storaged: add storaged native daemon

Storaged periodically checks the devices' disk stats and eMMC lifetime.
This information is logged to both the kernel and event log. By event
logging, the data can be aggregated by the checkin service and then
sent to Sawmill.

It also periodically traverses /proc/[pid], maintaining the I/O usage of
all tasks (all sampled tasks). The task I/O data can be reported using
the --dump option.

Storaged is booted when the device boots up and requires the permission
for reading /proc/diskstats, /proc/[pid]/io and debugfs(eMMC ext_csd).
For detailed description, please refer to go/storaged.

Bug: 28826771
Change-Id: I774b1a44cc30092bac1bfcbc08bf487295e7cfed

3 files changed, 445 insertions, 0 deletions

diff --git a/storaged/include/storaged.h b/storaged/include/storaged.h
new file mode 100644
index 000000000..eb827cf9d
--- /dev/null
+++ b/storaged/include/storaged.h
@@ -0,0 +1,340 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef _STORAGED_H_
+#define _STORAGED_H_
+
+#define DEBUG
+
+#include <queue>
+#include <semaphore.h>
+#include <stdint.h>
+#include <string>
+#include <syslog.h>
+#include <unordered_map>
+#include <vector>
+
+#define FRIEND_TEST(test_case_name, test_name) \
+friend class test_case_name##_##test_name##_Test
+
+/* For debug */
+#ifdef DEBUG
+#define debuginfo(fmt, ...) \
+ do {printf("%s():\t" fmt "\t[%s:%d]\n", __FUNCTION__, ##__VA_ARGS__, __FILE__, __LINE__);} \
+ while(0)
+#else
+#define debuginfo(...)
+#endif
+
+#define KMSG_PRIORITY(PRI)                            \
+    '<',                                              \
+    '0' + LOG_MAKEPRI(LOG_DAEMON, LOG_PRI(PRI)) / 10, \
+    '0' + LOG_MAKEPRI(LOG_DAEMON, LOG_PRI(PRI)) % 10, \
+    '>'
+
+static char kmsg_error_prefix[] = { KMSG_PRIORITY(LOG_ERR),
+    's', 't', 'o', 'r', 'a', 'g', 'e', 'd', ':', '\0' };
+
+static char kmsg_info_prefix[] = { KMSG_PRIORITY(LOG_INFO),
+    's', 't', 'o', 'r', 'a', 'g', 'e', 'd', ':', '\0' };
+
+static char kmsg_warning_prefix[] = { KMSG_PRIORITY(LOG_WARNING),
+    's', 't', 'o', 'r', 'a', 'g', 'e', 'd', ':', '\0' };
+
+// number of attributes diskstats has
+#define DISK_STATS_SIZE ( 11 )
+// maximum size limit of a stats file
+#define DISK_STATS_FILE_MAX_SIZE ( 256 )
+#define DISK_STATS_IO_IN_FLIGHT_IDX ( 8 )
+struct disk_stats {
+    /* It will be extremely unlikely for any of the following entries to overflow.
+     * For read_bytes(which will be greater than any of the following entries), it
+     * will take 27 years to overflow uint64_t at the reading rate of 20GB/s, which
+     * is the peak memory transfer rate for current memory.
+     * The diskstats entries (first 11) need to be at top in this structure _after_
+     * compiler's optimization.
