[PATCH 1/2] 9pnet: allow making incomplete read requests

A user doesn't necessarily want to wait for all the requested data to
be available, since the waiting time is unbounded.

Signed-off-by: Sergey Alirzaev <redacted>
---
 include/net/9p/client.h |   2 +
 net/9p/client.c         | 133 ++++++++++++++++++++++------------------
 2 files changed, 76 insertions(+), 59 deletions(-)

diff --git a/include/net/9p/client.h b/include/net/9p/client.h
index acc60d8a3b3b..f6c890e94f87 100644
--- a/include/net/9p/client.h
+++ b/include/net/9p/client.h

@@ -200,6 +200,8 @@ int p9_client_fsync(struct p9_fid *fid, int datasync);
 int p9_client_remove(struct p9_fid *fid);
 int p9_client_unlinkat(struct p9_fid *dfid, const char *name, int flags);
 int p9_client_read(struct p9_fid *fid, u64 offset, struct iov_iter *to, int *err);
+int p9_client_read_once(struct p9_fid *fid, u64 offset, struct iov_iter *to,
+		int *err);
 int p9_client_write(struct p9_fid *fid, u64 offset, struct iov_iter *from, int *err);
 int p9_client_readdir(struct p9_fid *fid, char *data, u32 count, u64 offset);
 int p9dirent_read(struct p9_client *clnt, char *buf, int len,

diff --git a/net/9p/client.c b/net/9p/client.c
index 1d48afc7033c..186f5b44aa01 100644
--- a/net/9p/client.c
+++ b/net/9p/client.c

@@ -1548,83 +1548,98 @@ EXPORT_SYMBOL(p9_client_unlinkat);
 
 int
 p9_client_read(struct p9_fid *fid, u64 offset, struct iov_iter *to, int *err)
+{
+	int total = 0;
+	*err = 0;
+
+	while (iov_iter_count(to)) {
+		int count;
+
+		count = p9_client_read_once(fid, offset, to, err);
+		if (!count || *err)
+			break;
+		offset += count;
+		total += count;
+	}
+	return total;
+}
+EXPORT_SYMBOL(p9_client_read);
+
+int
+p9_client_read_once(struct p9_fid *fid, u64 offset, struct iov_iter *to,
+		    int *err)
 {
 	struct p9_client *clnt = fid->clnt;
 	struct p9_req_t *req;
 	int total = 0;
-	*err = 0;
+	int count = iov_iter_count(to);
+	int rsize, non_zc = 0;
+	char *dataptr;
 
+	*err = 0;
 	p9_debug(P9_DEBUG_9P, ">>> TREAD fid %d offset %llu %d\n",
 		   fid->fid, (unsigned long long) offset, (int)iov_iter_count(to));
 
-	while (iov_iter_count(to)) {
-		int count = iov_iter_count(to);
-		int rsize, non_zc = 0;
-		char *dataptr;
+	rsize = fid->iounit;
+	if (!rsize || rsize > clnt->msize - P9_IOHDRSZ)
+		rsize = clnt->msize - P9_IOHDRSZ;
 
-		rsize = fid->iounit;
-		if (!rsize || rsize > clnt->msize-P9_IOHDRSZ)
-			rsize = clnt->msize - P9_IOHDRSZ;
+	if (count < rsize)
+		rsize = count;
 
-		if (count < rsize)
-			rsize = count;
+	/* Don't bother zerocopy for small IO (< 1024) */
+	if (clnt->trans_mod->zc_request && rsize > 1024) {
+		/* response header len is 11
+		 * PDU Header(7) + IO Size (4)
+		 */
+		req = p9_client_zc_rpc(clnt, P9_TREAD, to, NULL, rsize,
+				       0, 11, "dqd", fid->fid,
+				       offset, rsize);
+	} else {
+		non_zc = 1;
+		req = p9_client_rpc(clnt, P9_TREAD, "dqd", fid->fid, offset,
+				    rsize);
+	}
+	if (IS_ERR(req)) {
+		*err = PTR_ERR(req);
+		return total;
+	}
 
-		/* Don't bother zerocopy for small IO (< 1024) */
-		if (clnt->trans_mod->zc_request && rsize > 1024) {
-			/*
-			 * response header len is 11
-			 * PDU Header(7) + IO Size (4)
-			 */
-			req = p9_client_zc_rpc(clnt, P9_TREAD, to, NULL, rsize,
-					       0, 11, "dqd", fid->fid,
-					       offset, rsize);
-		} else {
-			non_zc = 1;
-			req = p9_client_rpc(clnt, P9_TREAD, "dqd", fid->fid, offset,
-					    rsize);
-		}
-		if (IS_ERR(req)) {
-			*err = PTR_ERR(req);
-			break;
-		}
+	*err = p9pdu_readf(&req->rc, clnt->proto_version,
+			   "D", &count, &dataptr);
+	if (*err) {
+		trace_9p_protocol_dump(clnt, &req->rc);
+		p9_tag_remove(clnt, req);
+		return total;
+	}
+	if (rsize < count) {
+		pr_err("bogus RREAD count (%d > %d)\n", count, rsize);
+		count = rsize;
+	}
 
