dd: handle stdout redirected to seekable file

[jfinkels]

This pull request fixes a bug in dd where null bytes would be unintentionally written if stdout were redirected to a seekable file. For example, before this commit, if dd were invoked from the command-line as

dd if=infile bs=1 count=10 seek=5 > /dev/sda1

then five zeros would be written to /dev/sda1 before copying ten bytes of infile to /dev/sda1. After this commit, dd will correctly seek five bytes forward in /dev/sda1 before copying the ten bytes of infile. I wasn't sure how to write a test for this.

In order to accomplish this, the first commit moves the argument parsing out of the Input::new() and Output::new() functions. This was needed in order to be able to determine whether we need to use the Output<File> specialization even when no outfile is specified in the command-line arguments. The code is a bit messy, but I think it should work.

Fixes #3542.

[jhscheer]

There's similar functionality in other uutils, e.g. tail. We should think about a follow-up PR to move this kind of checks to a central file to reduce code duplication.

[jhscheer]

I know you had serious problems with this on your system before, so sorry for asking. Did you try to run tests/dd/skip-seek-past-dev.sh as root with the changes from this PR?

[jfinkels]

I did not try to run that test yet. I thought the CI system executes that test because of the issues we had in #3442, but now it looks like the test is skipped? From the logs:

2022-08-27T21:23:35.8791953Z skip-seek-past-dev.sh: skipped test: must be run as root
2022-08-27T21:23:35.8830694Z SKIP: tests/dd/skip-seek-past-dev.sh

[jhscheer]

Yes, sadly the CI doesn't run any GNU tests which requires root. That's why I asked.

[jfinkels]

Okay, I'll install a virtual machine and try it there.

[jfinkels]

I checked out the branch from pull request #3882 (which includes the changes from this branch as well as the change to df that triggered the issue before) and tested by running

sudo bash util/run-gnu-test.sh tests/dd/skip-seek-past-dev.sh

the relevant part of the error log is:

--- err_ok	2022-08-30 19:04:36.700725532 -0400
+++ err	2022-08-30 19:04:36.596723639 -0400
@@ -1,3 +1 @@
-dd: 'standard output': cannot seek: Invalid argument
-0+0 records in
-0+0 records out
+dd: failed to seek in output file: Invalid input
FAIL tests/dd/skip-seek-past-dev.sh (exit status: 1)

The message "failed to seek in output file: Invalid input" comes from

coreutils/src/uu/dd/src/dd.rs

Lines 543 to 544 in 282774a

	dst.seek(io::SeekFrom::Start(i))
	.map_err_context(|| "failed to seek in output file".to_string())?;

which is in the implementation for Output<File>. That's what I expected, so that's good!