[Qemu-devel] [PATCH v2 0/2] Block: don't do copy-on-read in before_write_notifier

[Qemu-devel] [PATCH v2 0/2] Block: don't do copy-on-read in before_write_notifier

[Wen Congyang]

If we do copy-on-read in before_write_notifier, qemu will be crashed. The
backtrace is:
Program terminated with signal 6, Aborted.
#0  0x00007f5d69207b55 in raise () from /lib64/libc.so.6
(gdb) bt
#0  0x00007f5d69207b55 in raise () from /lib64/libc.so.6
#1  0x00007f5d69209131 in abort () from /lib64/libc.so.6
#2  0x00007f5d69200a10 in __assert_fail () from /lib64/libc.so.6
#3  0x00007f5d6cb3a0d5 in wait_serialising_requests (self=0x7f5d36feeae0) at block/io.c:452
#4  0x00007f5d6cb3ae79 in bdrv_aligned_preadv (bs=0x7f5d6d570170, req=0x7f5d36feeae0, offset=93585408, bytes=65536, align=512, qiov=0x7f5d36feec90, flags=1)
    at block/io.c:847
#5  0x00007f5d6cb3b3cc in bdrv_co_do_preadv (bs=0x7f5d6d570170, offset=93585408, bytes=65536, qiov=0x7f5d36feec90, flags=BDRV_REQ_COPY_ON_READ)
    at block/io.c:971
#6  0x00007f5d6cb3b497 in bdrv_co_do_readv (bs=0x7f5d6d570170, sector_num=182784, nb_sectors=128, qiov=0x7f5d36feec90, flags=0) at block/io.c:993
#7  0x00007f5d6cb3b504 in bdrv_co_readv (bs=0x7f5d6d570170, sector_num=182784, nb_sectors=128, qiov=0x7f5d36feec90) at block/io.c:1002
#8  0x00007f5d6c8eaf42 in backup_do_cow (bs=0x7f5d6d570170, sector_num=182784, nb_sectors=8, error_is_read=0x0) at block/backup.c:132
#9  0x00007f5d6c8eb6cf in backup_before_write_notify (notifier=0x7f5d435fcf30, opaque=0x7f5d36feee40) at block/backup.c:193
#10 0x00007f5d6cb9ec33 in notifier_with_return_list_notify (list=0x7f5d6d5731f8, data=0x7f5d36feee40) at util/notify.c:65
#11 0x00007f5d6cb3bcc8 in bdrv_aligned_pwritev (bs=0x7f5d6d570170, req=0x7f5d36feee40, offset=93585408, bytes=4096, qiov=0x7f5d6eef4cf8, flags=0)
    at block/io.c:1126
#12 0x00007f5d6cb3c79e in bdrv_co_do_pwritev (bs=0x7f5d6d570170, offset=93585408, bytes=4096, qiov=0x7f5d6eef4cf8, flags=0) at block/io.c:1364
#13 0x00007f5d6cb3c867 in bdrv_co_do_writev (bs=0x7f5d6d570170, sector_num=182784, nb_sectors=8, qiov=0x7f5d6eef4cf8, flags=0) at block/io.c:1388
#14 0x00007f5d6cb3ea09 in bdrv_co_do_rw (opaque=0x7f5d38465ee0) at block/io.c:2123
#15 0x00007f5d6caeb03d in coroutine_trampoline (i0=944146112, i1=32605) at coroutine-ucontext.c:80
#16 0x00007f5d69218be0 in __correctly_grouped_prefixwc () from /lib64/libc.so.6
#17 0x0000000000000000 in ?? ()


Wen Congyang (2):
  block: Introduce a new API bdrv_co_no_copy_on_readv()
  Backup: don't do copy-on-read in before_write_notifier

 block/backup.c        | 20 ++++++++++++++------
 block/io.c            | 12 +++++++++++-
 include/block/block.h |  9 ++++++---
 trace-events          |  1 +
 4 files changed, 32 insertions(+), 10 deletions(-)

-- 
2.4.3

[Qemu-devel] [PATCH v2 1/2] block: Introduce a new API bdrv_co_no_copy_on_readv()

[Wen Congyang]

In some cases, we need to disable copy-on-read, and just
read the data.

