Merge remote-tracking branch 'remotes/famz/tags/docker-and-block-pull-request' into staging

# gpg: Signature made Fri 16 Jun 2017 01:18:46 BST
# gpg:                using RSA key 0xCA35624C6A9171C6
# gpg: Good signature from "Fam Zheng <famz@redhat.com>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg:          It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 5003 7CB7 9706 0F76 F021  AD56 CA35 624C 6A91 71C6

* remotes/famz/tags/docker-and-block-pull-request: (23 commits)
  block: make accounting thread-safe
  block: split BlockAcctStats creation and setup
  block: introduce block_account_one_io
  block: protect modification of dirty bitmaps with a mutex
  migration/block: reset dirty bitmap before reading
  block: introduce dirty_bitmap_mutex
  block: protect tracked_requests and flush_queue with reqs_lock
  block: access write_gen with atomics
  block: use Stat64 for wr_highest_offset
  util: add stats64 module
  throttle-groups: protect throttled requests with a CoMutex
  throttle-groups: do not use qemu_co_enter_next
  throttle-groups: only start one coroutine from drained_begin
  block: access io_plugged with atomic ops
  block: access wakeup with atomic ops
  block: access serialising_in_flight with atomic ops
  block: access io_limits_disabled with atomic ops
  block: access quiesce_counter with atomic ops
  block: access copy_on_read with atomic ops
  docker: Add flex and bison to centos6 image
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

6 files changed, 266 insertions, 39 deletions

diff --git a/include/block/accounting.h b/include/block/accounting.h
index 20891639d5..b833d26d6c 100644
--- a/include/block/accounting.h
+++ b/include/block/accounting.h
@@ -26,8 +26,10 @@
 #define BLOCK_ACCOUNTING_H
 
 #include "qemu/timed-average.h"
+#include "qemu/thread.h"
 
 typedef struct BlockAcctTimedStats BlockAcctTimedStats;
+typedef struct BlockAcctStats BlockAcctStats;
 
 enum BlockAcctType {
     BLOCK_ACCT_READ,
@@ -37,12 +39,14 @@ enum BlockAcctType {
 };
 
 struct BlockAcctTimedStats {
+    BlockAcctStats *stats;
     TimedAverage latency[BLOCK_MAX_IOTYPE];
     unsigned interval_length; /* in seconds */
     QSLIST_ENTRY(BlockAcctTimedStats) entries;
 };
 
-typedef struct BlockAcctStats {
+struct BlockAcctStats {
+    QemuMutex lock;
     uint64_t nr_bytes[BLOCK_MAX_IOTYPE];
     uint64_t nr_ops[BLOCK_MAX_IOTYPE];
     uint64_t invalid_ops[BLOCK_MAX_IOTYPE];
@@ -53,7 +57,7 @@ typedef struct BlockAcctStats {
     QSLIST_HEAD(, BlockAcctTimedStats) intervals;
     bool account_invalid;
     bool account_failed;
-} BlockAcctStats;
+};
 
 typedef struct BlockAcctCookie {
     int64_t bytes;
@@ -61,7 +65,8 @@ typedef struct BlockAcctCookie {
     enum BlockAcctType type;
 } BlockAcctCookie;
 
-void block_acct_init(BlockAcctStats *stats, bool account_invalid,
+void block_acct_init(BlockAcctStats *stats);
+void block_acct_setup(BlockAcctStats *stats, bool account_invalid,
                      bool account_failed);
 void block_acct_cleanup(BlockAcctStats *stats);
 void block_acct_add_interval(BlockAcctStats *stats, unsigned interval_length);
diff --git a/include/block/block.h b/include/block/block.h
index 9b355e92d8..a4f09df95a 100644
--- a/include/block/block.h
+++ b/include/block/block.h
@@ -402,7 +402,8 @@ void bdrv_drain_all(void);
          * block_job_defer_to_main_loop for how to do it). \
          */                                                \
         assert(!bs_->wakeup);                              \
-        bs_->wakeup = true;                                \
+        /* Set bs->wakeup before evaluating cond.  */      \
+        atomic_mb_set(&bs_->wakeup, true);                 \
         while (busy_) {                                    \
             if ((cond)) {                                  \
                 waited_ = busy_ = true;                    \
@@ -414,7 +415,7 @@ void bdrv_drain_all(void);
                 waited_ |= busy_;                          \
             }                                              \
         }                                                  \
-        bs_->wakeup = false;                               \
+        atomic_set(&bs_->wakeup, false);                   \
     }                                                      \
     waited_; })
 
