innodb_fix_misc.patch

   1 # name       : innodb_fix_misc.patch
   2 # introduced : 11 or before
   3 # maintainer : Yasufumi
   4 #
   5 # Bug fix for
   6 # http://bugs.mysql.com/56433 (always: because good for all users, and safe)
   7 # and http://bugs.mysql.com/51325 (optional: innodb_lazy_drop_table)
   8 # were added. They may be removed in the future when will be fixed officially.
   9 #
  10 #!!! notice !!!
  11 # Any small change to this file in the main branch
  12 # should be done or reviewed by the maintainer!
  13 --- a/storage/innobase/buf/buf0buf.c
  14 +++ b/storage/innobase/buf/buf0buf.c
  15 @@ -1981,6 +1981,27 @@
  16  #endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
  17         }
  18
  19 +       if (UNIV_UNLIKELY(bpage->space_was_being_deleted)) {
  20 +               /* This page is obsoleted, should discard and retry */
  21 +               rw_lock_s_unlock(&buf_pool->page_hash_latch);
  22 +
  23 +               mutex_enter(&buf_pool->LRU_list_mutex);
  24 +               block_mutex = buf_page_get_mutex_enter(bpage);
  25 +
  26 +               if (UNIV_UNLIKELY(!block_mutex)) {
  27 +                       mutex_exit(&buf_pool->LRU_list_mutex);
  28 +                       goto lookup;
  29 +               }
  30 +
  31 +               buf_LRU_free_block(bpage, TRUE, TRUE);
  32 +
  33 +               mutex_exit(&buf_pool->LRU_list_mutex);
  34 +               mutex_exit(block_mutex);
  35 +               block_mutex = NULL;
  36 +
  37 +               goto lookup;
  38 +       }
  39 +
  40         if (UNIV_UNLIKELY(!bpage->zip.data)) {
  41                 /* There is no compressed page. */
  42  err_exit:
  43 @@ -2489,6 +2510,27 @@
  44                 block = (buf_block_t*) buf_page_hash_get_low(
  45                         buf_pool, space, offset, fold);
  46                 if (block) {
  47 +                       if (UNIV_UNLIKELY(block->page.space_was_being_deleted)) {
  48 +                               /* This page is obsoleted, should discard and retry */
  49 +                               rw_lock_s_unlock(&buf_pool->page_hash_latch);
  50 +
  51 +                               mutex_enter(&buf_pool->LRU_list_mutex);
  52 +                               block_mutex = buf_page_get_mutex_enter((buf_page_t*)block);
  53 +
  54 +                               if (UNIV_UNLIKELY(!block_mutex)) {
  55 +                                       mutex_exit(&buf_pool->LRU_list_mutex);
  56 +                                       goto loop;
  57 +                               }
  58 +
  59 +                               buf_LRU_free_block((buf_page_t*)block, TRUE, TRUE);
  60 +
  61 +                               mutex_exit(&buf_pool->LRU_list_mutex);
  62 +                               mutex_exit(block_mutex);
  63 +                               block_mutex = NULL;
  64 +
  65 +                               goto loop;
  66 +                       }
  67 +
  68                         block_mutex = buf_page_get_mutex_enter((buf_page_t*)block);
  69                         ut_a(block_mutex);
  70                 }
  71 @@ -3411,11 +3453,28 @@
  72
  73         fold = buf_page_address_fold(space, offset);
  74
  75 +retry:
  76         //buf_pool_mutex_enter(buf_pool);
  77         mutex_enter(&buf_pool->LRU_list_mutex);
  78         rw_lock_x_lock(&buf_pool->page_hash_latch);
  79
  80         watch_page = buf_page_hash_get_low(buf_pool, space, offset, fold);
  81 +
  82 +       if (UNIV_UNLIKELY(watch_page && watch_page->space_was_being_deleted)) {
  83 +               mutex_t*        block_mutex = buf_page_get_mutex_enter(watch_page);
  84 +
  85 +               /* This page is obsoleted, should discard and retry */
  86 +               rw_lock_x_unlock(&buf_pool->page_hash_latch);
  87 +               ut_a(block_mutex);
  88 +
  89 +               buf_LRU_free_block(watch_page, TRUE, TRUE);
  90 +
  91 +               mutex_exit(&buf_pool->LRU_list_mutex);
  92 +               mutex_exit(block_mutex);
  93 +
  94 +               goto retry;
  95 +       }
  96 +
  97         if (watch_page && !buf_pool_watch_is_sentinel(buf_pool, watch_page)) {
  98                 /* The page is already in the buffer pool. */
  99                 watch_page = NULL;
 100 @@ -3546,6 +3605,7 @@
 101                 bpage->state    = BUF_BLOCK_ZIP_PAGE;
 102                 bpage->space    = space;
 103                 bpage->offset   = offset;
 104 +               bpage->space_was_being_deleted = FALSE;
 105
 106  #ifdef UNIV_DEBUG
 107                 bpage->in_page_hash = FALSE;
 108 @@ -3630,6 +3690,7 @@
 109
 110         fold = buf_page_address_fold(space, offset);
 111
 112 +retry:
 113         //buf_pool_mutex_enter(buf_pool);
 114         mutex_enter(&buf_pool->LRU_list_mutex);
 115         rw_lock_x_lock(&buf_pool->page_hash_latch);
 116 @@ -3637,6 +3698,21 @@
 117         block = (buf_block_t*) buf_page_hash_get_low(
 118                 buf_pool, space, offset, fold);
 119
 120 +       if (UNIV_UNLIKELY(block && block->page.space_was_being_deleted)) {
