truncate() error values (was Re: Floppy disk strange behavior)

[Alexander Viro]

On Wed, 10 Jan 2001 [e-mail] wrote:

> >> dd: advancing past 1 blocks in output file `/dev/fd0': Permission denied
> 
> > dd bug. It tries to ftruncate() the output file and gets all upset when
> > kernel refuses to truncate a block device (surprise, surprise).
> 
> Yes. But EPERM means that something is wrong with privileges.
	EACCES, actually.
> One would expect EINVAL or so when something is wrong with the
> way the routine was called.

Oh, agreed. Patch I've sent to Linus and posted on l-k more than a month ago:

diff -urN rc12-pre5/fs/open.c rc12-pre5-truncate/fs/open.c
--- rc12-pre5/fs/open.c	Wed Nov 29 21:37:31 2000
+++ rc12-pre5-truncate/fs/open.c	Tue Dec  5 14:53:59 2000
@@ -102,7 +102,12 @@
 		goto out;
 	inode = nd.dentry->d_inode;
 
-	error = -EACCES;
+	/* For directories it's -EISDIR, for other non-regulars - -EINVAL */
+	error = -EISDIR;
+	if (S_ISDIR(inode->i_mode))
+		goto dput_and_out;
+
+	error = -EINVAL;
 	if (!S_ISREG(inode->i_mode))
 		goto dput_and_out;
 
@@ -163,7 +168,7 @@
 		goto out;
 	dentry = file->f_dentry;
 	inode = dentry->d_inode;
-	error = -EACCES;
+	error = -EINVAL;
 	if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
 		goto out_putf;
 	error = -EPERM;

That's test12-pre5. And yes, the case of files opened with O_RDONLY should
be -EINVAL according to
	* our own manpages
	* POSIX
	* *BSD behaviour
	* Solaris behaviour
	* manpages of all Unices I've seen.
-EACCES is wrong here. More logical, but...

Linus, I realize that we have a moratorium on patches. It's certainly not
a critical one, but it probably ought to be applied at some point. Could
you tell what time would be OK for resubmitting it? All it changes is
	* error returned for truncate() on directories (EACCES -> EISDIR)
	* error returned for truncate() on other non-file (EACCES -> EINVAL)
	* error returned for ftruncate() on O_RDONLY fds (EACCES -> EINVAL)
	* error returned for ftruncate() on non-files (EACCES -> EINVAL)
In all cases new error values are what we are required to do by POSIX, SuS,
our own manpages, manpages of other Unices and actual behaviour of other
Unices.