From patchwork Thu Jan 7 14:05:28 2016 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Yan, Zheng" X-Patchwork-Id: 7977651 Return-Path: X-Original-To: patchwork-ceph-devel@patchwork.kernel.org Delivered-To: patchwork-parsemail@patchwork2.web.kernel.org Received: from mail.kernel.org (mail.kernel.org [198.145.29.136]) by patchwork2.web.kernel.org (Postfix) with ESMTP id B1219BEEE5 for ; Thu, 7 Jan 2016 14:06:24 +0000 (UTC) Received: from mail.kernel.org (localhost [127.0.0.1]) by mail.kernel.org (Postfix) with ESMTP id 8B532200DE for ; Thu, 7 Jan 2016 14:06:23 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id B632C2017E for ; Thu, 7 Jan 2016 14:06:13 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752735AbcAGOGL (ORCPT ); Thu, 7 Jan 2016 09:06:11 -0500 Received: from mx1.redhat.com ([209.132.183.28]:56693 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752607AbcAGOGK (ORCPT ); Thu, 7 Jan 2016 09:06:10 -0500 Received: from int-mx09.intmail.prod.int.phx2.redhat.com (int-mx09.intmail.prod.int.phx2.redhat.com [10.5.11.22]) by mx1.redhat.com (Postfix) with ESMTPS id 83B90C0C0FBC; Thu, 7 Jan 2016 14:06:10 +0000 (UTC) Received: from localhost.localdomain (vpn1-4-93.pek2.redhat.com [10.72.4.93]) by int-mx09.intmail.prod.int.phx2.redhat.com (8.14.4/8.14.4) with ESMTP id u07E5cWn003807; Thu, 7 Jan 2016 09:06:07 -0500 From: "Yan, Zheng" To: ceph-devel@vger.kernel.org Cc: idryomov@gmail.com, "Yan, Zheng" Subject: [PATCH 6/6] ceph: scattered page writeback Date: Thu, 7 Jan 2016 22:05:28 +0800 Message-Id: <1452175528-20751-7-git-send-email-zyan@redhat.com> In-Reply-To: <1452175528-20751-1-git-send-email-zyan@redhat.com> References: <1452175528-20751-1-git-send-email-zyan@redhat.com> X-Scanned-By: MIMEDefang 2.68 on 10.5.11.22 Sender: ceph-devel-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: ceph-devel@vger.kernel.org X-Spam-Status: No, score=-6.9 required=5.0 tests=BAYES_00, RCVD_IN_DNSWL_HI, RP_MATCHES_RCVD, UNPARSEABLE_RELAY autolearn=unavailable version=3.3.1 X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on mail.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP This patch makes ceph_writepages_start() try using single OSD request to write all dirty pages within a strip. When a nonconsecutive dirty page is found, ceph_writepages_start() tries starting a new write operation to existing OSD request. If it succeeds, it uses the new operation to writeback the dirty page. Signed-off-by: Yan, Zheng --- fs/ceph/addr.c | 209 ++++++++++++++++++++++++++++++++++----------------------- 1 file changed, 125 insertions(+), 84 deletions(-) diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c index c222137..bda7f62 100644 --- a/fs/ceph/addr.c +++ b/fs/ceph/addr.c @@ -606,71 +606,71 @@ static void writepages_finish(struct ceph_osd_request *req, struct inode *inode = req->r_inode; struct ceph_inode_info *ci = ceph_inode(inode); struct ceph_osd_data *osd_data; - unsigned wrote; struct page *page; - int num_pages; - int i; + int num_pages, total_pages = 0; + int i, j; + int rc = req->r_result; struct ceph_snap_context *snapc = req->r_snapc; struct address_space *mapping = inode->i_mapping; - int rc = req->r_result; - u64 bytes = req->r_ops[0].extent.length; struct ceph_fs_client *fsc = ceph_inode_to_client(inode); - long writeback_stat; - unsigned issued = ceph_caps_issued(ci); + bool remove_page; - osd_data = osd_req_op_extent_osd_data(req, 0); - BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); - num_pages = calc_pages_for((u64)osd_data->alignment, - (u64)osd_data->length); - if (rc >= 0) { - /* - * Assume we wrote the pages we originally sent. The - * osd might reply with fewer pages if our writeback - * raced with a truncation and was adjusted at the osd, - * so don't believe the