diff --git a/include/block/block_int.h b/include/block/block_int.h
index cb78c4fa82..748970055e 100644
--- a/include/block/block_int.h
+++ b/include/block/block_int.h
@@ -29,6 +29,7 @@
 #include "qemu/option.h"
 #include "qemu/queue.h"
 #include "qemu/coroutine.h"
+#include "qemu/stats64.h"
 #include "qemu/timer.h"
 #include "qapi-types.h"
 #include "qemu/hbitmap.h"
@@ -595,11 +596,6 @@ struct BlockDriverState {
 
     /* Protected by AioContext lock */
 
-    /* If true, copy read backing sectors into image.  Can be >1 if more
-     * than one client has requested copy-on-read.
-     */
-    int copy_on_read;
-
     /* If we are reading a disk image, give its size in sectors.
      * Generally read-only; it is written to by load_snapshot and
      * save_snaphost, but the block layer is quiescent during those.
@@ -609,34 +605,57 @@ struct BlockDriverState {
     /* Callback before write request is processed */
     NotifierWithReturnList before_write_notifiers;
 
-    /* number of in-flight requests; overall and serialising */
-    unsigned int in_flight;
-    unsigned int serialising_in_flight;
+    /* threshold limit for writes, in bytes. "High water mark". */
+    uint64_t write_threshold_offset;
+    NotifierWithReturn write_threshold_notifier;
 
-    bool wakeup;
+    /* Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
+     * Reading from the list can be done with either the BQL or the
+     * dirty_bitmap_mutex.  Modifying a bitmap only requires
+     * dirty_bitmap_mutex.  */
+    QemuMutex dirty_bitmap_mutex;
+    QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
 
     /* Offset after the highest byte written to */
-    uint64_t wr_highest_offset;
+    Stat64 wr_highest_offset;
 
-    /* threshold limit for writes, in bytes. "High water mark". */
-    uint64_t write_threshold_offset;
-    NotifierWithReturn write_threshold_notifier;
+    /* If true, copy read backing sectors into image.  Can be >1 if more
+     * than one client has requested copy-on-read.  Accessed with atomic
+     * ops.
+     */
+    int copy_on_read;
 
-    /* counter for nested bdrv_io_plug */
-    unsigned io_plugged;
+    /* number of in-flight requests; overall and serialising.
+     * Accessed with atomic ops.
+     */
+    unsigned int in_flight;
+    unsigned int serialising_in_flight;
 
-    QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
-    CoQueue flush_queue;                  /* Serializing flush queue */
-    bool active_flush_req;                /* Flush request in flight? */
-    unsigned int write_gen;               /* Current data generation */
-    unsigned int flushed_gen;             /* Flushed write generation */
+    /* Internal to BDRV_POLL_WHILE and bdrv_wakeup.  Accessed with atomic
+     * ops.
+     */
+    bool wakeup;
 
-    QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
+    /* counter for nested bdrv_io_plug.
+     * Accessed with atomic ops.
+    */
+    unsigned io_plugged;
 
     /* do we need to tell the quest if we have a volatile write cache? */
     int enable_write_cache;
 
