From mboxrd@z Thu Jan 1 00:00:00 1970 From: Chao Yu Subject: Re: [PATCH RFC v4] f2fs: flush cp pack except cp pack 2 page at first Date: Thu, 1 Feb 2018 21:56:38 +0800 Message-ID: <349d782c-15e7-2460-fc4e-54d4d76f29da@kernel.org> References: <9047C53C18267742AB12E43B65C7F9F70BCE500E@dggemi505-mbx.china.huawei.com> <20180131222835.GE12901@jaegeuk-macbookpro.roam.corp.google.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Received: from sfi-mx-1.v28.ch3.sourceforge.com ([172.29.28.191] helo=mx.sourceforge.net) by sfs-ml-1.v29.ch3.sourceforge.com with esmtps (TLSv1.2:ECDHE-RSA-AES256-GCM-SHA384:256) (Exim 4.89) (envelope-from ) id 1ehFMQ-0007oO-SH for linux-f2fs-devel@lists.sourceforge.net; Thu, 01 Feb 2018 13:57:06 +0000 Received: from mail.kernel.org ([198.145.29.99]) by sfi-mx-1.v28.ch3.sourceforge.com with esmtps (TLSv1.2:ECDHE-RSA-AES256-GCM-SHA384:256) (Exim 4.89) id 1ehFMP-0007Hr-LW for linux-f2fs-devel@lists.sourceforge.net; Thu, 01 Feb 2018 13:57:06 +0000 In-Reply-To: <20180131222835.GE12901@jaegeuk-macbookpro.roam.corp.google.com> Content-Language: en-US List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: linux-f2fs-devel-bounces@lists.sourceforge.net To: Jaegeuk Kim , Chao Yu Cc: "linux-f2fs-devel@lists.sourceforge.net" , heyunlei , "linux-fsdevel@vger.kernel.org" , hutj , "Duwei (Device OS)" On 2018/2/1 6:28, Jaegeuk Kim wrote: > On 01/31, Chao Yu wrote: >> On 2018/1/31 14:39, Gaoxiang (OS) wrote: >>> Previously, we attempt to flush the whole cp pack in a single bio, >>> however, when suddenly powering off at this time, we could get into >>> an extreme scenario that cp pack 1 page and cp pack 2 page are updated >>> and latest, but payload or current summaries are still partially >>> outdated. (see reliable write in the UFS specification) >>> >>> This patch submits the whole cp pack except cp pack 2 page at first, >>> and then writes the cp pack 2 page with an extra independent >>> bio with pre-io barrier. >>> >>> Signed-off-by: Gao Xiang >>> Reviewed-by: Chao Yu >>> --- >>> Change log from v3: >>> - further review comments are applied from Jaegeuk and Chao >>> - Tested on this patch (without multiple-device): mount, boot Android with f2fs userdata and make fragment >>> - If any problem with this patch or I miss something, please kindly share your comments, thanks :) >>> Change log from v2: >>> - Apply the review comments from Chao >>> Change log from v1: >>> - Apply the review comments from Chao >>> - time data from "finish block_ops" to " finish checkpoint" (tested on ARM64 with TOSHIBA 128GB UFS): >>> Before patch: 0.002273 0.001973 0.002789 0.005159 0.002050 >>> After patch: 0.002502 0.001624 0.002487 0.003049 0.002696 >>> fs/f2fs/checkpoint.c | 67 ++++++++++++++++++++++++++++++++++++---------------- >>> 1 file changed, 46 insertions(+), 21 deletions(-) >>> >>> diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c >>> index 14d2fed..916dc72 100644 >>> --- a/fs/f2fs/checkpoint.c >>> +++ b/fs/f2fs/checkpoint.c >>> @@ -1158,6 +1158,39 @@ static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc) >>> spin_unlock_irqrestore(&sbi->cp_lock, flags); >>> } >>> >>> +static void commit_checkpoint(struct f2fs_sb_info *sbi, >>> + void *src, block_t blk_addr) >>> +{ >>> + struct writeback_control wbc = { >>> + .for_reclaim = 0, >>> + }; >>> + >>> + /* >>> + * pagevec_lookup_tag and lock_page again will take >>> + * some extra time. Therefore, update_meta_pages and >>> + * sync_meta_pages are combined in this function. >>> + */ >>> + struct page *page = grab_meta_page(sbi, blk_addr); >>> + int err; >>> + >>> + memcpy(page_address(page), src, PAGE_SIZE); >>> + set_page_dirty(page); >>> + >>> + f2fs_wait_on_page_writeback(page, META, true); >>> + f2fs_bug_on(sbi, PageWriteback(page)); >>> + if (unlikely(!clear_page_dirty_for_io(page))) >>> + f2fs_bug_on(sbi, 1); >>> + >>> + /* writeout cp pack 2 page */ >>> + err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO); >>> + f2fs_bug_on(sbi, err); >>> + >>> + f2fs_put_page(page, 0); >>> + >>> + /* submit checkpoint (with barrier if NOBARRIER is not set) */ >>> + f2fs_submit_merged_write(sbi, META_FLUSH); >>> +} >>> + >>> static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) >>> { >>> struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi); >>> @@ -1260,16 +1293,6 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) >>> } >>> } >>> >>> - /* need to wait for end_io results */ >>> - wait_on_all_pages_writeback(sbi); >>> - if (unlikely(f2fs_cp_error(sbi))) >>> - return -EIO; >>> - >>> - /* flush all device cache */ >>> - err = f2fs_flush_device_cache(sbi); >>> - if (err) >>> - return err; >>> - >>> /* write out checkpoint buffer at block 0 */ >>> update_meta_page(sbi, ckpt, start_blk++); >>> >>> @@ -1297,15 +1320,6 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) >>> start_blk += NR_CURSEG_NODE_TYPE; >>> } >>> >>> - /* writeout checkpoint block */ >>> - update_meta_page(sbi, ckpt, start_blk); >>> - >>> - /* wait for previous submitted node/meta pages writeback */ >>> - wait_on_all_pages_writeback(sbi); >>> - >>> - if (unlikely(f2fs_cp_error(sbi))) >>> - return -EIO; >>> - >>> filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX); >>> filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX); > > - remove Agreed. > >>> >>> @@ -1313,12 +1327,23 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc) >>> sbi->last_valid_block_count = sbi->total_valid_block_count; >>> percpu_counter_set(&sbi->alloc_valid_block_count, 0); >>> >>> - /* Here, we only have one bio having CP pack */ >>> - sync_meta_pages(sbi, META_FLUSH, LONG_MAX, FS_CP_META_IO); >>> + /* Here, we have one bio having CP pack except cp pack 2 page */ >>> + sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO); >>> + >>> + /* flush all device cache */ >>> + err = f2fs_flush_device_cache(sbi); >>> + if (err) >>> + return err; >>> >>> /* wait for previous submitted meta pages writeback */ >>> wait_on_all_pages_writeback(sbi); >> >> Move f2fs_flush_device_cache here? since meta area can cross the multiple >> devices, we should make sure all metadata were in device cache at least, and >> then trigger the flush. > > Agreed, and need to flush, only if we have multiple devices. f2fs_flush_device_cache is designed as only flushing devices except the first one when f2fs enables multiple device. Calling it directly will be OK. :) > >> >>> >>> + if (unlikely(f2fs_cp_error(sbi))) >>> + return -EIO; >>> + >>> + /* barrier and flush checkpoint cp pack 2 page if it can */ >>> + commit_checkpoint(sbi, ckpt, start_blk); >> >> Jaegeuk, are we really allow to make critical do_checkpoint which is on path of >> fsync()/sync() be asynchronous? > > Yeah, so we need to wait end_io on synchronous paths like f2fs_sync_fs(1). I think we should consider the case Xiang mentioned: 1. write async checkpoint #1 from gc; 2. write sync checkpoint #2 from sync_fs, end_io finished w/ error; -> cp #2 becomes corrupted 3. checkpoint #1's end_io finished w/ error; -> cp #1 becomes corrupted Then, entire filesystem becomes corrupted now. Thanks, > >> >> Thanks, >> >>> + >>> release_ino_entry(sbi, false); >>> >>> if (unlikely(f2fs_cp_error(sbi))) >>> ------------------------------------------------------------------------------ Check out the vibrant tech community on one of the world's most engaging tech sites, Slashdot.org! http://sdm.link/slashdot