reply. - */ - wrote = num_pages; - } else { - wrote = 0; + + dout("writepages_finish %p rc %d\n", inode, rc); + if (rc < 0) mapping_set_error(mapping, rc); - } - dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n", - inode, rc, bytes, wrote); - /* clean all pages */ - for (i = 0; i < num_pages; i++) { - page = osd_data->pages[i]; - BUG_ON(!page); - WARN_ON(!PageUptodate(page)); + /* + * We lost the cache cap, need to truncate the page before + * it is unlocked, otherwise we'd truncate it later in the + * page truncation thread, possibly losing some data that + * raced its way in + */ + remove_page = !(ceph_caps_issued(ci) & + (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)); - writeback_stat = - atomic_long_dec_return(&fsc->writeback_count); - if (writeback_stat < - CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb)) - clear_bdi_congested(&fsc->backing_dev_info, - BLK_RW_ASYNC); + /* clean all pages */ + for (i = 0; i < req->r_num_ops; i++) { + if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) + break; - ceph_put_snap_context(page_snap_context(page)); - page->private = 0; - ClearPagePrivate(page); - dout("unlocking %d %p\n", i, page); - end_page_writeback(page); + osd_data = osd_req_op_extent_osd_data(req, i); + BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); + num_pages = calc_pages_for((u64)osd_data->alignment, + (u64)osd_data->length); + total_pages += num_pages; + for (j = 0; j < num_pages; j++) { + page = osd_data->pages[j]; + BUG_ON(!page); + WARN_ON(!PageUptodate(page)); + + if (atomic_long_dec_return(&fsc->writeback_count) < + CONGESTION_OFF_THRESH( + fsc->mount_options->congestion_kb)) + clear_bdi_congested(&fsc->backing_dev_info, + BLK_RW_ASYNC); + + ceph_put_snap_context(page_snap_context(page)); + page->private = 0; + ClearPagePrivate(page); + dout("unlocking %p\n", page); + end_page_writeback(page); + + if (remove_page) + generic_error_remove_page(inode->i_mapping, + page); - /* - * We lost the cache cap, need to truncate the page before - * it is unlocked, otherwise we'd truncate it later in the - * page truncation thread, possibly losing some data that - * raced its way in - */ - if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) - generic_error_remove_page(inode->i_mapping, page); + unlock_page(page); + } + dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n", + inode, osd_data->length, rc >= 0 ? num_pages : 0); - unlock_page(page); + ceph_release_pages(osd_data->pages, num_pages); } - dout("%p wrote+cleaned %d pages\n", inode, wrote); - ceph_put_wrbuffer_cap_refs(ci, num_pages, snapc); - ceph_release_pages(osd_data->pages, num_pages); + ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc); + + osd_data = osd_req_op_extent_osd_data(req, 0); if (osd_data->pages_from_pool) mempool_free(osd_data->pages, ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool); @@ -778,17 +778,15 @@ retry: while (!done && index <= end) { unsigned i; int first; - pgoff_t next; - int pvec_pages, locked_pages; - struct page **pages = NULL; + pgoff_t strip_end = 0; + int num_ops = 0, max_ops = 0; + int pvec_pages, locked_pages = 0; + struct page **pages = NULL, **data_pages = NULL; mempool_t *pool = NULL; /* Becomes non-null if mempool used */ struct page *page; int want; - u64 offset, len; - long writeback_stat; + u64 offset = 0, len = 0; - next = 0; - locked_pages = 0; max_pages = max_pages_ever; get_more_pages: @@ -824,8 +822,8 @@ get_more_pages: unlock_page(page); break; } - if (next && (page->index != next)) { - dout("not consecutive %p\n", page); + if (strip_end && (page->index > strip_end)) { + dout("end of strip %p\n", page); unlock_page(page); break; } @@ -875,10 +873,11 @@ get_more_pages: /* prepare async write request */ offset = (u64)page_offset(page); len = wsize; + max_ops = CEPH_OSD_INITIAL_OP; req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino, - offset, &len, 0, - do_sync ? 