+    /* Accessed with atomic ops.  */
     int quiesce_counter;
+    unsigned int write_gen;               /* Current data generation */
+
+    /* Protected by reqs_lock.  */
+    CoMutex reqs_lock;
+    QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
+    CoQueue flush_queue;                  /* Serializing flush queue */
+    bool active_flush_req;                /* Flush request in flight? */
+
+    /* Only read/written by whoever has set active_flush_req to true.  */
+    unsigned int flushed_gen;             /* Flushed write generation */
 };
 
 struct BlockBackendRootState {
diff --git a/include/block/dirty-bitmap.h b/include/block/dirty-bitmap.h
index 9dea14ba03..ad6558af56 100644
--- a/include/block/dirty-bitmap.h
+++ b/include/block/dirty-bitmap.h
@@ -36,8 +36,6 @@ bool bdrv_dirty_bitmap_frozen(BdrvDirtyBitmap *bitmap);
 const char *bdrv_dirty_bitmap_name(const BdrvDirtyBitmap *bitmap);
 int64_t bdrv_dirty_bitmap_size(const BdrvDirtyBitmap *bitmap);
 DirtyBitmapStatus bdrv_dirty_bitmap_status(BdrvDirtyBitmap *bitmap);
-int bdrv_get_dirty(BlockDriverState *bs, BdrvDirtyBitmap *bitmap,
-                   int64_t sector);
 void bdrv_set_dirty_bitmap(BdrvDirtyBitmap *bitmap,
                            int64_t cur_sector, int64_t nr_sectors);
 void bdrv_reset_dirty_bitmap(BdrvDirtyBitmap *bitmap,
@@ -45,6 +43,9 @@ void bdrv_reset_dirty_bitmap(BdrvDirtyBitmap *bitmap,
 int bdrv_dirty_bitmap_get_meta(BlockDriverState *bs,
                                BdrvDirtyBitmap *bitmap, int64_t sector,
                                int nb_sectors);
+int bdrv_dirty_bitmap_get_meta_locked(BlockDriverState *bs,
+                                      BdrvDirtyBitmap *bitmap, int64_t sector,
+                                      int nb_sectors);
 void bdrv_dirty_bitmap_reset_meta(BlockDriverState *bs,
                                   BdrvDirtyBitmap *bitmap, int64_t sector,
                                   int nb_sectors);
@@ -52,11 +53,6 @@ BdrvDirtyBitmapIter *bdrv_dirty_meta_iter_new(BdrvDirtyBitmap *bitmap);
 BdrvDirtyBitmapIter *bdrv_dirty_iter_new(BdrvDirtyBitmap *bitmap,
                                          uint64_t first_sector);
 void bdrv_dirty_iter_free(BdrvDirtyBitmapIter *iter);
-int64_t bdrv_dirty_iter_next(BdrvDirtyBitmapIter *iter);
-void bdrv_set_dirty_iter(BdrvDirtyBitmapIter *hbi, int64_t sector_num);
-int64_t bdrv_get_dirty_count(BdrvDirtyBitmap *bitmap);
-int64_t bdrv_get_meta_dirty_count(BdrvDirtyBitmap *bitmap);
-void bdrv_dirty_bitmap_truncate(BlockDriverState *bs);
 
 uint64_t bdrv_dirty_bitmap_serialization_size(const BdrvDirtyBitmap *bitmap,
                                               uint64_t start, uint64_t count);
@@ -72,4 +68,19 @@ void bdrv_dirty_bitmap_deserialize_zeroes(BdrvDirtyBitmap *bitmap,
                                           bool finish);
 void bdrv_dirty_bitmap_deserialize_finish(BdrvDirtyBitmap *bitmap);
 