 121 +               mutex_t*        block_mutex = buf_page_get_mutex_enter((buf_page_t*)block);
 122 +
 123 +               /* This page is obsoleted, should discard and retry */
 124 +               rw_lock_x_unlock(&buf_pool->page_hash_latch);
 125 +               ut_a(block_mutex);
 126 +
 127 +               buf_LRU_free_block((buf_page_t*)block, TRUE, TRUE);
 128 +
 129 +               mutex_exit(&buf_pool->LRU_list_mutex);
 130 +               mutex_exit(block_mutex);
 131 +
 132 +               goto retry;
 133 +       }
 134 +
 135         if (block
 136             && buf_page_in_file(&block->page)
 137             && !buf_pool_watch_is_sentinel(buf_pool, &block->page)) {
 138 @@ -3990,8 +4066,11 @@
 139         }
 140
 141         if (io_type == BUF_IO_WRITE
 142 -           && (buf_page_get_state(bpage) == BUF_BLOCK_ZIP_DIRTY
 143 -               || buf_page_get_flush_type(bpage) == BUF_FLUSH_LRU)) {
 144 +           && (
 145 +#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
 146 +               buf_page_get_state(bpage) == BUF_BLOCK_ZIP_DIRTY ||
 147 +#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
 148 +               buf_page_get_flush_type(bpage) == BUF_FLUSH_LRU)) {
 149                 /* to keep consistency at buf_LRU_insert_zip_clean() */
 150                 have_LRU_mutex = TRUE; /* optimistic */
 151         }
 152 --- a/storage/innobase/buf/buf0flu.c
 153 +++ b/storage/innobase/buf/buf0flu.c
 154 @@ -439,7 +439,7 @@
 155
 156         if (UNIV_LIKELY(bpage->in_LRU_list && buf_page_in_file(bpage))) {
 157
 158 -               return(bpage->oldest_modification == 0
 159 +               return((bpage->oldest_modification == 0 || bpage->space_was_being_deleted)
 160                        && buf_page_get_io_fix(bpage) == BUF_IO_NONE
 161                        && bpage->buf_fix_count == 0);
 162         }
 163 @@ -481,6 +481,13 @@
 164             && buf_page_get_io_fix(bpage) == BUF_IO_NONE) {
 165                 ut_ad(bpage->in_flush_list);
 166
 167 +               if (bpage->space_was_being_deleted) {
 168 +                       /* should be removed from flush_list here */
 169 +                       /* because buf_flush_try_neighbors() cannot flush without fil_space_get_size(space) */
 170 +                       buf_flush_remove(bpage);
 171 +                       return(FALSE);
 172 +               }
 173 +
 174                 if (flush_type != BUF_FLUSH_LRU) {
 175
 176                         return(TRUE);
 177 --- a/storage/innobase/buf/buf0lru.c
 178 +++ b/storage/innobase/buf/buf0lru.c
 179 @@ -545,6 +545,62 @@
 180         }
 181  }
 182
 183 +/******************************************************************//**
 184 +*/
 185 +UNIV_INTERN
 186 +void
 187 +buf_LRU_mark_space_was_deleted(
 188 +/*===========================*/
 189 +       ulint   id)     /*!< in: space id */
 190 +{
 191 +       ulint   i;
 192 +
 193 +       for (i = 0; i < srv_buf_pool_instances; i++) {
 194 +               buf_pool_t*     buf_pool;
 195 +               buf_page_t*     bpage;
 196 +               buf_chunk_t*    chunk;
 197 +               ulint           j, k;
 198 +
 199 +               buf_pool = buf_pool_from_array(i);
 200 +
 201 +               mutex_enter(&buf_pool->LRU_list_mutex);
 202 +
 203 +               bpage = UT_LIST_GET_FIRST(buf_pool->LRU);
 204 +
 205 +               while (bpage != NULL) {
 206 +                       if (buf_page_get_space(bpage) == id) {
 207 +                               bpage->space_was_being_deleted = TRUE;
 208 +                       }
 209 +                       bpage = UT_LIST_GET_NEXT(LRU, bpage);
 210 +               }
 211 +
 212 +               mutex_exit(&buf_pool->LRU_list_mutex);
 213 +
 214 +               rw_lock_s_lock(&btr_search_latch);
 215 +               chunk = buf_pool->chunks;
 216 +               for (j = buf_pool->n_chunks; j--; chunk++) {
 217 +                       buf_block_t*    block   = chunk->blocks;
 218 +                       for (k = chunk->size; k--; block++) {
 219 +                               if (buf_block_get_state(block)
 220 +                                   != BUF_BLOCK_FILE_PAGE
 221 +                                   || !block->index
 222 +                                   || buf_page_get_space(&block->page) != id) {
 223 +                                       continue;
 224 +                               }
 225 +
 226 +                               rw_lock_s_unlock(&btr_search_latch);
 227 +
 228 +                               rw_lock_x_lock(&block->lock);
 229 +                               btr_search_drop_page_hash_index(block);
 230 +                               rw_lock_x_unlock(&block->lock);
 231 +
 232 +                               rw_lock_s_lock(&btr_search_latch);
 233 +                       }
 234 +               }
 235 +               rw_lock_s_unlock(&btr_search_latch);
 236 +       }
 237 +}
 238 +
 239  #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
 240  /********************************************************************//**
