int lwip_sock_read(void *conn, unsigned char *buff, unsigned long len) {
struct netbuf *new_buf=0;
unsigned char *data;
int data_len=0;
int ret,newret;
SYSCALL_DEBUG(" SOCK read :%x len:%d \n",buff,len);
mutexLock(g_netBH_lock);
ret=netconn_recv(conn, &new_buf);
mutexUnLock(g_netBH_lock);
if (ret!=ERR_OK){
SYSCALL_DEBUG(" Fail to recv data: %x newret:%x(%d) \n",ret,-ret,-ret);
return 0;
}
netbuf_data(new_buf,&data,&data_len);
SYSCALL_DEBUG(" SUCESS to recv data:%d ret:%d\n",data_len,ret);
if (data_len > 0){
ut_memcpy(buff,data,ut_min(data_len,len));
ret = ut_min(data_len,len);
}else{
ret = 0;
}
mutexLock(g_netBH_lock);
netbuf_delete(new_buf);
mutexUnLock(g_netBH_lock);
return ret;
}
int lwip_sock_read_from(void *conn, unsigned char *buff, unsigned long len,struct sockaddr *sockaddr, int addr_len) {
struct netbuf *new_buf=0;
unsigned char *data;
int data_len=0;
int ret=0;
SYSCALL_DEBUG(" SOCK recvfrom :%x len:%d \n",buff,len);
mutexLock(g_netBH_lock);
ret=netconn_recv(conn, &new_buf);
mutexUnLock(g_netBH_lock);
if (ret == ERR_TIMEOUT){
if (g_current_task->killed == 1){
return 0;
}
}
if (ret!=ERR_OK){
SYSCALL_DEBUG(" Fail to recvfrom data: %x newret:%x(%d) \n",ret,-ret,-ret);
return 0;
}
SYSCALL_DEBUG(" SUCESS to recv data:%d \n",ret);
netbuf_data(new_buf,&data,&data_len);
if (data_len > 0){
if (sockaddr != 0){
sockaddr->addr = new_buf->addr.addr;
sockaddr->sin_port = new_buf->port;
}
ut_memcpy(buff,data,ut_min(data_len,len));
ret = ut_min(data_len,len);
}else{
ret =0;
}
mutexLock(g_netBH_lock);
netbuf_delete(new_buf);
mutexUnLock(g_netBH_lock);
return ret;
}
/*************************************************************//**
Creates a hash table with at least n array cells. The actual number
of cells is chosen to be a prime number slightly bigger than n.
@return own: created table */
UNIV_INTERN
hash_table_t*
ha_create_func(
/*===========*/
ulint n, /*!< in: number of array cells */
#ifdef UNIV_SYNC_DEBUG
ulint mutex_level, /*!< in: level of the mutexes in the latching
order: this is used in the debug version */
#endif /* UNIV_SYNC_DEBUG */
ulint n_mutexes) /*!< in: number of mutexes to protect the
hash table: must be a power of 2, or 0 */
{
hash_table_t* table;
#ifndef UNIV_HOTBACKUP
ulint i;
#endif /* !UNIV_HOTBACKUP */
ut_ad(ut_is_2pow(n_mutexes));
table = hash_create(n);
#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
# ifndef UNIV_HOTBACKUP
table->adaptive = TRUE;
# endif /* !UNIV_HOTBACKUP */
#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
/* Creating MEM_HEAP_BTR_SEARCH type heaps can potentially fail,
but in practise it never should in this case, hence the asserts. */
if (n_mutexes == 0) {
table->heap = mem_heap_create_in_btr_search(
ut_min(4096, MEM_MAX_ALLOC_IN_BUF));
ut_a(table->heap);
return(table);
}
#ifndef UNIV_HOTBACKUP
hash_create_mutexes(table, n_mutexes, mutex_level);
table->heaps = mem_alloc(n_mutexes * sizeof(void*));
for (i = 0; i < n_mutexes; i++) {
table->heaps[i] = mem_heap_create_in_btr_search(4096);
ut_a(table->heaps[i]);
}
#endif /* !UNIV_HOTBACKUP */
return(table);
}
/**********************************************************************//**
Like ut_strlcpy, but if src doesn't fit in dst completely, copies the last
(size - 1) bytes of src, not the first.
