On Thu, 11 Aug 2011, Andreas Dilger wrote:
> On 2011-08-11, at 8:48 AM, Lukas Czerner wrote:
> > If e2fsprogs tools (mke2fs, e2fsck) is run on regular file instead of
> > on block device, we can use punch hole instead of regular discard
> > command which would not work on regular file anyway. This gives us
> > several advantages. First of all when e2fsck is run with '-E discard'
> > parameter it will punch out all ununsed space from the image, hence
> > trimming down the file system image. And secondly, when creating an
> > file system on regular file (with '-E discard' which is default), we
> > can use punch hole to clear the file content, hence we can skip inode
> > table initialization, because reads from sparse area returns zeros. This
> > will result in faster file system creation (without the need to specify
> > lazy_itable_init) and smaller images.
> >
> > This commit also fixes some tests that would wail due to mke2fs showing
> > discard progress, hence the output would differ.
> >
> > Signed-off-by: Lukas Czerner <lczerner@xxxxxxxxxx>
> > ---
> > lib/ext2fs/ext2_io.h | 1 +
> > lib/ext2fs/unix_io.c | 46 ++++++++++++++++++++++++++++++++++----
> > misc/mke2fs.c | 10 ++++++-
> > tests/f_resize_inode/expect | 1 +
> > tests/m_dasd_bs/expect.1 | 1 +
> > tests/m_extent_journal/expect.1 | 1 +
> > tests/m_large_file/expect.1 | 1 +
> > tests/m_meta_bg/expect.1 | 1 +
> > tests/m_no_opt/expect.1 | 1 +
> > tests/m_raid_opt/expect.1 | 1 +
> > tests/m_std/expect.1 | 1 +
> > tests/m_uninit/expect.1 | 1 +
> > 12 files changed, 59 insertions(+), 7 deletions(-)
> >
> > diff --git a/lib/ext2fs/ext2_io.h b/lib/ext2fs/ext2_io.h
> > index e71ada9..bcc2f87 100644
> > --- a/lib/ext2fs/ext2_io.h
> > +++ b/lib/ext2fs/ext2_io.h
> > @@ -30,6 +30,7 @@ typedef struct struct_io_stats *io_stats;
> >
> > #define CHANNEL_FLAGS_WRITETHROUGH 0x01
> > #define CHANNEL_FLAGS_DISCARD_ZEROES 0x02
> > +#define CHANNEL_FLAGS_BLOCK_DEVICE 0x04
> >
> > #define io_channel_discard_zeroes_data(i) (i->flags & CHANNEL_FLAGS_DISCARD_ZEROES)
> >
> > diff --git a/lib/ext2fs/unix_io.c b/lib/ext2fs/unix_io.c
> > index c1d0561..eb5df55 100644
> > --- a/lib/ext2fs/unix_io.c
> > +++ b/lib/ext2fs/unix_io.c
> > @@ -441,7 +441,11 @@ static errcode_t unix_open(const char *name, int flags, io_channel *channel)
> > struct unix_private_data *data = NULL;
> > errcode_t retval;
> > int open_flags, zeroes = 0;
> > +#ifdef HAVE_OPEN64
> > + struct stat64 st;
> > +#else
> > struct stat st;
> > +#endif
> > #ifdef __linux__
> > struct utsname ut;
> > #endif
> > @@ -482,11 +486,26 @@ static errcode_t unix_open(const char *name, int flags, io_channel *channel)
> > #endif
> > data->flags = flags;
> >
> > + st.st_mode = 0;
> > #ifdef HAVE_OPEN64
> > data->dev = open64(io->name, open_flags);
> > + stat64(io->name, &st);
> > #else
> > data->dev = open(io->name, open_flags);
> > + stat(io->name, &st);
> > #endif
>
> IMHO it would be nicer to have wrappers for open() and stat() instead
> of #ifdef spread around the code.
You're right. I'll fix that.
>
> > + /*
> > + * If the device is really a block device, then set the
> > + * appropriate flag, otherwise we can set DISCARD_ZEROES flag
> > + * because we are going to use punch hole instead of discard
> > + * and if it succeed, subsequent read from sparse area returns
> > + * zero.