 241  Insert a compressed block into buf_pool->zip_clean in the LRU order. */
 242 @@ -1499,6 +1555,10 @@
 243                 return(FALSE);
 244         }
 245
 246 +       if (bpage->space_was_being_deleted && bpage->oldest_modification != 0) {
 247 +               buf_flush_remove(bpage);
 248 +       }
 249 +
 250  #ifdef UNIV_IBUF_COUNT_DEBUG
 251         ut_a(ibuf_count_get(bpage->space, bpage->offset) == 0);
 252  #endif /* UNIV_IBUF_COUNT_DEBUG */
 253 --- a/storage/innobase/fil/fil0fil.c
 254 +++ b/storage/innobase/fil/fil0fil.c
 255 @@ -254,6 +254,7 @@
 256  struct fil_system_struct {
 257  #ifndef UNIV_HOTBACKUP
 258         mutex_t         mutex;          /*!< The mutex protecting the cache */
 259 +       mutex_t         file_extend_mutex;
 260  #endif /* !UNIV_HOTBACKUP */
 261         hash_table_t*   spaces;         /*!< The hash table of spaces in the
 262                                         system; they are hashed on the space
 263 @@ -863,7 +864,7 @@
 264         ut_ad(node && system);
 265         ut_ad(mutex_own(&(system->mutex)));
 266         ut_a(node->open);
 267 -       ut_a(node->n_pending == 0);
 268 +       ut_a(node->n_pending == 0 || node->space->is_being_deleted);
 269         ut_a(node->n_pending_flushes == 0);
 270         ut_a(node->modification_counter == node->flush_counter
 271              || srv_fast_shutdown == 2);
 272 @@ -877,7 +878,7 @@
 273         ut_a(system->n_open > 0);
 274         system->n_open--;
 275
 276 -       if (node->space->purpose == FIL_TABLESPACE && !trx_sys_sys_space(node->space->id)) {
 277 +       if (node->n_pending == 0 && node->space->purpose == FIL_TABLESPACE && !trx_sys_sys_space(node->space->id)) {
 278                 ut_a(UT_LIST_GET_LEN(system->LRU) > 0);
 279
 280                 /* The node is in the LRU list, remove it */
 281 @@ -1076,7 +1077,7 @@
 282         ut_ad(node && system && space);
 283         ut_ad(mutex_own(&(system->mutex)));
 284         ut_a(node->magic_n == FIL_NODE_MAGIC_N);
 285 -       ut_a(node->n_pending == 0);
 286 +       ut_a(node->n_pending == 0 || space->is_being_deleted);
 287
 288         if (node->open) {
 289                 /* We fool the assertion in fil_node_close_file() to think
 290 @@ -1598,6 +1599,8 @@
 291
 292         mutex_create(fil_system_mutex_key,
 293                      &fil_system->mutex, SYNC_ANY_LATCH);
 294 +       mutex_create(fil_system_mutex_key,
 295 +                    &fil_system->file_extend_mutex, SYNC_OUTER_ANY_LATCH);
 296
 297         fil_system->spaces = hash_create(hash_size);
 298         fil_system->name_hash = hash_create(hash_size);
 299 @@ -2352,7 +2355,11 @@
 300         completely and permanently. The flag is_being_deleted also prevents
 301         fil_flush() from being applied to this tablespace. */
 302
 303 +       if (srv_lazy_drop_table) {
 304 +               buf_LRU_mark_space_was_deleted(id);
 305 +       } else {
 306         buf_LRU_invalidate_tablespace(id);
 307 +       }
 308  #endif
 309         /* printf("Deleting tablespace %s id %lu\n", space->name, id); */
 310
 311 @@ -4739,6 +4746,10 @@
 312         ulint           page_size;
 313         ibool           success         = TRUE;
 314
 315 +       /* file_extend_mutex is for http://bugs.mysql.com/56433 */
 316 +       /* to protect from the other fil_extend_space_to_desired_size() */
 317 +       /* during temprary releasing &fil_system->mutex */
 318 +       mutex_enter(&fil_system->file_extend_mutex);
 319         fil_mutex_enter_and_prepare_for_io(space_id);
 320
 321         space = fil_space_get_by_id(space_id);
 322 @@ -4750,6 +4761,7 @@
 323                 *actual_size = space->size;
 324
 325                 mutex_exit(&fil_system->mutex);
 326 +               mutex_exit(&fil_system->file_extend_mutex);
 327
 328                 return(TRUE);
 329         }
 330 @@ -4782,6 +4794,8 @@
 331                 offset_low  = ((start_page_no - file_start_page_no)
 332                                % (4096 * ((1024 * 1024) / page_size)))
 333                         * page_size;
 334 +
 335 +               mutex_exit(&fil_system->mutex);
 336  #ifdef UNIV_HOTBACKUP
 337                 success = os_file_write(node->name, node->handle, buf,
 338                                         offset_low, offset_high,
 339 @@ -4791,8 +4805,10 @@
 340                                  node->name, node->handle, buf,
 341                                  offset_low, offset_high,
 342                                  page_size * n_pages,
 343 -                                NULL, NULL, NULL);
 344 +                                NULL, NULL, space_id, NULL);
 345  #endif
 346 +               mutex_enter(&fil_system->mutex);
 347 +
 348                 if (success) {
 349                         node->size += n_pages;
 350                         space->size += n_pages;
 351 @@ -4838,6 +4854,7 @@
 352         printf("Extended %s to %lu, actual size %lu pages\n", space->name,
 353         size_after_extend, *actual_size); */
 354         mutex_exit(&fil_system->mutex);
 355 +       mutex_exit(&fil_system->file_extend_mutex);
