Skip to contentSkip to conversation

backup-top filter driver for backup

PATCH Superseded Vladimir Sementsov-Ogievskiy 2 participants 39 messages Aug 26, 2019 – Sep 9, 2019 Block layer

v9 of 11 Superseded Latest is v11 →

lore
Share & tools
lore
#0 lore raw

v9 cover backup-top filter driver for backup

Hi all!

These series introduce backup-top driver. It's a filter-node, which do copy-before-write operation. Mirror uses filter-node for handling guest writes, let's move to filter-node (from write-notifiers) for backup too.

I decided to postpone all these "preparation" and "improvement" series, as it seems we will never do the core thing in such way.

v9:

01: rebased, it's from "backup improvements" series

02-05: new patches, introduction of block-copy

06: unchanged

07-08: new iotests preparation

10-11: unchanged

12:
backup-top filter is simplified: it don't do copy operation by hands but through block-copy, and it don't have second "target" child anymore. So the state is reduced to only two fields:
block-copy state and "active".

What else?
1. filter-node-name parameter added
2. "active" parameter. I failed on my way to make bdrv_append firstly do bdrv_replace_node and then bdrv_set_backing_hd, as it leads to much problems, block-layer is unprepared for child-less filters. So, it's a lot simpler to cheat a bit with permissions, and don't require unshared WRITE during appending.
3. filter is not finished in this patch, as block_copy is not finished too, you can see a TODO and abort() in backup_top_cbw.

13:
- filter-node-name added
- backup_clean simplified: don't bother with keeping block job blk on backup-top filter, it's not necessary.
- locking is all done in block-copy code
- itests changed in different manner
- I'm afraid that's not a full list, patch really changed significantly.

The series also needs one small fix:
"[PATCH] block: fix permission update in bdrv_replace_node"
Based-on: <[email protected]>

Vladimir Sementsov-Ogievskiy (13):
block/backup: fix backup_cow_with_offload for last cluster block/backup: split shareable copying part from backup_do_cow block/backup: introduce BlockCopyState
block/backup: adjust block-copy functions style
block: move block_copy from block/backup.c to separate file
block: teach bdrv_debug_breakpoint skip filters with backing
iotests: prepare 124 and 257 bitmap querying for backup-top filter
iotests: 257: drop unused Drive.device field
iotests: 257: drop device_add
block/io: refactor wait_serialising_requests
block: add lock/unlock range functions
block: introduce backup-top filter driver
block/backup: use backup-top instead of write notifiers

 qapi/block-core.json          |   8 +-
 block/backup-top.h            |  37 ++
 include/block/block-copy.h    |  59 +++
 include/block/block_int.h     |   5 +
 block.c                       |  34 +-
 block/backup-top.c            | 241 ++++++++++++
 block/backup.c                | 411 +++----------------
 block/block-copy.c            | 320 +++++++++++++++
 block/io.c                    |  68 +++-
 block/replication.c           |   2 +-
 blockdev.c                    |   1 +
 block/Makefile.objs           |   3 +
 block/trace-events            |  14 +-
 tests/qemu-iotests/056        |   2 +-
 tests/qemu-iotests/124        |   3 +-
 tests/qemu-iotests/257        |  81 ++--
 tests/qemu-iotests/257.out    | 714 ++++++++++++++--------------------
 tests/qemu-iotests/iotests.py |  22 ++
 18 files changed, 1163 insertions(+), 862 deletions(-)
 create mode 100644 block/backup-top.h
 create mode 100644 include/block/block-copy.h
 create mode 100644 block/backup-top.c
 create mode 100644 block/block-copy.c
Signature · 1 line
2.18.0
#1 lore raw

PATCH v9 1/13 block/backup: fix backup_cow_with_offload for last cluster

We shouldn't try to copy bytes beyond EOF. Fix it.

---
 block/backup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
block/backup.c +1 -1
index 2baf7bed65..33b144305f 100644
@@ -161,7 +161,7 @@ static int coroutine_fn backup_cow_with_offload(BackupBlockJob *job,
 
     assert(QEMU_IS_ALIGNED(job->copy_range_size, job->cluster_size));
     assert(QEMU_IS_ALIGNED(start, job->cluster_size));
-    nbytes = MIN(job->copy_range_size, end - start);
+    nbytes = MIN(job->copy_range_size, MIN(end - start, job->len - start));
     nr_clusters = DIV_ROUND_UP(nbytes, job->cluster_size);
     bdrv_reset_dirty_bitmap(job->copy_bitmap, start,
                             job->cluster_size * nr_clusters);
Signature · 1 line
2.18.0
Reviewed-by on v9 1/13
#2 lore raw

Re: [Qemu-devel] [PATCH v9 01/13] block/backup: fix backup_cow_with_offload for last cluster

Quoted 12 lines · Vladimir Sementsov-Ogievskiy

We shouldn't try to copy bytes beyond EOF. Fix it.

---
 block/backup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
block/backup.c +0 -0
index 2baf7bed65..33b144305f 100644

block/backup.c static int coroutine_fn backup_cow_with_offload(BackupBlockJob *job, · line 164

Quoted 5 lines
block/backup.c +1 -1
@@ -161,7 +161,7 @@ static int coroutine_fn backup_cow_with_offload(BackupBlockJob *job,
 
     assert(QEMU_IS_ALIGNED(job->copy_range_size, job->cluster_size));
     assert(QEMU_IS_ALIGNED(start, job->cluster_size));
-    nbytes = MIN(job->copy_range_size, end - start);
+    nbytes = MIN(job->copy_range_size, MIN(end - start, job->len - start));

Might be easier to read as MIN(end, job->len) - start, but either way:

Quoted 4 lines
block/backup.c +0 -0
     nr_clusters = DIV_ROUND_UP(nbytes, job->cluster_size);
     bdrv_reset_dirty_bitmap(job->copy_bitmap, start,
                             job->cluster_size * nr_clusters);
#3 lore raw

PATCH v9 2/13 block/backup: split shareable copying part from backup_do_cow

In reply to Vladimir Sementsov-Ogievskiy

Split copying logic which will be shared with backup-top filter.

---
 block/backup.c | 47 ++++++++++++++++++++++++++++++++---------------
 1 file changed, 32 insertions(+), 15 deletions(-)
block/backup.c +32 -15
index 33b144305f..13a1d80157 100644
@@ -248,26 +248,18 @@ static int64_t backup_bitmap_reset_unallocated(BackupBlockJob *s,
     return ret;
 }
 
-static int coroutine_fn backup_do_cow(BackupBlockJob *job,
-                                      int64_t offset, uint64_t bytes,
-                                      bool *error_is_read,
-                                      bool is_write_notifier)
+static int coroutine_fn backup_do_copy(BackupBlockJob *job,
+                                       int64_t offset, uint64_t bytes,
+                                       bool *error_is_read,
+                                       bool is_write_notifier)
 {
-    CowRequest cow_request;
     int ret = 0;
-    int64_t start, end; /* bytes */
+    int64_t start = offset, end = bytes + offset; /* bytes */
     void *bounce_buffer = NULL;
     int64_t status_bytes;
 
-    qemu_co_rwlock_rdlock(&job->flush_rwlock);
-
-    start = QEMU_ALIGN_DOWN(offset, job->cluster_size);
-    end = QEMU_ALIGN_UP(bytes + offset, job->cluster_size);
-
-    trace_backup_do_cow_enter(job, start, offset, bytes);
-
-    wait_for_overlapping_requests(job, start, end);
-    cow_request_begin(&cow_request, job, start, end);
+    assert(QEMU_IS_ALIGNED(start, job->cluster_size));
+    assert(QEMU_IS_ALIGNED(end, job->cluster_size));
 
     while (start < end) {
         int64_t dirty_end;
@@ -326,6 +318,31 @@ static int coroutine_fn backup_do_cow(BackupBlockJob *job,
         qemu_vfree(bounce_buffer);
     }
 
+    return ret;
+}
+
+static int coroutine_fn backup_do_cow(BackupBlockJob *job,
+                                      int64_t offset, uint64_t bytes,
+                                      bool *error_is_read,
+                                      bool is_write_notifier)
+{
+    CowRequest cow_request;
+    int ret = 0;
+    int64_t start, end; /* bytes */
+
+    qemu_co_rwlock_rdlock(&job->flush_rwlock);
+
+    start = QEMU_ALIGN_DOWN(offset, job->cluster_size);
+    end = QEMU_ALIGN_UP(bytes + offset, job->cluster_size);
+
+    trace_backup_do_cow_enter(job, start, offset, bytes);
+
+    wait_for_overlapping_requests(job, start, end);
+    cow_request_begin(&cow_request, job, start, end);
+
+    ret = backup_do_copy(job, start, end - start, error_is_read,
+                         is_write_notifier);
+
     cow_request_end(&cow_request);
 
     trace_backup_do_cow_return(job, offset, bytes, ret);
Signature · 1 line
2.18.0
#4 lore raw

Re: [Qemu-devel] [PATCH v9 02/13] block/backup: split shareable copying part from backup_do_cow

Quoted 18 lines · Vladimir Sementsov-Ogievskiy

Split copying logic which will be shared with backup-top filter.

---
 block/backup.c | 47 ++++++++++++++++++++++++++++++++---------------
 1 file changed, 32 insertions(+), 15 deletions(-)
block/backup.c +0 -3
index 33b144305f..13a1d80157 100644
@@ -248,26 +248,18 @@ static int64_t backup_bitmap_reset_unallocated(BackupBlockJob *s,
     return ret;
 }
 
-static int coroutine_fn backup_do_cow(BackupBlockJob *job,
-                                      int64_t offset, uint64_t bytes,
-                                      bool *error_is_read,

block/backup.c static int64_t backup_bitmap_reset_unallocated(BackupBlockJob *s, · line 257

Quoted 10 lines
block/backup.c +5 -3
-                                      bool is_write_notifier)
+static int coroutine_fn backup_do_copy(BackupBlockJob *job,
+                                       int64_t offset, uint64_t bytes,
+                                       bool *error_is_read,
+                                       bool is_write_notifier)
 {
-    CowRequest cow_request;
     int ret = 0;
-    int64_t start, end; /* bytes */
+    int64_t start = offset, end = bytes + offset; /* bytes */

Maybe just rename the “offset” parameter to “start”, replace the “bytes” parameter by an “end” parameter, and drop this line?

Max

Quoted 17 lines
block/backup.c +2 -9
     void *bounce_buffer = NULL;
     int64_t status_bytes;
 
-    qemu_co_rwlock_rdlock(&job->flush_rwlock);
-
-    start = QEMU_ALIGN_DOWN(offset, job->cluster_size);
-    end = QEMU_ALIGN_UP(bytes + offset, job->cluster_size);
-
-    trace_backup_do_cow_enter(job, start, offset, bytes);
-
-    wait_for_overlapping_requests(job, start, end);
-    cow_request_begin(&cow_request, job, start, end);
+    assert(QEMU_IS_ALIGNED(start, job->cluster_size));
+    assert(QEMU_IS_ALIGNED(end, job->cluster_size));
 
     while (start < end) {
         int64_t dirty_end;
#5 lore raw

Re: [Qemu-devel] [PATCH v9 02/13] block/backup: split shareable copying part from backup_do_cow

Quoted 16 lines · Max Reitz
Quoted 15 lines · Vladimir Sementsov-Ogievskiy

Split copying logic which will be shared with backup-top filter.

---
  block/backup.c | 47 ++++++++++++++++++++++++++++++++---------------
  1 file changed, 32 insertions(+), 15 deletions(-)
block/backup.c +0 -0
index 33b144305f..13a1d80157 100644
@@ -248,26 +248,18 @@ static int64_t backup_bitmap_reset_unallocated(BackupBlockJob *s,
      return ret;
  }
 
Quoted 17 lines

block/backup.c static int64_t backup_bitmap_reset_unallocated(BackupBlockJob *s, · line 257

Quoted 13 lines
block/backup.c +5 -6
-static int coroutine_fn backup_do_cow(BackupBlockJob *job,
-                                      int64_t offset, uint64_t bytes,
-                                      bool *error_is_read,
-                                      bool is_write_notifier)
+static int coroutine_fn backup_do_copy(BackupBlockJob *job,
+                                       int64_t offset, uint64_t bytes,
+                                       bool *error_is_read,
+                                       bool is_write_notifier)
  {
-    CowRequest cow_request;
      int ret = 0;
-    int64_t start, end; /* bytes */
+    int64_t start = offset, end = bytes + offset; /* bytes */

Maybe just rename the “offset” parameter to “start”, replace the “bytes” parameter by an “end” parameter, and drop this line?

I really want final block_copy have more common in block-layer offset+bytes interface. So better to refactor a bit the function itself, but I'd prefer to do it as a follow-up and keep this patch simpler..

Quoted 19 lines
Quoted 18 lines
block/backup.c +2 -9
      void *bounce_buffer = NULL;
      int64_t status_bytes;
 
-    qemu_co_rwlock_rdlock(&job->flush_rwlock);
-
-    start = QEMU_ALIGN_DOWN(offset, job->cluster_size);
-    end = QEMU_ALIGN_UP(bytes + offset, job->cluster_size);
-
-    trace_backup_do_cow_enter(job, start, offset, bytes);
-
-    wait_for_overlapping_requests(job, start, end);
-    cow_request_begin(&cow_request, job, start, end);
+    assert(QEMU_IS_ALIGNED(start, job->cluster_size));
+    assert(QEMU_IS_ALIGNED(end, job->cluster_size));
 
      while (start < end) {
          int64_t dirty_end;
Signature · 2 lines
Best regards,
Vladimir
#6 lore raw

Re: [Qemu-devel] [PATCH v9 02/13] block/backup: split shareable copying part from backup_do_cow

Quoted 35 lines · Vladimir Sementsov-Ogievskiy
Quoted 16 lines · Max Reitz
Quoted 15 lines · Vladimir Sementsov-Ogievskiy

Split copying logic which will be shared with backup-top filter.

---
  block/backup.c | 47 ++++++++++++++++++++++++++++++++---------------
  1 file changed, 32 insertions(+), 15 deletions(-)
block/backup.c +0 -0
index 33b144305f..13a1d80157 100644
@@ -248,26 +248,18 @@ static int64_t backup_bitmap_reset_unallocated(BackupBlockJob *s,
      return ret;
  }
 
Quoted 17 lines

block/backup.c static int64_t backup_bitmap_reset_unallocated(BackupBlockJob *s, · line 257

Quoted 13 lines
block/backup.c +5 -6
-static int coroutine_fn backup_do_cow(BackupBlockJob *job,
-                                      int64_t offset, uint64_t bytes,
-                                      bool *error_is_read,
-                                      bool is_write_notifier)
+static int coroutine_fn backup_do_copy(BackupBlockJob *job,
+                                       int64_t offset, uint64_t bytes,
+                                       bool *error_is_read,
+                                       bool is_write_notifier)
  {
-    CowRequest cow_request;
      int ret = 0;
-    int64_t start, end; /* bytes */
+    int64_t start = offset, end = bytes + offset; /* bytes */

Maybe just rename the “offset” parameter to “start”, replace the “bytes” parameter by an “end” parameter, and drop this line?

Quoted 3 lines

I really want final block_copy have more common in block-layer offset+bytes interface. So better to refactor a bit the function itself, but I'd prefer to do it as a follow-up and keep this patch simpler..

OK, but s/offset/start/ should still be possible.

Max

#7 lore raw

PATCH v9 3/13 block/backup: introduce BlockCopyState

In reply to Vladimir Sementsov-Ogievskiy

Split copying code part from backup to "block-copy", including separate state structure and function renaming. This is needed to share it with backup-top filter driver in further commits.

Notes:

1. As BlockCopyState keeps own BlockBackend objects, remaining job->common.blk users only use it to get bs by blk_bs() call, so clear job->commen.blk permissions set in block_job_create.

2. Rename s/initializing_bitmap/skip_unallocated/ to sound a bit better as interface to BlockCopyState

3. Split is not very clean: there left some duplicated fields, backup code uses some BlockCopyState fields directly, let's postpone it for further improvements and keep this comment simpler for review.

---
 block/backup.c     | 324 +++++++++++++++++++++++++++------------------
 block/trace-events |  12 +-
 2 files changed, 200 insertions(+), 136 deletions(-)
block/backup.c +194 -130
index 13a1d80157..f52ac622e0 100644
@@ -35,12 +35,35 @@ typedef struct CowRequest {
     CoQueue wait_queue; /* coroutines blocked on this request */
 } CowRequest;
 
+/*
+ * ProgressCallbackFunc
+ *
+ * Called when some progress is done in context of BlockCopyState:
+ *  1. When some bytes copied, called with @bytes > 0.
+ *  2. When some bytes resetted from copy_bitmap, called with @bytes = 0 (user
+ *     may recalculate remaining bytes from copy_bitmap dirty count.
+ */
+typedef void (*ProgressCallbackFunc)(int64_t bytes, void *opaque);
+typedef struct BlockCopyState {
+    BlockBackend *source;
+    BlockBackend *target;
+    BdrvDirtyBitmap *copy_bitmap;
+    int64_t cluster_size;
+    bool use_copy_range;
+    int64_t copy_range_size;
+    uint64_t len;
+
+    BdrvRequestFlags write_flags;
+    bool skip_unallocated;
+
+    ProgressCallbackFunc progress_callback;
+    void *progress_opaque;
+} BlockCopyState;
+
 typedef struct BackupBlockJob {
     BlockJob common;
-    BlockBackend *target;
 
     BdrvDirtyBitmap *sync_bitmap;
-    BdrvDirtyBitmap *copy_bitmap;
 
     MirrorSyncMode sync_mode;
     BitmapSyncMode bitmap_mode;
@@ -53,11 +76,7 @@ typedef struct BackupBlockJob {
     NotifierWithReturn before_write;
     QLIST_HEAD(, CowRequest) inflight_reqs;
 
-    bool use_copy_range;
-    int64_t copy_range_size;
-
-    BdrvRequestFlags write_flags;
-    bool initializing_bitmap;
+    BlockCopyState *bcs;
 } BackupBlockJob;
 
 static const BlockJobDriver backup_job_driver;
@@ -99,9 +118,83 @@ static void cow_request_end(CowRequest *req)
     qemu_co_queue_restart_all(&req->wait_queue);
 }
 
