Skip to contentSkip to conversation

block: Fix hangs in synchronous APIs with iothreads

PATCH In review Kevin Wolf 2 participants 2 messages Jan 7, 2019 – Jan 8, 2019 Block layer

lore
Share & tools
lore

Kevin Wolf

author
#0 lore raw

PATCH 1/1 block: Fix hangs in synchronous APIs with iothreads

In the block layer, synchronous APIs are often implemented by creating a coroutine that calls the asynchronous coroutine-based implementation and then waiting for completion with BDRV_POLL_WHILE().

For this to work with iothreads (more specifically, when the synchronous API is called in a thread that is not the home thread of the block device, so that the coroutine will run in a different thread), we must make sure to call aio_wait_kick() at the end of the operation. Many places are missing this, so that BDRV_POLL_WHILE() keeps hanging even if the condition has long become false.

Note that bdrv_dec_in_flight() involves an aio_wait_kick() call. This corresponds to the BDRV_POLL_WHILE() in the drain functions, but it is generally not enough for most other operations because they haven't set the return value in the coroutine entry stub yet. To avoid race conditions there, we need to kick after setting the return value.

The race window is small enough that the problem doesn't usually surface in the common path. However, it does surface and causes easily reproducible hangs if the operation can return early before even calling bdrv_inc/dec_in_flight, which many of them do (trivial error or no-op success paths).

The bug in bdrv_truncate(), bdrv_check() and bdrv_invalidate_cache() is slightly different: These functions even neglected to schedule the coroutine in the home thread of the node. This avoids the hang, but is obviously wrong, too. Fix those to schedule the coroutine in the right AioContext in addition to adding aio_wait_kick() calls.

---
 block.c                     |   6 +-
 block/block-backend.c       |   5 +
 block/io.c                  |   8 +-
 block/nbd-client.c          |   1 +
 block/nvme.c                |   1 +
 block/qcow2.c               |   1 +
 block/qed.c                 |   1 +
 tests/test-block-iothread.c | 372 ++++++++++++++++++++++++++++++++++++
 tests/Makefile.include      |   2 +
 9 files changed, 394 insertions(+), 3 deletions(-)
 create mode 100644 tests/test-block-iothread.c
block.c +4 -2
index 4f5ff2cc12..92a3b45a78 100644
@@ -3725,6 +3725,7 @@ static void bdrv_check_co_entry(void *opaque)
 {
     CheckCo *cco = opaque;
     cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
+    aio_wait_kick();
 }
 
 int bdrv_check(BlockDriverState *bs,
@@ -3743,7 +3744,7 @@ int bdrv_check(BlockDriverState *bs,
         bdrv_check_co_entry(&cco);
     } else {
         co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
-        qemu_coroutine_enter(co);
+        bdrv_coroutine_enter(bs, co);
         BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
     }
 
@@ -4560,6 +4561,7 @@ static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
     InvalidateCacheCo *ico = opaque;
     bdrv_co_invalidate_cache(ico->bs, ico->errp);
     ico->done = true;
+    aio_wait_kick();
 }
 
 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
@@ -4576,7 +4578,7 @@ void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
         bdrv_invalidate_cache_co_entry(&ico);
     } else {
         co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
-        qemu_coroutine_enter(co);
+        bdrv_coroutine_enter(bs, co);
         BDRV_POLL_WHILE(bs, !ico.done);
     }
 }
block/block-backend.c +5 -0
index 60d37a0c3d..3a3d888633 100644
@@ -1220,6 +1220,7 @@ static void blk_read_entry(void *opaque)
 
     rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, qiov->size,
                               qiov, rwco->flags);
+    aio_wait_kick();
 }
 
 static void blk_write_entry(void *opaque)
@@ -1229,6 +1230,7 @@ static void blk_write_entry(void *opaque)
 
     rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, qiov->size,
                                qiov, rwco->flags);
+    aio_wait_kick();
 }
 
 static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
@@ -1540,6 +1542,7 @@ static void blk_ioctl_entry(void *opaque)
 
     rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset,
                              qiov->iov[0].iov_base);
+    aio_wait_kick();
 }
 