 356
 357         fil_flush(space_id, TRUE);
 358
 359 @@ -5200,6 +5217,22 @@
 360                 srv_data_written+= len;
 361         }
 362
 363 +       /* if the table space was already deleted, space might not exist already. */
 364 +       if (message
 365 +           && space_id < SRV_LOG_SPACE_FIRST_ID
 366 +           && ((buf_page_t*)message)->space_was_being_deleted) {
 367 +
 368 +               if (mode == OS_AIO_NORMAL) {
 369 +                       buf_page_io_complete(message);
 370 +                       return(DB_SUCCESS); /*fake*/
 371 +               }
 372 +               if (type == OS_FILE_READ) {
 373 +                       return(DB_TABLESPACE_DELETED);
 374 +               } else {
 375 +                       return(DB_SUCCESS); /*fake*/
 376 +               }
 377 +       }
 378 +
 379         /* Reserve the fil_system mutex and make sure that we can open at
 380         least one file while holding it, if the file is not already open */
 381
 382 @@ -5341,10 +5374,24 @@
 383  #else
 384         /* Queue the aio request */
 385         ret = os_aio(type, mode | wake_later, node->name, node->handle, buf,
 386 -                    offset_low, offset_high, len, node, message, trx);
 387 +                    offset_low, offset_high, len, node, message, space_id, trx);
 388  #endif
 389         } /**/
 390
 391 +       /* if the table space was already deleted, space might not exist already. */
 392 +       if (message
 393 +           && space_id < SRV_LOG_SPACE_FIRST_ID
 394 +           && ((buf_page_t*)message)->space_was_being_deleted) {
 395 +
 396 +               if (mode == OS_AIO_SYNC) {
 397 +                       if (type == OS_FILE_READ) {
 398 +                               return(DB_TABLESPACE_DELETED);
 399 +                       } else {
 400 +                               return(DB_SUCCESS); /*fake*/
 401 +                       }
 402 +               }
 403 +       }
 404 +
 405         ut_a(ret);
 406
 407         if (mode == OS_AIO_SYNC) {
 408 @@ -5444,6 +5491,7 @@
 409         fil_node_t*     fil_node;
 410         void*           message;
 411         ulint           type;
 412 +       ulint           space_id = 0;
 413
 414         ut_ad(fil_validate_skip());
 415
 416 @@ -5451,10 +5499,10 @@
 417                 srv_set_io_thread_op_info(segment, "native aio handle");
 418  #ifdef WIN_ASYNC_IO
 419                 ret = os_aio_windows_handle(segment, 0, &fil_node,
 420 -                                           &message, &type);
 421 +                                           &message, &type, &space_id);
 422  #elif defined(LINUX_NATIVE_AIO)
 423                 ret = os_aio_linux_handle(segment, &fil_node,
 424 -                                         &message, &type);
 425 +                                         &message, &type, &space_id);
 426  #else
 427                 ut_error;
 428                 ret = 0; /* Eliminate compiler warning */
 429 @@ -5463,7 +5511,22 @@
 430                 srv_set_io_thread_op_info(segment, "simulated aio handle");
 431
 432                 ret = os_aio_simulated_handle(segment, &fil_node,
 433 -                                             &message, &type);
 434 +                                             &message, &type, &space_id);
 435 +       }
 436 +
 437 +       /* if the table space was already deleted, fil_node might not exist already. */
 438 +       if (message
 439 +           && space_id < SRV_LOG_SPACE_FIRST_ID
 440 +           && ((buf_page_t*)message)->space_was_being_deleted) {
 441 +
 442 +               /* intended not to be uncompress read page */
 443 +               ut_a(buf_page_get_io_fix(message) == BUF_IO_WRITE
 444 +                    || !buf_page_get_zip_size(message)
 445 +                    || buf_page_get_state(message) != BUF_BLOCK_FILE_PAGE);
 446 +
 447 +               srv_set_io_thread_op_info(segment, "complete io for buf page");
 448 +               buf_page_io_complete(message);
 449 +               return;
 450         }
 451
 452         ut_a(ret);
 453 --- a/storage/innobase/handler/ha_innodb.cc
 454 +++ b/storage/innobase/handler/ha_innodb.cc
 455 @@ -12189,6 +12189,12 @@
 456    "except for the deletion.",
 457    NULL, NULL, 0, &corrupt_table_action_typelib);
 458
 459 +static MYSQL_SYSVAR_ULONG(lazy_drop_table, srv_lazy_drop_table,
 460 +  PLUGIN_VAR_RQCMDARG,
 461 +  "At deleting tablespace, only miminum needed processes at the time are done. "
 462 +  "e.g. for http://bugs.mysql.com/51325",
 463 +  NULL, NULL, 0, 0, 1, 0);
 464 +
 465  static struct st_mysql_sys_var* innobase_system_variables[]= {
 466    MYSQL_SYSVAR(page_size),
 467    MYSQL_SYSVAR(log_block_size),
 468 @@ -12286,6 +12292,7 @@
 469    MYSQL_SYSVAR(flush_checkpoint_debug),
 470  #endif
 471    MYSQL_SYSVAR(corrupt_table_action),
 472 +  MYSQL_SYSVAR(lazy_drop_table),
 473    NULL
 474  };
 475
 476 @@ -12295,7 +12302,7 @@
 477    &innobase_storage_engine,
 478    innobase_hton_name,
 479    plugin_author,
 480 -  "Supports transactions, row-level locking, and foreign keys",
