ReplicatedPG: ensure that ec appends are aligned

When not writing out the last chunk, _write_copy_chunk
will trim the chunk to a stripe boundary for ec pools
and adjust cursor.data_offset to compensate.

Signed-off-by: Samuel Just <sam.just@inktank.com>

diff --git a/src/osd/ReplicatedPG.cc b/src/osd/ReplicatedPG.cc
index 9c48c18be4e60a56b21a924b0fbd8b058ae410fe..538296768eab2e3ae3cfc55da2a7bc0df41762dd 100644 (file)
--- a/src/osd/ReplicatedPG.cc
+++ b/src/osd/ReplicatedPG.cc
@@ -5379,7 +5379,7 @@ void ReplicatedPG::_copy_some(ObjectContextRef obc, CopyOpRef cop)
     // it already!
     assert(cop->cursor.is_initial());
   }
-  op.copy_get(&cop->cursor, cct->_conf->osd_copyfrom_max_chunk,
+  op.copy_get(&cop->cursor, get_copy_chunk_size(),
              &cop->results.object_size, &cop->results.mtime,
              &cop->results.category,
              &cop->attrs, &cop->data, &cop->omap_header, &cop->omap,
@@ -5509,6 +5509,27 @@ void ReplicatedPG::_write_copy_chunk(CopyOpRef cop, PGBackend::PGTransaction *t)
     cop->attrs.clear();
   }
   if (!cop->temp_cursor.data_complete) {
+    assert(cop->data.length() + cop->temp_cursor.data_offset ==
+          cop->cursor.data_offset);
+    if (pool.info.requires_aligned_append() &&
+       !cop->cursor.data_complete) {
+      /**
+       * Trim off the unaligned bit at the end, we'll adjust cursor.data_offset
+       * to pick it up on the next pass.
+       */
+      assert(cop->temp_cursor.data_offset %
+            pool.info.required_alignment() == 0);
+      if (cop->data.length() % pool.info.required_alignment() != 0) {
+       uint64_t to_trim =
+         cop->data.length() % pool.info.required_alignment();
+       bufferlist bl;
+       bl.substr_of(cop->data, 0, cop->data.length() - to_trim);
+       cop->data.swap(bl);
+       cop->cursor.data_offset -= to_trim;
+       assert(cop->data.length() + cop->temp_cursor.data_offset ==
+              cop->cursor.data_offset);
+      }
+    }
     t->append(
       cop->results.temp_oid,
       cop->temp_cursor.data_offset,

diff --git a/src/osd/ReplicatedPG.h b/src/osd/ReplicatedPG.h
index a8aec999dd3f6ba42648481079c5fa333b0852fd..c53d4b9b92ce036cfe02b74af5401d448c21435c 100644 (file)
--- a/src/osd/ReplicatedPG.h
+++ b/src/osd/ReplicatedPG.h
@@ -1119,6 +1119,16 @@ protected:
                  bool mirror_snapset);
   void process_copy_chunk(hobject_t oid, tid_t tid, int r);
   void _write_copy_chunk(CopyOpRef cop, PGBackend::PGTransaction *t);
+  uint64_t get_copy_chunk_size() const {
+    uint64_t size = cct->_conf->osd_copyfrom_max_chunk;
+    if (pool.info.requires_aligned_append()) {
+      uint64_t alignment = pool.info.required_alignment();
+      if (size % alignment) {
+       size += alignment - (size % alignment);
+      }
+    }
+    return size;
+  }
   void _copy_some(ObjectContextRef obc, CopyOpRef cop);
   void _build_finish_copy_transaction(CopyOpRef cop,
                                       PGBackend::PGTransaction *t);