+static void block_copy_state_free(BlockCopyState *s)
+{
+    if (!s) {
+        return;
+    }
+
+    bdrv_release_dirty_bitmap(blk_bs(s->source), s->copy_bitmap);
+    blk_unref(s->source);
+    s->source = NULL;
+    blk_unref(s->target);
+    s->target = NULL;
+    g_free(s);
+}
+
+static BlockCopyState *block_copy_state_new(
+        BlockDriverState *source, BlockDriverState *target,
+        int64_t cluster_size, BdrvRequestFlags write_flags,
+        ProgressCallbackFunc progress_callback, void *progress_opaque,
+        Error **errp)
+{
+    BlockCopyState *s;
+    int ret;
+    uint64_t no_resize = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
+                         BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD;
+    BdrvDirtyBitmap *copy_bitmap =
+            bdrv_create_dirty_bitmap(source, cluster_size, NULL, errp);
+
+    if (!copy_bitmap) {
+        return NULL;
+    }
+    bdrv_disable_dirty_bitmap(copy_bitmap);
+
+    s = g_new0(BlockCopyState, 1);
+    *s = (BlockCopyState) {
+        .source = blk_new(bdrv_get_aio_context(source),
+                          BLK_PERM_CONSISTENT_READ, no_resize),
+        .target = blk_new(bdrv_get_aio_context(target),
+                          BLK_PERM_WRITE, no_resize),
+        .copy_bitmap = copy_bitmap,
+        .cluster_size = cluster_size,
+        .len = bdrv_dirty_bitmap_size(copy_bitmap),
+        .write_flags = write_flags,
+        .use_copy_range = !(write_flags & BDRV_REQ_WRITE_COMPRESSED),
+        .progress_callback = progress_callback,
+        .progress_opaque = progress_opaque,
+    };
+
+    s->copy_range_size = QEMU_ALIGN_UP(MIN(blk_get_max_transfer(s->source),
+                                           blk_get_max_transfer(s->target)),
+                                       s->cluster_size),
+
+    blk_set_disable_request_queuing(s->source, true);
+    blk_set_allow_aio_context_change(s->source, true);
+    blk_set_disable_request_queuing(s->target, true);
+    blk_set_allow_aio_context_change(s->target, true);
+
+    ret = blk_insert_bs(s->source, source, errp);
+    if (ret < 0) {
+        goto fail;
+    }
+
+    ret = blk_insert_bs(s->target, target, errp);
+    if (ret < 0) {
+        goto fail;
+    }
+
+    return s;
+
+fail:
+    block_copy_state_free(s);
+
+    return NULL;
+}
+
 /* Copy range to target with a bounce buffer and return the bytes copied. If
  * error occurred, return a negative error number */
-static int coroutine_fn backup_cow_with_bounce_buffer(BackupBlockJob *job,
+static int coroutine_fn block_copy_with_bounce_buffer(BlockCopyState *s,
                                                       int64_t start,
                                                       int64_t end,
                                                       bool is_write_notifier,
@@ -109,30 +202,29 @@ static int coroutine_fn backup_cow_with_bounce_buffer(BackupBlockJob *job,
                                                       void **bounce_buffer)
 {
     int ret;
-    BlockBackend *blk = job->common.blk;
     int nbytes;
     int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
 
-    assert(QEMU_IS_ALIGNED(start, job->cluster_size));
-    bdrv_reset_dirty_bitmap(job->copy_bitmap, start, job->cluster_size);
-    nbytes = MIN(job->cluster_size, job->len - start);
+    assert(QEMU_IS_ALIGNED(start, s->cluster_size));
+    bdrv_reset_dirty_bitmap(s->copy_bitmap, start, s->cluster_size);
+    nbytes = MIN(s->cluster_size, s->len - start);
     if (!*bounce_buffer) {
-        *bounce_buffer = blk_blockalign(blk, job->cluster_size);
+        *bounce_buffer = blk_blockalign(s->source, s->cluster_size);
     }
 
-    ret = blk_co_pread(blk, start, nbytes, *bounce_buffer, read_flags);
+    ret = blk_co_pread(s->source, start, nbytes, *bounce_buffer, read_flags);
     if (ret < 0) {
-        trace_backup_do_cow_read_fail(job, start, ret);
+        trace_block_copy_with_bounce_buffer_read_fail(s, start, ret);
         if (error_is_read) {
             *error_is_read = true;
         }
         goto fail;
     }
 
-    ret = blk_co_pwrite(job->target, start, nbytes, *bounce_buffer,
-                        job->write_flags);
+    ret = blk_co_pwrite(s->target, start, nbytes, *bounce_buffer,
+                        s->write_flags);
     if (ret < 0) {
-        trace_backup_do_cow_write_fail(job, start, ret);
+        trace_block_copy_with_bounce_buffer_write_fail(s, start, ret);
         if (error_is_read) {
             *error_is_read = false;
         }
@@ -141,36 +233,35 @@ static int coroutine_fn backup_cow_with_bounce_buffer(BackupBlockJob *job,
 
     return nbytes;
 fail:
-    bdrv_set_dirty_bitmap(job->copy_bitmap, start, job->cluster_size);
+    bdrv_set_dirty_bitmap(s->copy_bitmap, start, s->cluster_size);
     return ret;
 
 }
 
 /* Copy range to target and return the bytes copied. If error occurred, return a
  * negative error number. */
-static int coroutine_fn backup_cow_with_offload(BackupBlockJob *job,
+static int coroutine_fn block_copy_with_offload(BlockCopyState *s,
                                                 int64_t start,
                                                 int64_t end,
                                                 bool is_write_notifier)
 {
     int ret;
     int nr_clusters;
-    BlockBackend *blk = job->common.blk;
     int nbytes;
     int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
 
-    assert(QEMU_IS_ALIGNED(job->copy_range_size, job->cluster_size));
-    assert(QEMU_IS_ALIGNED(start, job->cluster_size));
-    nbytes = MIN(job->copy_range_size, MIN(end - start, job->len - start));
-    nr_clusters = DIV_ROUND_UP(nbytes, job->cluster_size);
-    bdrv_reset_dirty_bitmap(job->copy_bitmap, start,
-                            job->cluster_size * nr_clusters);
-    ret = blk_co_copy_range(blk, start, job->target, start, nbytes,
-                            read_flags, job->write_flags);
+    assert(QEMU_IS_ALIGNED(s->copy_range_size, s->cluster_size));
+    assert(QEMU_IS_ALIGNED(start, s->cluster_size));
+    nbytes = MIN(s->copy_range_size, MIN(end - start, s->len - start));
+    nr_clusters = DIV_ROUND_UP(nbytes, s->cluster_size);
+    bdrv_reset_dirty_bitmap(s->copy_bitmap, start,
+                            s->cluster_size * nr_clusters);
+    ret = blk_co_copy_range(s->source, start, s->target, start, nbytes,
+                            read_flags, s->write_flags);
     if (ret < 0) {
-        trace_backup_do_cow_copy_range_fail(job, start, ret);
-        bdrv_set_dirty_bitmap(job->copy_bitmap, start,
-                              job->cluster_size * nr_clusters);
+        trace_block_copy_with_offload_fail(s, start, ret);
+        bdrv_set_dirty_bitmap(s->copy_bitmap, start,
+                              s->cluster_size * nr_clusters);
         return ret;
     }
 
@@ -181,10 +272,10 @@ static int coroutine_fn backup_cow_with_offload(BackupBlockJob *job,
  * Check if the cluster starting at offset is allocated or not.
  * return via pnum the number of contiguous clusters sharing this allocation.
  */
-static int backup_is_cluster_allocated(BackupBlockJob *s, int64_t offset,
-                                       int64_t *pnum)
+static int block_copy_is_cluster_allocated(BlockCopyState *s, int64_t offset,
+                                           int64_t *pnum)
 {
-    BlockDriverState *bs = blk_bs(s->common.blk);
+    BlockDriverState *bs = blk_bs(s->source);
     int64_t count, total_count = 0;
     int64_t bytes = s->len - offset;
     int ret;
@@ -225,13 +316,13 @@ static int backup_is_cluster_allocated(BackupBlockJob *s, int64_t offset,
  * @return 0 when the cluster at @offset was unallocated,
  *         1 otherwise, and -ret on error.
  */
-static int64_t backup_bitmap_reset_unallocated(BackupBlockJob *s,
-                                               int64_t offset, int64_t *count)
+static int64_t block_copy_reset_unallocated(BlockCopyState *s,
+                                            int64_t offset, int64_t *count)
 {
     int ret;
-    int64_t clusters, bytes, estimate;
+    int64_t clusters, bytes;
 
-    ret = backup_is_cluster_allocated(s, offset, &clusters);
+    ret = block_copy_is_cluster_allocated(s, offset, &clusters);
     if (ret < 0) {
         return ret;
     }
@@ -240,46 +331,45 @@ static int64_t backup_bitmap_reset_unallocated(BackupBlockJob *s,
 
     if (!ret) {
         bdrv_reset_dirty_bitmap(s->copy_bitmap, offset, bytes);
-        estimate = bdrv_get_dirty_count(s->copy_bitmap);
-        job_progress_set_remaining(&s->common.job, estimate);
+        s->progress_callback(0, s->progress_opaque);
     }
 
     *count = bytes;
     return ret;
 }
 
-static int coroutine_fn backup_do_copy(BackupBlockJob *job,
-                                       int64_t offset, uint64_t bytes,
-                                       bool *error_is_read,
-                                       bool is_write_notifier)
+static int coroutine_fn block_copy(BlockCopyState *s,
+                                   int64_t offset, uint64_t bytes,
+                                   bool *error_is_read,
+                                   bool is_write_notifier)
 {
     int ret = 0;
     int64_t start = offset, end = bytes + offset; /* bytes */
     void *bounce_buffer = NULL;
     int64_t status_bytes;
 
-    assert(QEMU_IS_ALIGNED(start, job->cluster_size));
-    assert(QEMU_IS_ALIGNED(end, job->cluster_size));
+    assert(QEMU_IS_ALIGNED(start, s->cluster_size));
+    assert(QEMU_IS_ALIGNED(end, s->cluster_size));
 
     while (start < end) {
         int64_t dirty_end;
 
-        if (!bdrv_dirty_bitmap_get(job->copy_bitmap, start)) {
-            trace_backup_do_cow_skip(job, start);
-            start += job->cluster_size;
+        if (!bdrv_dirty_bitmap_get(s->copy_bitmap, start)) {
+            trace_block_copy_skip(s, start);
+            start += s->cluster_size;
             continue; /* already copied */
         }
 
-        dirty_end = bdrv_dirty_bitmap_next_zero(job->copy_bitmap, start,
+        dirty_end = bdrv_dirty_bitmap_next_zero(s->copy_bitmap, start,
                                                 (end - start));
         if (dirty_end < 0) {
             dirty_end = end;
         }
 
-        if (job->initializing_bitmap) {
-            ret = backup_bitmap_reset_unallocated(job, start, &status_bytes);
+        if (s->skip_unallocated) {
+            ret = block_copy_reset_unallocated(s, start, &status_bytes);
             if (ret == 0) {
-                trace_backup_do_cow_skip_range(job, start, status_bytes);
+                trace_block_copy_skip_range(s, start, status_bytes);
                 start += status_bytes;
                 continue;
             }
@@ -287,17 +377,17 @@ static int coroutine_fn backup_do_copy(BackupBlockJob *job,
             dirty_end = MIN(dirty_end, start + status_bytes);
         }
 
-        trace_backup_do_cow_process(job, start);
+        trace_block_copy_process(s, start);
 
-        if (job->use_copy_range) {
-            ret = backup_cow_with_offload(job, start, dirty_end,
+        if (s->use_copy_range) {
+            ret = block_copy_with_offload(s, start, dirty_end,
                                           is_write_notifier);
             if (ret < 0) {
-                job->use_copy_range = false;
+                s->use_copy_range = false;
             }
         }
-        if (!job->use_copy_range) {
-            ret = backup_cow_with_bounce_buffer(job, start, dirty_end,
+        if (!s->use_copy_range) {
+            ret = block_copy_with_bounce_buffer(s, start, dirty_end,
                                                 is_write_notifier,
                                                 error_is_read, &bounce_buffer);
         }
@@ -305,12 +395,8 @@ static int coroutine_fn backup_do_copy(BackupBlockJob *job,
             break;
         }
 
-        /* Publish progress, guest I/O counts as progress too.  Note that the
-         * offset field is an opaque progress value, it is not a disk offset.
-         */
         start += ret;
-        job->bytes_read += ret;
-        job_progress_update(&job->common.job, ret);
+        s->progress_callback(ret, s->progress_opaque);
         ret = 0;
     }
 
@@ -321,6 +407,20 @@ static int coroutine_fn backup_do_copy(BackupBlockJob *job,
     return ret;
 }
 
+static void backup_progress_callback(int64_t bytes, void *opaque)
+{
+    BackupBlockJob *s = opaque;
+    uint64_t estimate;
+
+    if (bytes > 0) {
+        s->bytes_read += bytes;
+        job_progress_update(&s->common.job, bytes);
+    } else {
+        estimate = bdrv_get_dirty_count(s->bcs->copy_bitmap);
+        job_progress_set_remaining(&s->common.job, estimate);
+    }
+}
+
 static int coroutine_fn backup_do_cow(BackupBlockJob *job,
                                       int64_t offset, uint64_t bytes,
                                       bool *error_is_read,
@@ -340,8 +440,8 @@ static int coroutine_fn backup_do_cow(BackupBlockJob *job,
     wait_for_overlapping_requests(job, start, end);
     cow_request_begin(&cow_request, job, start, end);
 
-    ret = backup_do_copy(job, start, end - start, error_is_read,
-                         is_write_notifier);
+    ret = block_copy(job->bcs, start, end - start, error_is_read,
+                     is_write_notifier);
 
     cow_request_end(&cow_request);
 
@@ -391,7 +491,7 @@ static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret)
 
     if (ret < 0 && job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS) {
         /* If we failed and synced, merge in the bits we didn't copy: */
-        bdrv_dirty_bitmap_merge_internal(bm, job->copy_bitmap,
+        bdrv_dirty_bitmap_merge_internal(bm, job->bcs->copy_bitmap,
                                          NULL, true);
     }
 }
@@ -415,16 +515,9 @@ static void backup_abort(Job *job)
 static void backup_clean(Job *job)
 {
     BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
-    BlockDriverState *bs = blk_bs(s->common.blk);
 
-    if (s->copy_bitmap) {
-        bdrv_release_dirty_bitmap(bs, s->copy_bitmap);
-        s->copy_bitmap = NULL;
-    }
-
-    assert(s->target);
-    blk_unref(s->target);
-    s->target = NULL;
+    block_copy_state_free(s->bcs);
+    s->bcs = NULL;
 }
 
 void backup_do_checkpoint(BlockJob *job, Error **errp)
@@ -439,7 +532,7 @@ void backup_do_checkpoint(BlockJob *job, Error **errp)
         return;
     }
 
-    bdrv_set_dirty_bitmap(backup_job->copy_bitmap, 0, backup_job->len);
+    bdrv_set_dirty_bitmap(backup_job->bcs->copy_bitmap, 0, backup_job->len);
 }
 
 static void backup_drain(BlockJob *job)
@@ -449,8 +542,8 @@ static void backup_drain(BlockJob *job)
     /* Need to keep a reference in case blk_drain triggers execution
      * of backup_complete...
      */
-    if (s->target) {
-        BlockBackend *target = s->target;
+    if (s->bcs && s->bcs->target) {
+        BlockBackend *target = s->bcs->target;
         blk_ref(target);
         blk_drain(target);
         blk_unref(target);
@@ -497,7 +590,7 @@ static int coroutine_fn backup_loop(BackupBlockJob *job)
     BdrvDirtyBitmapIter *bdbi;
     int ret = 0;
 
-    bdbi = bdrv_dirty_iter_new(job->copy_bitmap);
+    bdbi = bdrv_dirty_iter_new(job->bcs->copy_bitmap);
     while ((offset = bdrv_dirty_iter_next(bdbi)) != -1) {
         do {
             if (yield_and_check(job)) {
@@ -524,7 +617,7 @@ static void backup_init_copy_bitmap(BackupBlockJob *job)
     uint64_t estimate;
 
     if (job->sync_mode == MIRROR_SYNC_MODE_BITMAP) {
-        ret = bdrv_dirty_bitmap_merge_internal(job->copy_bitmap,
+        ret = bdrv_dirty_bitmap_merge_internal(job->bcs->copy_bitmap,
                                                job->sync_bitmap,
                                                NULL, true);
         assert(ret);
@@ -534,12 +627,12 @@ static void backup_init_copy_bitmap(BackupBlockJob *job)
              * We can't hog the coroutine to initialize this thoroughly.
              * Set a flag and resume work when we are able to yield safely.
              */
-            job->initializing_bitmap = true;
+            job->bcs->skip_unallocated = true;
         }
-        bdrv_set_dirty_bitmap(job->copy_bitmap, 0, job->len);
+        bdrv_set_dirty_bitmap(job->bcs->copy_bitmap, 0, job->len);
     }
 
-    estimate = bdrv_get_dirty_count(job->copy_bitmap);
+    estimate = bdrv_get_dirty_count(job->bcs->copy_bitmap);
     job_progress_set_remaining(&job->common.job, estimate);
 }
 
@@ -567,14 +660,14 @@ static int coroutine_fn backup_run(Job *job, Error **errp)
                 goto out;
             }
 
-            ret = backup_bitmap_reset_unallocated(s, offset, &count);
+            ret = block_copy_reset_unallocated(s->bcs, offset, &count);
             if (ret < 0) {
                 goto out;
             }
 
             offset += count;
         }
-        s->initializing_bitmap = false;
+        s->bcs->skip_unallocated = false;
     }
 
     if (s->sync_mode == MIRROR_SYNC_MODE_NONE) {
@@ -663,9 +756,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
 {
     int64_t len;
     BackupBlockJob *job = NULL;
-    int ret;
     int64_t cluster_size;
-    BdrvDirtyBitmap *copy_bitmap = NULL;
 
     assert(bs);
     assert(target);
@@ -730,57 +821,34 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
         goto error;
     }
 
-    copy_bitmap = bdrv_create_dirty_bitmap(bs, cluster_size, NULL, errp);
-    if (!copy_bitmap) {
-        goto error;
-    }
-    bdrv_disable_dirty_bitmap(copy_bitmap);
-
     /* job->len is fixed, so we can't allow resize */
-    job = block_job_create(job_id, &backup_job_driver, txn, bs,
-                           BLK_PERM_CONSISTENT_READ,
-                           BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
-                           BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD,
+    job = block_job_create(job_id, &backup_job_driver, txn, bs, 0, BLK_PERM_ALL,
                            speed, creation_flags, cb, opaque, errp);
     if (!job) {
         goto error;
     }
 
-    /* The target must match the source in size, so no resize here either */
-    job->target = blk_new(job->common.job.aio_context,
-                          BLK_PERM_WRITE,
-                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
-                          BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD);
-    ret = blk_insert_bs(job->target, target, errp);
-    if (ret < 0) {
-        goto error;
-    }
-    blk_set_disable_request_queuing(job->target, true);
-
     job->on_source_error = on_source_error;
     job->on_target_error = on_target_error;
     job->sync_mode = sync_mode;
     job->sync_bitmap = sync_bitmap;
     job->bitmap_mode = bitmap_mode;
 
-    /*
-     * Set write flags:
-     * 1. Detect image-fleecing (and similar) schemes
-     * 2. Handle compression
-     */
-    job->write_flags =
-        (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |
-        (compress ? BDRV_REQ_WRITE_COMPRESSED : 0);
+    job->bcs = block_copy_state_new(
+            bs, target, cluster_size,
+            /*
+             * Set write flags:
+             * 1. Detect image-fleecing (and similar) schemes
+             * 2. Handle compression
+             */
+            (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |
+            (compress ? BDRV_REQ_WRITE_COMPRESSED : 0),
+            backup_progress_callback, job, errp);
+    if (!job->bcs) {
+        goto error;
+    }
 
     job->cluster_size = cluster_size;
-    job->copy_bitmap = copy_bitmap;
-    copy_bitmap = NULL;
-    job->use_copy_range = !compress; /* compression isn't supported for it */
-    job->copy_range_size = MIN_NON_ZERO(blk_get_max_transfer(job->common.blk),
-                                        blk_get_max_transfer(job->target));
-    job->copy_range_size = MAX(job->cluster_size,
-                               QEMU_ALIGN_UP(job->copy_range_size,
-                                             job->cluster_size));
 
     /* Required permissions are already taken with target's blk_new() */
     block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
@@ -790,10 +858,6 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
     return &job->common;
 
  error:
-    if (copy_bitmap) {
-        assert(!job || !job->copy_bitmap);
-        bdrv_release_dirty_bitmap(bs, copy_bitmap);
-    }
     if (sync_bitmap) {
         bdrv_reclaim_dirty_bitmap(bs, sync_bitmap, NULL);
     }
block/trace-events +6 -6
index 04209f058d..ad1454f539 100644
@@ -40,12 +40,12 @@ mirror_yield_in_flight(void *s, int64_t offset, int in_flight) "s %p offset %" P
 # backup.c
 backup_do_cow_enter(void *job, int64_t start, int64_t offset, uint64_t bytes) "job %p start %" PRId64 " offset %" PRId64 " bytes %" PRIu64
 backup_do_cow_return(void *job, int64_t offset, uint64_t bytes, int ret) "job %p offset %" PRId64 " bytes %" PRIu64 " ret %d"
-backup_do_cow_skip(void *job, int64_t start) "job %p start %"PRId64
-backup_do_cow_skip_range(void *job, int64_t start, uint64_t bytes) "job %p start %"PRId64" bytes %"PRId64
-backup_do_cow_process(void *job, int64_t start) "job %p start %"PRId64
-backup_do_cow_read_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
-backup_do_cow_write_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
-backup_do_cow_copy_range_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
+block_copy_skip(void *job, int64_t start) "job %p start %"PRId64
+block_copy_skip_range(void *job, int64_t start, uint64_t bytes) "job %p start %"PRId64" bytes %"PRId64
+block_copy_process(void *job, int64_t start) "job %p start %"PRId64
+block_copy_with_bounce_buffer_read_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
+block_copy_with_bounce_buffer_write_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
+block_copy_with_offload_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
 
 # ../blockdev.c
 qmp_block_job_cancel(void *job) "job %p"
Signature · 1 line
2.18.0
#8 lore raw

Re: [Qemu-devel] [PATCH v9 03/13] block/backup: introduce BlockCopyState

Quoted 7 lines · Vladimir Sementsov-Ogievskiy

Split copying code part from backup to "block-copy", including separate state structure and function renaming. This is needed to share it with backup-top filter driver in further commits.