@return strlen(src) */
UNIV_INTERN
ulint
ut_strlcpy_rev(
/*===========*/
char* dst, /*!< in: destination buffer */
const char* src, /*!< in: source buffer */
ulint size) /*!< in: size of destination buffer */
{
ulint src_size = strlen(src);
if (size != 0) {
ulint n = ut_min(src_size, size - 1);
memcpy(dst, src + src_size - n, n + 1);
}
return(src_size);
}
/*******************************************************************//**
Formats the raw data in "data" (in InnoDB on-disk format) that is of
type DATA_INT using "prtype" and writes the result to "buf".
If the data is in unknown format, then nothing is written to "buf",
0 is returned and "format_in_hex" is set to TRUE, otherwise
"format_in_hex" is left untouched.
Not more than "buf_size" bytes are written to "buf".
The result is always '\0'-terminated (provided buf_size > 0) and the
number of bytes that were written to "buf" is returned (including the
terminating '\0').
@return number of bytes that were written */
static
ulint
row_raw_format_int(
/*===============*/
const char* data, /*!< in: raw data */
ulint data_len, /*!< in: raw data length
in bytes */
ulint prtype, /*!< in: precise type */
char* buf, /*!< out: output buffer */
ulint buf_size, /*!< in: output buffer size
in bytes */
ibool* format_in_hex) /*!< out: should the data be
formated in hex */
{
ulint ret;
if (data_len <= sizeof(ullint)) {
ullint value;
ibool unsigned_type = prtype & DATA_UNSIGNED;
value = mach_read_int_type((const byte*) data,
data_len, unsigned_type);
if (unsigned_type) {
ret = ut_snprintf(buf, buf_size, "%llu",
value) + 1;
} else {
ret = ut_snprintf(buf, buf_size, "%lld",
(long long) value) + 1;
}
} else {
*format_in_hex = TRUE;
ret = 0;
}
return(ut_min(ret, buf_size));
}
/*************************************************************//**
Print a dfield value using ut_print_buf. */
static
void
dfield_print_raw(
/*=============*/
FILE* f, /*!< in: output stream */
const dfield_t* dfield) /*!< in: dfield */
{
ulint len = dfield_get_len(dfield);
if (!dfield_is_null(dfield)) {
ulint print_len = ut_min(len, 1000);
ut_print_buf(f, dfield_get_data(dfield), print_len);
if (len != print_len) {
fprintf(f, "(total %lu bytes%s)",
(ulong) len,
dfield_is_ext(dfield) ? ", external" : "");
}
} else {
fputs(" SQL NULL", f);
}
}
/********************************************************************//**
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);
}
big_rec_t*
dtuple_convert_big_rec(
/*===================*/
/* out, own: created big record vector,
NULL if we are not able to shorten
the entry enough, i.e., if there are
too many short fields in entry */
dict_index_t* index, /* in: index */
dtuple_t* entry, /* in: index entry */
ulint* ext_vec,/* in: array of externally stored fields,
or NULL: if a field already is externally
stored, then we cannot move it to the vector
this function returns */
ulint n_ext_vec)/* in: number of elements is ext_vec */
{
mem_heap_t* heap;
big_rec_t* vector;
dfield_t* dfield;
ulint size;
ulint n_fields;
ulint longest;
ulint longest_i = ULINT_MAX;
ibool is_externally_stored;
ulint i;
ulint j;
ut_a(dtuple_check_typed_no_assert(entry));
size = rec_get_converted_size(index, entry);
if (UNIV_UNLIKELY(size > 1000000000)) {
fprintf(stderr,
"InnoDB: Warning: tuple size very big: %lu\n", (ulong) size);
fputs("InnoDB: Tuple contents: ", stderr);
dtuple_print(stderr, entry);
putc('\n', stderr);
}
heap = mem_heap_create(size + dtuple_get_n_fields(entry)
* sizeof(big_rec_field_t) + 1000);
vector = mem_heap_alloc(heap, sizeof(big_rec_t));
vector->heap = heap;
vector->fields = mem_heap_alloc(heap, dtuple_get_n_fields(entry)
* sizeof(big_rec_field_t));
/* Decide which fields to shorten: the algorithm is to look for
the longest field whose type is DATA_BLOB */
n_fields = 0;
while (rec_get_converted_size(index, entry)