+     */
+    uint64_t read_ios;       // number of read I/Os processed
+    uint64_t read_merges;    // number of read I/Os merged with in-queue I/Os
+    uint64_t read_sectors;   // number of sectors read
+    uint64_t read_ticks;     // total wait time for read requests
+    uint64_t write_ios;      // number of write I/Os processed
+    uint64_t write_merges;   // number of write I/Os merged with in-queue I/Os
+    uint64_t write_sectors;  // number of sectors written
+    uint64_t write_ticks;    // total wait time for write requests
+    uint64_t io_in_flight;   // number of I/Os currently in flight
+    uint64_t io_ticks;       // total time this block device has been active
+    uint64_t io_in_queue;    // total wait time for all requests
+
+    uint64_t start_time;     // monotonic time accounting starts
+    uint64_t end_time;       // monotonic time accounting ends
+    uint32_t counter;        // private counter for accumulate calculations
+    double   io_avg;         // average io_in_flight for accumulate calculations
+};
+
+#define MMC_VER_STR_LEN ( 8 )   // maximum length of the MMC version string
+// minimum size of a ext_csd file
+#define EXT_CSD_FILE_MIN_SIZE ( 1024 )
+struct emmc_info {
+    int eol;                        // pre-eol (end of life) information
+    int lifetime_a;                 // device life time estimation (type A)
+    int lifetime_b;                 // device life time estimation (type B)
+    char mmc_ver[MMC_VER_STR_LEN];  // device version string
+};
+
+struct disk_perf {
+    uint32_t read_perf;         // read speed (kbytes/s)
+    uint32_t read_ios;          // read I/Os per second
+    uint32_t write_perf;        // write speed (kbytes/s)
+    uint32_t write_ios;         // write I/Os per second
+    uint32_t queue;             // I/Os in queue
+};
+
+#define CMD_MAX_LEN ( 64 )
+struct task_info {
+    uint32_t pid;                   // task id
+    uint64_t rchar;                 // characters read
+    uint64_t wchar;                 // characters written
+    uint64_t syscr;                 // read syscalls
+    uint64_t syscw;                 // write syscalls
+    uint64_t read_bytes;            // bytes read (from storage layer)
+    uint64_t write_bytes;           // bytes written (to storage layer)
+    uint64_t cancelled_write_bytes; // cancelled write byte by truncate
+
+    uint64_t starttime;             // start time of task
+
+    char cmd[CMD_MAX_LEN];          // filename of the executable
+};
+
+class lock_t {
+    sem_t* mSem;
+public:
+    lock_t(sem_t* sem) {
+        mSem = sem;
+        sem_wait(mSem);
+    }
+    ~lock_t() {
+        sem_post(mSem);
+    }
+};
+
+class tasks_t {
+private:
+    FRIEND_TEST(storaged_test, tasks_t);
+    sem_t mSem;
+    // hashmap for all running tasks w/ pid as key
+    std::unordered_map<uint32_t, struct task_info> mRunning;
+    // hashmap for all tasks that have been killed (categorized by cmd) w/ cmd as key
+    std::unordered_map<std::string, struct task_info> mOld;
+    std::unordered_map<std::uint32_t, struct task_info> get_running_tasks();
+public:
+    tasks_t() {
+        sem_init(&mSem, 0, 1); // TODO: constructor don't have a return value, what if sem_init fails
+    }
+
+    ~tasks_t() {
+        sem_destroy(&mSem);
+    }
+
+    void update_running_tasks(void);
+    std::vector<struct task_info> get_tasks(void);
+};
+
+class stream_stats {
+private:
+    double mSum;
+    double mSquareSum;
+    uint32_t mCnt;
+public:
+    stream_stats() : mSum(0), mSquareSum(0), mCnt(0) {};
+    ~stream_stats() {};
+    double get_mean() {
+        return mSum / mCnt;
+    }
+    double get_std() {
+        return sqrt(mSquareSum / mCnt - mSum * mSum / (mCnt * mCnt));
+    }
+    void add(uint32_t num) {
+        mSum += (double)num;
+        mSquareSum += (double)num * (double)num;
+        mCnt++;
+    }
+    void evict(uint32_t num) {
+        if (mSum < num || mSquareSum < (double)num * (double)num) return;
+        mSum -= (double)num;
+        mSquareSum -= (double)num * (double)num;
+        mCnt--;
+    }
+};
+
+#define MMC_DISK_STATS_PATH "/sys/block/mmcblk0/stat"
+#define SDA_DISK_STATS_PATH "/sys/block/sda/stat"
+#define EMMC_ECSD_PATH "/d/mmc0/mmc0:0001/ext_csd"
+class disk_stats_monitor {
+private:
+    FRIEND_TEST(storaged_test, disk_stats_monitor);
+    const char* DISK_STATS_PATH;