 int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf)
@@ -1586,6 +1589,7 @@ static void blk_flush_entry(void *opaque)
 {
     BlkRwCo *rwco = opaque;
     rwco->ret = blk_co_flush(rwco->blk);
+    aio_wait_kick();
 }
 
 int blk_flush(BlockBackend *blk)
@@ -2018,6 +2022,7 @@ static void blk_pdiscard_entry(void *opaque)
     QEMUIOVector *qiov = rwco->iobuf;
 
     rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, qiov->size);
+    aio_wait_kick();
 }
 
 int blk_pdiscard(BlockBackend *blk, int64_t offset, int bytes)
block/io.c +7 -1
index bd9d688f8b..213ca03d8d 100644
@@ -806,6 +806,7 @@ static void coroutine_fn bdrv_rw_co_entry(void *opaque)
                                     rwco->qiov->size, rwco->qiov,
                                     rwco->flags);
     }
+    aio_wait_kick();
 }
 
 /*
@@ -2279,6 +2280,7 @@ static void coroutine_fn bdrv_block_status_above_co_entry(void *opaque)
                                            data->offset, data->bytes,
                                            data->pnum, data->map, data->file);
     data->done = true;
+    aio_wait_kick();
 }
 
 /*
@@ -2438,6 +2440,7 @@ static void coroutine_fn bdrv_co_rw_vmstate_entry(void *opaque)
 {
     BdrvVmstateCo *co = opaque;
     co->ret = bdrv_co_rw_vmstate(co->bs, co->qiov, co->pos, co->is_read);
+    aio_wait_kick();
 }
 
 static inline int
@@ -2559,6 +2562,7 @@ static void coroutine_fn bdrv_flush_co_entry(void *opaque)
     FlushCo *rwco = opaque;
 
     rwco->ret = bdrv_co_flush(rwco->bs);
+    aio_wait_kick();
 }
 
 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
@@ -2704,6 +2708,7 @@ static void coroutine_fn bdrv_pdiscard_co_entry(void *opaque)
     DiscardCo *rwco = opaque;
 
     rwco->ret = bdrv_co_pdiscard(rwco->child, rwco->offset, rwco->bytes);
+    aio_wait_kick();
 }
 
 int coroutine_fn bdrv_co_pdiscard(BdrvChild *child, int64_t offset, int bytes)
@@ -3217,6 +3222,7 @@ static void coroutine_fn bdrv_truncate_co_entry(void *opaque)
     TruncateCo *tco = opaque;
     tco->ret = bdrv_co_truncate(tco->child, tco->offset, tco->prealloc,
                                 tco->errp);
+    aio_wait_kick();
 }
 
 int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc,
@@ -3236,7 +3242,7 @@ int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc,
         bdrv_truncate_co_entry(&tco);
     } else {
         co = qemu_coroutine_create(bdrv_truncate_co_entry, &tco);
-        qemu_coroutine_enter(co);
+        bdrv_coroutine_enter(child->bs, co);
         BDRV_POLL_WHILE(child->bs, tco.ret == NOT_DONE);
     }
 
block/nbd-client.c +1 -0
index fc5b7eda8e..a6b5605c0b 100644
@@ -116,6 +116,7 @@ static coroutine_fn void nbd_read_reply_entry(void *opaque)
     s->quit = true;
     nbd_recv_coroutines_wake_all(s);
     s->read_reply_co = NULL;
+    aio_wait_kick();
 }
 
 static int nbd_co_send_request(BlockDriverState *bs,
block/nvme.c +1 -0
index 29294038fc..36779fdd3d 100644
@@ -390,6 +390,7 @@ static void nvme_cmd_sync_cb(void *opaque, int ret)
 {
     int *pret = opaque;
     *pret = ret;
+    aio_wait_kick();
 }
 
 static int nvme_cmd_sync(BlockDriverState *bs, NVMeQueuePair *q,
block/qcow2.c +1 -0
index 4897abae5e..8c91b92865 100644
@@ -1671,6 +1671,7 @@ static int qcow2_open(BlockDriverState *bs, QDict *options, int flags,
         /* From bdrv_co_create.  */
         qcow2_open_entry(&qoc);
     } else {
+        assert(qemu_get_current_aio_context() == qemu_get_aio_context());
         qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry, &qoc));
         BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
     }
block/qed.c +1 -0
index 9377c0b7ad..1280870024 100644
@@ -559,6 +559,7 @@ static int bdrv_qed_open(BlockDriverState *bs, QDict *options, int flags,
     if (qemu_in_coroutine()) {
         bdrv_qed_open_entry(&qoc);
     } else {
+        assert(qemu_get_current_aio_context() == qemu_get_aio_context());
         qemu_coroutine_enter(qemu_coroutine_create(bdrv_qed_open_entry, &qoc));
         BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS);
     }
tests/test-block-iothread.c new file +372 -0