Notes:

1. As BlockCopyState keeps own BlockBackend objects, remaining

I suppose these should be BdrvChild objects at some point, but doing it now would just mean effectively duplicating code from block-backend.c. (“now” = before we have a backup-top filter to attach the children to.)

Quoted 7 lines

job->common.blk users only use it to get bs by blk_bs() call, so clear job->commen.blk permissions set in block_job_create.

2. Rename s/initializing_bitmap/skip_unallocated/ to sound a bit better as interface to BlockCopyState

3. Split is not very clean: there left some duplicated fields, backup

Are there any but cluster_size and len (and source, in a sense)?

Quoted 14 lines

code uses some BlockCopyState fields directly, let's postpone it for further improvements and keep this comment simpler for review.

---
 block/backup.c     | 324 +++++++++++++++++++++++++++------------------
 block/trace-events |  12 +-
 2 files changed, 200 insertions(+), 136 deletions(-)
block/backup.c +0 -0
index 13a1d80157..f52ac622e0 100644

block/backup.c typedef struct CowRequest { · line 43

Quoted 9 lines
block/backup.c +6 -0
@@ -35,12 +35,35 @@ typedef struct CowRequest {
     CoQueue wait_queue; /* coroutines blocked on this request */
 } CowRequest;
 
+/*
+ * ProgressCallbackFunc
+ *
+ * Called when some progress is done in context of BlockCopyState:
+ *  1. When some bytes copied, called with @bytes > 0.
+ *  2. When some bytes resetted from copy_bitmap, called with @bytes = 0 (user

*reset

block/backup.c typedef struct CowRequest { · line 46

Quoted 3 lines
block/backup.c +3 -0
+ *     may recalculate remaining bytes from copy_bitmap dirty count.
+ */
+typedef void (*ProgressCallbackFunc)(int64_t bytes, void *opaque);

Maybe there should be two callbacks instead, one for “We’ve actively made progress” (bytes > 0) and one for “The expected length has changed” (bytes == 0)?

block/backup.c typedef struct CowRequest { · line 57

Quoted 11 lines
block/backup.c +11 -0
+typedef struct BlockCopyState {
+    BlockBackend *source;
+    BlockBackend *target;
+    BdrvDirtyBitmap *copy_bitmap;
+    int64_t cluster_size;
+    bool use_copy_range;
+    int64_t copy_range_size;
+    uint64_t len;
+
+    BdrvRequestFlags write_flags;
+    bool skip_unallocated;

The rename seems reasonable, although I think this should get a comment, because it doesn’t mean just to skip unallocated clusters; it also means to clear unallocated clusters from the bitmap.

Quoted 14 lines
block/backup.c +5 -2
+
+    ProgressCallbackFunc progress_callback;
+    void *progress_opaque;
+} BlockCopyState;
+
 typedef struct BackupBlockJob {
     BlockJob common;
-    BlockBackend *target;
 
     BdrvDirtyBitmap *sync_bitmap;
-    BdrvDirtyBitmap *copy_bitmap;
 
     MirrorSyncMode sync_mode;
     BitmapSyncMode bitmap_mode;

[...]

Quoted 5 lines
block/backup.c +1 -0
@@ -99,9 +118,83 @@ static void cow_request_end(CowRequest *req)
     qemu_co_queue_restart_all(&req->wait_queue);
 }
 
+static void block_copy_state_free(BlockCopyState *s)

block/backup.c static void cow_request_end(CowRequest *req) · line 131

Quoted 10 lines
block/backup.c +10 -0
+{
+    if (!s) {
+        return;
+    }
+
+    bdrv_release_dirty_bitmap(blk_bs(s->source), s->copy_bitmap);
+    blk_unref(s->source);
+    s->source = NULL;
+    blk_unref(s->target);
+    s->target = NULL;

I’m not quite sure why you NULL these pointers when you free the whole object next anyway.

Quoted 6 lines
block/backup.c +6 -0
+    g_free(s);
+}
+
+static BlockCopyState *block_copy_state_new(
+        BlockDriverState *source, BlockDriverState *target,
+        int64_t cluster_size, BdrvRequestFlags write_flags,

block/backup.c static void cow_request_end(CowRequest *req) · line 147

Quoted 10 lines
block/backup.c +10 -0
+        ProgressCallbackFunc progress_callback, void *progress_opaque,
+        Error **errp)
+{
+    BlockCopyState *s;
+    int ret;
+    uint64_t no_resize = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
+                         BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD;
+    BdrvDirtyBitmap *copy_bitmap =
+            bdrv_create_dirty_bitmap(source, cluster_size, NULL, errp);
+

This probably were easier to read if you didn’t initialize copy_bitmap with the bdrv_create_dirty_bitmap() call but instead moved that call right above the if () here (it still fits on a single line).

block/backup.c static void cow_request_end(CowRequest *req) · line 153

Quoted 6 lines
block/backup.c +6 -0
+    if (!copy_bitmap) {
+        return NULL;
+    }
+    bdrv_disable_dirty_bitmap(copy_bitmap);
+
+    s = g_new0(BlockCopyState, 1);

With the following compound literal, you don’t need to allocate zero-initialized memory here.

block/backup.c static void cow_request_end(CowRequest *req) · line 158

Quoted 5 lines
block/backup.c +5 -0
+    *s = (BlockCopyState) {
+        .source = blk_new(bdrv_get_aio_context(source),
+                          BLK_PERM_CONSISTENT_READ, no_resize),
+        .target = blk_new(bdrv_get_aio_context(target),
+                          BLK_PERM_WRITE, no_resize),

Maybe we want to assert that source’s and target’s context are the same?

block/backup.c static void cow_request_end(CowRequest *req) · line 170

Quoted 12 lines
block/backup.c +12 -0
+        .copy_bitmap = copy_bitmap,
+        .cluster_size = cluster_size,
+        .len = bdrv_dirty_bitmap_size(copy_bitmap),
+        .write_flags = write_flags,
+        .use_copy_range = !(write_flags & BDRV_REQ_WRITE_COMPRESSED),
+        .progress_callback = progress_callback,
+        .progress_opaque = progress_opaque,
+    };
+
+    s->copy_range_size = QEMU_ALIGN_UP(MIN(blk_get_max_transfer(s->source),
+                                           blk_get_max_transfer(s->target)),
+                                       s->cluster_size),

Nice simplification.

block/backup.c static void cow_request_end(CowRequest *req) · line 175

Quoted 5 lines
block/backup.c +5 -0
+
+    blk_set_disable_request_queuing(s->source, true);
+    blk_set_allow_aio_context_change(s->source, true);
+    blk_set_disable_request_queuing(s->target, true);
+    blk_set_allow_aio_context_change(s->target, true);

Hm. Doesn’t creating new BBs here mean that we have to deal with the fallout of changing the AioContext on either of them somewhere?

[...]

block/backup.c static void backup_drain(BlockJob *job) · line 545

Quoted 7 lines
block/backup.c +1 -2
@@ -449,8 +542,8 @@ static void backup_drain(BlockJob *job)
     /* Need to keep a reference in case blk_drain triggers execution
      * of backup_complete...
      */
-    if (s->target) {
-        BlockBackend *target = s->target;
+    if (s->bcs && s->bcs->target) {

bcs->target should always be non-NULL, shouldn’t it?

Quoted 4 lines
block/backup.c +1 -0
+        BlockBackend *target = s->bcs->target;
         blk_ref(target);
         blk_drain(target);
         blk_unref(target);

[...]

Quoted 1 line
block/backup.c +0 -0
@@ -730,57 +821,34 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,

[...]

Quoted 7 lines
block/backup.c +0 -7
-    /*
-     * Set write flags:
-     * 1. Detect image-fleecing (and similar) schemes
-     * 2. Handle compression
-     */
-    job->write_flags =
-        (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |

block/backup.c BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs, · line 845

Quoted 10 lines
block/backup.c +9 -1
-        (compress ? BDRV_REQ_WRITE_COMPRESSED : 0);
+    job->bcs = block_copy_state_new(
+            bs, target, cluster_size,
+            /*
+             * Set write flags:
+             * 1. Detect image-fleecing (and similar) schemes
+             * 2. Handle compression
+             */
+            (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |
+            (compress ? BDRV_REQ_WRITE_COMPRESSED : 0),

This is a bit hard to read. Why not add a dedicated variable for it?

Quoted 17 lines
block/backup.c +4 -8
+            backup_progress_callback, job, errp);
+    if (!job->bcs) {
+        goto error;
+    }
 
     job->cluster_size = cluster_size;
-    job->copy_bitmap = copy_bitmap;
-    copy_bitmap = NULL;
-    job->use_copy_range = !compress; /* compression isn't supported for it */
-    job->copy_range_size = MIN_NON_ZERO(blk_get_max_transfer(job->common.blk),
-                                        blk_get_max_transfer(job->target));
-    job->copy_range_size = MAX(job->cluster_size,
-                               QEMU_ALIGN_UP(job->copy_range_size,
-                                             job->cluster_size));
 
     /* Required permissions are already taken with target's blk_new() */
     block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,

[...]

Quoted 10 lines
block/trace-events +0 -2
index 04209f058d..ad1454f539 100644
@@ -40,12 +40,12 @@ mirror_yield_in_flight(void *s, int64_t offset, int in_flight) "s %p offset %" P
 # backup.c
 backup_do_cow_enter(void *job, int64_t start, int64_t offset, uint64_t bytes) "job %p start %" PRId64 " offset %" PRId64 " bytes %" PRIu64
 backup_do_cow_return(void *job, int64_t offset, uint64_t bytes, int ret) "job %p offset %" PRId64 " bytes %" PRIu64 " ret %d"
-backup_do_cow_skip(void *job, int64_t start) "job %p start %"PRId64
-backup_do_cow_skip_range(void *job, int64_t start, uint64_t bytes) "job %p start %"PRId64" bytes %"PRId64

block/trace-events mirror_yield_in_flight(void *s, int64_t offset, int in_flight) "s %p offset %" P · line 48

Quoted 10 lines
block/trace-events +6 -4
-backup_do_cow_process(void *job, int64_t start) "job %p start %"PRId64
-backup_do_cow_read_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
-backup_do_cow_write_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
-backup_do_cow_copy_range_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
+block_copy_skip(void *job, int64_t start) "job %p start %"PRId64
+block_copy_skip_range(void *job, int64_t start, uint64_t bytes) "job %p start %"PRId64" bytes %"PRId64
+block_copy_process(void *job, int64_t start) "job %p start %"PRId64
+block_copy_with_bounce_buffer_read_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
+block_copy_with_bounce_buffer_write_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
+block_copy_with_offload_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"

The pointer is no longer a job pointer, though.

Max

Quoted 4 lines
block/trace-events +0 -0
 # ../blockdev.c
 qmp_block_job_cancel(void *job) "job %p"
 
#9 lore raw

Re: [Qemu-devel] [PATCH v9 03/13] block/backup: introduce BlockCopyState

Thanks for reviewing!

Quoted 5 lines · Max Reitz
Quoted 4 lines · Vladimir Sementsov-Ogievskiy

Split copying code part from backup to "block-copy", including separate state structure and function renaming. This is needed to share it with backup-top filter driver in further commits.

Quoted 7 lines
Quoted 3 lines

Notes:

1. As BlockCopyState keeps own BlockBackend objects, remaining

I suppose these should be BdrvChild objects at some point, but doing it now would just mean effectively duplicating code from block-backend.c. (“now” = before we have a backup-top filter to attach the children to.)

How much is it bad to not do it, but leave them to be block-backends in block-copy state? They'll connected anyway through the job, as they all are in job.nodes.

We have block-backends in jobs currently, is it bad?

Quoted 10 lines
Quoted 8 lines

job->common.blk users only use it to get bs by blk_bs() call, so clear job->commen.blk permissions set in block_job_create.

2. Rename s/initializing_bitmap/skip_unallocated/ to sound a bit better as interface to BlockCopyState

3. Split is not very clean: there left some duplicated fields, backup

Are there any but cluster_size and len (and source, in a sense)?

Seems no more

Quoted 27 lines
Quoted 15 lines

code uses some BlockCopyState fields directly, let's postpone it for further improvements and keep this comment simpler for review.

---
  block/backup.c     | 324 +++++++++++++++++++++++++++------------------
  block/trace-events |  12 +-
  2 files changed, 200 insertions(+), 136 deletions(-)
block/backup.c +0 -0
index 13a1d80157..f52ac622e0 100644

block/backup.c typedef struct CowRequest { · line 43

Quoted 9 lines
block/backup.c +6 -0
@@ -35,12 +35,35 @@ typedef struct CowRequest {
      CoQueue wait_queue; /* coroutines blocked on this request */
  } CowRequest;
 
+/*
+ * ProgressCallbackFunc
+ *
+ * Called when some progress is done in context of BlockCopyState:
+ *  1. When some bytes copied, called with @bytes > 0.
+ *  2. When some bytes resetted from copy_bitmap, called with @bytes = 0 (user

*reset

Quoted 7 lines

block/backup.c typedef struct CowRequest { · line 46

Quoted 3 lines
block/backup.c +3 -0
+ *     may recalculate remaining bytes from copy_bitmap dirty count.
+ */
+typedef void (*ProgressCallbackFunc)(int64_t bytes, void *opaque);

Maybe there should be two callbacks instead, one for “We’ve actively made progress” (bytes > 0) and one for “The expected length has changed” (bytes == 0)?

I thought, that there are already too many parameters in block_copy_state_new(). But I agree with you, as actually it led to two callbacks in a one with just if-else to distinguish them. Will do.

Quoted 38 lines

block/backup.c typedef struct CowRequest { · line 57

Quoted 12 lines
block/backup.c +11 -0
+typedef struct BlockCopyState {
+    BlockBackend *source;
+    BlockBackend *target;
+    BdrvDirtyBitmap *copy_bitmap;
+    int64_t cluster_size;
+    bool use_copy_range;
+    int64_t copy_range_size;
+    uint64_t len;
+
+    BdrvRequestFlags write_flags;
+    bool skip_unallocated;

The rename seems reasonable, although I think this should get a comment, because it doesn’t mean just to skip unallocated clusters; it also means to clear unallocated clusters from the bitmap.

Quoted 14 lines
block/backup.c +5 -2
+
+    ProgressCallbackFunc progress_callback;
+    void *progress_opaque;
+} BlockCopyState;
+
  typedef struct BackupBlockJob {
      BlockJob common;
-    BlockBackend *target;
 
      BdrvDirtyBitmap *sync_bitmap;
-    BdrvDirtyBitmap *copy_bitmap;
 
      MirrorSyncMode sync_mode;
      BitmapSyncMode bitmap_mode;

[...]

Quoted 4 lines
block/backup.c +0 -0
@@ -99,9 +118,83 @@ static void cow_request_end(CowRequest *req)
      qemu_co_queue_restart_all(&req->wait_queue);
  }
 
Quoted 14 lines

block/backup.c static void cow_request_end(CowRequest *req) · line 131

Quoted 11 lines
block/backup.c +11 -0
+static void block_copy_state_free(BlockCopyState *s)
+{
+    if (!s) {
+        return;
+    }
+
+    bdrv_release_dirty_bitmap(blk_bs(s->source), s->copy_bitmap);
+    blk_unref(s->source);
+    s->source = NULL;
+    blk_unref(s->target);
+    s->target = NULL;

I’m not quite sure why you NULL these pointers when you free the whole object next anyway.

it is for backup_drain, I'm afraid of some yield during blk_unref (and seems it's unsafe anyway, as I zero reference after calling blk_unref). Anyway,
backup_drain will be dropped in "[PATCH v3] job: drop job_drain", I'll drop "= NULL" here now and workaround backup_drain in backup_clean with corresponding comment.

Quoted 32 lines
Quoted 7 lines
block/backup.c +6 -0
+    g_free(s);
+}
+
+static BlockCopyState *block_copy_state_new(
+        BlockDriverState *source, BlockDriverState *target,
+        int64_t cluster_size, BdrvRequestFlags write_flags,

block/backup.c static void cow_request_end(CowRequest *req) · line 147

Quoted 10 lines
block/backup.c +10 -0
+        ProgressCallbackFunc progress_callback, void *progress_opaque,
+        Error **errp)
+{
+    BlockCopyState *s;
+    int ret;
+    uint64_t no_resize = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
+                         BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD;
+    BdrvDirtyBitmap *copy_bitmap =
+            bdrv_create_dirty_bitmap(source, cluster_size, NULL, errp);
+

This probably were easier to read if you didn’t initialize copy_bitmap with the bdrv_create_dirty_bitmap() call but instead moved that call right above the if () here (it still fits on a single line).

block/backup.c static void cow_request_end(CowRequest *req) · line 153

Quoted 6 lines
block/backup.c +6 -0
+    if (!copy_bitmap) {
+        return NULL;
+    }
+    bdrv_disable_dirty_bitmap(copy_bitmap);
+
+    s = g_new0(BlockCopyState, 1);

With the following compound literal, you don’t need to allocate zero-initialized memory here.