>= ut_min(page_get_free_space_of_empty(
index->table->comp) / 2,
REC_MAX_DATA_SIZE)) {
longest = 0;
for (i = dict_index_get_n_unique_in_tree(index);
i < dtuple_get_n_fields(entry); i++) {
/* Skip over fields which already are externally
stored */
is_externally_stored = FALSE;
if (ext_vec) {
for (j = 0; j < n_ext_vec; j++) {
if (ext_vec[j] == i) {
is_externally_stored = TRUE;
}
}
}
if (!is_externally_stored) {
dfield = dtuple_get_nth_field(entry, i);
if (dfield->len != UNIV_SQL_NULL &&
dfield->len > longest) {
longest = dfield->len;
longest_i = i;
}
}
}
/* We do not store externally fields which are smaller than
DICT_MAX_INDEX_COL_LEN */
ut_a(DICT_MAX_INDEX_COL_LEN > REC_1BYTE_OFFS_LIMIT);
if (longest < BTR_EXTERN_FIELD_REF_SIZE + 10
+ DICT_MAX_INDEX_COL_LEN) {
/* Cannot shorten more */
mem_heap_free(heap);
return(NULL);
}
/* Move data from field longest_i to big rec vector;
we do not let data size of the remaining entry
drop below 128 which is the limit for the 2-byte
offset storage format in a physical record. This
we accomplish by storing 128 bytes of data in entry
itself, and only the remaining part to big rec vec.
We store the first bytes locally to the record. Then
we can calculate all ordering fields in all indexes
from locally stored data. */
dfield = dtuple_get_nth_field(entry, longest_i);
vector->fields[n_fields].field_no = longest_i;
ut_a(dfield->len > DICT_MAX_INDEX_COL_LEN);
vector->fields[n_fields].len = dfield->len
- DICT_MAX_INDEX_COL_LEN;
vector->fields[n_fields].data = mem_heap_alloc(heap,
vector->fields[n_fields].len);
/* Copy data (from the end of field) to big rec vector */
ut_memcpy(vector->fields[n_fields].data,
((byte*)dfield->data) + dfield->len
- vector->fields[n_fields].len,
vector->fields[n_fields].len);
dfield->len = dfield->len - vector->fields[n_fields].len
+ BTR_EXTERN_FIELD_REF_SIZE;
/* Set the extern field reference in dfield to zero */
memset(((byte*)dfield->data)
+ dfield->len - BTR_EXTERN_FIELD_REF_SIZE,
0, BTR_EXTERN_FIELD_REF_SIZE);
n_fields++;
}
vector->n_fields = n_fields;
return(vector);
}
/*******************************************************************//**
Formats the raw data in "data" (in InnoDB on-disk format) using
"dict_field" and writes the result to "buf".
Not more than "buf_size" bytes are written to "buf".
The result is always NUL-terminated (provided buf_size is positive) and the
number of bytes that were written to "buf" is returned (including the
terminating NUL).
@return number of bytes that were written */
UNIV_INTERN
ulint
row_raw_format(
/*===========*/
const char* data, /*!< in: raw data */
ulint data_len, /*!< in: raw data length
in bytes */
const dict_field_t* dict_field, /*!< in: index field */
char* buf, /*!< out: output buffer */
ulint buf_size) /*!< in: output buffer size
in bytes */
{
ulint mtype;
ulint prtype;
ulint ret;
ibool format_in_hex;
if (buf_size == 0) {
return(0);
}
if (data_len == UNIV_SQL_NULL) {
ret = ut_snprintf((char*) buf, buf_size, "NULL") + 1;
return(ut_min(ret, buf_size));
}
mtype = dict_field->col->mtype;
prtype = dict_field->col->prtype;
format_in_hex = FALSE;
switch (mtype) {
case DATA_INT:
ret = row_raw_format_int(data, data_len, prtype,
buf, buf_size, &format_in_hex);
if (format_in_hex) {
goto format_in_hex;
}
break;
case DATA_CHAR:
case DATA_VARCHAR:
case DATA_MYSQL:
case DATA_VARMYSQL:
ret = row_raw_format_str(data, data_len, prtype,
buf, buf_size, &format_in_hex);
if (format_in_hex) {
goto format_in_hex;
}
break;
/* XXX support more data types */
default:
format_in_hex:
if (UNIV_LIKELY(buf_size > 2)) {
memcpy(buf, "0x", 2);
buf += 2;
buf_size -= 2;
ret = 2 + ut_raw_to_hex(data, data_len,
buf, buf_size);
} else {
buf[0] = '\0';
ret = 1;
}
}
return(ret);
}
/********************************************************************//**
Flush pages from flash cache.
