Page MenuHomeFreeBSD
Differential D51138

Mention write(2) can return EINVAL in when the target is a block device
ClosedPublic
Actions

Authored by felix.the.red_gmail.com on Jul 3 2025, 12:20 AM.
Tags
None
Referenced Files
Unknown Object (File)
Thu, Jul 31, 10:47 AM
Unknown Object (File)
Mon, Jul 28, 11:25 PM
Unknown Object (File)
Mon, Jul 28, 8:10 PM
Unknown Object (File)
Mon, Jul 28, 5:36 PM
Unknown Object (File)
Mon, Jul 28, 4:07 PM
Unknown Object (File)
Mon, Jul 28, 3:52 PM
Unknown Object (File)
Mon, Jul 28, 11:51 AM
Unknown Object (File)
Mon, Jul 28, 8:29 AM
View All 27 Files
Subscribers
imp
pauamma_gundo.com
ziaee

Details

Reviewers
imp
pauamma_gundo.com
ziaee
Group Reviewers
manpages
Commits
rG137fcd6dbc52: write.2: Add EINVAL to ERRORS
rGbd173a540c0e: write.2: Add EINVAL to ERRORS
Summary

lib/libsys/write.2: Add EINVAL entry for when nbytes (write size)
is not a multiple of the sector's block size.

From the original PR:
When coming from Linux, people are surprised by the fact
that write(2) to a device must be aligned to block size.
Writing a non-aligned block onto a raw device is a case
where EINVAL is also returned but the manpage does not mention this.

PR:227185

Diff Detail

Repository
rG FreeBSD src repository
Lint
Lint Not Applicable
Unit
Tests Not Applicable

Event Timeline

felix.the.red_gmail.com created this revision.Jul 3 2025, 12:20 AM
Herald added subscribers: ziaee, imp. · View Herald TranscriptJul 3 2025, 12:20 AM
felix.the.red_gmail.com requested review of this revision.Jul 3 2025, 12:20 AM
imp accepted this revision.Jul 3 2025, 8:21 PM

Yea, when this write(2) man page was written, this wasn't a restriction because we had both char and block devices for the disks and the former allowed any offset / size, while the latter had the restriction. Then phk rewrote the block layer and enforced this restriction.... So this has been the case for maybe 20 years now?

This revision is now accepted and ready to land.Jul 3 2025, 8:21 PM
ziaee added a comment.Jul 3 2025, 8:26 PM

Awesome. Thank you both so much, I'll merge this now!

ziaee added a comment.Jul 3 2025, 8:37 PM

Wait, should we have two EINVAL entries and not rewrite them both into one?

pauamma_gundo.com added a subscriber: pauamma_gundo.com.Jul 3 2025, 8:55 PM
In D51138#1167528, @ziaee wrote:

Wait, should we have two EINVAL entries and not rewrite them both into one?

Apparently allowed. See EPIPE earlier in that manual page.

pauamma_gundo.com accepted this revision.Jul 3 2025, 8:58 PM
imp added a comment.Jul 3 2025, 9:02 PM
In D51138#1167534, @pauamma_gundo.com wrote:
In D51138#1167528, @ziaee wrote:

Wait, should we have two EINVAL entries and not rewrite them both into one?

Apparently allowed. See EPIPE earlier in that manual page.

Happens all the time for EINVAL too and matches the POSIX docs.

ziaee accepted this revision.Jul 3 2025, 10:09 PM

Awesome, thank you so much, just wanted to check! I'll merge this when I get home, unless someone else wants to!

Closed by commit rGbd173a540c0e: write.2: Add EINVAL to ERRORS (authored by felix.the.red_gmail.com, committed by ziaee). · Explain WhyJul 4 2025, 5:19 AM
This revision was automatically updated to reflect the committed changes.
ziaee added a commit: rGbd173a540c0e: write.2: Add EINVAL to ERRORS.
ziaee added a commit: rG137fcd6dbc52: write.2: Add EINVAL to ERRORS.Mon, Jul 14, 5:27 PM

Revision Contents
Changeset List

PathSize
lib/
libsys/
3 lines

Diff 157925

View Options

lib/libsys/write.2

Loading...