> > + */
> > + if (S_ISBLK(st.st_mode))
> > + io->flags |= CHANNEL_FLAGS_BLOCK_DEVICE;
> > + else
> > + io->flags |= CHANNEL_FLAGS_DISCARD_ZEROES;
> > +
> > if (data->dev < 0) {
> > retval = errno;
> > goto cleanup;
> > @@ -552,7 +571,11 @@ static errcode_t unix_open(const char *name, int flags, io_channel *channel)
> > (ut.release[2] == '4') && (ut.release[3] == '.') &&
> > (ut.release[4] == '1') && (ut.release[5] >= '0') &&
> > (ut.release[5] < '8')) &&
> > - (fstat(data->dev, &st) == 0) &&
> > +#ifdef HAVE_OPEN64
> > + (stat64(io->name, &st) == 0) &&
> > +#else
> > + (stat(io->name, &st) == 0) &&
> > +#endif
> > (S_ISBLK(st.st_mode))) {
> > struct rlimit rlim;
> >
> > @@ -857,7 +880,9 @@ static errcode_t unix_set_option(io_channel channel, const char *option,
> > }
> >
> > #if defined(__linux__) && !defined(BLKDISCARD)
> > -#define BLKDISCARD _IO(0x12,119)
> > +#define BLKDISCARD _IO(0x12,119)
> > +#define FALLOC_FL_KEEP_SIZE 0x01
> > +#define FALLOC_FL_PUNCH_HOLE 0x02
>
> Should these get their own #ifndef FALLOC_FL_KEEP_SIZE and also
> #ifndef FALLOC_FL_PUNCH_HOLE? Otherwise the compiler may complain
> if they are defined multiple times. FALLOC_FL_PUNCH_HOLE was added
> a lot later than the others.
good point, thanks.
>
> > @@ -872,10 +897,21 @@ static errcode_t unix_discard(io_channel channel, unsigned long long block,
> > data = (struct unix_private_data *) channel->private_data;
> > EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
> >
> > - range[0] = (__uint64_t)(block) * channel->block_size;
> > - range[1] = (__uint64_t)(count) * channel->block_size;
> > + if (channel->flags & CHANNEL_FLAGS_BLOCK_DEVICE) {
> > + range[0] = (__uint64_t)(block) * channel->block_size;
> > + range[1] = (__uint64_t)(count) * channel->block_size;
> >
> > - ret = ioctl(data->dev, BLKDISCARD, &range);
> > + ret = ioctl(data->dev, BLKDISCARD, &range);
> > + } else {
> > + /*
> > + * If we are not on block device, try to use punch hole
> > + * to reclaim free space.
> > + */
> > + ret = fallocate(data->dev,
> > + FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
> > + (off_t)(block) * channel->block_size,
> > + (off_t)(count) * channel->block_size);
> > + }
> > if (ret < 0)
> > return errno;
>
>
> This will fail on kernels/userspace that don't have fallocate() or
> FALLOC_FL_PUNCH_HOLE (which is very new). Maybe it is better to wrap
> this whole thing in "#ifdef FALLOC_FL_PUNCH_HOLE" so that it is only
> compiled on systems that have such support, and convert EOPNOTSUPP into
> EXT2_ET_UNIMPLEMENTED.
Will do.
>
> > diff --git a/tests/f_resize_inode/expect b/tests/f_resize_inode/expect
> > index a396927..cfc699c 100644
> > --- a/tests/f_resize_inode/expect
> > +++ b/tests/f_resize_inode/expect
> > @@ -1,4 +1,5 @@
> > mke2fs -F -O resize_inode -o Linux -b 1024 -g 1024 test.img 16384
> > +Discarding device blocks: 1024/16384 done
>
> What happens on devices/kernels that don't support discard? In newer
> e2fsprogs this message isn't printed at all if the device doesn't
> support discard. It seems better to have the run_e2fsck script
> strip out these lines so that the tests pass regardless of whether
> discard is working or not.