372 lines not on this page · raw message

tests/Makefile.include +2 -0
index fb0b449c02..7a8bcc2da7 100644
@@ -70,6 +70,7 @@ check-unit-y += tests/test-bdrv-drain$(EXESUF)
 check-unit-y += tests/test-blockjob$(EXESUF)
 check-unit-y += tests/test-blockjob-txn$(EXESUF)
 check-unit-y += tests/test-block-backend$(EXESUF)
+check-unit-y += tests/test-block-iothread$(EXESUF)
 check-unit-y += tests/test-image-locking$(EXESUF)
 check-unit-y += tests/test-x86-cpuid$(EXESUF)
 # all code tested by test-x86-cpuid is inside topology.h
@@ -538,6 +539,7 @@ tests/test-bdrv-drain$(EXESUF): tests/test-bdrv-drain.o $(test-block-obj-y) $(te
 tests/test-blockjob$(EXESUF): tests/test-blockjob.o $(test-block-obj-y) $(test-util-obj-y)
 tests/test-blockjob-txn$(EXESUF): tests/test-blockjob-txn.o $(test-block-obj-y) $(test-util-obj-y)
 tests/test-block-backend$(EXESUF): tests/test-block-backend.o $(test-block-obj-y) $(test-util-obj-y)
+tests/test-block-iothread$(EXESUF): tests/test-block-iothread.o $(test-block-obj-y) $(test-util-obj-y)
 tests/test-image-locking$(EXESUF): tests/test-image-locking.o $(test-block-obj-y) $(test-util-obj-y)
 tests/test-thread-pool$(EXESUF): tests/test-thread-pool.o $(test-block-obj-y)
 tests/test-iov$(EXESUF): tests/test-iov.o $(test-util-obj-y)
Signature · 1 line
2.20.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 31 lines · Kevin Wolf

In the block layer, synchronous APIs are often implemented by creating a coroutine that calls the asynchronous coroutine-based implementation and then waiting for completion with BDRV_POLL_WHILE().

For this to work with iothreads (more specifically, when the synchronous API is called in a thread that is not the home thread of the block device, so that the coroutine will run in a different thread), we must make sure to call aio_wait_kick() at the end of the operation. Many places are missing this, so that BDRV_POLL_WHILE() keeps hanging even if the condition has long become false.

Note that bdrv_dec_in_flight() involves an aio_wait_kick() call. This corresponds to the BDRV_POLL_WHILE() in the drain functions, but it is generally not enough for most other operations because they haven't set the return value in the coroutine entry stub yet. To avoid race conditions there, we need to kick after setting the return value.

The race window is small enough that the problem doesn't usually surface in the common path. However, it does surface and causes easily reproducible hangs if the operation can return early before even calling bdrv_inc/dec_in_flight, which many of them do (trivial error or no-op success paths).

The bug in bdrv_truncate(), bdrv_check() and bdrv_invalidate_cache() is slightly different: These functions even neglected to schedule the coroutine in the home thread of the node. This avoids the hang, but is obviously wrong, too. Fix those to schedule the coroutine in the right AioContext in addition to adding aio_wait_kick() calls.

Quoted 12 lines
---
 block.c                     |   6 +-
 block/block-backend.c       |   5 +
 block/io.c                  |   8 +-
 block/nbd-client.c          |   1 +
 block/nvme.c                |   1 +
 block/qcow2.c               |   1 +
 block/qed.c                 |   1 +
 tests/test-block-iothread.c | 372 ++++++++++++++++++++++++++++++++++++
 tests/Makefile.include      |   2 +
 9 files changed, 394 insertions(+), 3 deletions(-)
 create mode 100644 tests/test-block-iothread.c