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
---
 block/io.c            | 12 +++++++++++-
 include/block/block.h |  9 ++++++---
 trace-events          |  1 +
 3 files changed, 18 insertions(+), 4 deletions(-)

diff --git a/block/io.c b/block/io.c
index d4bc83b..94e18e6 100644
--- a/block/io.c
+++ b/block/io.c
@@ -932,7 +932,8 @@ static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
         return ret;
     }
 
-    if (bs->copy_on_read) {
+    /* Don't do copy-on-read if we read data before write operation */
+    if (bs->copy_on_read && !(flags & BDRV_REQ_NO_COPY_ON_READ)) {
         flags |= BDRV_REQ_COPY_ON_READ;
     }
 
@@ -1001,6 +1002,15 @@ int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
     return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
 }
 
+int coroutine_fn bdrv_co_no_copy_on_readv(BlockDriverState *bs,
+    int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
+{
+    trace_bdrv_co_no_copy_on_readv(bs, sector_num, nb_sectors);
+
+    return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
+                            BDRV_REQ_NO_COPY_ON_READ);
+}
+
 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 {
diff --git a/include/block/block.h b/include/block/block.h
index 608cd4e..c80d95d 100644
--- a/include/block/block.h
+++ b/include/block/block.h
@@ -51,15 +51,16 @@ typedef struct BlockFragInfo {
 } BlockFragInfo;
 
 typedef enum {
-    BDRV_REQ_COPY_ON_READ = 0x1,
-    BDRV_REQ_ZERO_WRITE   = 0x2,
+    BDRV_REQ_COPY_ON_READ       = 0x1,
+    BDRV_REQ_ZERO_WRITE         = 0x2,
     /* The BDRV_REQ_MAY_UNMAP flag is used to indicate that the block driver
      * is allowed to optimize a write zeroes request by unmapping (discarding)
      * blocks if it is guaranteed that the result will read back as
      * zeroes. The flag is only passed to the driver if the block device is
      * opened with BDRV_O_UNMAP.
      */
-    BDRV_REQ_MAY_UNMAP    = 0x4,
+    BDRV_REQ_MAY_UNMAP          = 0x4,
+    BDRV_REQ_NO_COPY_ON_READ    = 0x8,
 } BdrvRequestFlags;
 
 typedef struct BlockSizes {
@@ -253,6 +254,8 @@ int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
     int nb_sectors, QEMUIOVector *qiov);
 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
+int coroutine_fn bdrv_co_no_copy_on_readv(BlockDriverState *bs,
+    int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
     int nb_sectors, QEMUIOVector *qiov);
 /*
diff --git a/trace-events b/trace-events
index 0a82f0c..3a7d163 100644
--- a/trace-events
+++ b/trace-events
@@ -69,6 +69,7 @@ bdrv_aio_write_zeroes(void *bs, int64_t sector_num, int nb_sectors, int flags, v
 bdrv_lock_medium(void *bs, bool locked) "bs %p locked %d"
 bdrv_co_readv(void *bs, int64_t sector_num, int nb_sector) "bs %p sector_num %"PRId64" nb_sectors %d"
 bdrv_co_copy_on_readv(void *bs, int64_t sector_num, int nb_sector) "bs %p sector_num %"PRId64" nb_sectors %d"
+bdrv_co_no_copy_on_readv(void *bs, int64_t sector_num, int nb_sector) "bs %p sector_num %"PRId64" nb_sectors %d"
 bdrv_co_writev(void *bs, int64_t sector_num, int nb_sector) "bs %p sector_num %"PRId64" nb_sectors %d"
 bdrv_co_write_zeroes(void *bs, int64_t sector_num, int nb_sector, int flags) "bs %p sector_num %"PRId64" nb_sectors %d flags %#x"
 bdrv_co_io_em(void *bs, int64_t sector_num, int nb_sectors, int is_write, void *acb) "bs %p sector_num %"PRId64" nb_sectors %d is_write %d acb %p"
-- 
2.4.3

[Qemu-devel] [PATCH v2 2/2] Backup: don't do copy-on-read in before_write_notifier

[Wen Congyang]

We will copy data in before_write_notifier to do backup.
It is a nested I/O request, so we cannot do copy-on-read.