Quoted 7 lines

block/backup.c static void cow_request_end(CowRequest *req) · line 158

Quoted 5 lines
block/backup.c +5 -0
+    *s = (BlockCopyState) {
+        .source = blk_new(bdrv_get_aio_context(source),
+                          BLK_PERM_CONSISTENT_READ, no_resize),
+        .target = blk_new(bdrv_get_aio_context(target),
+                          BLK_PERM_WRITE, no_resize),

Maybe we want to assert that source’s and target’s context are the same?

Context may change, so no reason to check it here. It'd better be asserted in block_copy() before copying, I wanted to do it, but forget, will add.

Quoted 15 lines

block/backup.c static void cow_request_end(CowRequest *req) · line 170

Quoted 13 lines
block/backup.c +12 -0
+        .copy_bitmap = copy_bitmap,
+        .cluster_size = cluster_size,
+        .len = bdrv_dirty_bitmap_size(copy_bitmap),
+        .write_flags = write_flags,
+        .use_copy_range = !(write_flags & BDRV_REQ_WRITE_COMPRESSED),
+        .progress_callback = progress_callback,
+        .progress_opaque = progress_opaque,
+    };
+
+    s->copy_range_size = QEMU_ALIGN_UP(MIN(blk_get_max_transfer(s->source),
+                                           blk_get_max_transfer(s->target)),
+                                       s->cluster_size),

Nice simplification. >

Quoted 8 lines

block/backup.c static void cow_request_end(CowRequest *req) · line 175

Quoted 5 lines
block/backup.c +5 -0
+
+    blk_set_disable_request_queuing(s->source, true);
+    blk_set_allow_aio_context_change(s->source, true);
+    blk_set_disable_request_queuing(s->target, true);
+    blk_set_allow_aio_context_change(s->target, true);

Hm. Doesn’t creating new BBs here mean that we have to deal with the fallout of changing the AioContext on either of them somewhere?

In backup context, backup job is responsible for keeping source and target bs in same context, so I think allowing blk to change aio context and assert in block_copy() that context is the same should be enough for now.

I'll add a comment on this here.

Quoted 12 lines

[...]

block/backup.c static void backup_drain(BlockJob *job) · line 545

Quoted 7 lines
block/backup.c +1 -2
@@ -449,8 +542,8 @@ static void backup_drain(BlockJob *job)
      /* Need to keep a reference in case blk_drain triggers execution
       * of backup_complete...
       */
-    if (s->target) {
-        BlockBackend *target = s->target;
+    if (s->bcs && s->bcs->target) {

bcs->target should always be non-NULL, shouldn’t it?

But if we intersect with some yield in cleanu-up procedure... Anyway, backup_drain will be dropped soon I hope.

Quoted 83 lines
Quoted 5 lines
block/backup.c +1 -0
+        BlockBackend *target = s->bcs->target;
          blk_ref(target);
          blk_drain(target);
          blk_unref(target);

[...]

Quoted 1 line
block/backup.c +0 -0
@@ -730,57 +821,34 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,

[...]

Quoted 7 lines
block/backup.c +0 -7
-    /*
-     * Set write flags:
-     * 1. Detect image-fleecing (and similar) schemes
-     * 2. Handle compression
-     */
-    job->write_flags =
-        (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |

block/backup.c BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs, · line 845

Quoted 10 lines
block/backup.c +9 -1
-        (compress ? BDRV_REQ_WRITE_COMPRESSED : 0);
+    job->bcs = block_copy_state_new(
+            bs, target, cluster_size,
+            /*
+             * Set write flags:
+             * 1. Detect image-fleecing (and similar) schemes
+             * 2. Handle compression
+             */
+            (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |
+            (compress ? BDRV_REQ_WRITE_COMPRESSED : 0),

This is a bit hard to read. Why not add a dedicated variable for it?

Quoted 17 lines
block/backup.c +4 -8
+            backup_progress_callback, job, errp);
+    if (!job->bcs) {
+        goto error;
+    }
 
      job->cluster_size = cluster_size;
-    job->copy_bitmap = copy_bitmap;
-    copy_bitmap = NULL;
-    job->use_copy_range = !compress; /* compression isn't supported for it */
-    job->copy_range_size = MIN_NON_ZERO(blk_get_max_transfer(job->common.blk),
-                                        blk_get_max_transfer(job->target));
-    job->copy_range_size = MAX(job->cluster_size,
-                               QEMU_ALIGN_UP(job->copy_range_size,
-                                             job->cluster_size));
 
      /* Required permissions are already taken with target's blk_new() */
      block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,

[...]

Quoted 10 lines
block/trace-events +0 -2
index 04209f058d..ad1454f539 100644
@@ -40,12 +40,12 @@ mirror_yield_in_flight(void *s, int64_t offset, int in_flight) "s %p offset %" P
  # backup.c
  backup_do_cow_enter(void *job, int64_t start, int64_t offset, uint64_t bytes) "job %p start %" PRId64 " offset %" PRId64 " bytes %" PRIu64
  backup_do_cow_return(void *job, int64_t offset, uint64_t bytes, int ret) "job %p offset %" PRId64 " bytes %" PRIu64 " ret %d"
-backup_do_cow_skip(void *job, int64_t start) "job %p start %"PRId64
-backup_do_cow_skip_range(void *job, int64_t start, uint64_t bytes) "job %p start %"PRId64" bytes %"PRId64

block/trace-events mirror_yield_in_flight(void *s, int64_t offset, int in_flight) "s %p offset %" P · line 48

Quoted 10 lines
block/trace-events +6 -4
-backup_do_cow_process(void *job, int64_t start) "job %p start %"PRId64
-backup_do_cow_read_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
-backup_do_cow_write_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
-backup_do_cow_copy_range_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
+block_copy_skip(void *job, int64_t start) "job %p start %"PRId64
+block_copy_skip_range(void *job, int64_t start, uint64_t bytes) "job %p start %"PRId64" bytes %"PRId64
+block_copy_process(void *job, int64_t start) "job %p start %"PRId64
+block_copy_with_bounce_buffer_read_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
+block_copy_with_bounce_buffer_write_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"
+block_copy_with_offload_fail(void *job, int64_t start, int ret) "job %p start %"PRId64" ret %d"

The pointer is no longer a job pointer, though.

Max

Quoted 4 lines
block/trace-events +0 -0
  # ../blockdev.c
  qmp_block_job_cancel(void *job) "job %p"
 
Signature · 2 lines
Best regards,
Vladimir
#10 lore raw

Re: [Qemu-devel] [PATCH v9 03/13] block/backup: introduce BlockCopyState

Quoted 12 lines · Vladimir Sementsov-Ogievskiy

Thanks for reviewing!

Quoted 9 lines · Max Reitz
Quoted 7 lines · Vladimir Sementsov-Ogievskiy

Split copying code part from backup to "block-copy", including separate state structure and function renaming. This is needed to share it with backup-top filter driver in further commits.

Notes:

1. As BlockCopyState keeps own BlockBackend objects, remaining

Quoted 8 lines
Quoted 3 lines

I suppose these should be BdrvChild objects at some point, but doing it now would just mean effectively duplicating code from block-backend.c. (“now” = before we have a backup-top filter to attach the children to.)

How much is it bad to not do it, but leave them to be block-backends in block-copy state? They'll connected anyway through the job, as they all are in job.nodes.

We have block-backends in jobs currently, is it bad?

Yes. First of all, it’s simply not how it should be. It’s ugly.

Second, it does produce tangible problems. One thing that comes to mind is that we only need BB.disable_request_queuing because these BlockBackends do not have a clear connection to the block job, which means that the job may want to perform requests on drained BBs.

If source and target were children of the filter node, draining them would first drain the job, and only then would they be marked as quiesced, thus solving the problem (as far as I remember).

Quoted 11 lines
Quoted 9 lines
Quoted 7 lines

job->common.blk users only use it to get bs by blk_bs() call, so clear job->commen.blk permissions set in block_job_create.

2. Rename s/initializing_bitmap/skip_unallocated/ to sound a bit better as interface to BlockCopyState

3. Split is not very clean: there left some duplicated fields, backup

Are there any but cluster_size and len (and source, in a sense)?

Seems no more

Good, I was just asking because duplicated fields may be difficult to keep in sync and so on.

[...]

Quoted 19 lines
Quoted 4 lines
Quoted 4 lines
block/backup.c +0 -0
@@ -99,9 +118,83 @@ static void cow_request_end(CowRequest *req)
      qemu_co_queue_restart_all(&req->wait_queue);
  }
 
Quoted 14 lines

block/backup.c static void cow_request_end(CowRequest *req) · line 131

Quoted 11 lines
block/backup.c +11 -0
+static void block_copy_state_free(BlockCopyState *s)
+{
+    if (!s) {
+        return;
+    }
+
+    bdrv_release_dirty_bitmap(blk_bs(s->source), s->copy_bitmap);
+    blk_unref(s->source);
+    s->source = NULL;
+    blk_unref(s->target);
+    s->target = NULL;

I’m not quite sure why you NULL these pointers when you free the whole object next anyway.

Quoted 5 lines

it is for backup_drain, I'm afraid of some yield during blk_unref (and seems it's unsafe anyway, as I zero reference after calling blk_unref). Anyway,
backup_drain will be dropped in "[PATCH v3] job: drop job_drain", I'll drop "= NULL" here now and workaround backup_drain in backup_clean with corresponding comment.

OK.

[...]

Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
block/backup.c +5 -0
+
+    blk_set_disable_request_queuing(s->source, true);
+    blk_set_allow_aio_context_change(s->source, true);
+    blk_set_disable_request_queuing(s->target, true);
+    blk_set_allow_aio_context_change(s->target, true);
Quoted 6 lines
Quoted 2 lines

Hm. Doesn’t creating new BBs here mean that we have to deal with the fallout of changing the AioContext on either of them somewhere?

In backup context, backup job is responsible for keeping source and target bs in same context, so I think allowing blk to change aio context and assert in block_copy() that context is the same should be enough for now.

Hm, OK, and the backup job takes care of that through child_job_set_aio_ctx() in blockjob.c.

But it should still be noted that on master, if you try to move e.g. the source to a new context (by attaching a device to it), this happens:

qemu-system-x86_64: Cannot change iothread of active block backend

This comes from blk_root_can_set_aio_ctx(); but at the same time, it will happily allow the context change if blk->allow_aio_context_change is set – which is precisely what you do here.

So with this patch, the change is allowed; but the target is moved to the new context, too.

So it should be noted that this is a change in behavior. (Well, at least I wanted to note it here.)

Quoted 1 line

I'll add a comment on this here.

By the way, this is another problem that we wouldn’t have if source and target were BdrvChildren of backup-top.

(The problem being that the BBs’ contexts are kept in sync indirectly through the node list attached to the job.)

Max

#11 lore raw

PATCH v9 4/13 block/backup: adjust block-copy functions style

In reply to Vladimir Sementsov-Ogievskiy

Fix comment style and reflow arguments in same manner like block_copy_state_new.

---
 block/backup.c | 40 +++++++++++++++++++---------------------
 1 file changed, 19 insertions(+), 21 deletions(-)
block/backup.c +19 -21
index f52ac622e0..d9192ac778 100644
@@ -192,14 +192,13 @@ fail:
     return NULL;
 }
 
-/* Copy range to target with a bounce buffer and return the bytes copied. If
- * error occurred, return a negative error number */
-static int coroutine_fn block_copy_with_bounce_buffer(BlockCopyState *s,
-                                                      int64_t start,
-                                                      int64_t end,
-                                                      bool is_write_notifier,
-                                                      bool *error_is_read,
-                                                      void **bounce_buffer)
+/*
+ * Copy range to target with a bounce buffer and return the bytes copied. If
+ * error occurred, return a negative error number
+ */
+static int coroutine_fn block_copy_with_bounce_buffer(
+        BlockCopyState *s, int64_t start, int64_t end, bool is_write_notifier,
+        bool *error_is_read, void **bounce_buffer)
 {
     int ret;
     int nbytes;
@@ -238,12 +237,12 @@ fail:
 
 }
 
-/* Copy range to target and return the bytes copied. If error occurred, return a
- * negative error number. */
-static int coroutine_fn block_copy_with_offload(BlockCopyState *s,
-                                                int64_t start,
-                                                int64_t end,
-                                                bool is_write_notifier)
+/*
+ * Copy range to target and return the bytes copied. If error occurred, return a
+ * negative error number.
+ */
+static int coroutine_fn block_copy_with_offload(
+        BlockCopyState *s, int64_t start, int64_t end, bool is_write_notifier)
 {
     int ret;
     int nr_clusters;
@@ -310,14 +309,14 @@ static int block_copy_is_cluster_allocated(BlockCopyState *s, int64_t offset,
     }
 }
 
-/**
+/*
  * Reset bits in copy_bitmap starting at offset if they represent unallocated
  * data in the image. May reset subsequent contiguous bits.
  * @return 0 when the cluster at @offset was unallocated,
  *         1 otherwise, and -ret on error.
  */
-static int64_t block_copy_reset_unallocated(BlockCopyState *s,
-                                            int64_t offset, int64_t *count)
+static int64_t block_copy_reset_unallocated(
+        BlockCopyState *s, int64_t offset, int64_t *count)
 {
     int ret;
     int64_t clusters, bytes;
@@ -338,10 +337,9 @@ static int64_t block_copy_reset_unallocated(BlockCopyState *s,
     return ret;
 }
 
-static int coroutine_fn block_copy(BlockCopyState *s,
-                                   int64_t offset, uint64_t bytes,
-                                   bool *error_is_read,
-                                   bool is_write_notifier)
+static int coroutine_fn block_copy(
+        BlockCopyState *s, int64_t offset, uint64_t bytes, bool *error_is_read,
+        bool is_write_notifier)
 {
     int ret = 0;
     int64_t start = offset, end = bytes + offset; /* bytes */
Signature · 1 line
2.18.0
#12 lore raw

Re: [Qemu-devel] [PATCH v9 04/13] block/backup: adjust block-copy functions style

Quoted 2 lines · Vladimir Sementsov-Ogievskiy

Fix comment style and reflow arguments in same manner like block_copy_state_new.

I like the current function header style better.

Max

Quoted 5 lines
---
 block/backup.c | 40 +++++++++++++++++++---------------------
 1 file changed, 19 insertions(+), 21 deletions(-)
#13 lore raw

Re: [Qemu-devel] [PATCH v9 04/13] block/backup: adjust block-copy functions style

Quoted 6 lines · Max Reitz
Quoted 2 lines · Vladimir Sementsov-Ogievskiy

Fix comment style and reflow arguments in same manner like block_copy_state_new.

I like the current function header style better.

OK, not critical at all for me, let's keep current. I'll instead fix all comments style here.

Signature · 2 lines
Best regards,
Vladimir
#14 lore raw

PATCH v9 5/13 block: move block_copy from block/backup.c to separate file

In reply to Vladimir Sementsov-Ogievskiy

Split block_copy to separate file, to be cleanly shared with backup-top filter driver in further commits.

It's a clean movement, the only change is drop "static" from interface functions.

---
 include/block/block-copy.h |  59 +++++++
 block/backup.c             | 313 +------------------------------------
 block/block-copy.c         | 307 ++++++++++++++++++++++++++++++++++++
 block/Makefile.objs        |   1 +
 block/trace-events         |   2 +
 5 files changed, 370 insertions(+), 312 deletions(-)
 create mode 100644 include/block/block-copy.h
 create mode 100644 block/block-copy.c
include/block/block-copy.h new file +59 -0
new file mode 100644
index 0000000000..0dd7a3f7bf
@@ -0,0 +1,59 @@
+/*
+ * block_copy API
+ *
+ * Copyright (C) 2013 Proxmox Server Solutions
+ * Copyright (c) 2019 Virtuozzo International GmbH.
+ *
+ * Authors:
+ *  Dietmar Maurer ([email protected])
+ *  Vladimir Sementsov-Ogievskiy <[email protected]>
+ *
+ * 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 BLOCK_COPY_H
+#define BLOCK_COPY_H
+
+#include "block/block.h"
+
+/*
+ * ProgressCallbackFunc
+ *
+ * Called when some progress is done in context of BlockCopyState:
+ *  1. When some bytes copied, called with @bytes > 0.
+ *  2. When some bytes resetted from copy_bitmap, called with @bytes = 0 (user
+ *     may recalculate remaining bytes from copy_bitmap dirty count.
+ */
+typedef void (*ProgressCallbackFunc)(int64_t bytes, void *opaque);
+typedef struct BlockCopyState {
+    BlockBackend *source;
+    BlockBackend *target;
+    BdrvDirtyBitmap *copy_bitmap;
+    int64_t cluster_size;
+    bool use_copy_range;
+    int64_t copy_range_size;
+    uint64_t len;
+
+    BdrvRequestFlags write_flags;
+    bool skip_unallocated;
+
+    ProgressCallbackFunc progress_callback;
+    void *progress_opaque;
+} BlockCopyState;
+
+BlockCopyState *block_copy_state_new(
+        BlockDriverState *source, BlockDriverState *target,
+        int64_t cluster_size, BdrvRequestFlags write_flags,
+        ProgressCallbackFunc progress_callback, void *progress_opaque,
+        Error **errp);
+
+void block_copy_state_free(BlockCopyState *s);
+
+int64_t block_copy_reset_unallocated(BlockCopyState *s,
+                                     int64_t offset, int64_t *count);
+
+int coroutine_fn block_copy(BlockCopyState *s, int64_t offset, uint64_t bytes,
+                            bool *error_is_read, bool is_write_notifier);
+
+#endif /* BLOCK_COPY_H */
block/backup.c +1 -312

331 lines not on this page · raw message

block/block-copy.c new file +307 -0

307 lines not on this page · raw message

block/Makefile.objs +1 -0
index 35f3bca4d9..0b5c635fb2 100644
@@ -37,6 +37,7 @@ block-obj-y += write-threshold.o
 block-obj-y += backup.o
 block-obj-$(CONFIG_REPLICATION) += replication.o
 block-obj-y += throttle.o copy-on-read.o
+block-obj-y += block-copy.o
 
 block-obj-y += crypto.o
 
block/trace-events +2 -0
index ad1454f539..ad5261836a 100644
@@ -40,6 +40,8 @@ mirror_yield_in_flight(void *s, int64_t offset, int in_flight) "s %p offset %" P
 # backup.c
 backup_do_cow_enter(void *job, int64_t start, int64_t offset, uint64_t bytes) "job %p start %" PRId64 " offset %" PRId64 " bytes %" PRIu64
 backup_do_cow_return(void *job, int64_t offset, uint64_t bytes, int ret) "job %p offset %" PRId64 " bytes %" PRIu64 " ret %d"
+
+# block-copy.c
 block_copy_skip(void *job, int64_t start) "job %p start %"PRId64
 block_copy_skip_range(void *job, int64_t start, uint64_t bytes) "job %p start %"PRId64" bytes %"PRId64
 block_copy_process(void *job, int64_t start) "job %p start %"PRId64
Signature · 1 line
2.18.0
Reviewed-by on v9 5/13
#15 lore raw

Re: [Qemu-devel] [PATCH v9 05/13] block: move block_copy from block/backup.c to separate file

Quoted 7 lines · Vladimir Sementsov-Ogievskiy

Split block_copy to separate file, to be cleanly shared with backup-top filter driver in further commits.

