[LU-14633] change ksocklnd over to use iov_iter_for_each_range()

• Type: Improvement
• Resolution: Unresolved
• Priority: Minor
• Labels: upstream
• Assignee: WC Triage
• Reporter: Andreas Dilger

Description

It came up in a discussion today on IRC that Lustre-staging was using iov_iter_for_each_range(), which Al Viro apparently added specifically for use by ksocknal_lib_recv() to clean up the code there.

This was done in upstream kernel commits:

• v4.14-rc4-19-g6d1ff4d6f37d "lnet_return_rx_credits_locked: don't abuse list_entry"
• v4.14-rc4-20-g09cf698a5942 "new primitive: iov_iter_for_each_range()"
• v4.14-rc4-23-g33ec9c5cae25 "lustre: switch struct ksock_conn to iov_iter"
• v4.14-rc4-24-g8ececffa12f5 "switch ksocknal_lib_recv_...() to use of iov_iter_for_each_range()"

It isn't clear why these changes weren't part of the patches being merged into master, but they look like a reasonable cleanup (remove a few hundred lines of code in total).

Since the iov_iter_for_each_range() is only available starting in 4.14 (it isn't in RHEL7.x), (and may be on the verge of being removed upstream, based on the IRC discussion) it makes sense to have a configure check and compat version of iov_iter_for_each_range() for kernels that do not provide it, and we can reintroduce it (if needed) when Lustre is pushed upstream.

Issue Links

• is related to: LU-12511 Prepare lustre for adoption into the linux kernel
• is related to: LU-12678 LNet simplification work from linux client

Comments

[James A Simmons]

The reason I haven't push this work is due to the iov changes for this lacking in RHLE7. Its going to take some work to make that to happen.

[Andreas Dilger]

James, Neil, I came across a few patches that Al Viro did to ksocklnd upstream that are missing from master that may be worthwhile to merge, so Al doesn't become unhappy about the regression on the future submission.