The steps to reproduce it:
1. -drive copy-on-read=on,...  // qemu option
2. drive_backup -f disk0 /path_to_backup.img // monitor command

Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Tested-by: Jeff Cody <jcody@redhat.com>
---
 block/backup.c | 20 ++++++++++++++------
 1 file changed, 14 insertions(+), 6 deletions(-)

diff --git a/block/backup.c b/block/backup.c
index 965654d..5696431 100644
--- a/block/backup.c
+++ b/block/backup.c
@@ -89,7 +89,8 @@ static void cow_request_end(CowRequest *req)
 
 static int coroutine_fn backup_do_cow(BlockDriverState *bs,
                                       int64_t sector_num, int nb_sectors,
-                                      bool *error_is_read)
+                                      bool *error_is_read,
+                                      bool is_write_notifier)
 {
     BackupBlockJob *job = (BackupBlockJob *)bs->job;
     CowRequest cow_request;
@@ -129,8 +130,14 @@ static int coroutine_fn backup_do_cow(BlockDriverState *bs,
         iov.iov_len = n * BDRV_SECTOR_SIZE;
         qemu_iovec_init_external(&bounce_qiov, &iov, 1);
 
-        ret = bdrv_co_readv(bs, start * BACKUP_SECTORS_PER_CLUSTER, n,
-                            &bounce_qiov);
+        if (is_write_notifier) {
+            ret = bdrv_co_no_copy_on_readv(bs,
+                                           start * BACKUP_SECTORS_PER_CLUSTER,
+                                           n, &bounce_qiov);
+        } else {
+            ret = bdrv_co_readv(bs, start * BACKUP_SECTORS_PER_CLUSTER, n,
+                                &bounce_qiov);
+        }
         if (ret < 0) {
             trace_backup_do_cow_read_fail(job, start, ret);
             if (error_is_read) {
@@ -190,7 +197,7 @@ static int coroutine_fn backup_before_write_notify(
     assert((req->offset & (BDRV_SECTOR_SIZE - 1)) == 0);
     assert((req->bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
 
-    return backup_do_cow(req->bs, sector_num, nb_sectors, NULL);
+    return backup_do_cow(req->bs, sector_num, nb_sectors, NULL, true);
 }
 
 static void backup_set_speed(BlockJob *job, int64_t speed, Error **errp)
@@ -303,7 +310,8 @@ static int coroutine_fn backup_run_incremental(BackupBlockJob *job)
                     return ret;
                 }
                 ret = backup_do_cow(bs, cluster * BACKUP_SECTORS_PER_CLUSTER,
-                                    BACKUP_SECTORS_PER_CLUSTER, &error_is_read);
+                                    BACKUP_SECTORS_PER_CLUSTER, &error_is_read,
+                                    false);
                 if ((ret < 0) &&
                     backup_error_action(job, error_is_read, -ret) ==
                     BLOCK_ERROR_ACTION_REPORT) {
@@ -408,7 +416,7 @@ static void coroutine_fn backup_run(void *opaque)
             }
             /* FULL sync mode we copy the whole drive. */
             ret = backup_do_cow(bs, start * BACKUP_SECTORS_PER_CLUSTER,
-                    BACKUP_SECTORS_PER_CLUSTER, &error_is_read);
+                    BACKUP_SECTORS_PER_CLUSTER, &error_is_read, false);
             if (ret < 0) {
                 /* Depending on error action, fail now or retry cluster */
                 BlockErrorAction action =
-- 
2.4.3

Re: [Qemu-devel] [PATCH v2 0/2] Block: don't do copy-on-read in before_write_notifier

[Jeff Cody]