It's a clean movement, the only change is drop "static" from interface functions.

Quoted 9 lines
---
 include/block/block-copy.h |  59 +++++++
 block/backup.c             | 313 +------------------------------------
 block/block-copy.c         | 307 ++++++++++++++++++++++++++++++++++++
 block/Makefile.objs        |   1 +
 block/trace-events         |   2 +
 5 files changed, 370 insertions(+), 312 deletions(-)
 create mode 100644 include/block/block-copy.h
 create mode 100644 block/block-copy.c

May change depending on changes to the preceding patches, but FWIW

#16 lore raw

Re: [Qemu-devel] [PATCH v9 05/13] block: move block_copy from block/backup.c to separate file

Quoted 18 lines · Max Reitz
Quoted 14 lines · Vladimir Sementsov-Ogievskiy

Split block_copy to separate file, to be cleanly shared with backup-top filter driver in further commits.

It's a clean movement, the only change is drop "static" from interface functions.

---
  include/block/block-copy.h |  59 +++++++
  block/backup.c             | 313 +------------------------------------
  block/block-copy.c         | 307 ++++++++++++++++++++++++++++++++++++
  block/Makefile.objs        |   1 +
  block/trace-events         |   2 +
  5 files changed, 370 insertions(+), 312 deletions(-)
Quoted 2 lines

create mode 100644 include/block/block-copy.h
create mode 100644 block/block-copy.c

Quoted 4 lines

May change depending on changes to the preceding patches, but FWIW

If you don't mind, I'll keep it until the patch doing simple nochange movement.

Signature · 2 lines
Best regards,
Vladimir
#17 lore raw

PATCH v9 6/13 block: teach bdrv_debug_breakpoint skip filters with backing

In reply to Vladimir Sementsov-Ogievskiy

Teach bdrv_debug_breakpoint and bdrv_debug_remove_breakpoint skip filters with backing. This is needed to implement and use in backup job it's own backup_top filter driver (like mirror already has one), and without this improvement, breakpoint removal will fail at least in 55 iotest.

---
 block.c | 34 ++++++++++++++++++++++++++--------
 1 file changed, 26 insertions(+), 8 deletions(-)
block.c +26 -8
index 5944124845..1c7c199849 100644
@@ -5164,14 +5164,35 @@ void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
     bs->drv->bdrv_debug_event(bs, event);
 }
 
-int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
-                          const char *tag)
+static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
 {
     while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
-        bs = bs->file ? bs->file->bs : NULL;
+        if (bs->file) {
+            bs = bs->file->bs;
+            continue;
+        }
+
+        if (bs->drv->is_filter && bs->backing) {
+            bs = bs->backing->bs;
+            continue;
+        }
+
+        break;
     }
 
     if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
+        assert(bs->drv->bdrv_debug_remove_breakpoint);
+        return bs;
+    }
+
+    return NULL;
+}
+
+int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
+                          const char *tag)
+{
+    bs = bdrv_find_debug_node(bs);
+    if (bs) {
         return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
     }
 
@@ -5180,11 +5201,8 @@ int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
 
 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
 {
-    while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
-        bs = bs->file ? bs->file->bs : NULL;
-    }
-
-    if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
+    bs = bdrv_find_debug_node(bs);
+    if (bs) {
         return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
     }
 
Signature · 1 line
2.18.0
Reviewed-by on v9 6/13
#18 lore raw

Re: [Qemu-devel] [PATCH v9 06/13] block: teach bdrv_debug_breakpoint skip filters with backing

Quoted 10 lines · Vladimir Sementsov-Ogievskiy

Teach bdrv_debug_breakpoint and bdrv_debug_remove_breakpoint skip filters with backing. This is needed to implement and use in backup job it's own backup_top filter driver (like mirror already has one), and without this improvement, breakpoint removal will fail at least in 55 iotest.

---
 block.c | 34 ++++++++++++++++++++++++++--------
 1 file changed, 26 insertions(+), 8 deletions(-)

FWIW

#19 lore raw

PATCH v9 7/13 iotests: prepare 124 and 257 bitmap querying for backup-top filter

In reply to Vladimir Sementsov-Ogievskiy

After backup-top filter appearing it's not possible to see dirty bitmaps in top node, so use node-name instead.

---
 tests/qemu-iotests/124        |   3 +-
 tests/qemu-iotests/257        |  39 +---
 tests/qemu-iotests/257.out    | 364 +++++++++++++---------------------
 tests/qemu-iotests/iotests.py |  22 ++
 4 files changed, 173 insertions(+), 255 deletions(-)
tests/qemu-iotests/124 +1 -2
index 3440f54781..8b6024769c 100755
@@ -749,8 +749,7 @@ class TestIncrementalBackupBlkdebug(TestIncrementalBackupBase):
 
         # Bitmap Status Check
         query = self.vm.qmp('query-block')
-        ret = [bmap for bmap in query['return'][0]['dirty-bitmaps']
-               if bmap.get('name') == bitmap.name][0]
+        ret = self.vm.get_bitmap(None, drive0['id'], bitmap.name)
         self.assert_qmp(ret, 'count', 458752)
         self.assert_qmp(ret, 'granularity', 65536)
         self.assert_qmp(ret, 'status', 'frozen')
tests/qemu-iotests/257 +10 -29
index c2a72c577a..8c3e96a7fc 100755
@@ -188,25 +188,6 @@ class Drive:
         self.size = size
         self.node = name
 
-def query_bitmaps(vm):
-    res = vm.qmp("query-block")
-    return {"bitmaps": {device['device'] or device['qdev']:
-                        device.get('dirty-bitmaps', []) for
-                        device in res['return']}}
-
-def get_bitmap(bitmaps, drivename, name, recording=None):
-    """
-    get a specific bitmap from the object returned by query_bitmaps.
-    :param recording: If specified, filter results by the specified value.
-    """
-    for bitmap in bitmaps['bitmaps'][drivename]:
-        if bitmap.get('name', '') == name:
-            if recording is None:
-                return bitmap
-            elif bitmap.get('recording') == recording:
-                return bitmap
-    return None
-
 def blockdev_backup(vm, device, target, sync, **kwargs):
     # Strip any arguments explicitly nulled by the caller:
     kwargs = {key: val for key, val in kwargs.items() if val is not None}
@@ -249,7 +230,7 @@ def perform_writes(drive, n):
             pattern.size)
         log(cmd)
         log(drive.vm.hmp_qemu_io(drive.name, cmd))
-    bitmaps = query_bitmaps(drive.vm)
+    bitmaps = drive.vm.query_bitmaps()
     log(bitmaps, indent=2)
     log('')
     return bitmaps
@@ -370,7 +351,7 @@ def test_bitmap_sync(bsync_mode, msync_mode='bitmap', failure=None):
         # 1 - Writes and Reference Backup
         bitmaps = perform_writes(drive0, 1)
         ebitmap.dirty_group(1)
-        bitmap = get_bitmap(bitmaps, drive0.device, 'bitmap0')
+        bitmap = vm.get_bitmap(bitmaps, drive0.node, 'bitmap0')
         ebitmap.compare(bitmap)
         reference_backup(drive0, 1, fbackup1)
 
@@ -388,11 +369,11 @@ def test_bitmap_sync(bsync_mode, msync_mode='bitmap', failure=None):
             log('')
             bitmaps = perform_writes(drive0, 2)
             # Named bitmap (static, should be unchanged)
-            ebitmap.compare(get_bitmap(bitmaps, drive0.device, 'bitmap0'))
+            ebitmap.compare(vm.get_bitmap(bitmaps, drive0.node, 'bitmap0'))
             # Anonymous bitmap (dynamic, shows new writes)
             anonymous = EmulatedBitmap()
             anonymous.dirty_group(2)
-            anonymous.compare(get_bitmap(bitmaps, drive0.device, '',
+            anonymous.compare(vm.get_bitmap(bitmaps, drive0.node, '',
                                          recording=True))
 
             # Simulate the order in which this will happen:
@@ -405,7 +386,7 @@ def test_bitmap_sync(bsync_mode, msync_mode='bitmap', failure=None):
         vm.run_job(job, auto_dismiss=True, auto_finalize=False,
                    pre_finalize=_callback,
                    cancel=(failure == 'simulated'))
-        bitmaps = query_bitmaps(vm)
+        bitmaps = vm.query_bitmaps()
         log(bitmaps, indent=2)
         log('')
 
@@ -423,29 +404,29 @@ def test_bitmap_sync(bsync_mode, msync_mode='bitmap', failure=None):
                 ebitmap.clear()
                 ebitmap.dirty_bits(range(fail_bit, SIZE // GRANULARITY))
 
-        ebitmap.compare(get_bitmap(bitmaps, drive0.device, 'bitmap0'))
+        ebitmap.compare(vm.get_bitmap(bitmaps, drive0.node, 'bitmap0'))
 
         # 2 - Writes and Reference Backup
         bitmaps = perform_writes(drive0, 3)
         ebitmap.dirty_group(3)
-        ebitmap.compare(get_bitmap(bitmaps, drive0.device, 'bitmap0'))
+        ebitmap.compare(vm.get_bitmap(bitmaps, drive0.node, 'bitmap0'))
         reference_backup(drive0, 2, fbackup2)
 
         # 2 - Bitmap Backup (In failure modes, this is a recovery.)
         job = backup(drive0, 2, bsync2, "bitmap",
                      bitmap="bitmap0", bitmap_mode=bsync_mode)
         vm.run_job(job, auto_dismiss=True, auto_finalize=False)
-        bitmaps = query_bitmaps(vm)
+        bitmaps = vm.query_bitmaps()
         log(bitmaps, indent=2)
         log('')
         if bsync_mode != 'never':
             ebitmap.clear()
-        ebitmap.compare(get_bitmap(bitmaps, drive0.device, 'bitmap0'))
+        ebitmap.compare(vm.get_bitmap(bitmaps, drive0.node, 'bitmap0'))
 
         log('--- Cleanup ---\n')
         vm.qmp_log("block-dirty-bitmap-remove",
                    node=drive0.name, name="bitmap0")
-        log(query_bitmaps(vm), indent=2)
+        log(vm.query_bitmaps(), indent=2)
         vm.shutdown()
         log('')
 
tests/qemu-iotests/257.out +140 -224

1,204 lines not on this page · raw message

tests/qemu-iotests/iotests.py +22 -0
index 84438e837c..9381964d9f 100644
@@ -643,6 +643,28 @@ class VM(qtest.QEMUQtestMachine):
                 return x
         return None
 
+    def query_bitmaps(self):
+        res = self.qmp("query-named-block-nodes")
+        return {"bitmaps": {device['node-name']: device['dirty-bitmaps']
+                                for device in res['return']
+                                    if 'dirty-bitmaps' in device}}
+
+    def get_bitmap(self, bitmaps, node_name, name, recording=None):
+        """
+        get a specific bitmap from the object returned by query_bitmaps.
+        :param recording: If specified, filter results by the specified value.
+        """
+        if bitmaps is None:
+            bitmaps = self.query_bitmaps()
+
+        for bitmap in bitmaps['bitmaps'][node_name]:
+            if bitmap.get('name', '') == name:
+                if recording is None:
+                    return bitmap
+                elif bitmap.get('recording') == recording:
+                    return bitmap
+        return None
+
 
 index_re = re.compile(r'([^\[]+)\[([^\]]+)\]')
 
Signature · 1 line
2.18.0
#20 lore raw

Re: [Qemu-devel] [PATCH v9 07/13] iotests: prepare 124 and 257 bitmap querying for backup-top filter

Quoted 16 lines · Vladimir Sementsov-Ogievskiy

After backup-top filter appearing it's not possible to see dirty bitmaps in top node, so use node-name instead.

---
 tests/qemu-iotests/124        |   3 +-
 tests/qemu-iotests/257        |  39 +---
 tests/qemu-iotests/257.out    | 364 +++++++++++++---------------------
 tests/qemu-iotests/iotests.py |  22 ++
 4 files changed, 173 insertions(+), 255 deletions(-)
tests/qemu-iotests/124 +0 -0
index 3440f54781..8b6024769c 100755

tests/qemu-iotests/124 class TestIncrementalBackupBlkdebug(TestIncrementalBackupBase): · line 755

Quoted 9 lines
tests/qemu-iotests/124 +1 -2
@@ -749,8 +749,7 @@ class TestIncrementalBackupBlkdebug(TestIncrementalBackupBase):
 
         # Bitmap Status Check
         query = self.vm.qmp('query-block')
-        ret = [bmap for bmap in query['return'][0]['dirty-bitmaps']
-               if bmap.get('name') == bitmap.name][0]
+        ret = self.vm.get_bitmap(None, drive0['id'], bitmap.name)
         self.assert_qmp(ret, 'count', 458752)
         self.assert_qmp(ret, 'granularity', 65536)
         self.assert_qmp(ret, 'status', 'frozen')

I see a couple more instances of querying the bitmaps through query-block here. Wouldn’t it make sense to replace them all with get_bitmap()?

[...]

tests/qemu-iotests/iotests.py class VM(qtest.QEMUQtestMachine): · line 650

Quoted 13 lines
tests/qemu-iotests/iotests.py +5 -0
index 84438e837c..9381964d9f 100644
@@ -643,6 +643,28 @@ class VM(qtest.QEMUQtestMachine):
                 return x
         return None
 
+    def query_bitmaps(self):
+        res = self.qmp("query-named-block-nodes")
+        return {"bitmaps": {device['node-name']: device['dirty-bitmaps']
+                                for device in res['return']
+                                    if 'dirty-bitmaps' in device}}

I’d leave the wrapping in {"bitmaps": x} to the callers, if they need it.

Quoted 5 lines
tests/qemu-iotests/iotests.py +5 -0
+
+    def get_bitmap(self, bitmaps, node_name, name, recording=None):
+        """
+        get a specific bitmap from the object returned by query_bitmaps.
+        :param recording: If specified, filter results by the specified value.

tests/qemu-iotests/iotests.py class VM(qtest.QEMUQtestMachine): · line 665

Quoted 10 lines
tests/qemu-iotests/iotests.py +10 -0
+        """
+        if bitmaps is None:
+            bitmaps = self.query_bitmaps()
+
+        for bitmap in bitmaps['bitmaps'][node_name]:
+            if bitmap.get('name', '') == name:
+                if recording is None:
+                    return bitmap
+                elif bitmap.get('recording') == recording:
+                    return bitmap

Maybe add a “break” or “return None” here?

(Yes, yes, you just moved existing code. Still.)

Max

Quoted 6 lines
tests/qemu-iotests/iotests.py +2 -0
+        return None
+
 
 index_re = re.compile(r'([^\[]+)\[([^\]]+)\]')
 
#21 lore raw

Re: [Qemu-devel] [PATCH v9 07/13] iotests: prepare 124 and 257 bitmap querying for backup-top filter

Quoted 67 lines · Max Reitz
Quoted 16 lines · Vladimir Sementsov-Ogievskiy

After backup-top filter appearing it's not possible to see dirty bitmaps in top node, so use node-name instead.

---
  tests/qemu-iotests/124        |   3 +-
  tests/qemu-iotests/257        |  39 +---
  tests/qemu-iotests/257.out    | 364 +++++++++++++---------------------
  tests/qemu-iotests/iotests.py |  22 ++
  4 files changed, 173 insertions(+), 255 deletions(-)
tests/qemu-iotests/124 +0 -0
index 3440f54781..8b6024769c 100755

tests/qemu-iotests/124 class TestIncrementalBackupBlkdebug(TestIncrementalBackupBase): · line 755

Quoted 9 lines
tests/qemu-iotests/124 +1 -2
@@ -749,8 +749,7 @@ class TestIncrementalBackupBlkdebug(TestIncrementalBackupBase):
 
          # Bitmap Status Check
          query = self.vm.qmp('query-block')
-        ret = [bmap for bmap in query['return'][0]['dirty-bitmaps']
-               if bmap.get('name') == bitmap.name][0]
+        ret = self.vm.get_bitmap(None, drive0['id'], bitmap.name)
          self.assert_qmp(ret, 'count', 458752)
          self.assert_qmp(ret, 'granularity', 65536)
          self.assert_qmp(ret, 'status', 'frozen')

I see a couple more instances of querying the bitmaps through query-block here. Wouldn’t it make sense to replace them all with get_bitmap()?

[...]

tests/qemu-iotests/iotests.py class VM(qtest.QEMUQtestMachine): · line 650

Quoted 13 lines
tests/qemu-iotests/iotests.py +5 -0
index 84438e837c..9381964d9f 100644
@@ -643,6 +643,28 @@ class VM(qtest.QEMUQtestMachine):
                  return x
          return None
 
+    def query_bitmaps(self):
+        res = self.qmp("query-named-block-nodes")
+        return {"bitmaps": {device['node-name']: device['dirty-bitmaps']
+                                for device in res['return']
+                                    if 'dirty-bitmaps' in device}}

I’d leave the wrapping in {"bitmaps": x} to the callers, if they need it.

Quoted 5 lines
tests/qemu-iotests/iotests.py +5 -0
+
+    def get_bitmap(self, bitmaps, node_name, name, recording=None):
+        """
+        get a specific bitmap from the object returned by query_bitmaps.
+        :param recording: If specified, filter results by the specified value.

tests/qemu-iotests/iotests.py class VM(qtest.QEMUQtestMachine): · line 665

Quoted 10 lines
tests/qemu-iotests/iotests.py +10 -0
+        """
+        if bitmaps is None:
+            bitmaps = self.query_bitmaps()
+
+        for bitmap in bitmaps['bitmaps'][node_name]:
+            if bitmap.get('name', '') == name:
+                if recording is None:
+                    return bitmap
+                elif bitmap.get('recording') == recording:
+                    return bitmap

Maybe add a “break” or “return None” here?

Quoted 2 lines

(Yes, yes, you just moved existing code. Still.)

No, as we may have several unnamed bitmaps, which should be selected by "recording"..

Quoted 9 lines
Quoted 7 lines
tests/qemu-iotests/iotests.py +2 -0
+        return None
+
 
  index_re = re.compile(r'([^\[]+)\[([^\]]+)\]')
 
Signature · 2 lines
Best regards,
Vladimir
#22 lore raw

Re: [Qemu-devel] [PATCH v9 07/13] iotests: prepare 124 and 257 bitmap querying for backup-top filter

Quoted 2 lines · Vladimir Sementsov-Ogievskiy
Quoted 1 line · Max Reitz

On 26.08.19 18:13, Vladimir Sementsov-Ogievskiy wrote:

[...]

