/********************************************************************//**
Test and dump block metadata to dump file if necessary. */
UNIV_INTERN
void
fc_flush_test_and_dump_blkmeta(
/*==========================*/
ulint last_time) /*!< in: the last time when
dump the block metadata */
{
ulint curr_time = 0;
curr_time = ut_time_ms();
if (((curr_time - FLASH_CACHE_DUMP_BLOCK_META_PERIOD) > last_time)) {
/*FIXME: seal with a function*/
flash_cache_mutex_enter();
rw_lock_s_lock(&fc->hash_rwlock);
fc_dump();
rw_lock_s_unlock(&fc->hash_rwlock);
flash_cache_log_mutex_enter();
fc_log_update(FALSE, FLASH_CACHE_LOG_UPDATE_DUMP);
fc_log_update_commit_status();
flash_cache_mutex_exit();
fc_log_commit();
flash_cache_log_mutex_exit();
srv_fc_flush_last_dump = ut_time_ms();
}
}
/*******************************************************************//**
Issue a shared/read lock on the tables cache. */
UNIV_INTERN
void
trx_i_s_cache_start_read(
/*=====================*/
trx_i_s_cache_t* cache) /*!< in: cache */
{
rw_lock_s_lock(&cache->rw_lock);
}
/*************************************************************************
Checks if possible foreign key constraints hold after a delete of the record
under pcur. NOTE that this function will temporarily commit mtr and lose
pcur position! */
static
ulint
row_upd_check_references_constraints(
/*=================================*/
/* out: DB_SUCCESS, DB_LOCK_WAIT, or an error
code */
btr_pcur_t* pcur, /* in: cursor positioned on a record; NOTE: the
cursor position is lost in this function! */
dict_table_t* table, /* in: table in question */
dict_index_t* index, /* in: index of the cursor */
que_thr_t* thr, /* in: query thread */
mtr_t* mtr) /* in: mtr */
{
dict_foreign_t* foreign;
mem_heap_t* heap;
dtuple_t* entry;
rec_t* rec;
ulint err;
rec = btr_pcur_get_rec(pcur);
heap = mem_heap_create(500);
entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec, heap);
mtr_commit(mtr);
mtr_start(mtr);
rw_lock_s_lock(&dict_foreign_key_check_lock);
foreign = UT_LIST_GET_FIRST(table->referenced_list);
while (foreign) {
if (foreign->referenced_index == index) {
err = row_ins_check_foreign_constraint(FALSE, foreign,
table, index, entry, thr);
if (err != DB_SUCCESS) {
rw_lock_s_unlock(&dict_foreign_key_check_lock);
mem_heap_free(heap);
return(err);
}
}
foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
}
rw_lock_s_unlock(&dict_foreign_key_check_lock);
mem_heap_free(heap);
return(DB_SUCCESS);
}
/*****************************************************************//**
Returns the value of ref_count. The value is protected by
btr_search_latch.
@return ref_count value. */
UNIV_INTERN
ulint
btr_search_info_get_ref_count(
/*==========================*/
btr_search_t* info) /*!< in: search info. */
{
ulint ret;
ut_ad(info);
#ifdef UNIV_SYNC_DEBUG
ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_SHARED));
ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_EX));
#endif /* UNIV_SYNC_DEBUG */
rw_lock_s_lock(&btr_search_latch);
ret = info->ref_count;
rw_lock_s_unlock(&btr_search_latch);
return(ret);
}
/********************************************************************//**
Applies linear read-ahead if in the buf_pool the page is a border page of
a linear read-ahead area and all the pages in the area have been accessed.
Does not read any page if the read-ahead mechanism is not activated. Note
that the algorithm looks at the 'natural' adjacent successor and
predecessor of the page, which on the leaf level of a B-tree are the next
and previous page in the chain of leaves. To know these, the page specified
in (space, offset) must already be present in the buf_pool. Thus, the
natural way to use this function is to call it when a page in the buf_pool
is accessed the first time, calling this function just after it has been
bufferfixed.
NOTE 1: as this function looks at the natural predecessor and successor
fields on the page, what happens, if these are not initialized to any
sensible value? No problem, before applying read-ahead we check that the
area to read is within the span of the space, if not, read-ahead is not
applied. An uninitialized value may result in a useless read operation, but
only very improbably.
NOTE 2: the calling thread may own latches on pages: to avoid deadlocks this
function must be written such that it cannot end up waiting for these
latches!
NOTE 3: the calling thread must want access to the page given: this rule is
set to prevent unintended read-aheads performed by ibuf routines, a situation
which could result in a deadlock if the OS does not support asynchronous io.
