[LU-6020] Bugfixes for GSS/Kerberos

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.8.0
• Labels: patch
• Severity: 3
• Assignee: WC Triage
• Reporter: Andrew Perepechko

Description

The attached file contains gss/kerberos fixes for Lustre 2.5 that we are using and a few public documents on gss/kerberos/rpcsec_gss/etc (for the reference).

I'll be porting the patches for Intel's repository and pushing for review wrt my schedule.

Issue Links

• is related to: LU-9279 coral-beta-combined build 124 kernel BUG at include/linux/scatterlist.h:65! invalid opcode: 0000 [#1] SMP
• is related to: LU-6490 builds on 3.12 fail in gss
• is related to: LU-1714 crash upon loading libcfs module

Comments

[Joseph Gmitter (Inactive)]

All patches have landed to master for 2.8

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/15546/
Subject: LU-6020 gss: add lsvcgssd init.d wrapper
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 4bfb99a83b632b0ad26df4544e556250683b4d4d

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/17319/
Subject: LU-6020 gss: properly map buffers to sg
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: ffaebaf0750fe218ed93532382f70f61177cc121

[Gerrit Updater]

Andrew Perepechko ([e-mail]) uploaded a new patch: http://review.whamcloud.com/17319
Subject: LU-6020 gss: properly map buffers to sg
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 84369c8b794f84b3db1dfb6e081f522317eb4915

[Andrew Perepechko]

The issue was discovered with Lustre 3.6-kernel-based client where even struct page dereference caused a kernel crash, since the page address itself did not have a mapping in page tables.

[Andrew Perepechko]

It looks like buf_to_sg() code is wrong.

A lot of buffer pointers that it gets as input are coming from vmalloc(), e.g. OBD_ALLOC_LARGE() in ptlrpc_add_rqs_to_pool().
sg_set_buf() uses virt_to_page() to map virtual addresses to struct page, which does not work for vmalloc addresses.
Lustre should get a fix similar to the one in upstream commit 0def10217e7b768a501d2c51ea6d5ee4332afe69.

Things may appear working just because for the non-accelerated encryption/decryption case page references ops used for
scatterwalk_pagedone()->flush_dcache_page() are, to the best of my knowledge, a no-op.

[Gerrit Updater]

Andrew Perepechko ([e-mail]) uploaded a new patch: http://review.whamcloud.com/15546
Subject: LU-6020 gss: lvcgssd wrapper
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 856197bc0693190ca93ebe998365a3b6b149382e

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/14020/
Subject: LU-6020 kerberos: readdir bulk replies are not wrapped
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 32992594f37b00f2dd2d73b7e0534fdbd83ce017

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/14018/
Subject: LU-6020 kerberos: bulk nob is not corrected on bulk writes
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 116c203e1fd74989e096909006cfc90e5986b709

[Gerrit Updater]

Andrew Perepechko ([e-mail]) uploaded a new patch: http://review.whamcloud.com/14020
Subject: LU-6020 kerberos: readdir bulk replies are not wrapped
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: d90ff20490d7a1ea958678ec0650fa0bae4dcba5