Ensure I/O buffers in fsck_ffs(8) are 64-byte aligned

[mckusick]

Summary

Various disk controllers require their buffers to be aligned to 64 bytes. For buffers allocated in structures, ensure that they are 64-byte aligned.

Create a new Balloc() function in fsck_ffs(8) that allocates 64-byte aligned memory to use when allocating memory that will be used for doing disk I/O. Go through and change all buffer allocations to call Balloc().

On 64-bit architectures, malloc always returns 64-bit aligned allocations, so this change should have no effect. But if a bigger alignment requirement (say 128) comes down in the future, it will be trivial to implement it.

Test Plan

The aligned I/O problems most recent showed up in Azure, so test that it is works on Azure.

[imp]

Same general comments s for the other review: don't hard code 64, document which controllers, and consider just doing this all page aligned.

sbin/fsck_ffs/suj.c

2277

And maybe make this allocated with mem_aligned() for a PAGE_SIZE alignment (or the sysconf returned value since arm64 is now both 4k and 64k page sized).

[mckusick]

Updates reflecting comments from kib.

[kib]

Please use git diff -U999999 ... or like to generate patches, so that the context is available.

sbin/fsck_ffs/fsck.h

433

static inline void *

439

It is simpler to just fail on aligned_alloc() failure without trying to free one buffer, IMO. Malloc() failures mean that UVA is exhausted and/or fragmented, and apps usually cannot get out of it.

[mckusick]

Relevant parts of these changes were incorporated in commit 772430dd6795585 as discussed in PR 267654 and review https://reviews.freebsd.org/D41724.