-		*err = p9pdu_readf(&req->rc, clnt->proto_version,
-				   "D", &count, &dataptr);
-		if (*err) {
-			trace_9p_protocol_dump(clnt, &req->rc);
-			p9_tag_remove(clnt, req);
-			break;
-		}
-		if (rsize < count) {
-			pr_err("bogus RREAD count (%d > %d)\n", count, rsize);
-			count = rsize;
-		}
+	p9_debug(P9_DEBUG_9P, "<<< RREAD count %d\n", count);
+	if (!count) {
+		p9_tag_remove(clnt, req);
+		return total;
+	}
 
-		p9_debug(P9_DEBUG_9P, "<<< RREAD count %d\n", count);
-		if (!count) {
-			p9_tag_remove(clnt, req);
-			break;
-		}
+	if (non_zc) {
+		int n = copy_to_iter(dataptr, count, to);
 
-		if (non_zc) {
-			int n = copy_to_iter(dataptr, count, to);
-			total += n;
-			offset += n;
-			if (n != count) {
-				*err = -EFAULT;
-				p9_tag_remove(clnt, req);
-				break;
-			}
-		} else {
-			iov_iter_advance(to, count);
-			total += count;
-			offset += count;
+		total += n;
+		if (n != count) {
+			*err = -EFAULT;
+			p9_tag_remove(clnt, req);
+			return total;
 		}
-		p9_tag_remove(clnt, req);
+	} else {
+		iov_iter_advance(to, count);
+		total += count;
 	}
+	p9_tag_remove(clnt, req);
 	return total;
 }
-EXPORT_SYMBOL(p9_client_read);
+EXPORT_SYMBOL(p9_client_read_once);
 
 int
 p9_client_write(struct p9_fid *fid, u64 offset, struct iov_iter *from, int *err)

-- 
2.25.0

A proper way to handle O_NONBLOCK would be making the requests and
responses happen asynchronously, but this would require serious code
refactoring.

Signed-off-by: Sergey Alirzaev <redacted>
---
 fs/9p/vfs_file.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/fs/9p/vfs_file.c b/fs/9p/vfs_file.c
index fe7f0bd2048e..92cd1d80218d 100644
--- a/fs/9p/vfs_file.c
+++ b/fs/9p/vfs_file.c

@@ -388,7 +388,10 @@ v9fs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 	p9_debug(P9_DEBUG_VFS, "count %zu offset %lld\n",
 		 iov_iter_count(to), iocb->ki_pos);
 
-	ret = p9_client_read(fid, iocb->ki_pos, to, &err);
+	if (iocb->ki_filp->f_flags & O_NONBLOCK)
+		ret = p9_client_read_once(fid, iocb->ki_pos, to, &err);
+	else
+		ret = p9_client_read(fid, iocb->ki_pos, to, &err);
 	if (!ret)
 		return err;
 

-- 
2.25.0

Sergey Alirzaev wrote on Wed, Feb 05, 2020:

A proper way to handle O_NONBLOCK would be making the requests and
responses happen asynchronously, but this would require serious code
refactoring.

FWIW I do have some async 9p code waiting (it's been sent to the list
ages ago but I never took the time to properly test it due to lack of
interest manifested), the problem here is more the caching model than a
synchronous issue, since in nocache mode (where this is used) there is
nowhere to fetch the data ahead of time.

If you're interested in that then please have a look at
https://lore.kernel.org/lkml/1544532108-21689-1-git-send-email-asmadeus@codewreck.org/

Signed-off-by: Sergey Alirzaev <redacted>

That aside, I guess, why not?
Will take when other patch gets addressed.

Thanks,
-- 
Dominique

Sergey Alirzaev wrote on Wed, Feb 05, 2020:

A user doesn't necessarily want to wait for all the requested data to
be available, since the waiting time is unbounded.

I'm not sure I agree on the argument there: the waiting time is
unbounded for a single request as well. What's your use case?

I think it would be better to describe what you really do with
O_NONBLOCK that requires this, and not just false theoritical
arguments.

Signed-off-by: Sergey Alirzaev <redacted>

Code-wise looks mostly good, just a nitpick about keeping the total
variable in p9_client_read_once inline.

total only makes sense in an iterating p9_client_read, I think it makes
code harder to read here (can basically use count or 0 directly now)

-	*err = 0;
+	int count = iov_iter_count(to);
+	int rsize, non_zc = 0;
+	char *dataptr;
 
+	*err = 0;
 	p9_debug(P9_DEBUG_9P, ">>> TREAD fid %d offset %llu %d\n",
 		   fid->fid, (unsigned long long) offset, (int)iov_iter_count(to));
 
-	while (iov_iter_count(to)) {
-		int count = iov_iter_count(to);
-		int rsize, non_zc = 0;
-		char *dataptr;
+	rsize = fid->iounit;
+	if (!rsize || rsize > clnt->msize - P9_IOHDRSZ)
+		rsize = clnt->msize - P9_IOHDRSZ;
 
