/*
* __wt_bm_read --
* Map or read address cookie referenced block into a buffer.
*/
int
__wt_bm_read(WT_BM *bm, WT_SESSION_IMPL *session,
WT_ITEM *buf, const uint8_t *addr, size_t addr_size)
{
WT_BLOCK *block;
wt_off_t offset;
uint32_t cksum, size;
bool mapped;
WT_UNUSED(addr_size);
block = bm->block;
/* Crack the cookie. */
WT_RET(__wt_block_buffer_to_addr(block, addr, &offset, &size, &cksum));
/*
* Map the block if it's possible.
*/
mapped = bm->map != NULL && offset + size <= (wt_off_t)bm->maplen;
if (mapped) {
buf->data = (uint8_t *)bm->map + offset;
buf->size = size;
WT_RET(__wt_mmap_preload(session, buf->data, buf->size));
WT_STAT_FAST_CONN_INCR(session, block_map_read);
WT_STAT_FAST_CONN_INCRV(session, block_byte_map_read, size);
return (0);
}
#ifdef HAVE_DIAGNOSTIC
/*
* In diagnostic mode, verify the block we're about to read isn't on
* the available list, or for live systems, the discard list.
*/
WT_RET(__wt_block_misplaced(
session, block, "read", offset, size, bm->is_live));
#endif
/* Read the block. */
WT_RET(__wt_block_read_off(session, block, buf, offset, size, cksum));
#ifdef HAVE_POSIX_FADVISE
/* Optionally discard blocks from the system's buffer cache. */
if (block->os_cache_max != 0 &&
(block->os_cache += size) > block->os_cache_max) {
WT_DECL_RET;
block->os_cache = 0;
/* Ignore EINVAL - some file systems don't support the flag. */
if ((ret = posix_fadvise(block->fh->fd,
(wt_off_t)0, (wt_off_t)0, POSIX_FADV_DONTNEED)) != 0 &&
ret != EINVAL)
WT_RET_MSG(
session, ret, "%s: posix_fadvise", block->name);
}
#endif
return (0);
}
/*
* __wt_bm_preload --
* Pre-load a page.
*/
int
__wt_bm_preload(
WT_BM *bm, WT_SESSION_IMPL *session, const uint8_t *addr, size_t addr_size)
{
WT_BLOCK *block;
WT_DECL_ITEM(tmp);
WT_DECL_RET;
wt_off_t offset;
uint32_t cksum, size;
bool mapped;
WT_UNUSED(addr_size);
block = bm->block;
/*
* Turn off pre-load when direct I/O is configured for the file,
* the kernel cache isn't interesting.
*/
if (block->fh->direct_io)
return (0);
WT_STAT_FAST_CONN_INCR(session, block_preload);
/* Crack the cookie. */
WT_RET(__wt_block_buffer_to_addr(block, addr, &offset, &size, &cksum));
/* Check for a mapped block. */
mapped = bm->map != NULL && offset + size <= (wt_off_t)bm->maplen;
if (mapped)
return (__wt_mmap_preload(
session, (uint8_t *)bm->map + offset, size));
#ifdef HAVE_POSIX_FADVISE
if (posix_fadvise(block->fh->fd,
(wt_off_t)offset, (wt_off_t)size, POSIX_FADV_WILLNEED) == 0)
return (0);
#endif
WT_RET(__wt_scr_alloc(session, size, &tmp));
ret = __wt_block_read_off(session, block, tmp, offset, size, cksum);
__wt_scr_free(session, &tmp);
return (ret);
}
/*
* __wt_bm_preload --
* Pre-load a page.
*/
int
__wt_bm_preload(WT_BM *bm,
WT_SESSION_IMPL *session, const uint8_t *addr, size_t addr_size)
{
WT_BLOCK *block;
WT_DECL_RET;
off_t offset;
uint32_t cksum, size;
int mapped;
WT_UNUSED(addr_size);
block = bm->block;
ret = EINVAL; /* Play games due to conditional compilation */
/* Crack the cookie. */
WT_RET(__wt_block_buffer_to_addr(block, addr, &offset, &size, &cksum));
/* Check for a mapped block. */
mapped = bm->map != NULL && offset + size <= (off_t)bm->maplen;
if (mapped)
WT_RET(__wt_mmap_preload(
session, (uint8_t *)bm->map + offset, size));
else {
#ifdef HAVE_POSIX_FADVISE
ret = posix_fadvise(block->fh->fd,
(off_t)offset, (off_t)size, POSIX_FADV_WILLNEED);
#endif
if (ret != 0) {
WT_DECL_ITEM(tmp);
WT_RET(__wt_scr_alloc(session, size, &tmp));
ret = __wt_block_read_off(
session, block, tmp, offset, size, cksum);
__wt_scr_free(&tmp);
WT_RET(ret);
}
}
WT_STAT_FAST_CONN_INCR(session, block_preload);
return (0);
}
/*
* __wt_bm_read --
* Map or read address cookie referenced block into a buffer.
*/
int
__wt_bm_read(WT_BM *bm, WT_SESSION_IMPL *session,
WT_ITEM *buf, const uint8_t *addr, uint32_t addr_size)
{
WT_BLOCK *block;
off_t offset;
uint32_t size, cksum;
int mapped;
WT_UNUSED(addr_size);
block = bm->block;
/* Crack the cookie. */
WT_RET(__wt_block_buffer_to_addr(block, addr, &offset, &size, &cksum));
/*
* Clear buffers previously used for mapped memory, we may be forced
* to read into this buffer.
*/
if (F_ISSET(buf, WT_ITEM_MAPPED))
__wt_buf_free(session, buf);
/*
* If we're going to be able to return mapped memory and the buffer
* has allocated memory, discard it.
*/
mapped = bm->map != NULL && offset + size <= (off_t)bm->maplen;
if (buf->mem != NULL && mapped)
__wt_buf_free(session, buf);
/* Map the block if it's possible. */
if (mapped) {
buf->mem = (uint8_t *)bm->map + offset;
buf->memsize = size;
buf->data = buf->mem;
buf->size = size;
F_SET(buf, WT_ITEM_MAPPED);
WT_RET(__wt_mmap_preload(session, buf->mem, buf->size));
WT_CSTAT_INCR(session, block_map_read);
WT_CSTAT_INCRV(session, block_byte_map_read, size);
return (0);
}
/* Read the block. */
WT_RET(__wt_block_read_off(session, block, buf, offset, size, cksum));
#ifdef HAVE_POSIX_FADVISE
/* Optionally discard blocks from the system's buffer cache. */
if (block->os_cache_max != 0 &&
(block->os_cache += size) > block->os_cache_max) {
WT_DECL_RET;
block->os_cache = 0;
if ((ret = posix_fadvise(block->fh->fd,
(off_t)0, (off_t)0, POSIX_FADV_DONTNEED)) != 0)
WT_RET_MSG(
session, ret, "%s: posix_fadvise", block->name);
}
#endif
return (0);
}