But that's not about discard at all. Since all the tests are done on
image files it will do punch hole instead of discard. But your point
remains valid, if the kernel does not support punch hole or even
fallocate it will not display the information and the check will fail.
Will fix that.
Thanks Andreas!
-Lukas
>
> > Filesystem label=
> > OS type: Linux
> > Block size=1024 (log=0)
> > diff --git a/tests/m_dasd_bs/expect.1 b/tests/m_dasd_bs/expect.1
> > index 31db54b..310fc95 100644
> > --- a/tests/m_dasd_bs/expect.1
> > +++ b/tests/m_dasd_bs/expect.1
> > @@ -1,3 +1,4 @@
> > +Discarding device blocks: 2048/32768 done
> > Filesystem label=
> > OS type: Linux
> > Block size=2048 (log=1)
> > diff --git a/tests/m_extent_journal/expect.1 b/tests/m_extent_journal/expect.1
> > index 88ea2d9..fb07588 100644
> > --- a/tests/m_extent_journal/expect.1
> > +++ b/tests/m_extent_journal/expect.1
> > @@ -1,3 +1,4 @@
> > +Discarding device blocks: 1024/65536 done
> > Filesystem label=
> > OS type: Linux
> > Block size=1024 (log=0)
> > diff --git a/tests/m_large_file/expect.1 b/tests/m_large_file/expect.1
> > index 2b40c84..3ebd03c 100644
> > --- a/tests/m_large_file/expect.1
> > +++ b/tests/m_large_file/expect.1
> > @@ -1,3 +1,4 @@
> > +Discarding device blocks: 4096/16384 done
> > Filesystem label=
> > OS type: Linux
> > Block size=4096 (log=2)
> > diff --git a/tests/m_meta_bg/expect.1 b/tests/m_meta_bg/expect.1
> > index f1c9cef..f947da9 100644
> > --- a/tests/m_meta_bg/expect.1
> > +++ b/tests/m_meta_bg/expect.1
> > @@ -1,3 +1,4 @@
> > +Discarding device blocks: 1024/131072 done
> > Filesystem label=
> > OS type: Linux
> > Block size=1024 (log=0)
> > diff --git a/tests/m_no_opt/expect.1 b/tests/m_no_opt/expect.1
> > index 7f3e5aa..81764d3 100644
> > --- a/tests/m_no_opt/expect.1
> > +++ b/tests/m_no_opt/expect.1
> > @@ -1,3 +1,4 @@
> > +Discarding device blocks: 1024/65536 done
> > Filesystem label=
> > OS type: Linux
> > Block size=1024 (log=0)
> > diff --git a/tests/m_raid_opt/expect.1 b/tests/m_raid_opt/expect.1
> > index 0c6acc1..c8667e8 100644
> > --- a/tests/m_raid_opt/expect.1
> > +++ b/tests/m_raid_opt/expect.1
> > @@ -1,3 +1,4 @@
> > +Discarding device blocks: 1024/131072 done
> > Filesystem label=
> > OS type: Linux
> > Block size=1024 (log=0)
> > diff --git a/tests/m_std/expect.1 b/tests/m_std/expect.1
> > index d0bf27c..a2b6c3f 100644
> > --- a/tests/m_std/expect.1
> > +++ b/tests/m_std/expect.1
> > @@ -1,3 +1,4 @@
> > +Discarding device blocks: 1024/65536 done
> > Filesystem label=
> > OS type: Linux
> > Block size=1024 (log=0)
> > diff --git a/tests/m_uninit/expect.1 b/tests/m_uninit/expect.1
> > index 173c072..72c652c 100644
> > --- a/tests/m_uninit/expect.1
> > +++ b/tests/m_uninit/expect.1
> > @@ -1,3 +1,4 @@
> > +Discarding device blocks: 1024/131072 done
> > Filesystem label=
> > OS type: Linux
> > Block size=1024 (log=0)
> > --
> > 1.7.4.4
> >
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> > the body of a message to majordomo@xxxxxxxxxxxxxxx
> > More majordomo info at http://vger.kernel.org/majordomo-info.html
>
>
> Cheers, Andreas
>
>
>
>
>
>
--
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