 481 +  "Percona-XtraDB, Supports transactions, row-level locking, and foreign keys",
 482    PLUGIN_LICENSE_GPL,
 483    innobase_init, /* Plugin Init */
 484    NULL, /* Plugin Deinit */
 485 --- a/storage/innobase/include/buf0buf.h
 486 +++ b/storage/innobase/include/buf0buf.h
 487 @@ -1475,6 +1475,7 @@
 488                                         0 if the block was never accessed
 489                                         in the buffer pool */
 490         /* @} */
 491 +       ibool           space_was_being_deleted;
 492         ibool           is_corrupt;
 493  # if defined UNIV_DEBUG_FILE_ACCESSES || defined UNIV_DEBUG
 494         ibool           file_page_was_freed;
 495 --- a/storage/innobase/include/buf0buf.ic
 496 +++ b/storage/innobase/include/buf0buf.ic
 497 @@ -426,6 +426,7 @@
 498         buf_block_set_state(block, BUF_BLOCK_FILE_PAGE);
 499         block->page.space = space;
 500         block->page.offset = page_no;
 501 +       block->page.space_was_being_deleted = FALSE;
 502  }
 503
 504  /*********************************************************************//**
 505 --- a/storage/innobase/include/buf0lru.h
 506 +++ b/storage/innobase/include/buf0lru.h
 507 @@ -73,6 +73,14 @@
 508  buf_LRU_invalidate_tablespace(
 509  /*==========================*/
 510         ulint   id);    /*!< in: space id */
 511 +
 512 +/******************************************************************//**
 513 +*/
 514 +UNIV_INTERN
 515 +void
 516 +buf_LRU_mark_space_was_deleted(
 517 +/*===========================*/
 518 +       ulint   id);    /*!< in: space id */
 519  #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
 520  /********************************************************************//**
 521  Insert a compressed block into buf_pool->zip_clean in the LRU order. */
 522 --- a/storage/innobase/include/os0file.h
 523 +++ b/storage/innobase/include/os0file.h
 524 @@ -280,9 +280,9 @@
 525         pfs_os_file_close_func(file, __FILE__, __LINE__)
 526
 527  # define os_aio(type, mode, name, file, buf, offset, offset_high,      \
 528 -               n, message1, message2, trx)                             \
 529 +               n, message1, message2, space_id, trx)                   \
 530         pfs_os_aio_func(type, mode, name, file, buf, offset,            \
 531 -                       offset_high, n, message1, message2, trx,        \
 532 +                       offset_high, n, message1, message2, space_id, trx,\
 533                         __FILE__, __LINE__)
 534
 535  # define os_file_read(file, buf, offset, offset_high, n)               \
 536 @@ -326,9 +326,9 @@
 537  # define os_file_close(file)   os_file_close_func(file)
 538
 539  # define os_aio(type, mode, name, file, buf, offset, offset_high,      \
 540 -              n, message1, message2, trx)                              \
 541 +              n, message1, message2, space_id, trx)                    \
 542         os_aio_func(type, mode, name, file, buf, offset, offset_high, n,\
 543 -                   message1, message2, trx)
 544 +                   message1, message2, space_id, trx)
 545
 546  # define os_file_read(file, buf, offset, offset_high, n)               \
 547         os_file_read_func(file, buf, offset, offset_high, n, NULL)
 548 @@ -757,6 +757,7 @@
 549                                 (can be used to identify a completed
 550                                 aio operation); ignored if mode is
 551                                  OS_AIO_SYNC */
 552 +       ulint           space_id,
 553         trx_t*          trx,
 554         const char*     src_file,/*!< in: file name where func invoked */
 555         ulint           src_line);/*!< in: line where the func invoked */
 556 @@ -1065,6 +1066,7 @@
 557                                 (can be used to identify a completed
 558                                 aio operation); ignored if mode is
 559                                 OS_AIO_SYNC */
 560 +       ulint           space_id,
 561         trx_t*          trx);
 562  /************************************************************************//**
 563  Wakes up all async i/o threads so that they know to exit themselves in
 564 @@ -1125,7 +1127,8 @@
 565                                 parameters are valid and can be used to
 566                                 restart the operation, for example */
 567         void**  message2,
 568 -       ulint*  type);          /*!< out: OS_FILE_WRITE or ..._READ */
 569 +       ulint*  type,           /*!< out: OS_FILE_WRITE or ..._READ */
 570 +       ulint*  space_id);
 571  #endif
 572
 573  /**********************************************************************//**
 574 @@ -1147,7 +1150,8 @@
 575                                 parameters are valid and can be used to
 576                                 restart the operation, for example */
 577         void**  message2,
 578 -       ulint*  type);          /*!< out: OS_FILE_WRITE or ..._READ */
 579 +       ulint*  type,           /*!< out: OS_FILE_WRITE or ..._READ */
 580 +       ulint*  space_id);