@return number of page read requests issued */
UNIV_INTERN
ulint
buf_read_ahead_linear(
/*==================*/
ulint space, /*!< in: space id */
ulint zip_size, /*!< in: compressed page size in bytes, or 0 */
ulint offset, /*!< in: page number; see NOTE 3 above */
ibool inside_ibuf, /*!< in: TRUE if we are inside ibuf routine */
trx_t* trx)
{
buf_pool_t* buf_pool = buf_pool_get(space, offset);
ib_int64_t tablespace_version;
buf_page_t* bpage;
buf_frame_t* frame;
buf_page_t* pred_bpage = NULL;
ulint pred_offset;
ulint succ_offset;
ulint count;
int asc_or_desc;
ulint new_offset;
ulint fail_count;
ulint ibuf_mode;
ulint low, high;
ulint err;
ulint i;
const ulint buf_read_ahead_linear_area
= BUF_READ_AHEAD_AREA(buf_pool);
ulint threshold;
if (!(srv_read_ahead & 2)) {
return(0);
}
if (UNIV_UNLIKELY(srv_startup_is_before_trx_rollback_phase)) {
/* No read-ahead to avoid thread deadlocks */
return(0);
}
low = (offset / buf_read_ahead_linear_area)
* buf_read_ahead_linear_area;
high = (offset / buf_read_ahead_linear_area + 1)
* buf_read_ahead_linear_area;
if ((offset != low) && (offset != high - 1)) {
/* This is not a border page of the area: return */
return(0);
}
if (ibuf_bitmap_page(zip_size, offset)
|| trx_sys_hdr_page(space, offset)) {
/* If it is an ibuf bitmap page or trx sys hdr, we do
no read-ahead, as that could break the ibuf page access
order */
return(0);
}
/* Remember the tablespace version before we ask te tablespace size
below: if DISCARD + IMPORT changes the actual .ibd file meanwhile, we
do not try to read outside the bounds of the tablespace! */
tablespace_version = fil_space_get_version(space);
buf_pool_mutex_enter(buf_pool);
if (high > fil_space_get_size(space)) {
buf_pool_mutex_exit(buf_pool);
/* The area is not whole, return */
return(0);
}
if (buf_pool->n_pend_reads
> buf_pool->curr_size / BUF_READ_AHEAD_PEND_LIMIT) {
buf_pool_mutex_exit(buf_pool);
return(0);
}
buf_pool_mutex_exit(buf_pool);
/* Check that almost all pages in the area have been accessed; if
offset == low, the accesses must be in a descending order, otherwise,
in an ascending order. */
asc_or_desc = 1;
if (offset == low) {
asc_or_desc = -1;
}
/* How many out of order accessed pages can we ignore
when working out the access pattern for linear readahead */
threshold = ut_min((64 - srv_read_ahead_threshold),
BUF_READ_AHEAD_AREA(buf_pool));
fail_count = 0;
rw_lock_s_lock(&buf_pool->page_hash_latch);
for (i = low; i < high; i++) {
bpage = buf_page_hash_get(buf_pool, space, i);
if (bpage == NULL || !buf_page_is_accessed(bpage)) {
/* Not accessed */
fail_count++;
} else if (pred_bpage) {
/* Note that buf_page_is_accessed() returns
the time of the first access. If some blocks
of the extent existed in the buffer pool at
the time of a linear access pattern, the first
access times may be nonmonotonic, even though
the latest access times were linear. The
threshold (srv_read_ahead_factor) should help
a little against this. */
int res = ut_ulint_cmp(
buf_page_is_accessed(bpage),
buf_page_is_accessed(pred_bpage));
/* Accesses not in the right order */
if (res != 0 && res != asc_or_desc) {
fail_count++;
}
}
if (fail_count > threshold) {
/* Too many failures: return */
//buf_pool_mutex_exit(buf_pool);
rw_lock_s_unlock(&buf_pool->page_hash_latch);
return(0);
}
if (bpage && buf_page_is_accessed(bpage)) {
pred_bpage = bpage;
}
}
/* If we got this far, we know that enough pages in the area have
