*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*
* lustre/llite/rw.c
*
unsigned long pages,
unsigned long pages_min)
{
- struct ll_ra_info *ra = &sbi->ll_ra_info;
- long ret;
+ struct ll_ra_info *ra = &sbi->ll_ra_info;
+ long ret;
+
ENTRY;
- /* If read-ahead pages left are less than 1M, do not do read-ahead,
- * otherwise it will form small read RPC(< 1M), which hurt server
- * performance a lot. */
+ /**
+ * Don't try readahead agreesively if we are limited
+ * LRU pages, otherwise, it could cause deadlock.
+ */
+ pages = min(sbi->ll_cache->ccc_lru_max >> 2, pages);
+
+ /*
+ * If read-ahead pages left are less than 1M, do not do read-ahead,
+ * otherwise it will form small read RPC(< 1M), which hurt server
+ * performance a lot.
+ */
ret = min(ra->ra_max_pages - atomic_read(&ra->ra_cur_pages),
pages);
- if (ret < 0 || ret < min_t(long, PTLRPC_MAX_BRW_PAGES, pages))
- GOTO(out, ret = 0);
+ if (ret < 0 || ret < min_t(long, PTLRPC_MAX_BRW_PAGES, pages))
+ GOTO(out, ret = 0);
if (atomic_add_return(ret, &ra->ra_cur_pages) > ra->ra_max_pages) {
atomic_sub(ret, &ra->ra_cur_pages);
cl_page_assume(env, io, page);
vpg = cl2vvp_page(cl_object_page_slice(clob, page));
if (!vpg->vpg_defer_uptodate && !PageUptodate(vmpage)) {
- vpg->vpg_defer_uptodate = 1;
- vpg->vpg_ra_used = 0;
+ if (hint == MAYNEED) {
+ vpg->vpg_defer_uptodate = 1;
+ vpg->vpg_ra_used = 0;
+ }
cl_page_list_add(queue, page);
} else {
/* skip completed pages */
struct cl_object *clob;
int ret = 0;
__u64 kms;
+ struct ll_sb_info *sbi;
+ struct ll_ra_info *ra;
+
ENTRY;
+ ENTRY;
+
clob = io->ci_obj;
inode = vvp_object_inode(clob);
+ sbi = ll_i2sbi(inode);
+ ra = &sbi->ll_ra_info;
+
+ /**
+ * In case we have a limited max_cached_mb, readahead
+ * should be stopped if it have run out of all LRU slots.
+ */
+ if (atomic_read(&ra->ra_cur_pages) >= sbi->ll_cache->ccc_lru_max) {
+ ll_ra_stats_inc(inode, RA_STAT_MAX_IN_FLIGHT);
+ RETURN(0);
+ }
memset(ria, 0, sizeof(*ria));
ret = ll_readahead_file_kms(env, io, &kms);
vio->vui_ra_start_idx + vio->vui_ra_pages - 1;
pages_min = vio->vui_ra_start_idx + vio->vui_ra_pages -
ria->ria_start_idx;
+ /**
+ * For performance reason, exceeding @ra_max_pages
+ * are allowed, but this should be limited with RPC
+ * size in case a large block size read issued. Trim
+ * to RPC boundary.
+ */
+ pages_min = min(pages_min, ras->ras_rpc_pages -
+ (ria->ria_start_idx % ras->ras_rpc_pages));
}
/* don't over reserved for mmap range read */
*/
static void ras_update(struct ll_sb_info *sbi, struct inode *inode,
struct ll_readahead_state *ras, pgoff_t index,
- enum ras_update_flags flags)
+ enum ras_update_flags flags, struct cl_io *io)
{
struct ll_ra_info *ra = &sbi->ll_ra_info;
bool hit = flags & LL_RAS_HIT;
if (ras->ras_no_miss_check)
GOTO(out_unlock, 0);
+ if (io && io->ci_rand_read)
+ GOTO(out_unlock, 0);
+
+ if (io && io->ci_seq_read) {
+ if (!hit) {
+ /* to avoid many small read RPC here */
+ ras->ras_window_pages = sbi->ll_ra_info.ra_range_pages;
+ ll_ra_stats_inc_sbi(sbi, RA_STAT_MMAP_RANGE_READ);
+ }
+ goto skip;
+ }
+
if (flags & LL_RAS_MMAP) {
unsigned long ra_pages;
flags |= LL_RAS_HIT;
if (!vio->vui_ra_valid)
flags |= LL_RAS_MMAP;
- ras_update(sbi, inode, ras, vvp_index(vpg), flags);
+ ras_update(sbi, inode, ras, vvp_index(vpg), flags, io);
}
cl_2queue_init(queue);
io_start_index = cl_index(io->ci_obj, io->u.ci_rw.crw_pos);
io_end_index = cl_index(io->ci_obj, io->u.ci_rw.crw_pos +
io->u.ci_rw.crw_count - 1);
- if (ll_readahead_enabled(sbi) && ras) {
+ if (ll_readahead_enabled(sbi) && ras && !io->ci_rand_read) {
pgoff_t skip_index = 0;
if (ras->ras_next_readahead_idx < vvp_index(vpg))
pgoff_t start_idx = ras_align(ras, ras->ras_next_readahead_idx);
pgoff_t end_idx = start_idx + pages - 1;
+ /**
+ * In case we have a limited max_cached_mb, readahead
+ * should be stopped if it have run out of all LRU slots.
+ */
+ if (atomic_read(&ra->ra_cur_pages) >= sbi->ll_cache->ccc_lru_max) {
+ ll_ra_stats_inc(inode, RA_STAT_MAX_IN_FLIGHT);
+ return 0;
+ }
+
throttle = min(ra->ra_async_pages_per_file_threshold,
ra->ra_max_pages_per_file);
/*
/* For fast read, it updates read ahead state only
* if the page is hit in cache because non cache page
* case will be handled by slow read later. */
- ras_update(sbi, inode, ras, vvp_index(vpg), flags);
+ ras_update(sbi, inode, ras, vvp_index(vpg), flags, io);
/* avoid duplicate ras_update() call */
vpg->vpg_ra_updated = 1;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff