[Qemu-devel] [PATCH 4/8] block: add a Rwlock to synchronous read/write drivers

[Paolo Bonzini <pbonzini@redhat.com>]

The big conversion of bdrv_read/write to coroutines caused the two
homonymous callbacks in BlockDriver to become reentrant.  It goes
like this:

1) bdrv_read is now called in a coroutine, and calls bdrv_read or
bdrv_pread.

2) the nested bdrv_read goes through the fast path in bdrv_rw_co_entry;

3) in the common case when the protocol is file, bdrv_co_do_readv calls
bdrv_co_readv_em (and from here goes to bdrv_co_io_em), which yields
until the AIO operation is complete;

4) if bdrv_read had been called from a bottom half, the main loop
is free to iterate again: a device model or another bottom half
can then come and call bdrv_read again.

This applies to all four of read/write/flush/discard.  It would also
apply to is_allocated, but it is not used from within coroutines:
besides qemu-img.c and qemu-io.c, which operate synchronously, the
only user is the monitor.  Copy-on-read will introduce a use in the
block layer, and will require converting it.

The solution is "simply" to convert all drivers to coroutines!  We
have nothing to do for the read-only drivers.  For the others, we
add a Rwlock that is taken around affected operations.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 block/cow.c   |    2 ++
 block/nbd.c   |    2 ++
 block/vmdk.c  |    2 ++
 block/vpc.c   |    2 ++
 block/vvfat.c |    2 ++
 5 files changed, 10 insertions(+), 0 deletions(-)

diff --git a/block/cow.c b/block/cow.c
index 4cf543c..d27e0aa 100644
--- a/block/cow.c
+++ b/block/cow.c
@@ -42,6 +42,7 @@ struct cow_header_v2 {
 };
 
 typedef struct BDRVCowState {
+    CoRwlock lock;
     int64_t cow_sectors_offset;
 } BDRVCowState;
 
@@ -84,6 +85,7 @@ static int cow_open(BlockDriverState *bs, int flags)
 
     bitmap_size = ((bs->total_sectors + 7) >> 3) + sizeof(cow_header);
     s->cow_sectors_offset = (bitmap_size + 511) & ~511;
+    qemu_co_rwlock_init(&s->lock);
     return 0;
  fail:
     return -1;
diff --git a/block/nbd.c b/block/nbd.c
index 76f04d8..ec8f086 100644
--- a/block/nbd.c
+++ b/block/nbd.c
@@ -47,6 +47,7 @@
 #endif
 
 typedef struct BDRVNBDState {
+    CoRwlock lock;
     int sock;
     uint32_t nbdflags;
     off_t size;
@@ -175,6 +176,7 @@ static int nbd_open(BlockDriverState *bs, const char* filename, int flags)
      */
     result = nbd_establish_connection(bs);
 
+    qemu_co_rwlock_init(&s->lock);
     return result;
 }
 
diff --git a/block/vmdk.c b/block/vmdk.c
index 12b38d2..6afd53e 100644
--- a/block/vmdk.c
+++ b/block/vmdk.c
@@ -90,6 +90,7 @@ typedef struct VmdkExtent {
 } VmdkExtent;
 
 typedef struct BDRVVmdkState {
+    CoRwlock lock;
     int desc_offset;
     bool cid_updated;
     uint32_t parent_cid;
@@ -644,6 +645,7 @@ static int vmdk_open(BlockDriverState *bs, int flags)
         goto fail;
     }
     s->parent_cid = vmdk_read_cid(bs, 1);
+    qemu_co_rwlock_init(&s->lock);
     return ret;
 
 fail:
diff --git a/block/vpc.c b/block/vpc.c
index 549a632..7220488 100644
--- a/block/vpc.c
+++ b/block/vpc.c
@@ -110,6 +110,7 @@ struct vhd_dyndisk_header {
 };
 
 typedef struct BDRVVPCState {
+    CoRwlock lock;
     uint8_t footer_buf[HEADER_SIZE];
     uint64_t free_data_block_offset;
     int max_table_entries;
@@ -226,6 +227,7 @@ static int vpc_open(BlockDriverState *bs, int flags)
     s->last_pagetable = -1;
 #endif
 
+    qemu_co_rwlock_init(&s->lock);
     return 0;
  fail:
     return err;
diff --git a/block/vvfat.c b/block/vvfat.c
index c567697..08a72ee 100644
--- a/block/vvfat.c
+++ b/block/vvfat.c
@@ -317,6 +317,7 @@ static void print_mapping(const struct mapping_t* mapping);
 /* here begins the real VVFAT driver */
 
 typedef struct BDRVVVFATState {
+    CoRwlock lock;
     BlockDriverState* bs; /* pointer to parent */
     unsigned int first_sectors_number; /* 1 for a single partition, 0x40 for a disk with partition table */
     unsigned char first_sectors[0x40*0x200];
@@ -1063,6 +1064,7 @@ DLOG(if (stderr == NULL) {
     }
 
     //    assert(is_consistent(s));
+    qemu_co_rwlock_init(&s->lock);
     return 0;
 }
 
-- 
1.7.6