On Thu, 11 Aug 2011, Andreas Dilger wrote:
> On 2011-08-11, at 9:10 AM, Lukas Czerner wrote:
> > On Mon, 18 Apr 2011, Lukas Czerner wrote:
> >> Currently concurrent reads/writes are atomic only wrt individual pages,
> >> however are not on the system call. This may cause read() to return data
> >> mixed from several different writes, which I do not think it is good
> >> approach. We might argue that application doing this is broken, but
> >> actually this is something we can easily do on filesystem level without
> >> significant performance issues, so we can be consistent. Also POSIX
> >> mentions this as well and XFS filesystem already has this feature.
> >>
> >> This commit adds new rw_semaphore into ext4_inode structure. We take
> >> read lock every time we read data from a file (via ext4_file_read() or
> >> ext4_file_splice_read()) and also when we write data in direct io mode,
> >> since in this mode application should know exactly what it is doing.
> >> Then we take write lock when we write into a file (via ext4_file_write()
> >> and ext4_file_splice_write()), except the direct io when we take read
> >> lock and unaligned direct io which is already serialized in different
> >> manner. Also we are locking ext4_truncate() as well so we are consistent
> >> and preserve atomicity.
> >>
> >> This should not have any significant performance impact since we still
> >> allow concurrent reads from the same inode and concurrent writes are
> >> serialized already by i_mutex. The only type of load which will feel the
> >> performance hit is the case of writing into an inode while reading from
> >> it and vice versa. In this case, if reads/writes are exclusive it might
> >> not need locking, however tracking this would be expensive.
> >
> > Anyone any thoughts on this one ?
>
> Rather than adding more global locking to the IO path, it would be much
> preferable IMHO to start looking at extent locks for file IO. At that
> point, a reader could get a read lock for the range of its syscall and
> get an atomic read, and other writers could write atomically to different
> parts of the file without contention (to the greatest degree possible).
>
> I wouldn't be surprised if there is already some code in the kernel that
> implements this.
Yes, that is what I was thinking as well. However I just wanted to
present the concept of this to know what people think about this.
Thanks!
-Lukas
>
> >> Signed-off-by: Lukas Czerner <lczerner@xxxxxxxxxx>
> >> ---
> >> fs/ext4/ext4.h | 5 ++++
> >> fs/ext4/file.c | 58 ++++++++++++++++++++++++++++++++++++++++++++++++++----
> >> fs/ext4/inode.c | 7 ++++++
> >> fs/ext4/super.c | 1 +
> >> 4 files changed, 66 insertions(+), 5 deletions(-)
> >>
> >> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> >> index 4daaf2b..037857c 100644
> >> --- a/fs/ext4/ext4.h
> >> +++ b/fs/ext4/ext4.h
> >> @@ -858,6 +858,11 @@ struct ext4_inode_info {
> >> */
> >> tid_t i_sync_tid;
> >> tid_t i_datasync_tid;
> >> +
> >> + /*
> >> + * Semaphore forcing read/write atomicity
> >> + */
> >> + struct rw_semaphore i_rwlock;
> >> };
> >>
> >> /*
> >> diff --git a/fs/ext4/file.c b/fs/ext4/file.c
> >> index 7b80d54..6c7ed94 100644
> >> --- a/fs/ext4/file.c
> >> +++ b/fs/ext4/file.c
> >> @@ -94,7 +94,7 @@ ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
> >> unsigned long nr_segs, loff_t pos)
> >> {
> >> struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
> >> - int unaligned_aio = 0;
> >> + int unaligned_aio = 0, direct_io = 0;
> >> int ret;
> >>
> >> /*
> >> @@ -117,6 +117,7 @@ ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
> >> } else if (unlikely((iocb->ki_filp->f_flags & O_DIRECT) &&
> >> !is_sync_kiocb(iocb))) {
> >> unaligned_aio = ext4_unaligned_aio(inode, iov, nr_segs, pos);
> >> + direct_io = 1;
> >> }
> >>
> >> /* Unaligned direct AIO must be serialized; see comment above */
> >> @@ -131,12 +132,19 @@ ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