@return number of pages have been flushed to tablespace */
UNIV_INTERN
ulint
fc_flush_to_disk(
/*==================*/
ibool do_full_io) /*!< in: whether do full io capacity */
{
ulint distance;
byte* page;
ulint ret;
ulint space;
ulint offset;
ulint page_type;
ulint i, j;
ulint pos;
ulint zip_size;
ulint block_offset, byte_offset;
ulint fc_size = fc_get_size();
ulint fc_blk_size = fc_get_block_size_byte();
ulint start_offset;
ulint data_size;
fc_block_t *flush_block = NULL;
ulint c_flush = 0;
ut_ad(!mutex_own(&fc->mutex));
ut_a(fc->flush_buf->free_pos == 0);
/* step 1: get the number of blocks need to flush to tablespace */
flash_cache_mutex_enter();
distance = fc_get_distance();
start_offset = fc->flush_off;
if ( distance == 0 ) {
flash_cache_mutex_exit();
return 0;
} else if ( recv_recovery_on ) {
if ( distance < (( 1.0 * srv_flash_cache_write_cache_pct /100 ) * fc_size)) {
fc->n_flush_cur = 0;
} else if ( distance < ( ( 1.0*srv_flash_cache_do_full_io_pct /100 ) * fc_size)) {
fc->n_flush_cur = ut_min(PCT_IO_FC(10), distance);
} else {
fc->n_flush_cur = ut_min(PCT_IO_FC(100), distance);
}
} else if ( distance < (( 1.0 * srv_flash_cache_write_cache_pct /100 ) * fc_size)
&& !do_full_io ) {
flash_cache_mutex_exit();
return 0;
} else if ( distance < (( 1.0 * srv_flash_cache_do_full_io_pct/100 ) * fc_size)
&& !do_full_io ) {
fc->n_flush_cur = PCT_IO_FC(srv_fc_write_cache_flush_pct);
} else {
ut_ad((distance > ( 1.0 * srv_flash_cache_do_full_io_pct/100 ) * fc_size)
|| do_full_io );
fc->n_flush_cur = ut_min(PCT_IO_FC(srv_fc_full_flush_pct), distance);
}
flash_cache_mutex_exit();
/* step 2: start to flush blocks use async io, set block io_fix IO_FIX_FLUSH */
i = 0;
while (i < fc->n_flush_cur) {
ulint b_space;
ulint b_offset;
ulint raw_zip_size;
ulint size;
ulint fil_offset;
#ifdef UNIV_FLASH_CACHE_TRACE
ulint is_v4_blk;
#endif
byte* page_io;
flash_cache_mutex_enter();
pos = ( start_offset + i ) % fc_size;
flush_block = fc_get_block(pos);
if (flush_block == NULL) {
i++;
flash_cache_mutex_exit();
continue;
}
/* we should get the mutex, as doublewrite may hit this block and invalid the block */
flash_block_mutex_enter(flush_block->fil_offset);
flash_cache_mutex_exit();
data_size = fc_block_get_data_size(flush_block);
if (flush_block->state != BLOCK_READY_FOR_FLUSH) {
/* if readonly or merge write or already flushed*/
ut_a (flush_block->state == BLOCK_NOT_USED
|| flush_block->state == BLOCK_READ_CACHE
|| flush_block->state == BLOCK_FLUSHED);
i += data_size;
flash_block_mutex_exit(flush_block->fil_offset);
if (flush_block->state == BLOCK_NOT_USED) {
//fc_block_detach(FALSE, flush_block);
fc_block_free(flush_block);
}
continue;
}
zip_size = fil_space_get_zip_size(flush_block->space);
if (zip_size == ULINT_UNDEFINED) {
/* table has been droped, just set it BLOCK_FLUSHED */
#ifdef UNIV_FLASH_CACHE_TRACE
ut_print_timestamp(fc->f_debug);
fprintf(fc->f_debug, "space:%lu is droped, the page(%lu, %lu) need not to be flushed.\n",
(ulong)flush_block->space, (ulong)flush_block->space, (ulong)flush_block->offset);
#endif
flush_block->state = BLOCK_FLUSHED;
i += data_size;
c_flush += data_size;
flash_block_mutex_exit(flush_block->fil_offset);
continue;
}
#ifdef UNIV_FLASH_CACHE_TRACE
if (flush_block->state != BLOCK_READY_FOR_FLUSH) {
fc_block_print(flush_block);
ut_error;
}
#endif
flush_block->io_fix |= IO_FIX_FLUSH;
/*
* we should set block state BLOCK_FLUSHED, if not, doublewrite may hit this block
* and invalid this block and reduce the dirty count, but when finish flush ,we will
* reduce the dirty count too, so it may reduce twice.
