| Message ID | 20230520000049.2226926-23-dhowells@redhat.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | splice: Kill ITER_PIPE | expand |
On 5/20/23 8:00 AM, David Howells wrote: > Provide a splice_read stub for ocfs2. This emits trace lines and does an > atime lock/update before calling filemap_splice_read(). Splicing from > direct I/O is handled by the caller. > > A couple of new tracepoints are added for this purpose. > > Signed-off-by: David Howells <dhowells@redhat.com> > cc: Christoph Hellwig <hch@lst.de> > cc: Al Viro <viro@zeniv.linux.org.uk> > cc: Jens Axboe <axboe@kernel.dk> > cc: Mark Fasheh <mark@fasheh.com> > cc: Joel Becker <jlbec@evilplan.org> > cc: Joseph Qi <joseph.qi@linux.alibaba.com> > cc: ocfs2-devel@oss.oracle.com > cc: linux-fsdevel@vger.kernel.org > cc: linux-block@vger.kernel.org > cc: linux-mm@kvack.org > --- > fs/ocfs2/file.c | 39 ++++++++++++++++++++++++++++++++++++++- > fs/ocfs2/ocfs2_trace.h | 3 +++ > 2 files changed, 41 insertions(+), 1 deletion(-) > > diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c > index efb09de4343d..f7e00b5689d5 100644 > --- a/fs/ocfs2/file.c > +++ b/fs/ocfs2/file.c > @@ -2581,6 +2581,43 @@ static ssize_t ocfs2_file_read_iter(struct kiocb *iocb, > return ret; > } > > +static ssize_t ocfs2_file_splice_read(struct file *in, loff_t *ppos, > + struct pipe_inode_info *pipe, > + size_t len, unsigned int flags) > +{ > + struct inode *inode = file_inode(in); > + ssize_t ret = 0; > + int lock_level = 0; > + > + trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry, > + (unsigned long long)OCFS2_I(inode)->ip_blkno, > + in->f_path.dentry->d_name.len, > + in->f_path.dentry->d_name.name, > + 0); Better also trace flags here. > + > + /* > + * We're fine letting folks race truncates and extending writes with > + * read across the cluster, just like they can locally. Hence no > + * rw_lock during read. > + * > + * Take and drop the meta data lock to update inode fields like i_size. > + * This allows the checks down below generic_file_splice_read() a Now it calls filemap_splice_read(). > + * chance of actually working. > + */ > + ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level, true); Since prototype is 'int wait', so directly passing '1' seems more appropriate. > + if (ret < 0) { > + if (ret != -EAGAIN) > + mlog_errno(ret); > + goto bail; > + } > + ocfs2_inode_unlock(inode, lock_level); > + Don't see direct IO logic now. Am I missing something? Thanks, Joseph > + ret = filemap_splice_read(in, ppos, pipe, len, flags); > + trace_filemap_splice_read_ret(ret); > +bail: > + return ret; > +} > + > /* Refer generic_file_llseek_unlocked() */ > static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence) > { > @@ -2744,7 +2781,7 @@ const struct file_operations ocfs2_fops = { > #endif > .lock = ocfs2_lock, > .flock = ocfs2_flock, > - .splice_read = generic_file_splice_read, > + .splice_read = ocfs2_file_splice_read, > .splice_write = iter_file_splice_write, > .fallocate = ocfs2_fallocate, > .remap_file_range = ocfs2_remap_file_range, > diff --git a/fs/ocfs2/ocfs2_trace.h b/fs/ocfs2/ocfs2_trace.h > index dc4bce1649c1..b8c3d1702076 100644 > --- a/fs/ocfs2/ocfs2_trace.h > +++ b/fs/ocfs2/ocfs2_trace.h > @@ -1319,6 +1319,8 @@ DEFINE_OCFS2_FILE_OPS(ocfs2_file_splice_write); > > DEFINE_OCFS2_FILE_OPS(ocfs2_file_read_iter); > > +DEFINE_OCFS2_FILE_OPS(ocfs2_file_splice_read); > + > DEFINE_OCFS2_ULL_ULL_ULL_EVENT(ocfs2_truncate_file); > > DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_truncate_file_error); > @@ -1470,6 +1472,7 @@ TRACE_EVENT(ocfs2_prepare_inode_for_write, > ); > > DEFINE_OCFS2_INT_EVENT(generic_file_read_iter_ret); > +DEFINE_OCFS2_INT_EVENT(filemap_splice_read_ret); > > /* End of trace events for fs/ocfs2/file.c. */ >
Joseph Qi <joseph.qi@linux.alibaba.com> wrote:
> Don't see direct IO logic now. Am I missing something?
