Uninitialized flush_counters_t::unzip_LRU_evicted use

[Laurynas Biveinis]

==5873== Thread 23:
==5873== Conditional jump or move depends on uninitialised value(s)
==5873== at 0xA15646: buf_do_LRU_batch(buf_pool_t*, unsigned long, bool, flush_counters_t*) (buf0flu.cc:1657)
==5873== by 0xA15A71: buf_flush_batch(buf_pool_t*, buf_flush_t, unsigned long, unsigned long, bool, flush_counters_t*) (buf0flu.cc:1790)
==5873== by 0xA15F04: buf_flush_LRU(buf_pool_t*, unsigned long, bool, flush_counters_t*) (buf0flu.cc:1954)
==5873== by 0xA1618C: buf_flush_LRU_tail() (buf0flu.cc:2259)
==5873== by 0xA16469: buf_flush_lru_manager_thread (buf0flu.cc:2850)
==5873== by 0x506A9D0: start_thread (in /lib64/libpthread-2.12.so)
==5873== by 0x634CB6C: clone (in /lib64/libc-2.12.so)
==5873== Uninitialised value was created by a stack allocation
==5873== at 0xA15F44: buf_flush_LRU_tail() (buf0flu.cc:2211)
==5873==

The cause is

 if (max > n->unzip_LRU_evicted) {

in buf_do_LRU_batch propagated from

  n->unzip_LRU_evicted
   += buf_free_from_unzip_LRU_list_batch(buf_pool, max);

in the same function, called from buf_flush_LRU, called from buf_flush_LRU_tail, which has a stack-allocated uninitialised flush_counters_t variable propagated down. Regression introduced by bug 1295268 fix.

[Laurynas Biveinis (laurynas-biveinis)]

Reproducible by valgrinding any of innodb_zip.innodb_bug36172 innodb_zip.innodb-create-options innodb_zip.innodb_cmp_per_index innodb_zip.innodb-wl5522-zip innodb_zip.innodb_index_large_prefix innodb_zip.innodb-zip MTR tests.

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-1509