[Tarantool-patches] [PATCH v4 1/2] vinyl: clean-up unprocessed read views in *_build_read_views()

[Nikita Pettik <korablev@tarantool.org>]

vy_write_iterator->read_views[i].history objects are allocated on
region (see vy_write_iterator_push_rv()) during building history of the
given key. However, in case of fail of vy_write_iterator_build_history()
region is truncated but pointers to vy_write_history objects are not
nullified. As a result, they may be accessed (for instance while
finalizing write_iterator object in  vy_write_iterator_stop) which in
turn may lead to crash, segfaul or disk formatting. The same may happen
if vy_read_view_merge() fails during processing of read view array.
Let's clean-up those objects in case of error takes place.

Part of #4864
---
 src/box/vy_write_iterator.c                   |  61 +++++++--
 src/errinj.h                                  |   1 +
 test/box/errinj.result                        |   1 +
 .../gh-4864-stmt-alloc-fail-compact.result    | 125 ++++++++++++++++++
 .../gh-4864-stmt-alloc-fail-compact.test.lua  |  53 ++++++++
 5 files changed, 227 insertions(+), 14 deletions(-)

diff --git a/src/box/vy_write_iterator.c b/src/box/vy_write_iterator.c
index 7a6a20627..7784dd13a 100644
--- a/src/box/vy_write_iterator.c
+++ b/src/box/vy_write_iterator.c
@@ -151,9 +151,11 @@ vy_read_view_stmt_destroy(struct vy_read_view_stmt *rv)
 	if (rv->tuple != NULL)
 		vy_stmt_unref_if_possible(rv->tuple);
 	rv->tuple = NULL;
-	if (rv->history != NULL)
-		vy_write_history_destroy(rv->history);
-	rv->history = NULL;
+	/*
+	 * History must be already cleaned up in
+	 * vy_write_iterator_build_read_views().
+	 */
+	assert(rv->history == NULL);
 }
 
 /* @sa vy_write_iterator.h */
@@ -790,8 +792,7 @@ next_lsn:
 	 * statement around if this is major compaction, because
 	 * there's no tuple it could overwrite.
 	 */
-	if (rc == 0 && stream->is_last_level &&
-	    stream->deferred_delete_stmt != NULL) {
+	if (stream->is_last_level && stream->deferred_delete_stmt != NULL) {
 		vy_stmt_unref_if_possible(stream->deferred_delete_stmt);
 		stream->deferred_delete_stmt = NULL;
 	}
@@ -834,6 +835,15 @@ vy_read_view_merge(struct vy_write_iterator *stream, struct tuple *hint,
 		rv->history = NULL;
 		return 0;
 	}
+#ifndef NDEBUG
+	struct errinj *inj =
+		errinj(ERRINJ_VY_READ_VIEW_MERGE_FAIL, ERRINJ_BOOL);
+	if (inj != NULL && inj->bparam) {
+		inj->bparam = false;
+		diag_set(OutOfMemory, 666, "malloc", "struct vy_stmt");
+		return -1;
+	}
+#endif
 	/*
 	 * Two possible hints to remove the current UPSERT.
 	 * 1. If the stream is working on the last level, we
@@ -940,6 +950,25 @@ vy_read_view_merge(struct vy_write_iterator *stream, struct tuple *hint,
 	return 0;
 }
 
+/**
+ * Clean up all histories related to given write iterator.
+ * Particular history is allocated using region, so single
+ * region truncation is enough to release all memory at once.
+ * Before that we should also unref tuples stored in those
+ * histories (which is done in vy_write_history_destroy()).
+ */
+static void
+vy_write_iterator_history_destroy(struct vy_write_iterator *stream,
+				  struct region *region, size_t used)
+{
+	for (int i = 0; i < stream->rv_count; ++i) {
+		if (stream->read_views[i].history != NULL)
+			vy_write_history_destroy(stream->read_views[i].history);
+		stream->read_views[i].history = NULL;
+	}
+	region_truncate(region, used);
+}
+
 /**
  * Split the current key into a sequence of read view
  * statements. @sa struct vy_write_iterator comment for details
@@ -960,9 +989,12 @@ vy_write_iterator_build_read_views(struct vy_write_iterator *stream, int *count)
 	struct region *region = &fiber()->gc;
 	size_t used = region_used(region);
 	stream->rv_used_count = 0;
+	int rc = 0;
 	if (vy_write_iterator_build_history(stream, &raw_count,
-					    &is_first_insert) != 0)
-		goto error;
+					    &is_first_insert) != 0) {
+		rc = -1;
+		goto cleanup;
+	}
 	if (raw_count == 0) {
 		/* A key is fully optimized. */
 		region_truncate(region, used);
@@ -983,8 +1015,10 @@ vy_write_iterator_build_read_views(struct vy_write_iterator *stream, int *count)
 		if (rv->history == NULL)
 			continue;
 		if (vy_read_view_merge(stream, hint, rv,
-				       is_first_insert) != 0)
-			goto error;
+				       is_first_insert) != 0) {
+			rc = -1;
+			goto cleanup;
+		}
 		assert(rv->history == NULL);
 		if (rv->tuple == NULL)
 			continue;
@@ -992,11 +1026,10 @@ vy_write_iterator_build_read_views(struct vy_write_iterator *stream, int *count)
 		++*count;
 		hint = rv->tuple;
 	}
-	region_truncate(region, used);
-	return 0;
-error:
-	region_truncate(region, used);
-	return -1;
+
+cleanup:
+	vy_write_iterator_history_destroy(stream, region, used);
+	return rc;
 }
 