+    struct disk_stats mPrevious;
+    struct disk_stats mAccumulate;
+    bool mStall;
+    std::queue<struct disk_perf> mBuffer;
+    struct {
+        stream_stats read_perf;           // read speed (bytes/s)
+        stream_stats read_ios;            // read I/Os per second
+        stream_stats write_perf;          // write speed (bytes/s)
+        stream_stats write_ios;           // write I/O per second
+        stream_stats queue;               // I/Os in queue
+    } mStats;
+    bool mValid;
+    const uint32_t mWindow;
+    const double mSigma;
+    struct disk_perf mMean;
+    struct disk_perf mStd;
+
+    void update_mean();
+    void update_std();
+    void add(struct disk_perf* perf);
+    void evict(struct disk_perf* perf);
+    bool detect(struct disk_perf* perf);
+
+    void update(struct disk_stats* stats);
+
+public:
+    disk_stats_monitor(uint32_t window_size = 5, double sigma = 1.0) :
+            mStall(false),
+            mValid(false),
+            mWindow(window_size),
+            mSigma(sigma) {
+        memset(&mPrevious, 0, sizeof(mPrevious));
+        memset(&mMean, 0, sizeof(mMean));
+        memset(&mStd, 0, sizeof(mStd));
+
+        if (access(MMC_DISK_STATS_PATH, R_OK) >= 0) {
+            DISK_STATS_PATH = MMC_DISK_STATS_PATH;
+        } else {
+            DISK_STATS_PATH = SDA_DISK_STATS_PATH;
+        }
+    }
+    void update(void);
+};
+
+class disk_stats_publisher {
+private:
+    FRIEND_TEST(storaged_test, disk_stats_publisher);
+    const char* DISK_STATS_PATH;
+    struct disk_stats mAccumulate;
+    struct disk_stats mPrevious;
+public:
+    disk_stats_publisher(void) {
+        memset(&mAccumulate, 0, sizeof(struct disk_stats));
+        memset(&mPrevious, 0, sizeof(struct disk_stats));
+
+        if (access(MMC_DISK_STATS_PATH, R_OK) >= 0) {
+            DISK_STATS_PATH = MMC_DISK_STATS_PATH;
+        } else {
+            DISK_STATS_PATH = SDA_DISK_STATS_PATH;
+        }
+    }
+
+    ~disk_stats_publisher(void) {}
+    void publish(void);
+    void update(void);
+};
+
+class emmc_info_t {
+private:
+    struct emmc_info mInfo;
+    bool mValid;
+    int mFdEmmc;
+public:
+    emmc_info_t(void) :
+            mValid(false),
+            mFdEmmc(-1) {
+        memset(&mInfo, 0, sizeof(struct emmc_info));
+    }
+    ~emmc_info_t(void) {}
+
+    void publish(void);
+    void update(void);
+    void set_emmc_fd(int fd) {
+        mFdEmmc = fd;
+    }
+};
+
+// Periodic chores intervals in seconds
+#define DEFAULT_PERIODIC_CHORES_INTERVAL_UNIT ( 20 )
+#define DEFAULT_PERIODIC_CHORES_INTERVAL_DISK_STATS_PUBLISH ( 60 )
+#define DEFAULT_PERIODIC_CHORES_INTERVAL_EMMC_INFO_PUBLISH ( 60 * 2 )
+
+struct storaged_config {
+    int periodic_chores_interval_unit;
+    int periodic_chores_interval_disk_stats_publish;
+    int periodic_chores_interval_emmc_info_publish;
+    bool proc_taskio_readable;  // are /proc/[pid]/{io, comm, cmdline, stat} all readable
+    bool emmc_available;        // whether eMMC est_csd file is readable
+    bool diskstats_available;   // whether diskstats is accessible
+};
+
+class storaged_t {
+private:
+    time_t mTimer;
+    storaged_config mConfig;
+    disk_stats_publisher mDiskStats;
+    disk_stats_monitor mDsm;
+    emmc_info_t mEmmcInfo;
+    tasks_t mTasks;
+    time_t mStarttime;
+public:
+    storaged_t(void);
+    ~storaged_t() {}
+    void event(void);
+    void pause(void) {
+        sleep(mConfig.periodic_chores_interval_unit);
+    }
+    void set_unit_interval(int unit) {
+        mConfig.periodic_chores_interval_unit = unit;
+    }
+    void set_diskstats_interval(int disk_stats) {
+        mConfig.periodic_chores_interval_disk_stats_publish = disk_stats;
+    }
+    void set_emmc_interval(int emmc_info) {
+        mConfig.periodic_chores_interval_emmc_info_publish = emmc_info;
+    }
+    std::vector<struct task_info> get_tasks(void) {
+        // There could be a race when get_tasks() and the main thread is updating at the same time
+        // While update_running_tasks() is updating the critical sections at the end of the function
+        // all together atomically, the final state of task_t can only be either the main thread's
+        // update or this update. Since the race can only occur when both threads are updating
+        // "simultaneously", either final state is acceptable.