 581  /**********************************************************************//**
 582  Validates the consistency of the aio system.
 583  @return        TRUE if ok */
 584 @@ -1226,7 +1230,8 @@
 585                                 aio operation failed, these output
 586                                 parameters are valid and can be used to
 587                                 restart the operation. */
 588 -       ulint*  type);          /*!< out: OS_FILE_WRITE or ..._READ */
 589 +       ulint*  type,           /*!< out: OS_FILE_WRITE or ..._READ */
 590 +       ulint*  space_id);
 591  #endif /* LINUX_NATIVE_AIO */
 592
 593  #ifndef UNIV_NONINL
 594 --- a/storage/innobase/include/os0file.ic
 595 +++ b/storage/innobase/include/os0file.ic
 596 @@ -229,6 +229,7 @@
 597                                 (can be used to identify a completed
 598                                 aio operation); ignored if mode is
 599                                  OS_AIO_SYNC */
 600 +       ulint           space_id,
 601         trx_t*          trx,
 602         const char*     src_file,/*!< in: file name where func invoked */
 603         ulint           src_line)/*!< in: line where the func invoked */
 604 @@ -245,7 +246,7 @@
 605                                    src_file, src_line);
 606
 607         result = os_aio_func(type, mode, name, file, buf, offset, offset_high,
 608 -                            n, message1, message2, trx);
 609 +                            n, message1, message2, space_id, trx);
 610
 611         register_pfs_file_io_end(locker, n);
 612
 613 --- a/storage/innobase/include/srv0srv.h
 614 +++ b/storage/innobase/include/srv0srv.h
 615 @@ -249,6 +249,8 @@
 616  extern ulint   srv_pass_corrupt_table;
 617
 618  extern ulint   srv_dict_size_limit;
 619 +
 620 +extern ulint   srv_lazy_drop_table;
 621  /*-------------------------------------------*/
 622
 623  extern ulint   srv_n_rows_inserted;
 624 --- a/storage/innobase/include/sync0sync.h
 625 +++ b/storage/innobase/include/sync0sync.h
 626 @@ -689,6 +689,7 @@
 627  #define        SYNC_BUF_POOL           150     /* Buffer pool mutex */
 628  #define        SYNC_BUF_FLUSH_LIST     145     /* Buffer flush list mutex */
 629  #define SYNC_DOUBLEWRITE       140
 630 +#define        SYNC_OUTER_ANY_LATCH    136
 631  #define        SYNC_ANY_LATCH          135
 632  #define        SYNC_MEM_HASH           131
 633  #define        SYNC_MEM_POOL           130
 634 --- a/storage/innobase/include/univ.i
 635 +++ b/storage/innobase/include/univ.i
 636 @@ -53,6 +53,11 @@
 637  #define INNODB_VERSION_MINOR   1
 638  #define INNODB_VERSION_BUGFIX  8
 639
 640 +#ifndef PERCONA_INNODB_VERSION
 641 +#define PERCONA_INNODB_VERSION 20.1
 642 +#endif
 643 +
 644 +
 645  /* The following is the InnoDB version as shown in
 646  SELECT plugin_version FROM information_schema.plugins;
 647  calculated in make_version_string() in sql/sql_show.cc like this:
 648 @@ -65,7 +70,8 @@
 649  #define INNODB_VERSION_STR                     \
 650         IB_TO_STR(INNODB_VERSION_MAJOR) "."     \
 651         IB_TO_STR(INNODB_VERSION_MINOR) "."     \
 652 -       IB_TO_STR(INNODB_VERSION_BUGFIX)
 653 +       IB_TO_STR(INNODB_VERSION_BUGFIX) "-"    \
 654 +       IB_TO_STR(PERCONA_INNODB_VERSION)
 655
 656  #define REFMAN "http://dev.mysql.com/doc/refman/"      \
 657         IB_TO_STR(MYSQL_MAJOR_VERSION) "."              \
 658 --- a/storage/innobase/os/os0file.c
 659 +++ b/storage/innobase/os/os0file.c
 660 @@ -180,6 +180,7 @@
 661                                         made and only the slot message
 662                                         needs to be passed to the caller
 663                                         of os_aio_simulated_handle */
 664 +       ulint           space_id;
 665         fil_node_t*     message1;       /*!< message which is given by the */
 666         void*           message2;       /*!< the requester of an aio operation
 667                                         and which can be used to identify
 668 @@ -3686,7 +3687,8 @@
 669                                 offset */
 670         ulint           offset_high, /*!< in: most significant 32 bits of
 671                                 offset */
 672 -       ulint           len)    /*!< in: length of the block to read or write */
 673 +       ulint           len,    /*!< in: length of the block to read or write */
 674 +       ulint           space_id)
 675  {
 676         os_aio_slot_t*  slot = NULL;
 677  #ifdef WIN_ASYNC_IO
 678 @@ -3775,6 +3777,7 @@
 679         slot->offset   = offset;
 680         slot->offset_high = offset_high;
 681         slot->io_already_done = FALSE;
 682 +       slot->space_id = space_id;
 683
 684  #ifdef WIN_ASYNC_IO
 685         control = &(slot->control);
 686 @@ -4062,6 +4065,7 @@
 687                                 (can be used to identify a completed
 688                                 aio operation); ignored if mode is
 689                                 OS_AIO_SYNC */
 690 +       ulint           space_id,
 691         trx_t*          trx)
 692  {
 693         os_aio_array_t* array;
 694 @@ -4157,7 +4161,7 @@
 695                 trx->io_read += n;
 696         }
 697         slot = os_aio_array_reserve_slot(type, array, message1, message2, file,
 698 -                                        name, buf, offset, offset_high, n);