+/* Functions that require manual locking.  */
+void bdrv_dirty_bitmap_lock(BdrvDirtyBitmap *bitmap);
+void bdrv_dirty_bitmap_unlock(BdrvDirtyBitmap *bitmap);
+int bdrv_get_dirty_locked(BlockDriverState *bs, BdrvDirtyBitmap *bitmap,
+                          int64_t sector);
+void bdrv_set_dirty_bitmap_locked(BdrvDirtyBitmap *bitmap,
+                                  int64_t cur_sector, int64_t nr_sectors);
+void bdrv_reset_dirty_bitmap_locked(BdrvDirtyBitmap *bitmap,
+                                    int64_t cur_sector, int64_t nr_sectors);
+int64_t bdrv_dirty_iter_next(BdrvDirtyBitmapIter *iter);
+void bdrv_set_dirty_iter(BdrvDirtyBitmapIter *hbi, int64_t sector_num);
+int64_t bdrv_get_dirty_count(BdrvDirtyBitmap *bitmap);
+int64_t bdrv_get_meta_dirty_count(BdrvDirtyBitmap *bitmap);
+void bdrv_dirty_bitmap_truncate(BlockDriverState *bs);
+
 #endif
diff --git a/include/qemu/stats64.h b/include/qemu/stats64.h
new file mode 100644
index 0000000000..4a357b3e9d
--- /dev/null
+++ b/include/qemu/stats64.h
@@ -0,0 +1,193 @@
+/*
+ * Atomic operations on 64-bit quantities.
+ *
+ * Copyright (C) 2017 Red Hat, Inc.
+ *
+ * Author: Paolo Bonzini <pbonzini@redhat.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#ifndef QEMU_STATS64_H
+#define QEMU_STATS64_H 1
+
+#include "qemu/atomic.h"
+
+/* This provides atomic operations on 64-bit type, using a reader-writer
+ * spinlock on architectures that do not have 64-bit accesses.  Even on
+ * those architectures, it tries hard not to take the lock.
+ */
+
+typedef struct Stat64 {
+#ifdef CONFIG_ATOMIC64
+    uint64_t value;
+#else
+    uint32_t low, high;
+    uint32_t lock;
+#endif
+} Stat64;
+
+#ifdef CONFIG_ATOMIC64
+static inline void stat64_init(Stat64 *s, uint64_t value)
+{
+    /* This is not guaranteed to be atomic! */
+    *s = (Stat64) { value };
+}
+
+static inline uint64_t stat64_get(const Stat64 *s)
+{
+    return atomic_read__nocheck(&s->value);
+}
+
+static inline void stat64_add(Stat64 *s, uint64_t value)
+{
+    atomic_add(&s->value, value);
+}
+
+static inline void stat64_min(Stat64 *s, uint64_t value)
+{
+    uint64_t orig = atomic_read__nocheck(&s->value);
+    while (orig > value) {
+        orig = atomic_cmpxchg__nocheck(&s->value, orig, value);
+    }
+}
+
+static inline void stat64_max(Stat64 *s, uint64_t value)
+{
+    uint64_t orig = atomic_read__nocheck(&s->value);
+    while (orig < value) {
+        orig = atomic_cmpxchg__nocheck(&s->value, orig, value);
+    }
+}
+#else
+uint64_t stat64_get(const Stat64 *s);
+bool stat64_min_slow(Stat64 *s, uint64_t value);
+bool stat64_max_slow(Stat64 *s, uint64_t value);
+bool stat64_add32_carry(Stat64 *s, uint32_t low, uint32_t high);
+
+static inline void stat64_init(Stat64 *s, uint64_t value)
+{
+    /* This is not guaranteed to be atomic! */
+    *s = (Stat64) { .low = value, .high = value >> 32, .lock = 0 };
+}
+
+static inline void stat64_add(Stat64 *s, uint64_t value)
+{
+    uint32_t low, high;
+    high = value >> 32;
+    low = (uint32_t) value;
+    if (!low) {
+        if (high) {
+            atomic_add(&s->high, high);
+        }
+        return;
+    }
+
+    for (;;) {
+        uint32_t orig = s->low;
+        uint32_t result = orig + low;
+        uint32_t old;
+
+        if (result < low || high) {
+            /* If the high part is affected, take the lock.  */
+            if (stat64_add32_carry(s, low, high)) {
+                return;
+            }
+            continue;
+        }
+
+        /* No carry, try with a 32-bit cmpxchg.  The result is independent of
+         * the high 32 bits, so it can race just fine with stat64_add32_carry
+         * and even stat64_get!
+         */