Quoted 18 lines
Quoted 18 lines
Quoted 5 lines
tests/qemu-iotests/iotests.py +5 -0
+
+    def get_bitmap(self, bitmaps, node_name, name, recording=None):
+        """
+        get a specific bitmap from the object returned by query_bitmaps.
+        :param recording: If specified, filter results by the specified value.

tests/qemu-iotests/iotests.py class VM(qtest.QEMUQtestMachine): · line 665

Quoted 10 lines
tests/qemu-iotests/iotests.py +10 -0
+        """
+        if bitmaps is None:
+            bitmaps = self.query_bitmaps()
+
+        for bitmap in bitmaps['bitmaps'][node_name]:
+            if bitmap.get('name', '') == name:
+                if recording is None:
+                    return bitmap
+                elif bitmap.get('recording') == recording:
+                    return bitmap

Maybe add a “break” or “return None” here?

Quoted 4 lines
Quoted 2 lines

(Yes, yes, you just moved existing code. Still.)

No, as we may have several unnamed bitmaps, which should be selected by "recording"..

Ah. OK. Now it all makes sense...

Max

#23 lore raw

PATCH v9 8/13 iotests: 257: drop unused Drive.device field

In reply to Vladimir Sementsov-Ogievskiy

After previous commit Drive.device is actually unused. Drop it together with .name property. While being here reuse .node in qmp commands instead of writing 'drive0' twice.

---
 tests/qemu-iotests/257 | 37 +++++++++++++++----------------------
 1 file changed, 15 insertions(+), 22 deletions(-)
tests/qemu-iotests/257 +15 -22
index 8c3e96a7fc..e0f521272b 100755
@@ -148,11 +148,6 @@ class Drive:
         self.fmt = None
         self.size = None
         self.node = None
-        self.device = None
-
-    @property
-    def name(self):
-        return self.node or self.device
 
     def img_create(self, fmt, size):
         self.fmt = fmt
@@ -201,7 +196,7 @@ def blockdev_backup(vm, device, target, sync, **kwargs):
 def blockdev_backup_mktarget(drive, target_id, filepath, sync, **kwargs):
     target_drive = Drive(filepath, vm=drive.vm)
     target_drive.create_target(target_id, drive.fmt, drive.size)
-    blockdev_backup(drive.vm, drive.name, target_id, sync, **kwargs)
+    blockdev_backup(drive.vm, drive.node, target_id, sync, **kwargs)
 
 def reference_backup(drive, n, filepath):
     log("--- Reference Backup #{:d} ---\n".format(n))
@@ -229,7 +224,7 @@ def perform_writes(drive, n):
             pattern.offset,
             pattern.size)
         log(cmd)
-        log(drive.vm.hmp_qemu_io(drive.name, cmd))
+        log(drive.vm.hmp_qemu_io(drive.node, cmd))
     bitmaps = drive.vm.query_bitmaps()
     log(bitmaps, indent=2)
     log('')
@@ -324,18 +319,17 @@ def test_bitmap_sync(bsync_mode, msync_mode='bitmap', failure=None):
                 }]
             }
 
+        drive0.node = 'drive0'
         vm.qmp_log('blockdev-add',
                    filters=[iotests.filter_qmp_testfiles],
-                   node_name="drive0",
+                   node_name=drive0.node,
                    driver=drive0.fmt,
                    file=file_config)
-        drive0.node = 'drive0'
-        drive0.device = 'device0'
         # Use share-rw to allow writes directly to the node;
         # The anonymous block-backend for this configuration prevents us
         # from using HMP's qemu-io commands to address the device.
-        vm.qmp_log("device_add", id=drive0.device,
-                   drive=drive0.name, driver="scsi-hd",
+        vm.qmp_log("device_add", id='device0',
+                   drive=drive0.node, driver="scsi-hd",
                    share_rw=True)
         log('')
 
@@ -343,7 +337,7 @@ def test_bitmap_sync(bsync_mode, msync_mode='bitmap', failure=None):
         perform_writes(drive0, 0)
         reference_backup(drive0, 0, fbackup0)
         log('--- Add Bitmap ---\n')
-        vm.qmp_log("block-dirty-bitmap-add", node=drive0.name,
+        vm.qmp_log("block-dirty-bitmap-add", node=drive0.node,
                    name="bitmap0", granularity=GRANULARITY)
         log('')
         ebitmap = EmulatedBitmap()
@@ -358,7 +352,7 @@ def test_bitmap_sync(bsync_mode, msync_mode='bitmap', failure=None):
         # 1 - Test Backup (w/ Optional induced failure)
         if failure == 'intermediate':
             # Activate blkdebug induced failure for second-to-next read
-            log(vm.hmp_qemu_io(drive0.name, 'flush'))
+            log(vm.hmp_qemu_io(drive0.node, 'flush'))
             log('')
         job = backup(drive0, 1, bsync1, msync_mode,
                      bitmap="bitmap0", bitmap_mode=bsync_mode)
@@ -425,7 +419,7 @@ def test_bitmap_sync(bsync_mode, msync_mode='bitmap', failure=None):
 
         log('--- Cleanup ---\n')
         vm.qmp_log("block-dirty-bitmap-remove",
-                   node=drive0.name, name="bitmap0")
+                   node=drive0.node, name="bitmap0")
         log(vm.query_bitmaps(), indent=2)
         vm.shutdown()
         log('')
@@ -465,22 +459,21 @@ def test_backup_api():
             'filename': drive0.path
         }
 
+        drive0.node = 'drive0'
         vm.qmp_log('blockdev-add',
                    filters=[iotests.filter_qmp_testfiles],
-                   node_name="drive0",
+                   node_name=drive0.node,
                    driver=drive0.fmt,
                    file=file_config)
-        drive0.node = 'drive0'
-        drive0.device = 'device0'
-        vm.qmp_log("device_add", id=drive0.device,
-                   drive=drive0.name, driver="scsi-hd")
+        vm.qmp_log("device_add", id='device0',
+                   drive=drive0.node, driver="scsi-hd")
         log('')
 
         target0 = Drive(backup_path, vm=vm)
         target0.create_target("backup_target", drive0.fmt, drive0.size)
         log('')
 
-        vm.qmp_log("block-dirty-bitmap-add", node=drive0.name,
+        vm.qmp_log("block-dirty-bitmap-add", node=drive0.node,
                    name="bitmap0", granularity=GRANULARITY)
         log('')
 
@@ -519,7 +512,7 @@ def test_backup_api():
             log("-- Sync mode {:s} tests --\n".format(sync_mode))
             for bitmap in (None, 'bitmap404', 'bitmap0'):
                 for policy in error_cases[sync_mode][bitmap]:
-                    blockdev_backup(drive0.vm, drive0.name, "backup_target",
+                    blockdev_backup(drive0.vm, drive0.node, "backup_target",
                                     sync_mode, job_id='api_job',
                                     bitmap=bitmap, bitmap_mode=policy)
                     log('')
Signature · 1 line
2.18.0
Reviewed-by on v9 8/13
#24 lore raw

Re: [Qemu-devel] [PATCH v9 08/13] iotests: 257: drop unused Drive.device field

Quoted 8 lines · Vladimir Sementsov-Ogievskiy

After previous commit Drive.device is actually unused. Drop it together with .name property. While being here reuse .node in qmp commands instead of writing 'drive0' twice.

---
 tests/qemu-iotests/257 | 37 +++++++++++++++----------------------
 1 file changed, 15 insertions(+), 22 deletions(-)
#25 lore raw

PATCH v9 9/13 iotests: 257: drop device_add

In reply to Vladimir Sementsov-Ogievskiy

SCSI devices are unused in test, drop them.

---
 tests/qemu-iotests/257     |  8 -------
 tests/qemu-iotests/257.out | 44 --------------------------------------
 2 files changed, 52 deletions(-)
tests/qemu-iotests/257 +0 -8
index e0f521272b..7bc6069e1b 100755
@@ -325,12 +325,6 @@ def test_bitmap_sync(bsync_mode, msync_mode='bitmap', failure=None):
                    node_name=drive0.node,
                    driver=drive0.fmt,
                    file=file_config)
-        # Use share-rw to allow writes directly to the node;
-        # The anonymous block-backend for this configuration prevents us
-        # from using HMP's qemu-io commands to address the device.
-        vm.qmp_log("device_add", id='device0',
-                   drive=drive0.node, driver="scsi-hd",
-                   share_rw=True)
         log('')
 
         # 0 - Writes and Reference Backup
@@ -465,8 +459,6 @@ def test_backup_api():
                    node_name=drive0.node,
                    driver=drive0.fmt,
                    file=file_config)
-        vm.qmp_log("device_add", id='device0',
-                   drive=drive0.node, driver="scsi-hd")
         log('')
 
         target0 = Drive(backup_path, vm=vm)
tests/qemu-iotests/257.out +0 -44

176 lines not on this page · raw message

Signature · 1 line
2.18.0
#27 lore raw

PATCH v9 10/13 block/io: refactor wait_serialising_requests

In reply to Vladimir Sementsov-Ogievskiy

Split out do_wait_serialising_requests with additional possibility to not actually wait but just check, that there is something to wait for.

---
 block/io.c | 24 ++++++++++++++++--------
 1 file changed, 16 insertions(+), 8 deletions(-)
block/io.c +16 -8
index 56bbf195bb..378180b274 100644
@@ -785,12 +785,13 @@ void bdrv_dec_in_flight(BlockDriverState *bs)
     bdrv_wakeup(bs);
 }
 
-static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
+static bool coroutine_fn do_wait_serialising_requests(BdrvTrackedRequest *self,
+                                                      bool wait)
 {
     BlockDriverState *bs = self->bs;
     BdrvTrackedRequest *req;
     bool retry;
-    bool waited = false;
+    bool found = false;
 
     if (!atomic_read(&bs->serialising_in_flight)) {
         return false;
@@ -816,11 +817,13 @@ static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
                  * will wait for us as soon as it wakes up, then just go on
                  * (instead of producing a deadlock in the former case). */
                 if (!req->waiting_for) {
-                    self->waiting_for = req;
-                    qemu_co_queue_wait(&req->wait_queue, &bs->reqs_lock);
-                    self->waiting_for = NULL;
-                    retry = true;
-                    waited = true;
+                    found = true;
+                    if (wait) {
+                        self->waiting_for = req;
+                        qemu_co_queue_wait(&req->wait_queue, &bs->reqs_lock);
+                        self->waiting_for = NULL;
+                        retry = true;
+                    }
                     break;
                 }
             }
@@ -828,7 +831,12 @@ static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
         qemu_co_mutex_unlock(&bs->reqs_lock);
     } while (retry);
 
-    return waited;
+    return found;
+}
+
+static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
+{
+    return do_wait_serialising_requests(self, true);
 }
 
 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
Signature · 1 line
2.18.0
#28 lore raw

PATCH v9 11/13 block: add lock/unlock range functions

In reply to Vladimir Sementsov-Ogievskiy

From: Vladimir Sementsov-Ogievskiy <[email protected]>

Introduce lock/unlock range functionality, based on serialized requests. This is needed to refactor backup, dropping local tracked-request-like synchronization.

---
 include/block/block_int.h |  4 ++++
 block/io.c                | 44 ++++++++++++++++++++++++++++++++++++++-
 2 files changed, 47 insertions(+), 1 deletion(-)
include/block/block_int.h +4 -0
index ceec8c2f56..0307cc78c6 100644
@@ -65,6 +65,7 @@ enum BdrvTrackedRequestType {
     BDRV_TRACKED_WRITE,
     BDRV_TRACKED_DISCARD,
     BDRV_TRACKED_TRUNCATE,
+    BDRV_TRACKED_LOCK,
 };
 
 typedef struct BdrvTrackedRequest {
@@ -947,6 +948,9 @@ int coroutine_fn bdrv_co_preadv(BdrvChild *child,
 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
     BdrvRequestFlags flags);
+void *coroutine_fn bdrv_co_try_lock(BlockDriverState *bs,
+                                    int64_t offset, unsigned int bytes);
+void coroutine_fn bdrv_co_unlock(void *opaque);
 
 static inline int coroutine_fn bdrv_co_pread(BdrvChild *child,
     int64_t offset, unsigned int bytes, void *buf, BdrvRequestFlags flags)
block/io.c +43 -1

65 lines not on this page · raw message

Signature · 1 line
2.18.0
Reviewed-by on v9 11/13
#29 lore raw

Re: [Qemu-devel] [PATCH v9 11/13] block: add lock/unlock range functions

Quoted 1 line · Vladimir Sementsov-Ogievskiy

From: Vladimir Sementsov-Ogievskiy <[email protected]>

Hm. :-)

Do you want to fix that?

Quoted 9 lines

Introduce lock/unlock range functionality, based on serialized requests. This is needed to refactor backup, dropping local tracked-request-like synchronization.

---
 include/block/block_int.h |  4 ++++
 block/io.c                | 44 ++++++++++++++++++++++++++++++++++++++-
 2 files changed, 47 insertions(+), 1 deletion(-)

Apart from that, I can’t see any changes from v8, so:

#30 lore raw

Re: [Qemu-devel] [PATCH v9 11/13] block: add lock/unlock range functions

Quoted 6 lines · Max Reitz
Quoted 1 line · Vladimir Sementsov-Ogievskiy

From: Vladimir Sementsov-Ogievskiy <[email protected]>

Hm. :-)

Do you want to fix that?

Yes.. Hmm seems like I was working from home at some moment

Quoted 15 lines
Quoted 10 lines

Introduce lock/unlock range functionality, based on serialized requests. This is needed to refactor backup, dropping local tracked-request-like synchronization.

---
  include/block/block_int.h |  4 ++++
  block/io.c                | 44 ++++++++++++++++++++++++++++++++++++++-
  2 files changed, 47 insertions(+), 1 deletion(-)

Apart from that, I can’t see any changes from v8, so:

Signature · 2 lines
Best regards,
Vladimir
#31 lore raw

PATCH v9 12/13 block: introduce backup-top filter driver

In reply to Vladimir Sementsov-Ogievskiy

Backup-top filter caches write operations and does copy-before-write operations.

The driver will be used in backup instead of write-notifiers.

---
 block/backup-top.h  |  37 +++++++
 block/backup-top.c  | 245 ++++++++++++++++++++++++++++++++++++++++++++
 block/Makefile.objs |   2 +
 3 files changed, 284 insertions(+)
 create mode 100644 block/backup-top.h
 create mode 100644 block/backup-top.c
block/backup-top.h new file +37 -0
new file mode 100644
index 0000000000..67de7a9133
@@ -0,0 +1,37 @@
+/*
+ * backup-top filter driver
+ *
+ * The driver performs Copy-Before-Write (CBW) operation: it is injected above
+ * some node, and before each write it copies _old_ data to the target node.
+ *
+ * Copyright (c) 2018-2019 Virtuozzo International GmbH.
+ *
+ * Author:
+ *  Sementsov-Ogievskiy Vladimir <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef BACKUP_TOP_H
+#define BACKUP_TOP_H
+
+#include "block/block_int.h"
+
+BlockDriverState *bdrv_backup_top_append(BlockDriverState *source,
+                                         const char *filter_node_name,
+                                         Error **errp);
+void bdrv_backup_top_set_bcs(BlockDriverState *bs, BlockCopyState *copy_state);
+void bdrv_backup_top_drop(BlockDriverState *bs);
+
+#endif /* BACKUP_TOP_H */
block/backup-top.c new file +245 -0

245 lines not on this page · raw message

block/Makefile.objs +2 -0
index 0b5c635fb2..6f348c56c9 100644
@@ -41,6 +41,8 @@ block-obj-y += block-copy.o
 
 block-obj-y += crypto.o
 
+block-obj-y += backup-top.o
+
 common-obj-y += stream.o
 
 nfs.o-libs         := $(LIBNFS_LIBS)
Signature · 1 line
2.18.0
#32 lore raw

Re: [Qemu-devel] [PATCH v9 12/13] block: introduce backup-top filter driver

Quoted 13 lines · Vladimir Sementsov-Ogievskiy

Backup-top filter caches write operations and does copy-before-write operations.

The driver will be used in backup instead of write-notifiers.

---
 block/backup-top.h  |  37 +++++++
 block/backup-top.c  | 245 ++++++++++++++++++++++++++++++++++++++++++++
 block/Makefile.objs |   2 +
 3 files changed, 284 insertions(+)
 create mode 100644 block/backup-top.h
 create mode 100644 block/backup-top.c

[...]

block/backup-top.c line 118

Quoted 10 lines
block/backup-top.c new file +10 -0
+static void backup_top_refresh_filename(BlockDriverState *bs)
+{
+    if (bs->backing == NULL) {
+        /*
+         * we can be here after failed bdrv_attach_child in
+         * bdrv_set_backing_hd
+         */
+        return;
+    }
+    bdrv_refresh_filename(bs->backing->bs);

bdrv_refresh_filename() should have done this already.

Quoted 3 lines
block/backup-top.c new file +3 -0
+    pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
+            bs->backing->bs->filename);
+}

[...]

Quoted 4 lines
block/backup-top.c new file +4 -0
+BlockDriverState *bdrv_backup_top_append(BlockDriverState *source,
+                                         const char *filter_node_name,
+                                         Error **errp)
+{

[...]

Quoted 1 line

+ assert(!(top->backing->shared_perm & BLK_PERM_WRITE));

Not wrong, but why?

Quoted 11 lines
block/backup-top.c new file +11 -0
+
+    return top;
+}
+
+void bdrv_backup_top_set_bcs(BlockDriverState *bs, BlockCopyState *copy_state)
+{
+    BDRVBackupTopState *s = bs->opaque;
+
+    assert(blk_bs(copy_state->source) == bs->backing->bs);
+    s->bcs = copy_state;
+}

block/backup-top.c line 236

Quoted 10 lines
block/backup-top.c new file +10 -0
+
+void bdrv_backup_top_drop(BlockDriverState *bs)
+{
+    AioContext *aio_context = bdrv_get_aio_context(bs);
+
+    aio_context_acquire(aio_context);
+
+    bdrv_drained_begin(bs);
+
+    bdrv_child_try_set_perm(bs->backing, 0, BLK_PERM_ALL, &error_abort);

I would prefer a state->active = false and bdrv_child_refresh_perms().

Max

#33 lore raw

PATCH v9 13/13 block/backup: use backup-top instead of write notifiers

In reply to Vladimir Sementsov-Ogievskiy

Drop write notifiers and use filter node instead.

= Changes =

1. add filter-node-name argument for backup qmp api. We have to do it in this commit, as 257 needs to be fixed.

2. there no move write notifiers here, so is_write_notifier parameter is dropped from block-copy paths.

3. Intersecting requests handling changed, now synchronization between backup-top, backup and guest writes are all done in block/block-copy.c and works as follows:

On copy operation, we work only with dirty areas. If bits are dirty it means that there are no requests intersecting with this area. We clear dirty bits and take bdrv range lock (bdrv_co_try_lock) on this area to prevent further operations from interaction with guest (only with guest, as neither backup nor backup-top will touch non-dirty area). If copy-operation failed we set dirty bits back together with releasing the lock.

The actual difference with old scheme is that on guest writes we don't lock the whole region but only dirty-parts, and to be more
precise: only dirty-part we are currently operate on. In old scheme guest write to non-dirty area (which may be safely ignored by backup) may wait for intersecting request, touching some other area which is dirty.

4. To sync with in-flight requests at job finish we now have drained removing of the filter, we don't need rw-lock.

= Notes =

Note the consequence of three objects appearing: backup-top, backup job and block-copy-state:

1. We want to insert backup-top before job creation, to behave similar with mirror and commit, where job is started upon filter.