 /**
diff --git a/src/errinj.h b/src/errinj.h
index 383dafcb5..b7550bb5e 100644
--- a/src/errinj.h
+++ b/src/errinj.h
@@ -128,6 +128,7 @@ struct errinj {
 	_(ERRINJ_DYN_MODULE_COUNT, ERRINJ_INT, {.iparam = 0}) \
 	_(ERRINJ_INDEX_RESERVE, ERRINJ_BOOL, {.bparam = false})\
 	_(ERRINJ_VY_STMT_ALLOC, ERRINJ_INT, {.iparam = -1})\
+	_(ERRINJ_VY_READ_VIEW_MERGE_FAIL, ERRINJ_BOOL, {.bparam = false})\
 
 ENUM0(errinj_id, ERRINJ_LIST);
 extern struct errinj errinjs[];
diff --git a/test/box/errinj.result b/test/box/errinj.result
index efbb4e85e..e1b9fbe2a 100644
--- a/test/box/errinj.result
+++ b/test/box/errinj.result
@@ -68,6 +68,7 @@ evals
   - ERRINJ_VY_READ_PAGE: false
   - ERRINJ_VY_READ_PAGE_DELAY: false
   - ERRINJ_VY_READ_PAGE_TIMEOUT: 0
+  - ERRINJ_VY_READ_VIEW_MERGE_FAIL: false
   - ERRINJ_VY_RUN_DISCARD: false
   - ERRINJ_VY_RUN_FILE_RENAME: false
   - ERRINJ_VY_RUN_WRITE: false
diff --git a/test/vinyl/gh-4864-stmt-alloc-fail-compact.result b/test/vinyl/gh-4864-stmt-alloc-fail-compact.result
index 1afc02bef..af116a4b4 100644
--- a/test/vinyl/gh-4864-stmt-alloc-fail-compact.result
+++ b/test/vinyl/gh-4864-stmt-alloc-fail-compact.result
@@ -121,3 +121,128 @@ errinj.set('ERRINJ_VY_STMT_ALLOC', -1)
 s:drop()
  | ---
  | ...
+
+-- All the same except for delayed vy_stmt_alloc() fail.
+-- Re-create space for the sake of test purity.
+--
+s = box.schema.space.create('test', {engine = 'vinyl'})
+ | ---
+ | ...
+_ = s:create_index('pk', {run_count_per_level = 100, page_size = 128, range_size = 1024})
+ | ---
+ | ...
+
+dump(true)
+ | ---
+ | ...
+dump()
+ | ---
+ | ...
+
+compact()
+ | ---
+ | ...
+
+dump()
+ | ---
+ | ...
+
+errinj = box.error.injection
+ | ---
+ | ...
+errinj.set('ERRINJ_VY_STMT_ALLOC', 5)
+ | ---
+ | - ok
+ | ...
+-- Compaction of first range fails, so it is re-scheduled and
+-- then successfully finishes at the second attempt.
+--
+compact()
+ | ---
+ | ...
+assert(s.index.pk:stat().range_count == 2)
+ | ---
+ | - true
+ | ...
+assert(s.index.pk:stat().run_count == 2)
+ | ---
+ | - true
+ | ...
+assert(errinj.get('ERRINJ_VY_STMT_ALLOC') == -1)
+ | ---
+ | - true
+ | ...
+errinj.set('ERRINJ_VY_STMT_ALLOC', -1)
+ | ---
+ | - ok
+ | ...
+-- Unthrottle scheduler to allow next dump.
+--
+errinj.set("ERRINJ_VY_SCHED_TIMEOUT", 0.0001)
+ | ---
+ | - ok
+ | ...
+
+s:drop()
+ | ---
+ | ...
+
+-- Once again but test that clean-up is made in case
+-- vy_read_view_merge() fails.
+--
+s = box.schema.space.create('test', {engine = 'vinyl'})
+ | ---
+ | ...
+_ = s:create_index('pk', {run_count_per_level = 100, page_size = 128, range_size = 1024})
+ | ---
+ | ...
+
+dump(true)
+ | ---
+ | ...
+dump()
+ | ---
+ | ...
+
+compact()
+ | ---
+ | ...
+
+dump()
+ | ---
+ | ...
+
+errinj = box.error.injection
+ | ---
+ | ...
+errinj.set('ERRINJ_VY_READ_VIEW_MERGE_FAIL', true)
+ | ---
+ | - ok
+ | ...
+compact()
+ | ---
+ | ...
+assert(s.index.pk:stat().range_count == 2)
+ | ---
+ | - true
+ | ...
+assert(s.index.pk:stat().run_count == 2)
+ | ---
+ | - true
+ | ...
+assert(errinj.get('ERRINJ_VY_READ_VIEW_MERGE_FAIL') == false)
+ | ---
+ | - true
+ | ...
+errinj.set('ERRINJ_VY_READ_VIEW_MERGE_FAIL', false)
+ | ---
+ | - ok
+ | ...
+s:drop()
+ | ---
+ | ...
+
+errinj.set("ERRINJ_VY_SCHED_TIMEOUT", 0)
+ | ---
+ | - ok
+ | ...
diff --git a/test/vinyl/gh-4864-stmt-alloc-fail-compact.test.lua b/test/vinyl/gh-4864-stmt-alloc-fail-compact.test.lua
index bf70bdf75..a68c73d32 100644
--- a/test/vinyl/gh-4864-stmt-alloc-fail-compact.test.lua
+++ b/test/vinyl/gh-4864-stmt-alloc-fail-compact.test.lua
@@ -53,3 +53,56 @@ assert(errinj.get('ERRINJ_VY_STMT_ALLOC') == -1)
 errinj.set('ERRINJ_VY_STMT_ALLOC', -1)
 