-		rsize = fid->iounit;
-		if (!rsize || rsize > clnt->msize-P9_IOHDRSZ)
-			rsize = clnt->msize - P9_IOHDRSZ;
+	if (count < rsize)
+		rsize = count;
 
-		if (count < rsize)
-			rsize = count;
+	/* Don't bother zerocopy for small IO (< 1024) */
+	if (clnt->trans_mod->zc_request && rsize > 1024) {
+		/* response header len is 11
+		 * PDU Header(7) + IO Size (4)
+		 */
+		req = p9_client_zc_rpc(clnt, P9_TREAD, to, NULL, rsize,
+				       0, 11, "dqd", fid->fid,
+				       offset, rsize);
+	} else {
+		non_zc = 1;
+		req = p9_client_rpc(clnt, P9_TREAD, "dqd", fid->fid, offset,
+				    rsize);
+	}
+	if (IS_ERR(req)) {
+		*err = PTR_ERR(req);
+		return total;
+	}
 
-		/* Don't bother zerocopy for small IO (< 1024) */
-		if (clnt->trans_mod->zc_request && rsize > 1024) {
-			/*
-			 * response header len is 11
-			 * PDU Header(7) + IO Size (4)
-			 */
-			req = p9_client_zc_rpc(clnt, P9_TREAD, to, NULL, rsize,
-					       0, 11, "dqd", fid->fid,
-					       offset, rsize);
-		} else {
-			non_zc = 1;
-			req = p9_client_rpc(clnt, P9_TREAD, "dqd", fid->fid, offset,
-					    rsize);
-		}
-		if (IS_ERR(req)) {
-			*err = PTR_ERR(req);
-			break;
-		}
+	*err = p9pdu_readf(&req->rc, clnt->proto_version,
+			   "D", &count, &dataptr);
+	if (*err) {
+		trace_9p_protocol_dump(clnt, &req->rc);
+		p9_tag_remove(clnt, req);
+		return total;
+	}
+	if (rsize < count) {
+		pr_err("bogus RREAD count (%d > %d)\n", count, rsize);
+		count = rsize;
+	}
 
-		*err = p9pdu_readf(&req->rc, clnt->proto_version,
-				   "D", &count, &dataptr);
-		if (*err) {
-			trace_9p_protocol_dump(clnt, &req->rc);
-			p9_tag_remove(clnt, req);
-			break;
-		}
-		if (rsize < count) {
-			pr_err("bogus RREAD count (%d > %d)\n", count, rsize);
-			count = rsize;
-		}
+	p9_debug(P9_DEBUG_9P, "<<< RREAD count %d\n", count);
+	if (!count) {
+		p9_tag_remove(clnt, req);
+		return total;
+	}
 
-		p9_debug(P9_DEBUG_9P, "<<< RREAD count %d\n", count);
-		if (!count) {
-			p9_tag_remove(clnt, req);
-			break;
-		}
+	if (non_zc) {
+		int n = copy_to_iter(dataptr, count, to);
 
-		if (non_zc) {
-			int n = copy_to_iter(dataptr, count, to);
-			total += n;
-			offset += n;
-			if (n != count) {
-				*err = -EFAULT;
-				p9_tag_remove(clnt, req);
-				break;
-			}
-		} else {
-			iov_iter_advance(to, count);
-			total += count;
-			offset += count;
+		total += n;
+		if (n != count) {
+			*err = -EFAULT;
+			p9_tag_remove(clnt, req);
+			return total;
 		}
-		p9_tag_remove(clnt, req);
+	} else {
+		iov_iter_advance(to, count);
+		total += count;
 	}
+	p9_tag_remove(clnt, req);
 	return total;
 }
-EXPORT_SYMBOL(p9_client_read);
+EXPORT_SYMBOL(p9_client_read_once);
 
 int
 p9_client_write(struct p9_fid *fid, u64 offset, struct iov_iter *from, int *err)

-- 
Dominique

On Wed, Feb 05, 2020 at 08:35:04AM +0100, Dominique Martinet wrote:

Sergey Alirzaev wrote on Wed, Feb 05, 2020:

quoted

A user doesn't necessarily want to wait for all the requested data to
be available, since the waiting time is unbounded.

I'm not sure I agree on the argument there: the waiting time is
unbounded for a single request as well. What's your use case?

I want to interface with synthetic file systems that represent arbitrary data streams.
The one where i've hit the problem is reading the log of a XMPP chat client that blocks if there's no new data available.

-- 
()  ascii ribbon campaign - against html mail
/\  http://arc.pasp.de/   - against proprietary attachments

l29ah@cock.li wrote on Wed, Feb 05, 2020:

On Wed, Feb 05, 2020 at 08:35:04AM +0100, Dominique Martinet wrote:

quoted

I'm not sure I agree on the argument there: the waiting time is
unbounded for a single request as well. What's your use case?

I want to interface with synthetic file systems that represent
arbitrary data streams.
The one where i've hit the problem is reading the log of a XMPP chat
client that blocks if there's no new data available.

Definitely a valid use case for 9p, please rephrase your commit message
to describe the problem a bit better.

I'll wait for a v2 removing the 'total' variable from
p9_client_read_once anyway, unless you disagree.


Thanks,
-- 
Dominique