linux-kernel - Re: pstore: fix compression

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Fri, 19 Feb 2021 01:54:23 +0100
From:   Jiri Bohac <jbohac@...e.cz>
To:     Kees Cook <keescook@...omium.org>
Cc:     Tony Luck <tony.luck@...el.com>,
        Anton Vorontsov <anton@...msg.org>,
        Colin Cross <ccross@...roid.com>,
        Matteo Croce <mcroce@...ux.microsoft.com>,
        linux-kernel@...r.kernel.org
Subject: Re: pstore: fix compression

On Thu, Feb 18, 2021 at 12:30:03PM -0800, Kees Cook wrote:
> Eek; thanks for the catch!

thanks for applying the fix;

BTW, with the compression broken, I was not getting any dmesg
stored in ERST at all, not even uncompressed. After instrumenting
the code with a lot of debug printks I found that writing
erst_erange.size worth of data into the ERST fails and the
maximum writeable size is 98 bytes smaler:

Details: 

	- erst_erange.size = 65536
	- this results in  erst_info.bufsize = 65336
	- pstore_compress() returned -EINVAL (because of the
	  just-fixed typo), zipped_len = -EINVAL.
	- pstore_dump calls copy_kmsg_to_buffer to only copy bufsize
	  bytes from big_oops_buf to psinfo->buf;
	  record.size = bufsize = 65336

	psinfo->write(&record) then fails with -EINVAL;
	by more tracing inside the ERST code I found the -EINVAL was
	produced by __erst_write_to_storage()
	after apei_exec_ctx_get_output() returned
	val=ERST_STATUS_FAILED=3 and this got translated into -EINVAL by
	erst_errno().

	Once the compression was fixed everything started working because
	the records are much smaller after the compression (~30kB).

	My next thought was to find the largest possible record that
	could be written successfully.
	I modified the ERST init code to decrease erst_info.bufsize by a
	value specified on the cmdline. The maximum writable record was
	65238 bytes long (i.e. erst_erange.size - sizeof(struct
	cper_pstore_record) - 98). With this hack I got
	65238 bytes of uncompressed dmesg stored to ERST.

Any idea what might be causing this?
As far as I can tell, there are no other records in the ERST
(checked through the erst-dbg interface).
Tested on a HPE ProLiant DL120 Gen10 server.

Thanks!


-- 
Jiri Bohac <jbohac@...e.cz>
SUSE Labs, Prague, Czechia