On Tue, Sep 08, 2015 at 11:28:31AM +0800, Wen Congyang wrote:
> If we do copy-on-read in before_write_notifier, qemu will be crashed. The
> backtrace is:
> Program terminated with signal 6, Aborted.
> #0  0x00007f5d69207b55 in raise () from /lib64/libc.so.6
> (gdb) bt
> #0  0x00007f5d69207b55 in raise () from /lib64/libc.so.6
> #1  0x00007f5d69209131 in abort () from /lib64/libc.so.6
> #2  0x00007f5d69200a10 in __assert_fail () from /lib64/libc.so.6
> #3  0x00007f5d6cb3a0d5 in wait_serialising_requests (self=0x7f5d36feeae0) at block/io.c:452
> #4  0x00007f5d6cb3ae79 in bdrv_aligned_preadv (bs=0x7f5d6d570170, req=0x7f5d36feeae0, offset=93585408, bytes=65536, align=512, qiov=0x7f5d36feec90, flags=1)
>     at block/io.c:847
> #5  0x00007f5d6cb3b3cc in bdrv_co_do_preadv (bs=0x7f5d6d570170, offset=93585408, bytes=65536, qiov=0x7f5d36feec90, flags=BDRV_REQ_COPY_ON_READ)
>     at block/io.c:971
> #6  0x00007f5d6cb3b497 in bdrv_co_do_readv (bs=0x7f5d6d570170, sector_num=182784, nb_sectors=128, qiov=0x7f5d36feec90, flags=0) at block/io.c:993
> #7  0x00007f5d6cb3b504 in bdrv_co_readv (bs=0x7f5d6d570170, sector_num=182784, nb_sectors=128, qiov=0x7f5d36feec90) at block/io.c:1002
> #8  0x00007f5d6c8eaf42 in backup_do_cow (bs=0x7f5d6d570170, sector_num=182784, nb_sectors=8, error_is_read=0x0) at block/backup.c:132
> #9  0x00007f5d6c8eb6cf in backup_before_write_notify (notifier=0x7f5d435fcf30, opaque=0x7f5d36feee40) at block/backup.c:193
> #10 0x00007f5d6cb9ec33 in notifier_with_return_list_notify (list=0x7f5d6d5731f8, data=0x7f5d36feee40) at util/notify.c:65
> #11 0x00007f5d6cb3bcc8 in bdrv_aligned_pwritev (bs=0x7f5d6d570170, req=0x7f5d36feee40, offset=93585408, bytes=4096, qiov=0x7f5d6eef4cf8, flags=0)
>     at block/io.c:1126
> #12 0x00007f5d6cb3c79e in bdrv_co_do_pwritev (bs=0x7f5d6d570170, offset=93585408, bytes=4096, qiov=0x7f5d6eef4cf8, flags=0) at block/io.c:1364
> #13 0x00007f5d6cb3c867 in bdrv_co_do_writev (bs=0x7f5d6d570170, sector_num=182784, nb_sectors=8, qiov=0x7f5d6eef4cf8, flags=0) at block/io.c:1388
> #14 0x00007f5d6cb3ea09 in bdrv_co_do_rw (opaque=0x7f5d38465ee0) at block/io.c:2123
> #15 0x00007f5d6caeb03d in coroutine_trampoline (i0=944146112, i1=32605) at coroutine-ucontext.c:80
> #16 0x00007f5d69218be0 in __correctly_grouped_prefixwc () from /lib64/libc.so.6
> #17 0x0000000000000000 in ?? ()
> 
> 
> Wen Congyang (2):
>   block: Introduce a new API bdrv_co_no_copy_on_readv()
>   Backup: don't do copy-on-read in before_write_notifier
> 
>  block/backup.c        | 20 ++++++++++++++------
>  block/io.c            | 12 +++++++++++-
>  include/block/block.h |  9 ++++++---
>  trace-events          |  1 +
>  4 files changed, 32 insertions(+), 10 deletions(-)
> 
> -- 
> 2.4.3
>

Thanks, applied to my block branch:

git://github.com/codyprime/qemu-kvm-jtc.git block

-Jeff