| Message ID | 20241018062638.2619214-2-chao@kernel.org (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [f2fs-dev,1/3] f2fs: zone: introduce first_zoned_segno in f2fs_sb_info | expand |
On 10/18, Chao Yu wrote: > blkzoned feature enables lfs mode by default, for data updates from > regular file, it only allow to use OPU, this patch changes to allow > to use IPU if regular file's data block locates in regular block > device, if data block locates in zoned block device, it still uses > OPU. > > Signed-off-by: Chao Yu <chao@kernel.org> > --- > fs/f2fs/data.c | 12 ++++++++++-- > fs/f2fs/f2fs.h | 5 +++++ > 2 files changed, 15 insertions(+), 2 deletions(-) > > diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c > index 949b6bb957c7..c6d0e4aa12e3 100644 > --- a/fs/f2fs/data.c > +++ b/fs/f2fs/data.c > @@ -2648,8 +2648,16 @@ bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio) > return false; > if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK)) > return true; > - if (f2fs_lfs_mode(sbi)) > - return true; > + if (f2fs_lfs_mode(sbi)) { > + /* no regular block device in mainarea */ > + if (!f2fs_blkzoned_has_regular_section(sbi)) > + return true; > + /* blkaddr locates in sequential zone */ > + if (fio && __is_valid_data_blkaddr(fio->old_blkaddr) && > + GET_SEGNO(sbi, fio->old_blkaddr) >= > + sbi->first_zoned_segno) What if there's another regular device beyond zoned partition? I prefer to keep the entire area as lfs mode to simplify the performance. Do we really need IPU on conventional partition? Why not asking to use file pinning, if someone needs IPU? > + return true; > + } > if (S_ISDIR(inode->i_mode)) > return true; > if (IS_NOQUOTA(inode)) > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h > index 55bcd411257a..6781b4fab43f 100644 > --- a/fs/f2fs/f2fs.h > +++ b/fs/f2fs/f2fs.h > @@ -4790,6 +4790,11 @@ static inline void f2fs_invalidate_internal_cache(struct f2fs_sb_info *sbi, > f2fs_invalidate_compress_page(sbi, blkaddr); > } > > +static inline bool f2fs_blkzoned_has_regular_section(struct f2fs_sb_info *sbi) > +{ > + return f2fs_sb_has_blkzoned(sbi) && sbi->first_zoned_segno; > +} > + > #define EFSBADCRC EBADMSG /* Bad CRC detected */ > #define EFSCORRUPTED EUCLEAN /* Filesystem is corrupted */ > > -- > 2.40.1
On 2024/10/22 7:18, Jaegeuk Kim wrote: > On 10/18, Chao Yu wrote: >> blkzoned feature enables lfs mode by default, for data updates from >> regular file, it only allow to use OPU, this patch changes to allow >> to use IPU if regular file's data block locates in regular block >> device, if data block locates in zoned block device, it still uses >> OPU. >> >> Signed-off-by: Chao Yu <chao@kernel.org> >> --- >> fs/f2fs/data.c | 12 ++++++++++-- >> fs/f2fs/f2fs.h | 5 +++++ >> 2 files changed, 15 insertions(+), 2 deletions(-) >> >> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c >> index 949b6bb957c7..c6d0e4aa12e3 100644 >> --- a/fs/f2fs/data.c >> +++ b/fs/f2fs/data.c >> @@ -2648,8 +2648,16 @@ bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio) >> return false; >> if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK)) >> return true; >> - if (f2fs_lfs_mode(sbi)) >> - return true; >> + if (f2fs_lfs_mode(sbi)) { >> + /* no regular block device in mainarea */ >> + if (!f2fs_blkzoned_has_regular_section(sbi)) >> + return true; >> + /* blkaddr locates in sequential zone */ >> + if (fio && __is_valid_data_blkaddr(fio->old_blkaddr) && >> + GET_SEGNO(sbi, fio->old_blkaddr) >= >> + sbi->first_zoned_segno) > > What if there's another regular device beyond zoned partition? With above logic, it will trigger OPU if data locates in regular device which is beyond zoned partition. > I prefer to keep the entire area as lfs mode to simplify the performance. Oh, what do you think of introducing another option mode=hybrid? so that we can provide an option to allow LFS/IPU/SSR in regular device beside we limit LFS mode in zoned device. > Do we really need IPU on conventional partition? Why not asking to use > file pinning, if someone needs IPU? One concern is size of pinned file should be multiple of section, pined file will cost too much space if section size is large. Another problem is flow of third-party applications can not be controlled by us, so it's not possible to execute pinfile & falloc after file creation. Thanks, > >> + return true; >> + } >> if (S_ISDIR(inode->i_mode)) >> return true; >> if (IS_NOQUOTA(inode)) >> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h >> index 55bcd411257a..6781b4fab43f 100644 >> --- a/fs/f2fs/f2fs.h >> +++ b/fs/f2fs/f2fs.h >> @@ -4790,6 +4790,11 @@ static inline void f2fs_invalidate_internal_cache(struct f2fs_sb_info *sbi, >> f2fs_invalidate_compress_page(sbi, blkaddr); >> } >> >> +static inline bool f2fs_blkzoned_has_regular_section(struct f2fs_sb_info *sbi) >> +{ >> + return f2fs_sb_has_blkzoned(sbi) && sbi->first_zoned_segno; >> +} >> + >> #define EFSBADCRC EBADMSG /* Bad CRC detected */ >> #define EFSCORRUPTED EUCLEAN /* Filesystem is corrupted */ >> >> -- >> 2.40.1
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c index 949b6bb957c7..c6d0e4aa12e3 100644 --- a/fs/f2fs/data.c +++ b/fs/f2fs/data.c @@ -2648,8 +2648,16 @@ bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio) return false; if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK)) return true; - if (f2fs_lfs_mode(sbi)) - return true; + if (f2fs_lfs_mode(sbi)) { + /* no regular block device in mainarea */ + if (!f2fs_blkzoned_has_regular_section(sbi)) + return true; + /* blkaddr locates in sequential zone */ + if (fio && __is_valid_data_blkaddr(fio->old_blkaddr) && + GET_SEGNO(sbi, fio->old_blkaddr) >= + sbi->first_zoned_segno) + return true; + } if (S_ISDIR(inode->i_mode)) return true; if (IS_NOQUOTA(inode)) diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h index 55bcd411257a..6781b4fab43f 100644 --- a/fs/f2fs/f2fs.h +++ b/fs/f2fs/f2fs.h @@ -4790,6 +4790,11 @@ static inline void f2fs_invalidate_internal_cache(struct f2fs_sb_info *sbi, f2fs_invalidate_compress_page(sbi, blkaddr); } +static inline bool f2fs_blkzoned_has_regular_section(struct f2fs_sb_info *sbi) +{ + return f2fs_sb_has_blkzoned(sbi) && sbi->first_zoned_segno; +} + #define EFSBADCRC EBADMSG /* Bad CRC detected */ #define EFSCORRUPTED EUCLEAN /* Filesystem is corrupted */
blkzoned feature enables lfs mode by default, for data updates from regular file, it only allow to use OPU, this patch changes to allow to use IPU if regular file's data block locates in regular block device, if data block locates in zoned block device, it still uses OPU. Signed-off-by: Chao Yu <chao@kernel.org> --- fs/f2fs/data.c | 12 ++++++++++-- fs/f2fs/f2fs.h | 5 +++++ 2 files changed, 15 insertions(+), 2 deletions(-)