+        mTasks.update_running_tasks();
+        return mTasks.get_tasks();
+    }
+
+    void set_privileged_fds(int fd_emmc) {
+        mEmmcInfo.set_emmc_fd(fd_emmc);
+    }
+
+    time_t get_starttime(void) {
+        return mStarttime;
+    }
+};
+
+// Eventlog tag
+// The content must match the definition in EventLogTags.logtags
+#define EVENTLOGTAG_DISKSTATS ( 2732 )
+#define EVENTLOGTAG_EMMCINFO ( 2733 )
+
+#endif /* _STORAGED_H_ */
diff --git a/storaged/include/storaged_service.h b/storaged/include/storaged_service.h
new file mode 100644
index 000000000..64a9c81c5
--- /dev/null
+++ b/storaged/include/storaged_service.h
@@ -0,0 +1,59 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef _STORAGED_SERVICE_H_
+#define _STORAGED_SERVICE_H_
+
+#include <vector>
+
+#include <binder/IInterface.h>
+#include <binder/IBinder.h>
+
+#include "storaged.h"
+
+using namespace android;
+
+// Interface
+class IStoraged : public IInterface {
+public:
+    enum {
+        DUMPTASKS = IBinder::FIRST_CALL_TRANSACTION,
+    };
+    // Request the service to run the test function
+    virtual std::vector<struct task_info> dump_tasks(const char* option) = 0;
+
+    DECLARE_META_INTERFACE(Storaged);
+};
+
+// Client
+class BpStoraged : public BpInterface<IStoraged> {
+public:
+    BpStoraged(const sp<IBinder>& impl) : BpInterface<IStoraged>(impl){};
+    virtual std::vector<struct task_info> dump_tasks(const char* option);
+};
+
+// Server
+class BnStoraged : public BnInterface<IStoraged> {
+    virtual status_t onTransact(uint32_t code, const Parcel& data, Parcel* reply, uint32_t flags = 0);
+};
+
+class Storaged : public BnStoraged {
+    virtual std::vector<struct task_info> dump_tasks(const char* option);
+};
+
+sp<IStoraged> get_storaged_service();
+
+#endif /* _STORAGED_SERVICE_H_ */
\ No newline at end of file
diff --git a/storaged/include/storaged_utils.h b/storaged/include/storaged_utils.h
new file mode 100644
index 000000000..b0e514605
--- /dev/null
+++ b/storaged/include/storaged_utils.h
@@ -0,0 +1,46 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef _STORAGED_UTILS_H_
+#define _STORAGED_UTILS_H_
+
+#include <stdint.h>
+#include <string>
+#include <unordered_map>
+#include <vector>
+
+#include "storaged.h"
+
+// Diskstats
+bool parse_disk_stats(const char* disk_stats_path, struct disk_stats* stats);
+struct disk_perf get_disk_perf(struct disk_stats* stats);
+struct disk_stats get_inc_disk_stats(struct disk_stats* prev, struct disk_stats* curr);
+void add_disk_stats(struct disk_stats* src, struct disk_stats* dst);
+bool parse_emmc_ecsd(int ext_csd_fd, struct emmc_info* info);
+
+// Task I/O
+bool parse_task_info(uint32_t pid, struct task_info* info);
+void sort_running_tasks_info(std::vector<struct task_info> &tasks);
+
+// Logging
+void log_console_running_tasks_info(std::vector<struct task_info> tasks);
+void log_kernel_disk_stats(struct disk_stats* stats, const char* type);
+void log_kernel_disk_perf(struct disk_perf* perf, const char* type);
+void log_kernel_emmc_info(struct emmc_info* info);
+
+void log_event_disk_stats(struct disk_stats* stats, const char* type);
+void log_event_emmc_info(struct emmc_info* info_);
+#endif /* _STORAGED_UTILS_H_ */
\ No newline at end of file