2. We also have to create block-copy-state after filter injection, as we don't want it's source child be replaced by fitler. Instead we want to keep BCS.source to be real source node, as we want to use bdrv_co_try_lock in CBW operations and it can't be used on filter, as on filter we already have in-flight (write) request from upper layer.

So, we firstly create inject backup-top, then create job and BCS. BCS is the latest just to not create extra variable for it. Finally we set bcs for backup-top filter.

= Iotest changes =

56: op-blocker doesn't shot now, as we set it on source, but then check on filter, when trying to start second backup, so error caught in test_dismiss_collision is changed. It's OK anyway, as this test-case seems to test that after some collision we can dismiss first job and successfully start the second one. So, the change is that collision is changed from op-blocker to file-posix locks.
However, it's obvious now that we'd better drop this op-blocker at all and add a test-case for two backups from one node (to different destinations) actually works. But not in these series.

257: The test wants to emulate guest write during backup. They should go to filter node, not to original source node, of course. Therefore we need to specify filter node name and use it.

---
 qapi/block-core.json       |   8 +-
 include/block/block-copy.h |   2 +-
 include/block/block_int.h  |   1 +
 block/backup-top.c         |  14 +-
 block/backup.c             | 113 +++-----------
 block/block-copy.c         |  55 ++++---
 block/replication.c        |   2 +-
 blockdev.c                 |   1 +
 tests/qemu-iotests/056     |   2 +-
 tests/qemu-iotests/257     |   7 +-
 tests/qemu-iotests/257.out | 306 ++++++++++++++++++-------------------
 11 files changed, 230 insertions(+), 281 deletions(-)
qapi/block-core.json +7 -1
index e6edd641f1..b5cd00c361 100644
@@ -1391,6 +1391,11 @@
 #                list without user intervention.
 #                Defaults to true. (Since 2.12)
 #
+# @filter-node-name: the node name that should be assigned to the
+#                    filter driver that the backup job inserts into the graph
+#                    above node specified by @drive. If this option is not given,
+#                    a node name is autogenerated. (Since: 4.2)
+#
 # Note: @on-source-error and @on-target-error only affect background
 # I/O.  If an error occurs during a guest write request, the device's
 # rerror/werror actions will be used.
@@ -1404,7 +1409,8 @@
             '*compress': 'bool',
             '*on-source-error': 'BlockdevOnError',
             '*on-target-error': 'BlockdevOnError',
-            '*auto-finalize': 'bool', '*auto-dismiss': 'bool' } }
+            '*auto-finalize': 'bool', '*auto-dismiss': 'bool',
+            '*filter-node-name': 'str' } }
 
 ##
 # @DriveBackup:
include/block/block-copy.h +1 -1
index 0dd7a3f7bf..c15d47a70d 100644
@@ -54,6 +54,6 @@ int64_t block_copy_reset_unallocated(BlockCopyState *s,
                                      int64_t offset, int64_t *count);
 
 int coroutine_fn block_copy(BlockCopyState *s, int64_t offset, uint64_t bytes,
-                            bool *error_is_read, bool is_write_notifier);
+                            bool *error_is_read);
 
 #endif /* BLOCK_COPY_H */
include/block/block_int.h +1 -0
index 0307cc78c6..27c4bf4673 100644
@@ -1179,6 +1179,7 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
                             BdrvDirtyBitmap *sync_bitmap,
                             BitmapSyncMode bitmap_mode,
                             bool compress,
+                            const char *filter_node_name,
                             BlockdevOnError on_source_error,
                             BlockdevOnError on_target_error,
                             int creation_flags,
block/backup-top.c +5 -9
index 41cb015452..cac46f05b1 100644
@@ -49,15 +49,11 @@ static coroutine_fn int backup_top_co_preadv(
 static coroutine_fn int backup_top_cbw(BlockDriverState *bs, uint64_t offset,
                                        uint64_t bytes)
 {
-    /*
-     * Here we'd like to use block_copy(), but it needs some additional
-     * synchronization mechanism to prevent intersecting guest writes during
-     * copy operation. The will appear in further commit (it should be done
-     * together with moving backup to using of backup-top and to the same
-     * synchronization mechanism), and for now it is a TODO.
-     */
-
-    abort();
+    BDRVBackupTopState *s = bs->opaque;
+    uint64_t end = QEMU_ALIGN_UP(offset + bytes, s->bcs->cluster_size);
+    uint64_t off = QEMU_ALIGN_DOWN(offset, s->bcs->cluster_size);
+
+    return block_copy(s->bcs, off, end - off, NULL);
 }
 
 static int coroutine_fn backup_top_co_pdiscard(BlockDriverState *bs,
block/backup.c +22 -91

222 lines not on this page · raw message

block/block-copy.c +34 -21

126 lines not on this page · raw message

block/replication.c +1 -1
index 936b2f8b5a..99532ce521 100644
@@ -543,7 +543,7 @@ static void replication_start(ReplicationState *rs, ReplicationMode mode,
 
         s->backup_job = backup_job_create(
                                 NULL, s->secondary_disk->bs, s->hidden_disk->bs,
-                                0, MIRROR_SYNC_MODE_NONE, NULL, 0, false,
+                                0, MIRROR_SYNC_MODE_NONE, NULL, 0, false, NULL,
                                 BLOCKDEV_ON_ERROR_REPORT,
                                 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
                                 backup_job_completed, bs, NULL, &local_err);
blockdev.c +1 -0
index fbef6845c8..f89e48fc79 100644
@@ -3601,6 +3601,7 @@ static BlockJob *do_backup_common(BackupCommon *backup,
     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
                             backup->sync, bmap, backup->bitmap_mode,
                             backup->compress,
+                            backup->filter_node_name,
                             backup->on_source_error,
                             backup->on_target_error,
                             job_flags, NULL, NULL, txn, errp);
tests/qemu-iotests/056 +1 -1
index e761e465ae..9c7c653320 100755
@@ -253,7 +253,7 @@ class BackupTest(iotests.QMPTestCase):
         res = self.vm.qmp('query-block-jobs')
         self.assert_qmp(res, 'return[0]/status', 'concluded')
         # Leave zombie job un-dismissed, observe a failure:
-        res = self.qmp_backup_and_wait(serror="Node 'drive0' is busy: block device is in use by block job: backup",
+        res = self.qmp_backup_and_wait(serror='Failed to get "write" lock',
                                        device='drive0', format=iotests.imgfmt,
                                        sync='full', target=self.dest_img,
                                        auto_dismiss=False)
tests/qemu-iotests/257 +4 -3
index 7bc6069e1b..2dbcc6e7e1 100755
@@ -190,6 +190,7 @@ def blockdev_backup(vm, device, target, sync, **kwargs):
                         device=device,
                         target=target,
                         sync=sync,
+                        filter_node_name='backup-top',
                         **kwargs)
     return result
 
@@ -216,7 +217,7 @@ def backup(drive, n, filepath, sync, **kwargs):
                              job_id=job_id, **kwargs)
     return job_id
 
-def perform_writes(drive, n):
+def perform_writes(drive, n, filter_node_name=None):
     log("--- Write #{:d} ---\n".format(n))
     for pattern in GROUPS[n].patterns:
         cmd = "write -P{:s} 0x{:07x} 0x{:x}".format(
@@ -224,7 +225,7 @@ def perform_writes(drive, n):
             pattern.offset,
             pattern.size)
         log(cmd)
-        log(drive.vm.hmp_qemu_io(drive.node, cmd))
+        log(drive.vm.hmp_qemu_io(filter_node_name or drive.node, cmd))
     bitmaps = drive.vm.query_bitmaps()
     log(bitmaps, indent=2)
     log('')
@@ -355,7 +356,7 @@ def test_bitmap_sync(bsync_mode, msync_mode='bitmap', failure=None):
             """Issue writes while the job is open to test bitmap divergence."""
             # Note: when `failure` is 'intermediate', this isn't called.
             log('')
-            bitmaps = perform_writes(drive0, 2)
+            bitmaps = perform_writes(drive0, 2, filter_node_name='backup-top')
             # Named bitmap (static, should be unchanged)
             ebitmap.compare(vm.get_bitmap(bitmaps, drive0.node, 'bitmap0'))
             # Anonymous bitmap (dynamic, shows new writes)
tests/qemu-iotests/257.out +153 -153

1,043 lines not on this page · raw message

Signature · 1 line
2.18.0
#34 lore raw

Re: [Qemu-devel] [PATCH v9 13/13] block/backup: use backup-top instead of write notifiers

Quoted 6 lines · Vladimir Sementsov-Ogievskiy

Drop write notifiers and use filter node instead.

= Changes =

1. add filter-node-name argument for backup qmp api. We have to do it in this commit, as 257 needs to be fixed.

I feel a bit bad about it not being an implicit node. But I know you don’t like them, they’re probably just broken altogether and because libvirt doesn’t use backup (yet), probably nobody cares.

Quoted 1 line

2. there no move write notifiers here, so is_write_notifier parameter

s/there/there are/, I suppose?

Quoted 26 lines

is dropped from block-copy paths.

3. Intersecting requests handling changed, now synchronization between backup-top, backup and guest writes are all done in block/block-copy.c and works as follows:

On copy operation, we work only with dirty areas. If bits are dirty it means that there are no requests intersecting with this area. We clear dirty bits and take bdrv range lock (bdrv_co_try_lock) on this area to prevent further operations from interaction with guest (only with guest, as neither backup nor backup-top will touch non-dirty area). If copy-operation failed we set dirty bits back together with releasing the lock.

The actual difference with old scheme is that on guest writes we don't lock the whole region but only dirty-parts, and to be more
precise: only dirty-part we are currently operate on. In old scheme guest write to non-dirty area (which may be safely ignored by backup) may wait for intersecting request, touching some other area which is dirty.

4. To sync with in-flight requests at job finish we now have drained removing of the filter, we don't need rw-lock.

= Notes =

Quoted 8 lines

Note the consequence of three objects appearing: backup-top, backup job and block-copy-state:

1. We want to insert backup-top before job creation, to behave similar with mirror and commit, where job is started upon filter.

2. We also have to create block-copy-state after filter injection, as we don't want it's source child be replaced by fitler. Instead we want

s/it's/its/, s/filter/filter/ (although “fitler” does have an amusing ring to it)

Quoted 3 lines

to keep BCS.source to be real source node, as we want to use bdrv_co_try_lock in CBW operations and it can't be used on filter, as on filter we already have in-flight (write) request from upper layer.

Reasonable, even more so as I suppose BCS.source should eventually be a BdrvChild of backup-top.

What looks wrong is that the sync_bitmap is created on the source (“bs” in backup_job_create()), but backup_cleanup_sync_bitmap() still assumes it is on blk_bs(job->common.blk) (which is no longer true).

Quoted 7 lines

So, we firstly create inject backup-top, then create job and BCS. BCS is the latest just to not create extra variable for it. Finally we set bcs for backup-top filter.

= Iotest changes =

56: op-blocker doesn't shot now, as we set it on source, but then check

s/shot/show/?

Quoted 5 lines

on filter, when trying to start second backup, so error caught in test_dismiss_collision is changed. It's OK anyway, as this test-case seems to test that after some collision we can dismiss first job and successfully start the second one. So, the change is that collision is changed from op-blocker to file-posix locks.

Well, but the op blocker belongs to the source, which should be covered by internal permissions. The fact that it now shows up as a file-posix error thus shows that the conflict also moves from the source to the target. It’s OK because we actually don’t have a conflict on the source.

But I wonder whether it would be better for test_dismiss_collision() to do a blockdev-backup instead where we can see the collision on the target.

Hm. On second thought, why do we even get a conflict on the target? block-copy does share the WRITE permission for it...

Quoted 9 lines

However, it's obvious now that we'd better drop this op-blocker at all and add a test-case for two backups from one node (to different destinations) actually works. But not in these series.

257: The test wants to emulate guest write during backup. They should go to filter node, not to original source node, of course. Therefore we need to specify filter node name and use it.

Quoted 13 lines
---
 qapi/block-core.json       |   8 +-
 include/block/block-copy.h |   2 +-
 include/block/block_int.h  |   1 +
 block/backup-top.c         |  14 +-
 block/backup.c             | 113 +++-----------
 block/block-copy.c         |  55 ++++---
 block/replication.c        |   2 +-
 blockdev.c                 |   1 +
 tests/qemu-iotests/056     |   2 +-
 tests/qemu-iotests/257     |   7 +-
 tests/qemu-iotests/257.out | 306 ++++++++++++++++++-------------------
 11 files changed, 230 insertions(+), 281 deletions(-)

Nice.

I don’t have much to object (for some reason, I feel a bit uneasy about dropping all the request waiting code, but I suppose that is indeed taken care of with the range locks now).

[...]

Quoted 4 lines
block/backup.c +0 -0
index d927c63e5a..259a165405 100644

[...]

block/backup.c BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs, · line 485

Quoted 7 lines
block/backup.c +3 -0
@@ -552,6 +480,9 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
         backup_clean(&job->common.job);
         job_early_fail(&job->common.job);
     }
+    if (backup_top) {
+        bdrv_backup_top_drop(backup_top);
+    }

This order is weird because backup-top still has a pointer to the BCS, but we have already freed it at this point.

Quoted 14 lines
block/backup.c +0 -0
     return NULL;
 }
block/block-copy.c +1 -1
index 6828c46ba0..f3102ec3ff 100644
@@ -98,27 +98,32 @@ fail:
  * error occurred, return a negative error number
  */
 static int coroutine_fn block_copy_with_bounce_buffer(
-        BlockCopyState *s, int64_t start, int64_t end, bool is_write_notifier,
+        BlockCopyState *s, int64_t start, int64_t end,
         bool *error_is_read, void **bounce_buffer)

block/block-copy.c fail: · line 110

Quoted 10 lines
block/block-copy.c +6 -2
 {
     int ret;
-    int nbytes;
-    int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
+    int nbytes = MIN(s->cluster_size, s->len - start);
+    void *lock = bdrv_co_try_lock(blk_bs(s->source), start, nbytes);
+
+    /*
+     * Function must be called only on full-dirty region, so nobody touching or
+     * touched these bytes. Therefore, we must successfully get lock.

Which is OK because backup-top does not share the WRITE permission. But it is organized a bit weirdly, because it’s actually block-copy here that relies on the WRITE permission not to be shared; which it itself cannot do because backup-top needs it.

This of course results from the fact that backup-top and block-copy should really use the same object to access the source, namely the backing BdrvChild of backup-top.

Maybe there should be a comment somewhere that points this out?

Quoted 2 lines
block/block-copy.c +2 -0
+     */
+    assert(lock);

[...]

Quoted 10 lines
block/block-copy.c +1 -3
@@ -128,13 +133,16 @@ static int coroutine_fn block_copy_with_bounce_buffer(
         if (error_is_read) {
             *error_is_read = false;
         }
-        goto fail;
+        goto out;
     }
 
-    return nbytes;
-fail:

block/block-copy.c static int coroutine_fn block_copy_with_bounce_buffer( · line 146

Quoted 10 lines
block/block-copy.c +7 -2
-    bdrv_set_dirty_bitmap(s->copy_bitmap, start, s->cluster_size);
-    return ret;
+out:
+    if (ret < 0) {
+        bdrv_set_dirty_bitmap(s->copy_bitmap, start, s->cluster_size);
+    }
+    bdrv_co_unlock(lock);
+
+    return ret < 0 ? ret : nbytes;
 

I feel like it would still be simpler to keep the separate fail path and just duplicate the bdrv_co_unlock(lock) call.

[...]

blockdev.c static BlockJob *do_backup_common(BackupCommon *backup, · line 3604

Quoted 9 lines
blockdev.c +1 -0
index fbef6845c8..f89e48fc79 100644
@@ -3601,6 +3601,7 @@ static BlockJob *do_backup_common(BackupCommon *backup,
     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
                             backup->sync, bmap, backup->bitmap_mode,
                             backup->compress,
+                            backup->filter_node_name,

Is this guaranteed to be NULL when not specified by the user?

Max

#35 lore raw

Re: [Qemu-devel] [PATCH v9 13/13] block/backup: use backup-top instead of write notifiers

Quoted 86 lines · Max Reitz
Quoted 6 lines · Vladimir Sementsov-Ogievskiy

Drop write notifiers and use filter node instead.

= Changes =

1. add filter-node-name argument for backup qmp api. We have to do it in this commit, as 257 needs to be fixed.

I feel a bit bad about it not being an implicit node. But I know you don’t like them, they’re probably just broken altogether and because libvirt doesn’t use backup (yet), probably nobody cares.

Quoted 1 line

2. there no move write notifiers here, so is_write_notifier parameter

s/there/there are/, I suppose?

Quoted 26 lines

is dropped from block-copy paths.

3. Intersecting requests handling changed, now synchronization between backup-top, backup and guest writes are all done in block/block-copy.c and works as follows:

On copy operation, we work only with dirty areas. If bits are dirty it means that there are no requests intersecting with this area. We clear dirty bits and take bdrv range lock (bdrv_co_try_lock) on this area to prevent further operations from interaction with guest (only with guest, as neither backup nor backup-top will touch non-dirty area). If copy-operation failed we set dirty bits back together with releasing the lock.

The actual difference with old scheme is that on guest writes we don't lock the whole region but only dirty-parts, and to be more
precise: only dirty-part we are currently operate on. In old scheme guest write to non-dirty area (which may be safely ignored by backup) may wait for intersecting request, touching some other area which is dirty.

4. To sync with in-flight requests at job finish we now have drained removing of the filter, we don't need rw-lock.

= Notes =

Quoted 8 lines

Note the consequence of three objects appearing: backup-top, backup job and block-copy-state:

1. We want to insert backup-top before job creation, to behave similar with mirror and commit, where job is started upon filter.

2. We also have to create block-copy-state after filter injection, as we don't want it's source child be replaced by fitler. Instead we want

s/it's/its/, s/filter/filter/ (although “fitler” does have an amusing ring to it)

Quoted 3 lines

to keep BCS.source to be real source node, as we want to use bdrv_co_try_lock in CBW operations and it can't be used on filter, as on filter we already have in-flight (write) request from upper layer.

Reasonable, even more so as I suppose BCS.source should eventually be a BdrvChild of backup-top.

What looks wrong is that the sync_bitmap is created on the source (“bs” in backup_job_create()), but backup_cleanup_sync_bitmap() still assumes it is on blk_bs(job->common.blk) (which is no longer true).

Quoted 7 lines

So, we firstly create inject backup-top, then create job and BCS. BCS is the latest just to not create extra variable for it. Finally we set bcs for backup-top filter.

= Iotest changes =

56: op-blocker doesn't shot now, as we set it on source, but then check

s/shot/show/?

Quoted 5 lines

on filter, when trying to start second backup, so error caught in test_dismiss_collision is changed. It's OK anyway, as this test-case seems to test that after some collision we can dismiss first job and successfully start the second one. So, the change is that collision is changed from op-blocker to file-posix locks.

Well, but the op blocker belongs to the source, which should be covered by internal permissions. The fact that it now shows up as a file-posix error thus shows that the conflict also moves from the source to the target. It’s OK because we actually don’t have a conflict on the source.

