vm_ooffset_t is now unsigned

[vangyzen]

Summary

vm_ooffset_t is now unsigned. Remove some tests for negative values,
or make other adjustments accordingly.

Test Plan

Dead code is hard to test.

[vangyzen]

sys/kern/vfs_vnops.c

2620–2630

I'm not sure how to update this comment. Suggestions welcome.

[kib]

sys/fs/tmpfs/tmpfs_subr.c

178

I think this should be size_t, and before subtraction, the check should be made to return 0.

sys/kern/vfs_vnops.c

2620–2630

I believe the comment is still valid, it depends on the vm_ooffset_t being type-punned for off_t. Check below (foff > OFF_MAX-size) ensures that we do not require fs to mmap pass OFF_MAX.

[vangyzen]

• CR feedback

[vangyzen]

sys/kern/vfs_vnops.c

2620–2630

Makes sense. Thank you.

[kib]

sys/fs/tmpfs/tmpfs_subr.c

183

You perform two reads of tmpfs_pages_reserved. If user changes the sysctl meanime, you might get overflow on the second calculation.

[vangyzen]

• fix TOCTOU bug using tmpfs_pages_reserved

[markj]

sys/fs/tmpfs/tmpfs_subr.c

183

To make this pedantically correct I believe you have to use atomic_load_long(&tmpfs_pages_reserved). There is nothing preventing the compiler from loading tmpfs_pages_reserved twice.