See that patch description ;-)
Provide a splice_read stub for ocfs2. This emits trace lines and does an
atime lock/update before calling filemap_splice_read(). Splicing from
direct I/O is handled by the caller.
David
On 5/22/23 2:28 PM, David Howells wrote: > Joseph Qi <joseph.qi@linux.alibaba.com> wrote: > >> Don't see direct IO logic now. Am I missing something? > > See that patch description ;-) > > Provide a splice_read stub for ocfs2. This emits trace lines and does an > atime lock/update before calling filemap_splice_read(). Splicing from > direct I/O is handled by the caller. > Oops, missed the patch 7 of the series since I've only received this one:( Have checked it on maillist, it's fine for me now. Thanks, Joseph
So something like the attached changes? Any suggestions as to how to improve
the comments?
David
---
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index f7e00b5689d5..86add13b5f23 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -2552,7 +2552,7 @@ static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
*
* Take and drop the meta data lock to update inode fields
* like i_size. This allows the checks down below
- * generic_file_read_iter() a chance of actually working.
+ * copy_splice_read() a chance of actually working.
*/
ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
!nowait);
@@ -2593,7 +2593,7 @@ static ssize_t ocfs2_file_splice_read(struct file *in, loff_t *ppos,
(unsigned long long)OCFS2_I(inode)->ip_blkno,
in->f_path.dentry->d_name.len,
in->f_path.dentry->d_name.name,
- 0);
+ flags);
/*
* We're fine letting folks race truncates and extending writes with
@@ -2601,10 +2601,10 @@ static ssize_t ocfs2_file_splice_read(struct file *in, loff_t *ppos,
* rw_lock during read.
*
* Take and drop the meta data lock to update inode fields like i_size.
- * This allows the checks down below generic_file_splice_read() a
- * chance of actually working.
+ * This allows the checks down below filemap_splice_read() a chance of
+ * actually working.
*/
- ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level, true);
+ ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level, 1);
if (ret < 0) {
if (ret != -EAGAIN)
mlog_errno(ret);
On 5/22/23 2:49 PM, David Howells wrote: > So something like the attached changes? Any suggestions as to how to improve > the comments? > Looks fine to me now. Thanks. Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com> > David > --- > diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c > index f7e00b5689d5..86add13b5f23 100644 > --- a/fs/ocfs2/file.c > +++ b/fs/ocfs2/file.c > @@ -2552,7 +2552,7 @@ static ssize_t ocfs2_file_read_iter(struct kiocb *iocb, > * > * Take and drop the meta data lock to update inode fields > * like i_size. This allows the checks down below > - * generic_file_read_iter() a chance of actually working. > + * copy_splice_read() a chance of actually working. > */ > ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level, > !nowait); > @@ -2593,7 +2593,7 @@ static ssize_t ocfs2_file_splice_read(struct file *in, loff_t *ppos, > (unsigned long long)OCFS2_I(inode)->ip_blkno, > in->f_path.dentry->d_name.len, > in->f_path.dentry->d_name.name, > - 0); > + flags); > > /* > * We're fine letting folks race truncates and extending writes with > @@ -2601,10 +2601,10 @@ static ssize_t ocfs2_file_splice_read(struct file *in, loff_t *ppos, > * rw_lock during read. > * > * Take and drop the meta data lock to update inode fields like i_size. > - * This allows the checks down below generic_file_splice_read() a > - * chance of actually working. > + * This allows the checks down below filemap_splice_read() a chance of > + * actually working. > */ > - ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level, true); > + ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level, 1); > if (ret < 0) { > if (ret != -EAGAIN) > mlog_errno(ret);
