[Libsysio-commit] HEAD: libsysio/drivers/native fs_native.c

[Lee W. <lw...@us...>]

Update of /cvsroot/libsysio/libsysio/drivers/native
In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv16325/drivers/native

Modified Files:
	fs_native.c 
Log Message:


+ Merged in changes from namespace_assembly branch (see .../misc/init-env.sh)
	This provoked a bunch of bugs. See below.

+ Fixed a bug in _sysio_enumerate_extents(). If the IO operation was
	short, it would go round the loop again, mistakenly trying
	to fill more of the extent.
+ In rw.c, fixed bugs in [p]{read,write}[vx] (all the synchronous routines)
	that improperly propagated error returns. They were returning -errno
	instead of setting errno and returning -1.
+ In fs_native.c:doiov, fixed a bug where a zero-length IO was improperly
	thought to be an error.
+ In lseek.c:_sysio_lseek, fixed final position check to properly determine
	{under,over}flow.
+ In link.c:link, fixed the existence check. No error is returned for
	nonexistent files when ND_NEGOK is specified. We're supposed
	to check whether it's a negative entry or not.
+ A new macro, I_GONE, was added to inode.h. This will *try* to
	kill an inode but if it can't, it becomes a zombie instead.
+ In unlink.c:unlink, link.c:link, rename.c:rename, the driver ops
	were being called but the actual operation in the internal path
	tree was not reflected. Also, for unlink and rename, use the
	new I_GONE macro on the destroyed inode.
+ In fs_native.c:native_inop_gone, close() was always called, even when
	the fildes was -1.
+ In fs_native.c:native_inop_gone, close() was called. We really meant
	to call syscal(SYS_close, ...);
+ In namei.c:_sysio_path_walk, fixed broken symlink handling. It wasn't
	following symlinks anywhere if ND_NOFOLLOW was set. That flag only
	means that the *last* component should not be followed.
+ In namei.c:_sysio_path_walk, fixed buffer overrun problem for very long
	symlinks.
+ In fs_incore.c, fixed dirop_{link,rename,unlink,rmdir} because they
	were manipulating the system path cache and shouldn't.
+ In mount.c:_sysio_unmount_all we were mistakenly releasing an FS root
	after a failed unmount attempt.
+ Fixes in test_regions.c free allocated memory at the end so valgrind
	doesn't show a leak.


Index: fs_native.c
===================================================================
RCS file: /cvsroot/libsysio/libsysio/drivers/native/fs_native.c,v
retrieving revision 1.31
retrieving revision 1.32
diff -u -w -b -B -p -r1.31 -r1.32
--- fs_native.c	8 Feb 2004 23:48:51 -0000	1.31
+++ fs_native.c	14 Feb 2004 19:42:58 -0000	1.32
@@ -608,8 +608,7 @@ native_iget(struct filesys *fs,
 			 * Cached inode has stale attrs
 			 * make way for the new one
 			 */
-			I_RELE(ino);
-			_sysio_i_undead(ino);
+			I_GONE(ino);
 			ino = NULL;
 		} else
 			/* 
@@ -1263,10 +1262,10 @@ doiov(const struct iovec *iov,
 						    _SYSIO_OFF_T,
 						    void *))dopio,
 				       nio);
-	if (cc > 0)
-		nio->nio_nino->ni_fpos += cc;
-	else
+	if (cc < 0)
 		cc = -errno;
+	else
+		nio->nio_nino->ni_fpos += cc;
 	return cc;
 
 #if !(defined(REDSTORM) || defined(MAX_IOVEC))
@@ -1567,8 +1566,8 @@ native_inop_gone(struct inode *ino)
 {
 	struct native_inode *nino = I2NI(ino);
 
-	if (nino->ni_fd)
-		(void )close(nino->ni_fd);
+	if (nino->ni_fd >= 0)
+		(void )syscall(SYS_close, nino->ni_fd);
 	free(ino->i_private);
 }