*/
flush_block->state = BLOCK_FLUSHED;
/* save the block info, as the block may be invalided by doublewrite after release mutex */
b_space = flush_block->space;
b_offset = flush_block->offset;
raw_zip_size = flush_block->raw_zip_size;
size = flush_block->size;
fil_offset = flush_block->fil_offset;
#ifdef UNIV_FLASH_CACHE_TRACE
is_v4_blk = flush_block->is_v4_blk;
#endif
/* release the block now, so read can hit in this blocks and read the data */
flash_block_mutex_exit(flush_block->fil_offset);
/*
* Only flush thread will update read_buf and flush_off/round.
* there only single flush thread no need to lock read_buf
*/
page = fc->flush_buf->buf + fc->flush_buf->free_pos * fc_blk_size;
if (raw_zip_size > 0) {
ut_a((size * fc_blk_size) == UNIV_PAGE_SIZE);
page_io = fc->flush_zip_read_buf;
} else {
page_io = page;
}
fc_io_offset(fil_offset, &block_offset, &byte_offset);
ret = fil_io(OS_FILE_READ, TRUE, FLASH_CACHE_SPACE, 0,
block_offset, byte_offset, data_size * fc_blk_size,
page_io, NULL);
if (ret != DB_SUCCESS) {
ut_print_timestamp(stderr);
fprintf(stderr, " InnoDB: Flash cache [Error]: unable to read page from flash cache.\n"
"flash cache flush offset is:%lu.\n", (ulong)(start_offset + i));
ut_error;
}
if ((flush_block != NULL) && (flush_block->state == BLOCK_NOT_USED)) {
goto skip;
}
/* decompress the compress data */
if (raw_zip_size > 0) {
#ifdef UNIV_FLASH_CACHE_TRACE
ulint blk_zip_size_byte;
if (is_v4_blk) {
blk_zip_size_byte = raw_zip_size * fc_get_block_size_byte();
} else {
blk_zip_size_byte = fc_block_compress_align(raw_zip_size) * fc_get_block_size_byte();
ut_a((ulint)mach_read_from_4(page_io + FC_ZIP_PAGE_ZIP_RAW_SIZE) == raw_zip_size);
}
ut_a(page_io);
ut_a(page);
ut_a((ulint)mach_read_from_4(page_io + FC_ZIP_PAGE_HEADER) == FC_ZIP_PAGE_CHECKSUM);
ut_a((ulint)mach_read_from_4(page_io + blk_zip_size_byte - FC_ZIP_PAGE_TAILER)
== FC_ZIP_PAGE_CHECKSUM);
ut_a((ulint)mach_read_from_4(page_io + FC_ZIP_PAGE_SIZE) == blk_zip_size_byte);
ut_a((ulint)mach_read_from_4(page_io + FC_ZIP_PAGE_ORIG_SIZE) == UNIV_PAGE_SIZE);
ut_a((ulint)mach_read_from_4(page_io + FC_ZIP_PAGE_SPACE) == b_space);
ut_a((ulint)mach_read_from_4(page_io + FC_ZIP_PAGE_OFFSET) == b_offset);
/* only qlz can do this check */
if (srv_flash_cache_compress_algorithm == FC_BLOCK_COMPRESS_QUICKLZ) {
if (is_v4_blk) {
ut_a(raw_zip_size * fc_get_block_size_byte()
>= (ulint)fc_qlz_size_compressed((const char *)(page_io + FC_ZIP_PAGE_DATA)));
} else {
ut_a(raw_zip_size
== (ulint)fc_qlz_size_compressed((const char *)(page_io + FC_ZIP_PAGE_DATA)));
}
ut_a(UNIV_PAGE_SIZE == fc_qlz_size_decompressed((const char *)(page_io + FC_ZIP_PAGE_DATA)));
}
#endif
fc_block_do_decompress(DECOMPRESS_FLUSH, page_io, raw_zip_size, page);
}