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c index efb09de4343d..f7e00b5689d5 100644 --- a/fs/ocfs2/file.c +++ b/fs/ocfs2/file.c @@ -2581,6 +2581,43 @@ static ssize_t ocfs2_file_read_iter(struct kiocb *iocb, return ret; } +static ssize_t ocfs2_file_splice_read(struct file *in, loff_t *ppos, + struct pipe_inode_info *pipe, + size_t len, unsigned int flags) +{ + struct inode *inode = file_inode(in); + ssize_t ret = 0; + int lock_level = 0; + + trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry, + (unsigned long long)OCFS2_I(inode)->ip_blkno, + in->f_path.dentry->d_name.len, + in->f_path.dentry->d_name.name, + 0); + + /* + * We're fine letting folks race truncates and extending writes with + * read across the cluster, just like they can locally. Hence no + * rw_lock during read. + * + * Take and drop the meta data lock to update inode fields like i_size. + * This allows the checks down below generic_file_splice_read() a + * chance of actually working. + */ + ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level, true); + if (ret < 0) { + if (ret != -EAGAIN) + mlog_errno(ret); + goto bail; + } + ocfs2_inode_unlock(inode, lock_level); + + ret = filemap_splice_read(in, ppos, pipe, len, flags); + trace_filemap_splice_read_ret(ret); +bail: + return ret; +} + /* Refer generic_file_llseek_unlocked() */ static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence) { @@ -2744,7 +2781,7 @@ const struct file_operations ocfs2_fops = { #endif .lock = ocfs2_lock, .flock = ocfs2_flock, - .splice_read = generic_file_splice_read, + .splice_read = ocfs2_file_splice_read, .splice_write = iter_file_splice_write, .fallocate = ocfs2_fallocate, .remap_file_range = ocfs2_remap_file_range, diff --git a/fs/ocfs2/ocfs2_trace.h b/fs/ocfs2/ocfs2_trace.h index dc4bce1649c1..b8c3d1702076 100644 --- a/fs/ocfs2/ocfs2_trace.h +++ b/fs/ocfs2/ocfs2_trace.h @@ -1319,6 +1319,8 @@ DEFINE_OCFS2_FILE_OPS(ocfs2_file_splice_write); DEFINE_OCFS2_FILE_OPS(ocfs2_file_read_iter); +DEFINE_OCFS2_FILE_OPS(ocfs2_file_splice_read); + DEFINE_OCFS2_ULL_ULL_ULL_EVENT(ocfs2_truncate_file); DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_truncate_file_error); @@ -1470,6 +1472,7 @@ TRACE_EVENT(ocfs2_prepare_inode_for_write, ); DEFINE_OCFS2_INT_EVENT(generic_file_read_iter_ret); +DEFINE_OCFS2_INT_EVENT(filemap_splice_read_ret); /* End of trace events for fs/ocfs2/file.c. */
Provide a splice_read stub for ocfs2. This emits trace lines and does an atime lock/update before calling filemap_splice_read(). Splicing from direct I/O is handled by the caller. A couple of new tracepoints are added for this purpose. Signed-off-by: David Howells <dhowells@redhat.com> cc: Christoph Hellwig <hch@lst.de> cc: Al Viro <viro@zeniv.linux.org.uk> cc: Jens Axboe <axboe@kernel.dk> cc: Mark Fasheh <mark@fasheh.com> cc: Joel Becker <jlbec@evilplan.org> cc: Joseph Qi <joseph.qi@linux.alibaba.com> cc: ocfs2-devel@oss.oracle.com cc: linux-fsdevel@vger.kernel.org cc: linux-block@vger.kernel.org cc: linux-mm@kvack.org --- fs/ocfs2/file.c | 39 ++++++++++++++++++++++++++++++++++++++- fs/ocfs2/ocfs2_trace.h | 3 +++ 2 files changed, 41 insertions(+), 1 deletion(-)