From mboxrd@z Thu Jan 1 00:00:00 1970 From: Chao Yu Subject: Re: [f2fs-dev] [PATCH 1/3] f2fs: turn off atomic writes when deteting abnormal behaviors Date: Fri, 27 Jul 2018 19:44:28 +0800 Message-ID: <65fc88dc-5d38-6323-9ba0-8776bcd6165f@kernel.org> References: <20180715011112.11475-1-jaegeuk@kernel.org> <29151284-ef34-2eb2-1060-7e214fe8834a@huawei.com> <20180723130346.GD19644@jaegeuk-macbookpro.roam.corp.google.com> <20180727091721.GA16155@jaegeuk-macbookpro.roam.corp.google.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20180727091721.GA16155@jaegeuk-macbookpro.roam.corp.google.com> Content-Language: en-US Sender: linux-kernel-owner@vger.kernel.org To: Jaegeuk Kim Cc: Chao Yu , linux-kernel@vger.kernel.org, linux-f2fs-devel@lists.sourceforge.net List-Id: linux-f2fs-devel.lists.sourceforge.net On 2018/7/27 17:17, Jaegeuk Kim wrote: > On 07/23, Chao Yu wrote: >> On 2018/7/23 21:03, Jaegeuk Kim wrote: >>> On 07/16, Chao Yu wrote: >>>> On 2018/7/15 9:11, Jaegeuk Kim wrote: >>>>> In order to prevent abusing atomic writes by abnormal users, we've added a >>>>> threshold, 20% over memory footprint, which disallows further atomic writes. >>>>> Previously, however, SQLite doesn't know the files became normal, so that >>>>> it could write stale data and commit on revoked normal database file. >>>>> >>>>> Once f2fs detects such the abnormal behavior, this patch simply disables >>>>> all the atomic operations such as: >>>>> - write_begin() gives EINVAL to avoid stale data writes, and SQLite will call >>>>> F2FS_IOC_ABORT_VOLATILE_WRITE to notify aborting the transaction, >>>>> - F2FS_IOC_START_ATOMIC_WRITE gives EINVAL for SQLite to fall back normal >>>>> journal_mode, >>>>> - F2FS_IOC_COMMIT_ATOMIC_WRITE gives EINVAL, if the file was revoked, so that >>>>> Framework retries to submit the transaction given propagated SQLite error. >>>>> >>>>> Note that, this patch also turns off atomic operations, if foreground GC tries >>>>> to move atomic files too frequently. >>>> >>>> Well, how about just keeping original implementation: shutdown atomic write for >>>> those files which are really affect fggc? Since intention of the original >>>> behavior is targeting to abnormal atomic write usage, e.g. open atomic_write >>>> file for very long time, then fggc will be blocked each time when moving its >>>> block. So shutdown it, fggc will recover. >>> >>> The point here is stopping sqlite to keep going wrong data writes even after >>> we already revoked blocks. >> >> Yes, that's correct, what I mean is that if we can do that with smaller >> granularity like just revoke blocks for file which is blocking fggc, it will >> affect system/sqlite flow much less than forcing closing all atomic_write. > > How about this? Yes, it looks good to me. :) > > From 64d2becb82a496c2e2c04abeed42efa3b401ee20 Mon Sep 17 00:00:00 2001 > From: Jaegeuk Kim > Date: Fri, 27 Jul 2018 18:15:11 +0900 > Subject: [PATCH] f2fs: don't allow any writes on aborted atomic writes > > In order to prevent abusing atomic writes by abnormal users, we've added a > threshold, 20% over memory footprint, which disallows further atomic writes. > Previously, however, SQLite doesn't know the files became normal, so that > it could write stale data and commit on revoked normal database file. > > Once f2fs detects such the abnormal behavior, this patch tries to avoid further > writes in write_begin(). > > Signed-off-by: Jaegeuk Kim Reviewed-by: Chao Yu Thanks, > --- > fs/f2fs/data.c | 3 ++- > fs/f2fs/file.c | 7 ++++++- > 2 files changed, 8 insertions(+), 2 deletions(-) > > diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c > index 22dd00c6e241..02ec2603725f 100644 > --- a/fs/f2fs/data.c > +++ b/fs/f2fs/data.c > @@ -2295,7 +2295,8 @@ static int f2fs_write_begin(struct file *file, struct address_space *mapping, > trace_f2fs_write_begin(inode, pos, len, flags); > > if (f2fs_is_atomic_file(inode) && > - !f2fs_available_free_memory(sbi, INMEM_PAGES)) { > + (is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST) || > + !f2fs_available_free_memory(sbi, INMEM_PAGES))) { > err = -ENOMEM; > drop_atomic = true; > goto fail; > diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c > index ff2cb8fb6934..c2c47f3248c4 100644 > --- a/fs/f2fs/file.c > +++ b/fs/f2fs/file.c > @@ -1708,8 +1708,11 @@ static int f2fs_ioc_start_atomic_write(struct file *filp) > > down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]); > > - if (f2fs_is_atomic_file(inode)) > + if (f2fs_is_atomic_file(inode)) { > + if (is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST)) > + ret = -EINVAL; > goto out; > + } > > ret = f2fs_convert_inline_inode(inode); > if (ret) > @@ -1871,6 +1874,8 @@ static int f2fs_ioc_abort_volatile_write(struct file *filp) > ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true); > } > > + clear_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST); > + > inode_unlock(inode); > > mnt_drop_write_file(filp); >