<bool()>& callback);
789 
790     // Return device string of a mapped image, or if it is not available, the mapped image path.
791     bool GetMappedImageDeviceStringOrPath(const std::string& device_name,
792                                           std::string* device_string_or_mapped_path);
793 
794     // Same as above, but for paths only (no major:minor device strings).
795     bool GetMappedImageDevicePath(const std::string& device_name, std::string* device_path);
796 
797     // Wait for a device to be created by ueventd (eg, its symlink or node to be populated).
798     // This is needed for any code that uses device-mapper path in first-stage init. If
799     // |timeout_ms| is empty or the given device is not a path, WaitForDevice immediately
800     // returns true.
801     bool WaitForDevice(const std::string& device, std::chrono::milliseconds timeout_ms);
802 
803     enum class InitTransition { SELINUX_DETACH, SECOND_STAGE };
804 
805     // Initiate the transition from first-stage to second-stage snapuserd. This
806     // process involves re-creating the dm-user table entries for each device,
807     // so that they connect to the new daemon. Once all new tables have been
808     // activated, we ask the first-stage daemon to cleanly exit.
809     //
810     // If the mode is SELINUX_DETACH, snapuserd_argv must be non-null and will
811     // be populated with a list of snapuserd arguments to pass to execve(). It
812     // is otherwise ignored.
813     bool PerformInitTransition(InitTransition transition,
814                                std::vector<std::string>* snapuserd_argv = nullptr);
815 
snapuserd_client()816     SnapuserdClient* snapuserd_client() const { return snapuserd_client_.get(); }
817 
818     // Helper of UpdateUsesCompression
819     bool UpdateUsesCompression(LockedFile* lock);
820     // Locked and unlocked functions to test whether the current update uses
821     // userspace snapshots.
822     bool UpdateUsesUserSnapshots(LockedFile* lock);
823 
824     // Check if io_uring API's need to be used
825     bool UpdateUsesIouring(LockedFile* lock);
826 
827     // Check if direct reads are enabled for the source image
828     bool UpdateUsesODirect(LockedFile* lock);
829 
830     // Wrapper around libdm, with diagnostics.
831     bool DeleteDeviceIfExists(const std::string& name,
832                               const std::chrono::milliseconds& timeout_ms = {});
833 
834     // Set read-ahead size during OTA
835     void SetReadAheadSize(const std::string& entry_block_device, off64_t size_kb);
836 
837     // Returns true post OTA reboot if legacy snapuserd is required
838     bool IsLegacySnapuserdPostReboot();
839 
840     android::dm::IDeviceMapper& dm_;
841     std::unique_ptr<IDeviceInfo> device_;
842     std::string metadata_dir_;
843     std::unique_ptr<IImageManager> images_;
844     bool use_first_stage_snapuserd_ = false;
845     bool in_factory_data_reset_ = false;
846     std::function<bool(const std::string&)> uevent_regen_callback_;
847     std::unique_ptr<SnapuserdClient> snapuserd_client_;
848     std::unique_ptr<LpMetadata> old_partition_metadata_;
849     std::optional<bool> is_snapshot_userspace_;
850     std::optional<bool> is_legacy_snapuserd_;
851 };
852 
853 }  // namespace snapshot
854 }  // namespace android
855 
856 #ifdef DEFINED_FRIEND_TEST
857 #undef DEFINED_FRIEND_TEST
858 #undef FRIEND_TEST
859 #endif
860