space = mach_read_from_4(page + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID);
offset = mach_read_from_4(page + FIL_PAGE_OFFSET);
if ((space != b_space) || (offset != b_offset)) {
ut_print_timestamp(stderr);
fc_block_print(flush_block);
ut_error;
}
if (buf_page_is_corrupted(page, zip_size)) {
buf_page_print(page, zip_size, BUF_PAGE_PRINT_NO_CRASH);
ut_error;
}
page_type = fil_page_get_type(page);
if (page_type == FIL_PAGE_INDEX) {
page_type = 1;
}
srv_flash_cache_flush_detail[page_type]++;
ret = fil_io(OS_FILE_WRITE | OS_AIO_SIMULATED_WAKE_LATER, FALSE, space,
zip_size, offset, 0, zip_size ? zip_size : UNIV_PAGE_SIZE, page, NULL);
if (ret != DB_SUCCESS && ret != DB_TABLESPACE_DELETED) {
ut_print_timestamp(stderr);
fc_block_print(flush_block);
ut_error;
}
/* add UNIV_PAGE_SIZE / fc_blk_size for safe */
fc->flush_buf->free_pos += UNIV_PAGE_SIZE / fc_blk_size;
skip:
i += data_size;
c_flush += data_size;
if ((fc->flush_buf->free_pos + UNIV_PAGE_SIZE / fc_blk_size) >= fc->flush_buf->size) {
/* FIXME: is it safe to change n_flush, as step 3 will use n_flush */
fc->n_flush_cur = i;
break;
}
}
/* ok, now flush all async io to disk */
fc_flush_sync_dbfile();
/* step 3: all the flush blocks have sync to disk, update the state and io_fix */
j = 0;
while (j < fc->n_flush_cur) {
flash_cache_mutex_enter();
pos = (start_offset + j) % fc_size;
flush_block = fc_get_block(pos);
if (flush_block == NULL) {
j++;
flash_cache_mutex_exit();
continue;
}
/* block state and io_fix may be changed by doublewrite and lru move */
flash_block_mutex_enter(flush_block->fil_offset);
flash_cache_mutex_exit();
if (flush_block->io_fix & IO_FIX_FLUSH) {
/* the block is already in BLOCK_FLUSHED state */
flush_block->io_fix &= ~IO_FIX_FLUSH;
}
data_size = fc_block_get_data_size(flush_block);
flash_block_mutex_exit(flush_block->fil_offset);
j += data_size;
}
/*
* i and j may be different, as the last been flushed block may be invalid by doublewrite,
* so maybe i > j
*/
/* add the actual flushed blocks */
srv_flash_cache_flush = srv_flash_cache_flush + c_flush;
/* step 4: update fc status and flush_off, and wake up threads that are sleep for space */
if (i > 0) {
ut_a(i >= c_flush);
flash_cache_mutex_enter();
/*
* it is safe to inc flush off and sub dirty blocks at this time,
* as fc_validate is not work
*/
fc_inc_flush_off(i);
flash_cache_log_mutex_enter();
fc_log->current_stat->flush_offset = fc->flush_off;
fc_log->current_stat->flush_round = fc->flush_round;
flash_cache_log_mutex_exit();
ut_a(srv_flash_cache_dirty >= c_flush);
srv_flash_cache_dirty -= c_flush;
srv_fc_flush_should_commit_log_flush++;
os_event_set(fc->wait_space_event);
fc->n_flush_cur = 0;
flash_cache_mutex_exit();
}
fc->flush_buf->free_pos = 0;
return c_flush;
}