been accessed in the right order: linear read-ahead can be sensible */
bpage = buf_page_hash_get(buf_pool, space, offset);
if (bpage == NULL) {
//buf_pool_mutex_exit(buf_pool);
rw_lock_s_unlock(&buf_pool->page_hash_latch);
return(0);
}
switch (buf_page_get_state(bpage)) {
case BUF_BLOCK_ZIP_PAGE:
frame = bpage->zip.data;
break;
case BUF_BLOCK_FILE_PAGE:
frame = ((buf_block_t*) bpage)->frame;
break;
default:
ut_error;
break;
}
/* Read the natural predecessor and successor page addresses from
the page; NOTE that because the calling thread may have an x-latch
on the page, we do not acquire an s-latch on the page, this is to
prevent deadlocks. Even if we read values which are nonsense, the
algorithm will work. */
pred_offset = fil_page_get_prev(frame);
succ_offset = fil_page_get_next(frame);
//buf_pool_mutex_exit(buf_pool);
rw_lock_s_unlock(&buf_pool->page_hash_latch);
if ((offset == low) && (succ_offset == offset + 1)) {
/* This is ok, we can continue */
new_offset = pred_offset;
} else if ((offset == high - 1) && (pred_offset == offset - 1)) {
/* This is ok, we can continue */
new_offset = succ_offset;
} else {
/* Successor or predecessor not in the right order */
return(0);
}
low = (new_offset / buf_read_ahead_linear_area)
* buf_read_ahead_linear_area;
high = (new_offset / buf_read_ahead_linear_area + 1)
* buf_read_ahead_linear_area;
if ((new_offset != low) && (new_offset != high - 1)) {
/* This is not a border page of the area: return */
return(0);
}
if (high > fil_space_get_size(space)) {
/* The area is not whole, return */
return(0);
}
/* If we got this far, read-ahead can be sensible: do it */
ibuf_mode = inside_ibuf
? BUF_READ_IBUF_PAGES_ONLY | OS_AIO_SIMULATED_WAKE_LATER
: BUF_READ_ANY_PAGE | OS_AIO_SIMULATED_WAKE_LATER;
count = 0;
/* Since Windows XP seems to schedule the i/o handler thread
very eagerly, and consequently it does not wait for the
full read batch to be posted, we use special heuristics here */
os_aio_simulated_put_read_threads_to_sleep();
for (i = low; i < high; i++) {
/* It is only sensible to do read-ahead in the non-sync
aio mode: hence FALSE as the first parameter */
if (!ibuf_bitmap_page(zip_size, i)) {
count += buf_read_page_low(
&err, FALSE,
ibuf_mode,
space, zip_size, FALSE, tablespace_version, i, trx);
if (err == DB_TABLESPACE_DELETED) {
ut_print_timestamp(stderr);
fprintf(stderr,
" InnoDB: Warning: in"
" linear readahead trying to access\n"
"InnoDB: tablespace %lu page %lu,\n"
"InnoDB: but the tablespace does not"
" exist or is just being dropped.\n",
(ulong) space, (ulong) i);
}
}
}
/* In simulated aio we wake the aio handler threads only after
queuing all aio requests, in native aio the following call does
nothing: */
os_aio_simulated_wake_handler_threads();
/* Flush pages from the end of the LRU list if necessary */
buf_flush_free_margin(buf_pool, TRUE);
#ifdef UNIV_DEBUG
if (buf_debug_prints && (count > 0)) {
fprintf(stderr,
"LINEAR read-ahead space %lu offset %lu pages %lu\n",
(ulong) space, (ulong) offset, (ulong) count);
}
#endif /* UNIV_DEBUG */
/* Read ahead is considered one I/O operation for the purpose of
LRU policy decision. */
buf_LRU_stat_inc_io();
buf_pool->stat.n_ra_pages_read += count;
return(count);
}
/*****************************************************************//**
Constructs the last committed version of a clustered index record,
which should be seen by a semi-consistent read.