Quoted 5 lines

But I wonder whether it would be better for test_dismiss_collision() to do a blockdev-backup instead where we can see the collision on the target.

Hm. On second thought, why do we even get a conflict on the target? block-copy does share the WRITE permission for it...

Not sure, but assume that this is because in file-posix.c in raw_co_create we do want RESIZE perm.

I can instead move this test to use specified job-id, to move the collision to "job-id already exists" error. Is it better?

I'm afraid that posix locks will not work if disable them in config.

Quoted 39 lines
Quoted 10 lines

However, it's obvious now that we'd better drop this op-blocker at all and add a test-case for two backups from one node (to different destinations) actually works. But not in these series.

257: The test wants to emulate guest write during backup. They should go to filter node, not to original source node, of course. Therefore we need to specify filter node name and use it.

Quoted 13 lines
---
  qapi/block-core.json       |   8 +-
  include/block/block-copy.h |   2 +-
  include/block/block_int.h  |   1 +
  block/backup-top.c         |  14 +-
  block/backup.c             | 113 +++-----------
  block/block-copy.c         |  55 ++++---
  block/replication.c        |   2 +-
  blockdev.c                 |   1 +
  tests/qemu-iotests/056     |   2 +-
  tests/qemu-iotests/257     |   7 +-
  tests/qemu-iotests/257.out | 306 ++++++++++++++++++-------------------
  11 files changed, 230 insertions(+), 281 deletions(-)

Nice.

I don’t have much to object (for some reason, I feel a bit uneasy about dropping all the request waiting code, but I suppose that is indeed taken care of with the range locks now).

[...]

Quoted 4 lines
block/backup.c +0 -0
index d927c63e5a..259a165405 100644

[...]

Quoted 4 lines

block/backup.c BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs, · line 482

Quoted 4 lines
block/backup.c +0 -0
@@ -552,6 +480,9 @@ BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
          backup_clean(&job->common.job);
          job_early_fail(&job->common.job);
      }

actually, here should be "else if"

Quoted 37 lines

block/backup.c BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs, · line 485

Quoted 3 lines
block/backup.c +3 -0
+    if (backup_top) {
+        bdrv_backup_top_drop(backup_top);
+    }

This order is weird because backup-top still has a pointer to the BCS, but we have already freed it at this point.

Quoted 14 lines
block/backup.c +0 -0
      return NULL;
  }
block/block-copy.c +1 -1
index 6828c46ba0..f3102ec3ff 100644
@@ -98,27 +98,32 @@ fail:
   * error occurred, return a negative error number
   */
  static int coroutine_fn block_copy_with_bounce_buffer(
-        BlockCopyState *s, int64_t start, int64_t end, bool is_write_notifier,
+        BlockCopyState *s, int64_t start, int64_t end,
          bool *error_is_read, void **bounce_buffer)

block/block-copy.c fail: · line 110

Quoted 10 lines
block/block-copy.c +6 -2
  {
      int ret;
-    int nbytes;
-    int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
+    int nbytes = MIN(s->cluster_size, s->len - start);
+    void *lock = bdrv_co_try_lock(blk_bs(s->source), start, nbytes);
+
+    /*
+     * Function must be called only on full-dirty region, so nobody touching or
+     * touched these bytes. Therefore, we must successfully get lock.

Which is OK because backup-top does not share the WRITE permission. But it is organized a bit weirdly, because it’s actually block-copy here that relies on the WRITE permission not to be shared; which it itself cannot do because backup-top needs it.

Quoted 5 lines

This of course results from the fact that backup-top and block-copy should really use the same object to access the source, namely the backing BdrvChild of backup-top.

Maybe there should be a comment somewhere that points this out?

But I wanted block-copy not to be directly connected to backup-top.

I think actually we rely on the fact that user of block-copy() guarantees that nobody is touching dirty areas (except block_copy). And with backup, this is done by backup-top filter. I'll add a comment on this.

Quoted 32 lines
Quoted 3 lines
block/block-copy.c +2 -0
+     */
+    assert(lock);

[...]

Quoted 10 lines
block/block-copy.c +1 -3
@@ -128,13 +133,16 @@ static int coroutine_fn block_copy_with_bounce_buffer(
          if (error_is_read) {
              *error_is_read = false;
          }
-        goto fail;
+        goto out;
      }
 
-    return nbytes;
-fail:

block/block-copy.c static int coroutine_fn block_copy_with_bounce_buffer( · line 146

Quoted 10 lines
block/block-copy.c +7 -2
-    bdrv_set_dirty_bitmap(s->copy_bitmap, start, s->cluster_size);
-    return ret;
+out:
+    if (ret < 0) {
+        bdrv_set_dirty_bitmap(s->copy_bitmap, start, s->cluster_size);
+    }
+    bdrv_co_unlock(lock);
+
+    return ret < 0 ? ret : nbytes;
 

I feel like it would still be simpler to keep the separate fail path and just duplicate the bdrv_co_unlock(lock) call.

[...]

Quoted 12 lines

blockdev.c static BlockJob *do_backup_common(BackupCommon *backup, · line 3604

Quoted 9 lines
blockdev.c +1 -0
index fbef6845c8..f89e48fc79 100644
@@ -3601,6 +3601,7 @@ static BlockJob *do_backup_common(BackupCommon *backup,
      job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
                              backup->sync, bmap, backup->bitmap_mode,
                              backup->compress,
+                            backup->filter_node_name,

Is this guaranteed to be NULL when not specified by the user?

As far as I know - yes, it is.

Signature · 2 lines
Best regards,
Vladimir
#36 lore raw

Re: [Qemu-devel] [PATCH v9 13/13] block/backup: use backup-top instead of write notifiers

Quoted 90 lines · Vladimir Sementsov-Ogievskiy
Quoted 89 lines · Max Reitz
Quoted 6 lines · Vladimir Sementsov-Ogievskiy

Drop write notifiers and use filter node instead.

= Changes =

1. add filter-node-name argument for backup qmp api. We have to do it in this commit, as 257 needs to be fixed.

I feel a bit bad about it not being an implicit node. But I know you don’t like them, they’re probably just broken altogether and because libvirt doesn’t use backup (yet), probably nobody cares.

Quoted 1 line

2. there no move write notifiers here, so is_write_notifier parameter

s/there/there are/, I suppose?

Quoted 26 lines

is dropped from block-copy paths.

3. Intersecting requests handling changed, now synchronization between backup-top, backup and guest writes are all done in block/block-copy.c and works as follows:

On copy operation, we work only with dirty areas. If bits are dirty it means that there are no requests intersecting with this area. We clear dirty bits and take bdrv range lock (bdrv_co_try_lock) on this area to prevent further operations from interaction with guest (only with guest, as neither backup nor backup-top will touch non-dirty area). If copy-operation failed we set dirty bits back together with releasing the lock.

The actual difference with old scheme is that on guest writes we don't lock the whole region but only dirty-parts, and to be more
precise: only dirty-part we are currently operate on. In old scheme guest write to non-dirty area (which may be safely ignored by backup) may wait for intersecting request, touching some other area which is dirty.

4. To sync with in-flight requests at job finish we now have drained removing of the filter, we don't need rw-lock.

= Notes =

Quoted 8 lines

Note the consequence of three objects appearing: backup-top, backup job and block-copy-state:

1. We want to insert backup-top before job creation, to behave similar with mirror and commit, where job is started upon filter.

2. We also have to create block-copy-state after filter injection, as we don't want it's source child be replaced by fitler. Instead we want

s/it's/its/, s/filter/filter/ (although “fitler” does have an amusing ring to it)

Quoted 3 lines

to keep BCS.source to be real source node, as we want to use bdrv_co_try_lock in CBW operations and it can't be used on filter, as on filter we already have in-flight (write) request from upper layer.

Reasonable, even more so as I suppose BCS.source should eventually be a BdrvChild of backup-top.

What looks wrong is that the sync_bitmap is created on the source (“bs” in backup_job_create()), but backup_cleanup_sync_bitmap() still assumes it is on blk_bs(job->common.blk) (which is no longer true).

Quoted 7 lines

So, we firstly create inject backup-top, then create job and BCS. BCS is the latest just to not create extra variable for it. Finally we set bcs for backup-top filter.

= Iotest changes =

56: op-blocker doesn't shot now, as we set it on source, but then check

s/shot/show/?

Quoted 5 lines

on filter, when trying to start second backup, so error caught in test_dismiss_collision is changed. It's OK anyway, as this test-case seems to test that after some collision we can dismiss first job and successfully start the second one. So, the change is that collision is changed from op-blocker to file-posix locks.

Well, but the op blocker belongs to the source, which should be covered by internal permissions. The fact that it now shows up as a file-posix error thus shows that the conflict also moves from the source to the target. It’s OK because we actually don’t have a conflict on the source.

But I wonder whether it would be better for test_dismiss_collision() to do a blockdev-backup instead where we can see the collision on the target.

Quoted 8 lines
Quoted 2 lines

Hm. On second thought, why do we even get a conflict on the target? block-copy does share the WRITE permission for it...

Not sure, but assume that this is because in file-posix.c in raw_co_create we do want RESIZE perm.

I can instead move this test to use specified job-id, to move the collision to "job-id already exists" error. Is it better?

It’s probably the only thing that actually makes sense.

Quoted 1 line

I'm afraid that posix locks will not work if disable them in config.

Yes (or if the environment doesn’t support them); which is why I suggested using blockdev-backup. (But that would have the same problem of “Where does the permission conflict even come from and is it inevitable?”)

[...]

Quoted 27 lines
Quoted 27 lines
Quoted 11 lines
block/block-copy.c +1 -1
index 6828c46ba0..f3102ec3ff 100644
@@ -98,27 +98,32 @@ fail:
   * error occurred, return a negative error number
   */
  static int coroutine_fn block_copy_with_bounce_buffer(
-        BlockCopyState *s, int64_t start, int64_t end, bool is_write_notifier,
+        BlockCopyState *s, int64_t start, int64_t end,
          bool *error_is_read, void **bounce_buffer)

block/block-copy.c fail: · line 110

Quoted 10 lines
block/block-copy.c +6 -2
  {
      int ret;
-    int nbytes;
-    int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
+    int nbytes = MIN(s->cluster_size, s->len - start);
+    void *lock = bdrv_co_try_lock(blk_bs(s->source), start, nbytes);
+
+    /*
+     * Function must be called only on full-dirty region, so nobody touching or
+     * touched these bytes. Therefore, we must successfully get lock.

Which is OK because backup-top does not share the WRITE permission. But it is organized a bit weirdly, because it’s actually block-copy here that relies on the WRITE permission not to be shared; which it itself cannot do because backup-top needs it.

Quoted 7 lines
Quoted 5 lines

This of course results from the fact that backup-top and block-copy should really use the same object to access the source, namely the backing BdrvChild of backup-top.

Maybe there should be a comment somewhere that points this out?

But I wanted block-copy not to be directly connected to backup-top.

But it doesn’t really work. There’s an obscure contract that already binds block-copy to backup-top.

Quoted 3 lines

I think actually we rely on the fact that user of block-copy() guarantees that nobody is touching dirty areas (except block_copy). And with backup, this is done by backup-top filter. I'll add a comment on this.

Yes, that’s the contract. And it’s weird, because block-copy has its own BB, so if it doesn’t want anyone to write to the source outside of its control, it should just unshare the WRITE permission – but it cannot do that, because its user needs WRITE access, and so it implicitly relies on that user to then unshare WRITE.

If both used the same single BdrvChild, it would be clear

(1) why the BdrvChild belongs to block-copy’s user (because block-copy’s user would need to own a node that then has this BdrvChild),

(2) why block-copy does not unshare the WRITE permission (it cannot, because it doesn’t create the BdrvChild),

(3) how block-copy could explicitly ensure that WRITE is unshared (by asserting !(child->shared_perm & WRITE)).

I’m not saying that we need to abandon having BBs right now, but I think there are a couple of cases which show why I say it’s uglier than using BdrvChildren instead.

Max

#37 lore raw

Re: [Qemu-devel] [PATCH v9 13/13] block/backup: use backup-top instead of write notifiers

Quoted 149 lines · Max Reitz
Quoted 90 lines · Vladimir Sementsov-Ogievskiy
Quoted 89 lines · Max Reitz
Quoted 6 lines · Vladimir Sementsov-Ogievskiy

Drop write notifiers and use filter node instead.

= Changes =

1. add filter-node-name argument for backup qmp api. We have to do it in this commit, as 257 needs to be fixed.

I feel a bit bad about it not being an implicit node. But I know you don’t like them, they’re probably just broken altogether and because libvirt doesn’t use backup (yet), probably nobody cares.

Quoted 1 line

2. there no move write notifiers here, so is_write_notifier parameter

s/there/there are/, I suppose?

Quoted 26 lines

is dropped from block-copy paths.

3. Intersecting requests handling changed, now synchronization between backup-top, backup and guest writes are all done in block/block-copy.c and works as follows:

On copy operation, we work only with dirty areas. If bits are dirty it means that there are no requests intersecting with this area. We clear dirty bits and take bdrv range lock (bdrv_co_try_lock) on this area to prevent further operations from interaction with guest (only with guest, as neither backup nor backup-top will touch non-dirty area). If copy-operation failed we set dirty bits back together with releasing the lock.

The actual difference with old scheme is that on guest writes we don't lock the whole region but only dirty-parts, and to be more
precise: only dirty-part we are currently operate on. In old scheme guest write to non-dirty area (which may be safely ignored by backup) may wait for intersecting request, touching some other area which is dirty.

4. To sync with in-flight requests at job finish we now have drained removing of the filter, we don't need rw-lock.

= Notes =

Quoted 8 lines

Note the consequence of three objects appearing: backup-top, backup job and block-copy-state:

1. We want to insert backup-top before job creation, to behave similar with mirror and commit, where job is started upon filter.

2. We also have to create block-copy-state after filter injection, as we don't want it's source child be replaced by fitler. Instead we want

s/it's/its/, s/filter/filter/ (although “fitler” does have an amusing ring to it)

Quoted 3 lines

to keep BCS.source to be real source node, as we want to use bdrv_co_try_lock in CBW operations and it can't be used on filter, as on filter we already have in-flight (write) request from upper layer.

Reasonable, even more so as I suppose BCS.source should eventually be a BdrvChild of backup-top.

What looks wrong is that the sync_bitmap is created on the source (“bs” in backup_job_create()), but backup_cleanup_sync_bitmap() still assumes it is on blk_bs(job->common.blk) (which is no longer true).

Quoted 7 lines

So, we firstly create inject backup-top, then create job and BCS. BCS is the latest just to not create extra variable for it. Finally we set bcs for backup-top filter.

= Iotest changes =

56: op-blocker doesn't shot now, as we set it on source, but then check

s/shot/show/?

Quoted 5 lines

on filter, when trying to start second backup, so error caught in test_dismiss_collision is changed. It's OK anyway, as this test-case seems to test that after some collision we can dismiss first job and successfully start the second one. So, the change is that collision is changed from op-blocker to file-posix locks.

Well, but the op blocker belongs to the source, which should be covered by internal permissions. The fact that it now shows up as a file-posix error thus shows that the conflict also moves from the source to the target. It’s OK because we actually don’t have a conflict on the source.

But I wonder whether it would be better for test_dismiss_collision() to do a blockdev-backup instead where we can see the collision on the target.

Quoted 8 lines
Quoted 2 lines

Hm. On second thought, why do we even get a conflict on the target? block-copy does share the WRITE permission for it...

Not sure, but assume that this is because in file-posix.c in raw_co_create we do want RESIZE perm.

I can instead move this test to use specified job-id, to move the collision to "job-id already exists" error. Is it better?

It’s probably the only thing that actually makes sense.

Quoted 1 line

I'm afraid that posix locks will not work if disable them in config.

Yes (or if the environment doesn’t support them); which is why I suggested using blockdev-backup. (But that would have the same problem of “Where does the permission conflict even come from and is it inevitable?”)

[...]

Quoted 27 lines
Quoted 27 lines
Quoted 11 lines
block/block-copy.c +1 -1
index 6828c46ba0..f3102ec3ff 100644
@@ -98,27 +98,32 @@ fail:
    * error occurred, return a negative error number
    */
   static int coroutine_fn block_copy_with_bounce_buffer(
-        BlockCopyState *s, int64_t start, int64_t end, bool is_write_notifier,
+        BlockCopyState *s, int64_t start, int64_t end,
           bool *error_is_read, void **bounce_buffer)

block/block-copy.c fail: · line 110

Quoted 10 lines
block/block-copy.c +6 -2
   {
       int ret;
-    int nbytes;
-    int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
+    int nbytes = MIN(s->cluster_size, s->len - start);
+    void *lock = bdrv_co_try_lock(blk_bs(s->source), start, nbytes);
+
+    /*
+     * Function must be called only on full-dirty region, so nobody touching or
+     * touched these bytes. Therefore, we must successfully get lock.

Which is OK because backup-top does not share the WRITE permission. But it is organized a bit weirdly, because it’s actually block-copy here that relies on the WRITE permission not to be shared; which it itself cannot do because backup-top needs it.

Quoted 7 lines
Quoted 5 lines

This of course results from the fact that backup-top and block-copy should really use the same object to access the source, namely the backing BdrvChild of backup-top.

Maybe there should be a comment somewhere that points this out?

But I wanted block-copy not to be directly connected to backup-top.

But it doesn’t really work. There’s an obscure contract that already binds block-copy to backup-top.

Quoted 6 lines
Quoted 3 lines

I think actually we rely on the fact that user of block-copy() guarantees that nobody is touching dirty areas (except block_copy). And with backup, this is done by backup-top filter. I'll add a comment on this.

Yes, that’s the contract. And it’s weird, because block-copy has its own BB, so if it doesn’t want anyone to write

doesn't want anyone to write to dirty areas, and we don't have such option in the permission system...

Quoted 13 lines

to the source outside of
its control, it should just unshare the WRITE permission – but it cannot do that, because its user needs WRITE access, and so it implicitly relies on that user to then unshare WRITE.

If both used the same single BdrvChild, it would be clear

(1) why the BdrvChild belongs to block-copy’s user (because block-copy’s user would need to own a node that then has this BdrvChild),

(2) why block-copy does not unshare the WRITE permission (it cannot, because it doesn’t create the BdrvChild),

Quoted 8 lines

(3) how block-copy could explicitly ensure that WRITE is unshared (by asserting !(child->shared_perm & WRITE)).

I’m not saying that we need to abandon having BBs right now, but I think there are a couple of cases which show why I say it’s uglier than using BdrvChildren instead.

OK. I'd prefer to move block-copy to BdrvChildren as follow-up, if you don't mind.

Signature · 2 lines
Best regards,
Vladimir
#38 lore raw

Re: [Qemu-devel] [PATCH v9 13/13] block/backup: use backup-top instead of write notifiers

Quoted 1 line · Vladimir Sementsov-Ogievskiy

02.09.2019 19:34, Max Reitz wrote:

[...]

Quoted 6 lines
Quoted 4 lines

I’m not saying that we need to abandon having BBs right now, but I think there are a couple of cases which show why I say it’s uglier than using BdrvChildren instead.

OK. I'd prefer to move block-copy to BdrvChildren as follow-up, if you don't mind.

Yep, sure, that’s A-OK with me.

Max