[00/14] xfsprogs: make buffer functions return error codes

Message

Darrick J. Wong Feb. 20, 2020, 1:44 a.m. UTC

Hi all,

Let's fix all the xfs read/get buffer functions to return the usual
integer error codes and pass the buffer pointer as a out-argument.  This
makes it so that we can return useful error output instead of making
callers infer ENOMEM or EAGAIN or whatever other reality they crave from
the NULL pointer that they get when things don't go perfectly.

FWIW, all XBF_TRYLOCK callers must now trigger retries if they receive
EAGAIN.  This may lead to a slight behavioral change in that TRYLOCK
callers will no longer retry for things like ENOMEM, though I didn't see
any obvious changes in user-visible behavior when running fstests.

After finishing the error code conversion, we straighten out the TRYLOCK
callers to remove all this null pointer checks in favor of explicit
EAGAIN checks; and then we change the buffer IO/corruption error
reporting so that we report whoever called the buffer code even when
reading a buffer in transaction context.

The userspace port of the kernel patchset was very difficult, so I am
submitting this series to avoid dumping the work on Eric.

If you're going to start using this mess, you probably ought to just
pull from my git trees, which are linked below.

This is an extraordinary way to destroy everything.  Enjoy!
Comments and questions are, as always, welcome.

--D

xfsprogs git tree:
https://git.kernel.org/cgit/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=buf-return-errorcodes

Comments

Darrick J. Wong Feb. 20, 2020, 1:53 a.m. UTC | #1

On Wed, Feb 19, 2020 at 05:44:29PM -0800, Darrick J. Wong wrote:
> Hi all,
> 
> Let's fix all the xfs read/get buffer functions to return the usual
> integer error codes and pass the buffer pointer as a out-argument.  This
> makes it so that we can return useful error output instead of making
> callers infer ENOMEM or EAGAIN or whatever other reality they crave from
> the NULL pointer that they get when things don't go perfectly.
> 
> FWIW, all XBF_TRYLOCK callers must now trigger retries if they receive
> EAGAIN.  This may lead to a slight behavioral change in that TRYLOCK
> callers will no longer retry for things like ENOMEM, though I didn't see
> any obvious changes in user-visible behavior when running fstests.
> 
> After finishing the error code conversion, we straighten out the TRYLOCK
> callers to remove all this null pointer checks in favor of explicit
> EAGAIN checks; and then we change the buffer IO/corruption error
> reporting so that we report whoever called the buffer code even when
> reading a buffer in transaction context.
> 
> The userspace port of the kernel patchset was very difficult, so I am
> submitting this series to avoid dumping the work on Eric.

I should've noted that the "xfsprogs: refactor buffer function names"
series applies against 5.5; this series is a backport of the second half
of the kernel 5.6 merge.  I didn't send that part because I didn't feel
that it was significant enough to increase the patchbomb spew.

SO... the order is:

1. "refactor buffer function names", which is the thread before this one
2. "sync with 5.6", available at [1]
3. "make buffer functions return error codes", which is this one

--D

[1] https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/tag/?h=libxfs-5.6-sync_2020-02-19

> If you're going to start using this mess, you probably ought to just
> pull from my git trees, which are linked below.
> 
> This is an extraordinary way to destroy everything.  Enjoy!
> Comments and questions are, as always, welcome.
> 
> --D
> 
> xfsprogs git tree:
> https://git.kernel.org/cgit/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=buf-return-errorcodes