@return DB_SUCCESS or DB_MISSING_HISTORY */
UNIV_INTERN
ulint
row_vers_build_for_semi_consistent_read(
/*====================================*/
const rec_t* rec, /*!< in: record in a clustered index; the
caller must have a latch on the page; this
latch locks the top of the stack of versions
of this records */
mtr_t* mtr, /*!< in: mtr holding the latch on rec */
dict_index_t* index, /*!< in: the clustered index */
ulint** offsets,/*!< in/out: offsets returned by
rec_get_offsets(rec, index) */
mem_heap_t** offset_heap,/*!< in/out: memory heap from which
the offsets are allocated */
mem_heap_t* in_heap,/*!< in: memory heap from which the memory for
*old_vers is allocated; memory for possible
intermediate versions is allocated and freed
locally within the function */
const rec_t** old_vers)/*!< out: rec, old version, or NULL if the
record does not exist in the view, that is,
it was freshly inserted afterwards */
{
const rec_t* version;
mem_heap_t* heap = NULL;
byte* buf;
ulint err;
trx_id_t rec_trx_id = ut_dulint_zero;
ut_ad(dict_index_is_clust(index));
ut_ad(mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_X_FIX)
|| mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_S_FIX));
#ifdef UNIV_SYNC_DEBUG
ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
#endif /* UNIV_SYNC_DEBUG */
ut_ad(rec_offs_validate(rec, index, *offsets));
rw_lock_s_lock(&(purge_sys->latch));
/* The S-latch on purge_sys prevents the purge view from
changing. Thus, if we have an uncommitted transaction at
this point, then purge cannot remove its undo log even if
the transaction could commit now. */
version = rec;
for (;;) {
trx_t* version_trx;
mem_heap_t* heap2;
rec_t* prev_version;
trx_id_t version_trx_id;
version_trx_id = row_get_rec_trx_id(version, index, *offsets);
if (rec == version) {
rec_trx_id = version_trx_id;
}
mutex_enter(&kernel_mutex);
version_trx = trx_get_on_id(version_trx_id);
if (version_trx
&& (version_trx->conc_state == TRX_COMMITTED_IN_MEMORY
|| version_trx->conc_state == TRX_NOT_STARTED)) {
version_trx = NULL;
}
mutex_exit(&kernel_mutex);
if (!version_trx) {
/* We found a version that belongs to a
committed transaction: return it. */
#if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG
ut_a(!rec_offs_any_null_extern(version, *offsets));
#endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */
if (rec == version) {
*old_vers = rec;
err = DB_SUCCESS;
break;
}
/* We assume that a rolled-back transaction stays in
TRX_ACTIVE state until all the changes have been
rolled back and the transaction is removed from
the global list of transactions. */
if (!ut_dulint_cmp(rec_trx_id, version_trx_id)) {
/* The transaction was committed while
we searched for earlier versions.
Return the current version as a
semi-consistent read. */
version = rec;
*offsets = rec_get_offsets(version,
index, *offsets,
ULINT_UNDEFINED,
offset_heap);
}
buf = mem_heap_alloc(in_heap, rec_offs_size(*offsets));
*old_vers = rec_copy(buf, version, *offsets);
rec_offs_make_valid(*old_vers, index, *offsets);
err = DB_SUCCESS;
break;
}
heap2 = heap;
heap = mem_heap_create(1024);
err = trx_undo_prev_version_build(rec, mtr, version, index,
*offsets, heap,
&prev_version);
if (heap2) {
mem_heap_free(heap2); /* free version */
}
if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
break;
}
if (prev_version == NULL) {
/* It was a freshly inserted version */
*old_vers = NULL;
err = DB_SUCCESS;
break;
}
version = prev_version;
*offsets = rec_get_offsets(version, index, *offsets,
ULINT_UNDEFINED, offset_heap);
#if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG
ut_a(!rec_offs_any_null_extern(version, *offsets));
#endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */
}/* for (;;) */
if (heap) {
mem_heap_free(heap);
}
rw_lock_s_unlock(&(purge_sys->latch));
return(err);
}
/*****************************************************************//**
Constructs the version of a clustered index record which a consistent
read should see. We assume that the trx id stored in rec is such that
the consistent read should not see rec in its present version.
