When doing 10+ concurrent dd's, I observed very bumpy commits submission (partly because the dd's are started at the same time, and hence reached 4MB to-commit pages at the same time). Basically we rely on the server to complete and return write/commit requests, and want both to progress smoothly and not consume too many pages. The write request wait queue is not enough as it's mainly network bounded. So add another commit request wait queue. Only async writes need to sleep on this queue. cc: Trond Myklebust Signed-off-by: Wu Fengguang --- fs/nfs/client.c | 1 fs/nfs/write.c | 51 ++++++++++++++++++++++++++++++------ include/linux/nfs_fs_sb.h | 2 + 3 files changed, 46 insertions(+), 8 deletions(-) --- linux-next.orig/fs/nfs/write.c 2010-12-13 21:46:21.000000000 +0800 +++ linux-next/fs/nfs/write.c 2010-12-13 21:46:21.000000000 +0800 @@ -516,7 +516,7 @@ nfs_mark_request_commit(struct nfs_page } static int -nfs_clear_request_commit(struct nfs_page *req) +nfs_clear_request_commit(struct inode *inode, struct nfs_page *req) { struct page *page = req->wb_page; @@ -554,7 +554,7 @@ nfs_mark_request_commit(struct nfs_page } static inline int -nfs_clear_request_commit(struct nfs_page *req) +nfs_clear_request_commit(struct inode *inode, struct nfs_page *req) { return 0; } @@ -599,8 +599,10 @@ nfs_scan_commit(struct inode *inode, str return 0; ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT); - if (ret > 0) + if (ret > 0) { nfsi->ncommit -= ret; + atomic_long_add(ret, &NFS_SERVER(inode)->in_commit); + } if (nfs_need_commit(NFS_I(inode))) __mark_inode_dirty(inode, I_DIRTY_DATASYNC); return ret; @@ -668,7 +670,7 @@ static struct nfs_page *nfs_try_to_updat spin_lock(&inode->i_lock); } - if (nfs_clear_request_commit(req) && + if (nfs_clear_request_commit(inode, req) && radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) NFS_I(inode)->ncommit--; @@ -1271,6 +1273,34 @@ int nfs_writeback_done(struct rpc_task * #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) +static void nfs_commit_wait(struct nfs_server *nfss) +{ + long limit = nfs_congestion_kb >> (PAGE_SHIFT - 10); + DEFINE_WAIT(wait); + + if (atomic_long_read(&nfss->in_commit) < limit) + return; + + for (;;) { + prepare_to_wait(&nfss->in_commit_wait, &wait, + TASK_UNINTERRUPTIBLE); + if (atomic_long_read(&nfss->in_commit) < limit) + break; + + io_schedule(); + } + finish_wait(&nfss->in_commit_wait, &wait); +} + +static void nfs_commit_wakeup(struct nfs_server *nfss) +{ + long limit = nfs_congestion_kb >> (PAGE_SHIFT - 10); + + if (atomic_long_read(&nfss->in_commit) < limit - limit / 8 && + waitqueue_active(&nfss->in_commit_wait)) + wake_up(&nfss->in_commit_wait); +} + static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) { if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) @@ -1376,6 +1406,7 @@ nfs_commit_list(struct inode *inode, str req = nfs_list_entry(head->next); nfs_list_remove_request(req); nfs_mark_request_commit(req); + atomic_long_dec(&NFS_SERVER(inode)->in_commit); dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); dec_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); @@ -1409,7 +1440,8 @@ static void nfs_commit_release(void *cal while (!list_empty(&data->pages)) { req = nfs_list_entry(data->pages.next); nfs_list_remove_request(req); - nfs_clear_request_commit(req); + nfs_clear_request_commit(data->inode, req); + atomic_long_dec(&NFS_SERVER(data->inode)->in_commit); dprintk("NFS: commit (%s/%lld %d@%lld)", req->wb_context->path.dentry->d_inode->i_sb->s_id, @@ -1438,6 +1470,7 @@ static void nfs_commit_release(void *cal nfs_clear_page_tag_locked(req); } nfs_commit_clear_lock(NFS_I(data->inode)); + nfs_commit_wakeup(NFS_SERVER(data->inode)); nfs_commitdata_release(calldata); } @@ -1452,11 +1485,13 @@ static const struct rpc_call_ops nfs_com int nfs_commit_inode(struct inode *inode, int how) { LIST_HEAD(head); - int may_wait = how & FLUSH_SYNC; + int sync = how & FLUSH_SYNC; int res = 0; - if (!nfs_commit_set_lock(NFS_I(inode), may_wait)) + if (!nfs_commit_set_lock(NFS_I(inode), sync)) goto out_mark_dirty; + if (!sync) + nfs_commit_wait(NFS_SERVER(inode)); spin_lock(&inode->i_lock); res = nfs_scan_commit(inode, &head, 0, 0); spin_unlock(&inode->i_lock); @@ -1464,7 +1499,7 @@ int nfs_commit_inode(struct inode *inode int error = nfs_commit_list(inode, &head, how); if (error < 0) return error; - if (may_wait) + if (sync) wait_on_bit(&NFS_I(inode)->flags, NFS_INO_COMMIT, nfs_wait_bit_killable, TASK_KILLABLE); --- linux-next.orig/include/linux/nfs_fs_sb.h 2010-12-13 21:46:21.000000000 +0800 +++ linux-next/include/linux/nfs_fs_sb.h 2010-12-13 21:46:21.000000000 +0800 @@ -107,6 +107,8 @@ struct nfs_server { struct backing_dev_info backing_dev_info; atomic_long_t writeback; /* number of writeback pages */ wait_queue_head_t writeback_wait[2]; + atomic_long_t in_commit; /* number of in-commit pages */ + wait_queue_head_t in_commit_wait; int flags; /* various flags */ unsigned int caps; /* server capabilities */ unsigned int rsize; /* read size */ --- linux-next.orig/fs/nfs/client.c 2010-12-13 21:46:21.000000000 +0800 +++ linux-next/fs/nfs/client.c 2010-12-13 21:46:21.000000000 +0800 @@ -1008,6 +1008,7 @@ static struct nfs_server *nfs_alloc_serv atomic_set(&server->active, 0); init_waitqueue_head(&server->writeback_wait[BLK_RW_SYNC]); init_waitqueue_head(&server->writeback_wait[BLK_RW_ASYNC]); + init_waitqueue_head(&server->in_commit_wait); server->io_stats = nfs_alloc_iostats(); if (!server->io_stats) { -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/