dd: fix GNU test 'dd/skip-seek-past-dev'

[bbara]

The GNU test suggests that GNU's variant is trying to seek stdin/stdout s.t. it can abort early if the respective fd is seekable and does not have enough length to do the requested skip/seek.

Draft for now to signalize that I am working on it.
There are some small open points, e.g. recognize this better:

# dd bs=1 skip=10000000000000000 count=0 status=noxfer < /dev/sda1
dd: 'standard input': cannot skip: Invalid argument
0+0 records in
0+0 records out
# echo $?
1

vs

# echo "abcd" | dd bs=1 skip=5 count=0 status=noxfer
0+0 records in
0+0 records out
# echo $?
0

and obviously add tests :)

[bbara]

Because there are currently some attempts to work around the missing Rust feature to seek stdin/stdout/stderr, I found this open issue (actually a comment for another workaround implementation: rust-lang/rust#72802 (comment)).

I am currently not sure why skipping a non-seekable stdin does not return exit code 1:
echo "abcd" | dd bs=1 skip=5 count=0 status=noxfer
But I haven't checked the docu yet

[github-actions]

GNU testsuite comparison:

Congrats! The gnu test tests/dd/no-allocate is no longer failing!

[jfinkels]

Hi @bbara, does this relate to issue #3008, for which pull request #4189 was just merged? I also just posted pull request #4542 to complement #4189.

[bbara]

@jfinkels: your implementation helps to fix my targeted problem. I had a similar implementation as you but observed a couple of illegal seeks and broken pipes. I think as long as there is no "safe Rust way" to do a seek on stdin/stdout, we should do some UNIX-specific workarounds to avoid the errors. I guess the correct way would be to find out if stdin is something "seekable" (e.g. a regular file), which could by done by using fstat on the fd of the stream, or using stat on /dev/std{in,out}.

I guess a stat would also help to solve my problem. My problem in short: if stdin or stdout has a known size, and we want so skip/seek past the size, a special error message is shown.

I tried to avoid to add too much UNIX-specific stuff, but I guess these problems/features are very UNIX-specific, therefore it should be ok.

[tertsdiepraam]

I tried to replicate the error messages you test for with GNU and I just get the error messages that we already have. So, I'm not sure this change makes sense? You mentioned some GNU test, which one is that? I'm using GNU coreutils 9.1 if that matters.

[tertsdiepraam]

For this case I get from GNU:

dd: invalid number: ‘10000000000000000000’: Value too large for defined data type

Which is more informative and has nothing to do with the length of the target file.

[bbara]

Hey, sorry was a little bit busy last week. The mentioned test is a "root test" and can be found here: tests/dd/skip-seek-past-dev.sh. It takes a block device as stdin/-out and therefore requires root. I try to "simulate" this by using a normal file, which unfortunately behaves not quite similar.

[bbara]

So the error message should come when the position to skip to is "past" the device. My idea was to take a "high enough number" to skip past "every" device, but it seem to work not always as expected.

[bbara]

I guess I will adapt my implementation to be more specific and only work with block devices.

[tertsdiepraam]

For this case I get:

dd: 'standard input': cannot skip to specified offset

[github-actions]

GNU testsuite comparison:

Congrats! The gnu test tests/dd/nocache is no longer failing!
GNU test failed: tests/dd/nocache_eof. tests/dd/nocache_eof is passing on 'main'. Maybe you have to rebase?

[sylvestre]

Sorry but Clippy is complaining about:

error: unreachable statement
    --> src\uu\dd\src/dd.rs:1119:5
     |
1117 |     return Ok(None);
     |     --------------- any code following this expression is unreachable
1118 |
1119 |     let ftype = file.metadata()?.file_type();
     |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ unreachable statement
     |
     = note: `-D unreachable-code` implied by `-D warnings`

error: unused variable: `ftype`
    --> src\uu\dd\src/dd.rs:1119:9
     |
1119 |     let ftype = file.metadata()?.file_type();
     |         ^^^^^ help: if this is intentional, prefix it with an underscore: `_ftype`
     |
     = note: `-D unused-variables` implied by `-D warnings`

error: unused variable: `file`
    --> src\uu\dd\src/dd.rs:1115:32
     |
1115 | fn try_get_len_of_block_device(file: &mut File) -> io::Result<Option<u64>> {
     |                                ^^^^ help: if this is intentional, prefix it with an underscore: `_file`

[sylvestre]

ping ?

[bbara]

ping ?

Heyho, sorry for the late response.

I will look into run_ucmd_as_root() to try to set up a test working with a block device. Otherwise the implementation might currently not make really sense.

[sylvestre]

seems that it is fixed :)

it was failing with but it now passes:

2023-09-23T17:30:53.6409474Z 
2023-09-23T17:30:53.6409714Z FAIL: tests/dd/skip-seek-past-dev
2023-09-23T17:30:53.6415226Z =================================
2023-09-23T17:30:53.6415418Z 
2023-09-23T17:30:53.6418620Z --- err_ok	2023-09-23 17:30:34.758495511 +0000
2023-09-23T17:30:53.6419008Z +++ err	2023-09-23 17:30:23.706344514 +0000
2023-09-23T17:30:53.6422417Z @@ -1,3 +0,0 @@
2023-09-23T17:30:53.6425434Z -dd: 'standard input': cannot skip: Invalid argument
2023-09-23T17:30:53.6433997Z -0+0 records in
2023-09-23T17:30:53.6440725Z -0+0 records out
2023-09-23T17:30:53.6444320Z --- err_ok	2023-09-23 17:30:36.238516527 +0000
2023-09-23T17:30:53.6449599Z +++ err	2023-09-23 17:30:36.238516527 +0000
2023-09-23T17:30:53.6451600Z @@ -1,3 +1 @@
2023-09-23T17:30:53.6455605Z -dd: 'standard output': cannot seek: Invalid argument
2023-09-23T17:30:53.6456385Z -0+0 records in
2023-09-23T17:30:53.6459838Z -0+0 records out
2023-09-23T17:30:53.6460182Z +dd: failed to seek in output file: Invalid input
2023-09-23T17:30:53.6460697Z FAIL tests/dd/skip-seek-past-dev.sh (exit status: 1)
2023-09-23T17:30:53.6462780Z