@return DB_SUCCESS or DB_MISSING_HISTORY */
UNIV_INTERN
ulint
row_vers_build_for_consistent_read(
/*===============================*/
const rec_t* rec, /*!< in: record in a clustered index; the
caller must have a latch on the page; this
latch locks the top of the stack of versions
of this records */
mtr_t* mtr, /*!< in: mtr holding the latch on rec */
dict_index_t* index, /*!< in: the clustered index */
ulint** offsets,/*!< in/out: offsets returned by
rec_get_offsets(rec, index) */
read_view_t* view, /*!< in: the consistent read view */
mem_heap_t** offset_heap,/*!< in/out: memory heap from which
the offsets are allocated */
mem_heap_t* in_heap,/*!< in: memory heap from which the memory for
*old_vers is allocated; memory for possible
intermediate versions is allocated and freed
locally within the function */
rec_t** old_vers)/*!< out, own: old version, or NULL if the
record does not exist in the view, that is,
it was freshly inserted afterwards */
{
const rec_t* version;
rec_t* prev_version;
trx_id_t trx_id;
mem_heap_t* heap = NULL;
byte* buf;
ulint err;
ut_ad(dict_index_is_clust(index));
ut_ad(mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_X_FIX)
|| mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_S_FIX));
#ifdef UNIV_SYNC_DEBUG
ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
#endif /* UNIV_SYNC_DEBUG */
ut_ad(rec_offs_validate(rec, index, *offsets));
trx_id = row_get_rec_trx_id(rec, index, *offsets);
ut_ad(!read_view_sees_trx_id(view, trx_id));
rw_lock_s_lock(&(purge_sys->latch));
version = rec;
for (;;) {
mem_heap_t* heap2 = heap;
trx_undo_rec_t* undo_rec;
roll_ptr_t roll_ptr;
undo_no_t undo_no;
heap = mem_heap_create(1024);
/* If we have high-granularity consistent read view and
creating transaction of the view is the same as trx_id in
the record we see this record only in the case when
undo_no of the record is < undo_no in the view. */
if (view->type == VIEW_HIGH_GRANULARITY
&& ut_dulint_cmp(view->creator_trx_id, trx_id) == 0) {
roll_ptr = row_get_rec_roll_ptr(version, index,
*offsets);
undo_rec = trx_undo_get_undo_rec_low(roll_ptr, heap);
undo_no = trx_undo_rec_get_undo_no(undo_rec);
mem_heap_empty(heap);
if (ut_dulint_cmp(view->undo_no, undo_no) > 0) {
/* The view already sees this version: we can
copy it to in_heap and return */
#if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG
ut_a(!rec_offs_any_null_extern(
version, *offsets));
#endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */
buf = mem_heap_alloc(in_heap,
rec_offs_size(*offsets));
*old_vers = rec_copy(buf, version, *offsets);
rec_offs_make_valid(*old_vers, index,
*offsets);
err = DB_SUCCESS;
break;
}
}
err = trx_undo_prev_version_build(rec, mtr, version, index,
*offsets, heap,
&prev_version);
if (heap2) {
mem_heap_free(heap2); /* free version */
}
if (err != DB_SUCCESS) {
break;
}
if (prev_version == NULL) {
/* It was a freshly inserted version */
*old_vers = NULL;
err = DB_SUCCESS;
break;
}
*offsets = rec_get_offsets(prev_version, index, *offsets,
ULINT_UNDEFINED, offset_heap);
#if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG
ut_a(!rec_offs_any_null_extern(prev_version, *offsets));
#endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */
trx_id = row_get_rec_trx_id(prev_version, index, *offsets);
if (read_view_sees_trx_id(view, trx_id)) {
/* The view already sees this version: we can copy
it to in_heap and return */
buf = mem_heap_alloc(in_heap, rec_offs_size(*offsets));
*old_vers = rec_copy(buf, prev_version, *offsets);
rec_offs_make_valid(*old_vers, index, *offsets);
err = DB_SUCCESS;
break;
}
version = prev_version;
}/* for (;;) */
mem_heap_free(heap);
rw_lock_s_unlock(&(purge_sys->latch));
return(err);
}
/*************************************************************************
Checks if index currently is mentioned as a referenced index in a foreign
key constraint. This function also loads into the dictionary cache the
possible referencing table. */
static
ibool
row_upd_index_is_referenced(
/*========================*/
/* out: TRUE if referenced; NOTE that since
we do not hold dict_foreign_key_check_lock
when leaving the function, it may be that
the referencing table has been dropped when
we leave this function: this function is only
for heuristic use! */
dict_index_t* index) /* in: index */
{
dict_table_t* table = index->table;
dict_foreign_t* foreign;
ulint phase = 1;
try_again:
if (!UT_LIST_GET_FIRST(table->referenced_list)) {
return(FALSE);
}
if (phase == 2) {
mutex_enter(&(dict_sys->mutex));
}
rw_lock_s_lock(&dict_foreign_key_check_lock);
foreign = UT_LIST_GET_FIRST(table->referenced_list);
while (foreign) {
if (foreign->referenced_index == index) {
if (foreign->foreign_table == NULL) {
if (phase == 2) {
dict_table_get_low(foreign->
foreign_table_name);
} else {
phase = 2;
rw_lock_s_unlock(
&dict_foreign_key_check_lock);
goto try_again;
}
}
rw_lock_s_unlock(&dict_foreign_key_check_lock);
if (phase == 2) {
mutex_exit(&(dict_sys->mutex));
}
return(TRUE);
}
foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
}
rw_lock_s_unlock(&dict_foreign_key_check_lock);
if (phase == 2) {
mutex_exit(&(dict_sys->mutex));
}
return(FALSE);
}
