I have made some inline comments. I think a partial vmcore shouldn't be removed because another panic might happen before a user tries to run savecore manually again.

In D2587#81386, @def wrote:

I have made some inline comments. I think a partial vmcore shouldn't be removed because another panic might happen before a user tries to run savecore manually again.

Hm, I'm not sure I agree. If the FS really doesn't have enough space, this will just cause savecore to consume all the remaining space with a kernel dump fragment, and I'm pretty sure a truncated vmcore is mostly useless anyway.

Generally I think it'd be wrong to entirely remove this checking. There are lots of systems in which /var/crash resides on a UFS FS and the space checking is mostly correct. In particular, Isilon has code to compress kernel dumps before writing them to the dumpdev (see D3054-59), so savecore will not create a sparse file in this case, and we frequently create large cores that cannot be extracted to /var/crash. Why not make the check conditional on a command-line option? This option could be set automatically if -z is specified.

Hm, I'm not sure I agree. If the FS really doesn't have enough space, this will just cause savecore to consume all the remaining space with a kernel dump fragment, and I'm pretty sure a truncated vmcore is mostly useless anyway.

Agreed. I think it would be better to remove a truncated vmcore. pjd suggested removing the partial vmcore and not clearing the dump from the dump device. I agree with him on that.

Generally I think it'd be wrong to entirely remove this checking. There are lots of systems in which /var/crash resides on a UFS FS and the space checking is mostly correct. In particular, Isilon has code to compress kernel dumps before writing them to the dumpdev (see D3054-59), so savecore will not create a sparse file in this case, and we frequently create large cores that cannot be extracted to /var/crash. Why not make the check conditional on a command-line option? This option could be set automatically if -z is specified.

The space checking is not mostly correct. savecore will always save either a compressed dump or a sparse dump. In my experiments even a sparse dump consumes far less disk space than its nominal size.

OTOH, if Isilon's compressed core dump code gets merged, then the space checking would be useful again. I don't like the idea of adding another command-line option though. How does this sound?

If core dump is compressed
   If /var/crash has sufficient space
       Write compressed core file
else
    if user specified "-z"
        Try to write compressed core file
    else
        Try to write sparse core file
If the write succeeded
    Clear the core from the dump device
else
    Delete the partial core file
    Log an error message

In D2587#81396, @asomers wrote:

Hm, I'm not sure I agree. If the FS really doesn't have enough space, this will just cause savecore to consume all the remaining space with a kernel dump fragment, and I'm pretty sure a truncated vmcore is mostly useless anyway.

Agreed. I think it would be better to remove a truncated vmcore. pjd suggested removing the partial vmcore and not clearing the dump from the dump device. I agree with him on that.

Generally I think it'd be wrong to entirely remove this checking. There are lots of systems in which /var/crash resides on a UFS FS and the space checking is mostly correct. In particular, Isilon has code to compress kernel dumps before writing them to the dumpdev (see D3054-59), so savecore will not create a sparse file in this case, and we frequently create large cores that cannot be extracted to /var/crash. Why not make the check conditional on a command-line option? This option could be set automatically if -z is specified.

The space checking is not mostly correct. savecore will always save either a compressed dump or a sparse dump. In my experiments even a sparse dump consumes far less disk space than its nominal size.

OTOH, if Isilon's compressed core dump code gets merged, then the space checking would be useful again. I don't like the idea of adding another command-line option though. How does this sound?

If core dump is compressed
   If /var/crash has sufficient space
       Write compressed core file
else
    if user specified "-z"
        Try to write compressed core file
    else
        Try to write sparse core file
If the write succeeded
    Clear the core from the dump device
else
    Delete the partial core file
    Log an error message

This makes sense to me. I do plan on merging the compression code before 11.0, so the checking will indeed be useful in that case.