> >> inode->i_ino, current->comm);
> >> mutex_lock(ext4_aio_mutex(inode));
> >> ext4_aiodio_wait(inode);
> >> - }
> >> + } else if (unlikely(direct_io))
> >> + down_read(&EXT4_I(inode)->i_rwlock);
> >> + else
> >> + down_write(&EXT4_I(inode)->i_rwlock);
> >>
> >> ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
> >>
> >> if (unaligned_aio)
> >> mutex_unlock(ext4_aio_mutex(inode));
> >> + else if (unlikely(direct_io))
> >> + up_read(&EXT4_I(inode)->i_rwlock);
> >> + else
> >> + up_write(&EXT4_I(inode)->i_rwlock);
> >>
> >> return ret;
> >> }
> >> @@ -252,11 +260,51 @@ loff_t ext4_llseek(struct file *file, loff_t offset, int origin)
> >> return offset;
> >> }
> >>
> >> +static ssize_t
> >> +ext4_file_read(struct kiocb *iocb, const struct iovec *iov,
> >> + unsigned long nr_segs, loff_t pos)
> >> +{
> >> + struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
> >> + ssize_t size;
> >> +
> >> + down_read(&EXT4_I(inode)->i_rwlock);
> >> + size = generic_file_aio_read(iocb, iov, nr_segs, pos);
> >> + up_read(&EXT4_I(inode)->i_rwlock);
> >> + return size;
> >> +}
> >> +
> >> +ssize_t ext4_file_splice_read(struct file *in, loff_t *ppos,
> >> + struct pipe_inode_info *pipe, size_t len,
> >> + unsigned int flags)
> >> +{
> >> + struct inode *inode = in->f_mapping->host;
> >> + ssize_t size;
> >> +
> >> + down_read(&EXT4_I(inode)->i_rwlock);
> >> + size = generic_file_splice_read(in, ppos, pipe, len, flags);
> >> + up_read(&EXT4_I(inode)->i_rwlock);
> >> + return size;
> >> +}
> >> +
> >> +ssize_t ext4_file_splice_write(struct pipe_inode_info *pipe,
> >> + struct file *out, loff_t *ppos, size_t len,
> >> + unsigned int flags)
> >> +{
> >> + struct inode *inode = out->f_mapping->host;
> >> + ssize_t size;
> >> +
> >> + down_write(&EXT4_I(inode)->i_rwlock);
> >> + size = generic_file_splice_write(pipe, out, ppos, len, flags);
> >> + up_write(&EXT4_I(inode)->i_rwlock);
> >> + return size;
> >> +}
> >> +
> >> +
> >> const struct file_operations ext4_file_operations = {
> >> .llseek = ext4_llseek,
> >> .read = do_sync_read,
> >> .write = do_sync_write,
> >> - .aio_read = generic_file_aio_read,
> >> + .aio_read = ext4_file_read,
> >> .aio_write = ext4_file_write,
> >> .unlocked_ioctl = ext4_ioctl,
> >> #ifdef CONFIG_COMPAT
> >> @@ -266,8 +314,8 @@ const struct file_operations ext4_file_operations = {
> >> .open = ext4_file_open,
> >> .release = ext4_release_file,
> >> .fsync = ext4_sync_file,
> >> - .splice_read = generic_file_splice_read,
> >> - .splice_write = generic_file_splice_write,
> >> + .splice_read = ext4_file_splice_read,
> >> + .splice_write = ext4_file_splice_write,
> >> .fallocate = ext4_fallocate,
> >> };
> >>
> >> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> >> index f2fa5e8..769ab0f 100644
> >> --- a/fs/ext4/inode.c
> >> +++ b/fs/ext4/inode.c
> >> @@ -4482,6 +4482,12 @@ void ext4_truncate(struct inode *inode)
> >> goto out_stop;
> >>
> >> /*
> >> + * We should block reads/writes to that inode so we are sure we are
> >> + * consistent and reads/writes remain atomic.
> >> + */
> >> + down_write(&EXT4_I(inode)->i_rwlock);
> >> +
> >> + /*
> >> * From here we block out all ext4_get_block() callers who want to
> >> * modify the block allocation tree.
> >> */
> >> @@ -4566,6 +4572,7 @@ do_indirects:
> >>
> >> out_unlock:
> >> up_write(&ei->i_data_sem);
> >> + up_write(&EXT4_I(inode)->i_rwlock);
> >> inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
> >> ext4_mark_inode_dirty(handle, inode);
> >>
> >> diff --git a/fs/ext4/super.c b/fs/ext4/super.c
> >> index 8553dfb..2dbe86a 100644
> >> --- a/fs/ext4/super.c
> >> +++ b/fs/ext4/super.c
> >> @@ -895,6 +895,7 @@ static void init_once(void *foo)
> >> init_rwsem(&ei->xattr_sem);
> >> #endif
> >> init_rwsem(&ei->i_data_sem);
> >> + init_rwsem(&ei->i_rwlock);
> >> inode_init_once(&ei->vfs_inode);
> >> }
> >>
> >>
> >
> > --
> > --
> > 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