2 : 1, + offset, &len, + 0, max_ops, CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK, @@ -890,9 +889,8 @@ get_more_pages: break; } - if (do_sync) - osd_req_op_init(req, 1, - CEPH_OSD_OP_STARTSYNC, 0); + strip_end = page->index + + ((len - 1) >> PAGE_CACHE_SHIFT); req->r_callback = writepages_finish; req->r_inode = inode; @@ -905,6 +903,48 @@ get_more_pages: pages = mempool_alloc(pool, GFP_NOFS); BUG_ON(!pages); } + + num_ops = 1; + data_pages = pages; + len = 0; + } else if (page->index != + (offset + len) >> PAGE_CACHE_SHIFT) { + u64 offset_inc; + bool new_op = true; + + if (num_ops + do_sync == CEPH_OSD_MAX_OP) { + new_op = false; + } else if (num_ops + do_sync == max_ops) { + max_ops = num_ops << 1; + max_ops = min(max_ops, CEPH_OSD_MAX_OP); + if (ceph_osdc_extend_request(req, + max_ops, + GFP_NOFS) < 0) + new_op = false; + } + if (!new_op) { + redirty_page_for_writepage(wbc, page); + unlock_page(page); + break; + } + + offset_inc = (u64)page_offset(page) - offset; + osd_req_op_extent_dup_last(req, offset_inc); + + dout("writepages got pages at %llu~%llu\n", + offset, len); + + osd_req_op_extent_update(req, num_ops - 1, len); + osd_req_op_extent_osd_data_pages(req, + num_ops - 1, + data_pages, + len, 0, + !!pool, false); + + num_ops++; + data_pages = pages + locked_pages; + offset = (u64)page_offset(page); + len = 0; } /* note position of first page in pvec */ @@ -913,9 +953,8 @@ get_more_pages: dout("%p will write page %p idx %lu\n", inode, page, page->index); - writeback_stat = - atomic_long_inc_return(&fsc->writeback_count); - if (writeback_stat > CONGESTION_ON_THRESH( + if (atomic_long_inc_return(&fsc->writeback_count) > + CONGESTION_ON_THRESH( fsc->mount_options->congestion_kb)) { set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC); @@ -924,7 +963,7 @@ get_more_pages: set_page_writeback(page); pages[locked_pages] = page; locked_pages++; - next = page->index + 1; + len += PAGE_CACHE_SIZE; } /* did we get anything? */ @@ -952,31 +991,34 @@ get_more_pages: } /* Format the osd request message and submit the write */ - offset = page_offset(pages[0]); - len = (u64)locked_pages << PAGE_CACHE_SHIFT; if (snap_size == -1) { + /* writepages_finish() clears writeback pages + * according to the data length, so make sure + * data length covers all locked pages */ + u64 min_len = len + 1 - PAGE_CACHE_SIZE; len = min(len, (u64)i_size_read(inode) - offset); - /* writepages_finish() clears writeback pages - * according to the data length, so make sure - * data length covers all locked pages */ - len = max(len, 1 + - ((u64)(locked_pages - 1) << PAGE_CACHE_SHIFT)); + len = max(len, min_len); } else { len = min(len, snap_size - offset); } - dout("writepages got %d pages at %llu~%llu\n", - locked_pages, offset, len); + dout("writepages got pages at %llu~%llu\n", offset, len); - osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, - !!pool, false); + osd_req_op_extent_osd_data_pages(req, num_ops - 1, data_pages, + len, 0, !!pool, false); + osd_req_op_extent_update(req, num_ops - 1, len); + if (do_sync) { + osd_req_op_init(req, num_ops, CEPH_OSD_OP_STARTSYNC, 0); + num_ops++; + } + BUG_ON(num_ops > max_ops); + + index = pages[locked_pages - 1]->index + 1; pages = NULL; /* request message now owns the pages array */ pool = NULL; /* Update the write op length in case we changed it */ - osd_req_op_extent_update(req, 0, len); - vino = ceph_vino(inode); ceph_osdc_build_request(req, offset, snapc, vino.snap, &inode->i_mtime); @@ -986,7 +1028,6 @@ get_more_pages: req = NULL; /* continue? */ - index = next; wbc->nr_to_write -= locked_pages; if (wbc->nr_to_write <= 0) done = 1;