 s:drop()
+
+-- All the same except for delayed vy_stmt_alloc() fail.
+-- Re-create space for the sake of test purity.
+--
+s = box.schema.space.create('test', {engine = 'vinyl'})
+_ = s:create_index('pk', {run_count_per_level = 100, page_size = 128, range_size = 1024})
+
+dump(true)
+dump()
+
+compact()
+
+dump()
+
+errinj = box.error.injection
+errinj.set('ERRINJ_VY_STMT_ALLOC', 5)
+-- Compaction of first range fails, so it is re-scheduled and
+-- then successfully finishes at the second attempt.
+--
+compact()
+assert(s.index.pk:stat().range_count == 2)
+assert(s.index.pk:stat().run_count == 2)
+assert(errinj.get('ERRINJ_VY_STMT_ALLOC') == -1)
+errinj.set('ERRINJ_VY_STMT_ALLOC', -1)
+-- Unthrottle scheduler to allow next dump.
+--
+errinj.set("ERRINJ_VY_SCHED_TIMEOUT", 0.0001)
+
+s:drop()
+
+-- Once again but test that clean-up is made in case
+-- vy_read_view_merge() fails.
+--
+s = box.schema.space.create('test', {engine = 'vinyl'})
+_ = s:create_index('pk', {run_count_per_level = 100, page_size = 128, range_size = 1024})
+
+dump(true)
+dump()
+
+compact()
+
+dump()
+
+errinj = box.error.injection
+errinj.set('ERRINJ_VY_READ_VIEW_MERGE_FAIL', true)
+compact()
+assert(s.index.pk:stat().range_count == 2)
+assert(s.index.pk:stat().run_count == 2)
+assert(errinj.get('ERRINJ_VY_READ_VIEW_MERGE_FAIL') == false)
+errinj.set('ERRINJ_VY_READ_VIEW_MERGE_FAIL', false)
+s:drop()
+
+errinj.set("ERRINJ_VY_SCHED_TIMEOUT", 0)
-- 
2.17.1