From: David van Moolenbroek <david@minix3.org>
Date: Mon, 2 Sep 2013 15:34:44 +0000 (+0200)
Subject: VFS: update filp_pos on chardev I/O (workaround)
X-Git-Tag: v3.3.0~598
X-Git-Url: http://zhaoyanbai.com/repos/?a=commitdiff_plain;h=refs%2Fchanges%2F60%2F960%2F2;p=minix.git

VFS: update filp_pos on chardev I/O (workaround)

Previously, reading from or writing to a character device would not
update the file position on the corresponding filp object.  Performing
this update correctly is not trivial: during and after the I/O
operation, the filp object must not be locked.  Ideally, read/write
requests on a filp that is already involved in a read/write operation,
should be queued.  For now, we optimistically update the file position
at the start of the I/O; this works under the assumptions listed in
the corresponding comment.

Change-Id: I172a61781850423709924390ae3df1f2d1f94707
---

diff --git a/servers/vfs/read.c b/servers/vfs/read.c
index fd3f9fcb6..fcd280aaf 100644
--- a/servers/vfs/read.c
+++ b/servers/vfs/read.c
@@ -177,9 +177,31 @@ int read_write(struct fproc *rfp, int rw_flag, struct filp *f,
 	r = dev_io(op, dev, for_e, buf, position, size, f->filp_flags,
 		   suspend_reopen);
 	if (r >= 0) {
+		/* This should no longer happen: all calls are asynchronous. */
+		printf("VFS: I/O to device %x succeeded immediately!?\n", dev);
 		cum_io = r;
-		position += cum_io;
+		position += r;
 		r = OK;
+	} else if (r == SUSPEND) {
+		/* FIXME: multiple read/write operations on a single filp
+		 * should be serialized. They currently aren't; in order to
+		 * achieve a similar effect, we optimistically advance the file
+		 * position here. This works under the following assumptions:
+		 * - character drivers that use the seek position at all,
+		 *   expose a view of a statically-sized range of bytes, i.e.,
+		 *   they are basically byte-granular block devices;
+		 * - if short I/O or an error is returned, all subsequent calls
+		 *   will return (respectively) EOF and an error;
+		 * - the application never checks its own file seek position,
+		 *   or does not care that it may end up having seeked beyond
+		 *   the number of bytes it has actually read;
+		 * - communication to the character driver is FIFO (this one
+		 *   is actually true! whew).
+		 * Many improvements are possible here, but in the end,
+		 * anything short of queuing concurrent operations will be
+		 * suboptimal - so we settle for this hack for now.
+		 */
+		position += size;
 	}
   } else if (S_ISBLK(vp->v_mode)) {	/* Block special files. */
 	if (vp->v_sdev == NO_DEV)