+        old = atomic_cmpxchg(&s->low, orig, result);
+        if (orig == old) {
+            return;
+        }
+    }
+}
+
+static inline void stat64_min(Stat64 *s, uint64_t value)
+{
+    uint32_t low, high;
+    uint32_t orig_low, orig_high;
+
+    high = value >> 32;
+    low = (uint32_t) value;
+    do {
+        orig_high = atomic_read(&s->high);
+        if (orig_high < high) {
+            return;
+        }
+
+        if (orig_high == high) {
+            /* High 32 bits are equal.  Read low after high, otherwise we
+             * can get a false positive (e.g. 0x1235,0x0000 changes to
+             * 0x1234,0x8000 and we read it as 0x1234,0x0000). Pairs with
+             * the write barrier in stat64_min_slow.
+             */
+            smp_rmb();
+            orig_low = atomic_read(&s->low);
+            if (orig_low <= low) {
+                return;
+            }
+
+            /* See if we were lucky and a writer raced against us.  The
+             * barrier is theoretically unnecessary, but if we remove it
+             * we may miss being lucky.
+             */
+            smp_rmb();
+            orig_high = atomic_read(&s->high);
+            if (orig_high < high) {
+                return;
+            }
+        }
+
+        /* If the value changes in any way, we have to take the lock.  */
+    } while (!stat64_min_slow(s, value));
+}
+
+static inline void stat64_max(Stat64 *s, uint64_t value)
+{
+    uint32_t low, high;
+    uint32_t orig_low, orig_high;
+
+    high = value >> 32;
+    low = (uint32_t) value;
+    do {
+        orig_high = atomic_read(&s->high);
+        if (orig_high > high) {
+            return;
+        }
+
+        if (orig_high == high) {
+            /* High 32 bits are equal.  Read low after high, otherwise we
+             * can get a false positive (e.g. 0x1234,0x8000 changes to
+             * 0x1235,0x0000 and we read it as 0x1235,0x8000). Pairs with
+             * the write barrier in stat64_max_slow.
+             */
+            smp_rmb();
+            orig_low = atomic_read(&s->low);
+            if (orig_low >= low) {
+                return;
+            }
+
+            /* See if we were lucky and a writer raced against us.  The
+             * barrier is theoretically unnecessary, but if we remove it
+             * we may miss being lucky.
+             */
+            smp_rmb();
+            orig_high = atomic_read(&s->high);
+            if (orig_high > high) {
+                return;
+            }
+        }
+
+        /* If the value changes in any way, we have to take the lock.  */
+    } while (!stat64_max_slow(s, value));
+}
+
+#endif
+
+#endif
diff --git a/include/sysemu/block-backend.h b/include/sysemu/block-backend.h
index 840ad6134c..999eb2333a 100644
--- a/include/sysemu/block-backend.h
+++ b/include/sysemu/block-backend.h
@@ -72,15 +72,13 @@ typedef struct BlockDevOps {
  * fields that must be public. This is in particular for QLIST_ENTRY() and
  * friends so that BlockBackends can be kept in lists outside block-backend.c */
 typedef struct BlockBackendPublic {
-    /* I/O throttling has its own locking, but also some fields are
-     * protected by the AioContext lock.
-     */
-
-    /* Protected by AioContext lock.  */
+    /* throttled_reqs_lock protects the CoQueues for throttled requests.  */
+    CoMutex      throttled_reqs_lock;
     CoQueue      throttled_reqs[2];
 
     /* Nonzero if the I/O limits are currently being ignored; generally
-     * it is zero.  */
+     * it is zero.  Accessed with atomic operations.
+     */
     unsigned int io_limits_disabled;
 
     /* The following fields are protected by the ThrottleGroup lock.