 699 +                                        name, buf, offset, offset_high, n, space_id);
 700         if (type == OS_FILE_READ) {
 701                 if (srv_use_native_aio) {
 702                         os_n_file_reads++;
 703 @@ -4276,7 +4280,8 @@
 704                                 parameters are valid and can be used to
 705                                 restart the operation, for example */
 706         void**  message2,
 707 -       ulint*  type)           /*!< out: OS_FILE_WRITE or ..._READ */
 708 +       ulint*  type,           /*!< out: OS_FILE_WRITE or ..._READ */
 709 +       ulint*  space_id)
 710  {
 711         ulint           orig_seg        = segment;
 712         os_aio_array_t* array;
 713 @@ -4355,6 +4360,7 @@
 714         *message2 = slot->message2;
 715
 716         *type = slot->type;
 717 +       *space_id = slot->space_id;
 718
 719         if (ret && len == slot->len) {
 720                 ret_val = TRUE;
 721 @@ -4583,7 +4589,8 @@
 722                                 aio operation failed, these output
 723                                 parameters are valid and can be used to
 724                                 restart the operation. */
 725 -       ulint*  type)           /*!< out: OS_FILE_WRITE or ..._READ */
 726 +       ulint*  type,           /*!< out: OS_FILE_WRITE or ..._READ */
 727 +       ulint*  space_id)
 728  {
 729         ulint           segment;
 730         os_aio_array_t* array;
 731 @@ -4656,6 +4663,7 @@
 732         *message2 = slot->message2;
 733
 734         *type = slot->type;
 735 +       *space_id = slot->space_id;
 736
 737         if ((slot->ret == 0) && (slot->n_bytes == (long)slot->len)) {
 738                 ret = TRUE;
 739 @@ -4709,7 +4717,8 @@
 740                                 parameters are valid and can be used to
 741                                 restart the operation, for example */
 742         void**  message2,
 743 -       ulint*  type)           /*!< out: OS_FILE_WRITE or ..._READ */
 744 +       ulint*  type,           /*!< out: OS_FILE_WRITE or ..._READ */
 745 +       ulint*  space_id)
 746  {
 747         os_aio_array_t* array;
 748         ulint           segment;
 749 @@ -5005,6 +5014,7 @@
 750         *message2 = slot->message2;
 751
 752         *type = slot->type;
 753 +       *space_id = slot->space_id;
 754
 755         os_mutex_exit(array->mutex);
 756
 757 --- a/storage/innobase/row/row0mysql.c
 758 +++ b/storage/innobase/row/row0mysql.c
 759 @@ -51,6 +51,7 @@
 760  #include "btr0sea.h"
 761  #include "fil0fil.h"
 762  #include "ibuf0ibuf.h"
 763 +#include "ha_prototypes.h"
 764
 765  /** Provide optional 4.x backwards compatibility for 5.0 and above */
 766  UNIV_INTERN ibool      row_rollback_on_timeout = FALSE;
 767 @@ -1192,6 +1193,13 @@
 768
 769         thr = que_fork_get_first_thr(prebuilt->ins_graph);
 770
 771 +       if (!prebuilt->mysql_has_locked && !(prebuilt->table->flags & (DICT_TF2_TEMPORARY << DICT_TF2_SHIFT))) {
 772 +               fprintf(stderr, "InnoDB: Error: row_insert_for_mysql is called without ha_innobase::external_lock()\n");
 773 +               if (trx->mysql_thd != NULL) {
 774 +                       innobase_mysql_print_thd(stderr, trx->mysql_thd, 600);
 775 +               }
 776 +       }
 777 +
 778         if (prebuilt->sql_stat_start) {
 779                 node->state = INS_NODE_SET_IX_LOCK;
 780                 prebuilt->sql_stat_start = FALSE;
 781 @@ -2576,10 +2584,29 @@
 782
 783                         err = DB_ERROR;
 784                 } else {
 785 +                       dict_index_t*   index;
 786 +
 787                         /* Set the flag which tells that now it is legal to
 788                         IMPORT a tablespace for this table */
 789                         table->tablespace_discarded = TRUE;
 790                         table->ibd_file_missing = TRUE;
 791 +
 792 +                       /* check adaptive hash entries */
 793 +                       index = dict_table_get_first_index(table);
 794 +                       while (index) {
 795 +                               ulint ref_count = btr_search_info_get_ref_count(index->search_info);
 796 +                               if (ref_count) {
 797 +                                       fprintf(stderr, "InnoDB: Warning:"
 798 +                                               " hash index ref_count (%lu) is not zero"
 799 +                                               " after fil_discard_tablespace().\n"
 800 +                                               "index: \"%s\""
 801 +                                               " table: \"%s\"\n",
 802 +                                               ref_count,
 803 +                                               index->name,
 804 +                                               table->name);
 805 +                               }
 806 +                               index = dict_table_get_next_index(index);
 807 +                       }
 808                 }
 809         }
 810
 811 @@ -2928,6 +2955,19 @@
 812                         table->space = space;
 813                         index = dict_table_get_first_index(table);
 814                         do {
 815 +                               ulint ref_count = btr_search_info_get_ref_count(index->search_info);
 816 +                               /* check adaptive hash entries */
 817 +                               if (ref_count) {
 818 +                                       fprintf(stderr, "InnoDB: Warning:"
 819 +                                               " hash index ref_count (%lu) is not zero"
 820 +                                               " after fil_discard_tablespace().\n"
 821 +                                               "index: \"%s\""
 822 +                                               " table: \"%s\"\n",
 823 +                                               ref_count,
 824 +                                               index->name,
 825 +                                               table->name);
 826 +                               }
 827 +
 828                                 index->space = space;
 829                                 index = dict_table_get_next_index(index);
 830                         } while (index);
 831 --- a/storage/innobase/row/row0sel.c
 832 +++ b/storage/innobase/row/row0sel.c
 833 @@ -3418,6 +3418,7 @@
 834         ulint           offsets_[REC_OFFS_NORMAL_SIZE];
 835         ulint*          offsets                         = offsets_;
 836         ibool           table_lock_waited               = FALSE;
 837 +       ibool           problematic_use                 = FALSE;
 838
 839         rec_offs_init(offsets_);
 840
 841 @@ -3796,6 +3797,17 @@
 842
 843         /* Do some start-of-statement preparations */
 844
 845 +       if (!prebuilt->mysql_has_locked) {
 846 +               if (!(prebuilt->table->flags & (DICT_TF2_TEMPORARY << DICT_TF2_SHIFT))) {
 847 +                       fprintf(stderr, "InnoDB: Error: row_search_for_mysql() is called without ha_innobase::external_lock()\n");
 848 +                       if (trx->mysql_thd != NULL) {
 849 +                               innobase_mysql_print_thd(stderr, trx->mysql_thd, 600);
 850 +                       }
 851 +               }
 852 +               problematic_use = TRUE;
 853 +       }
 854 +retry_check:
 855 +
 856         if (!prebuilt->sql_stat_start) {
 857                 /* No need to set an intention lock or assign a read view */
 858
 859 @@ -3806,6 +3818,18 @@
 860                               " perform a consistent read\n"
 861                               "InnoDB: but the read view is not assigned!\n",
 862                               stderr);
 863 +                       if (problematic_use) {
 864 +                               fprintf(stderr, "InnoDB: It may be caused by calling "
 865 +                                               "without ha_innobase::external_lock()\n"
 866 +                                               "InnoDB: For the first-aid, avoiding the crash. "
 867 +                                               "But it should be fixed ASAP.\n");
 868 +                               if (prebuilt->table->flags & (DICT_TF2_TEMPORARY << DICT_TF2_SHIFT)
 869 +                                   && trx->mysql_thd != NULL) {
 870 +                                       innobase_mysql_print_thd(stderr, trx->mysql_thd, 600);
 871 +                               }
 872 +                               prebuilt->sql_stat_start = TRUE;
 873 +                               goto retry_check;
 874 +                       }
 875                         trx_print(stderr, trx, 600);
 876                         fputc('\n', stderr);
 877                         ut_error;
 878 --- a/storage/innobase/srv/srv0srv.c
 879 +++ b/storage/innobase/srv/srv0srv.c
 880 @@ -447,6 +447,8 @@
 881  UNIV_INTERN ulint      srv_pass_corrupt_table = 0; /* 0:disable 1:enable */
 882
 883  UNIV_INTERN ulint      srv_dict_size_limit = 0;
 884 +
 885 +UNIV_INTERN ulint      srv_lazy_drop_table = 0;
 886  /*-------------------------------------------*/
 887  UNIV_INTERN ulong      srv_n_spin_wait_rounds  = 30;
 888  UNIV_INTERN ulong      srv_n_free_tickets_to_enter = 500;
 889 --- a/storage/innobase/srv/srv0start.c
 890 +++ b/storage/innobase/srv/srv0start.c
 891 @@ -2187,7 +2187,7 @@
 892         if (srv_print_verbose_log) {
 893                 ut_print_timestamp(stderr);
 894                 fprintf(stderr,
 895 -                       " InnoDB: %s started; "
 896 +                       " Percona XtraDB (http://www.percona.com) %s started; "
 897                         "log sequence number %llu\n",
 898                         INNODB_VERSION_STR, srv_start_lsn);
 899         }
 900 --- a/storage/innobase/sync/sync0sync.c
 901 +++ b/storage/innobase/sync/sync0sync.c
 902 @@ -1219,6 +1219,7 @@
 903         case SYNC_LOG:
 904         case SYNC_LOG_FLUSH_ORDER:
 905         case SYNC_ANY_LATCH:
 906 +       case SYNC_OUTER_ANY_LATCH:
 907         case SYNC_FILE_FORMAT_TAG:
 908         case SYNC_DOUBLEWRITE:
 909         case SYNC_SEARCH_SYS:
 910 --- a/storage/innobase/trx/trx0purge.c
 911 +++ b/storage/innobase/trx/trx0purge.c
 912 @@ -1149,8 +1149,7 @@
 913         /* If we cannot advance the 'purge view' because of an old
 914         'consistent read view', then the DML statements cannot be delayed.
 915         Also, srv_max_purge_lag <= 0 means 'infinity'. */
 916 -       if (srv_max_purge_lag > 0
 917 -           && !UT_LIST_GET_LAST(trx_sys->view_list)) {
 918 +       if (srv_max_purge_lag > 0) {
 919                 float   ratio = (float) trx_sys->rseg_history_len
 920                         / srv_max_purge_lag;
 921                 if (ratio > ULINT_MAX / 10000) {