fix users of cn_flags for 64 bits
AcceptedPublic
Actions

Authored by sigsys_gmail.com on Apr 4 2020, 4:40 AM.

Details

Reviewers

kib
markj

Summary

namei's componentname struct has a cn_flags field that was changed to 64 bits at some point to allow for more flags, but some users of this field still treat it as an int (or sometimes long).

This patch fixes all of those that I could find.

I added the ULL suffix to the individual flag macro literals to make masking operations just work.

I changed some variables to bool but I was careful to do so in a way that would still work if bool is typedef'd to int.

Diff Detail

Lint

Lint Skipped

Unit

Tests Skipped

Event Timeline

sigsys_gmail.com created this revision.Apr 4 2020, 4:40 AM

Herald added subscribers: asomers, delphij. · View Herald TranscriptApr 4 2020, 4:40 AM

Updated to use uint64_t instead of u_int64_t. No other changes.

kevans added reviewers: kib, markj.Apr 4 2020, 2:21 PM

kib accepted this revision.Apr 4 2020, 2:36 PM

kib added inline comments.

sys/sys/namei.h
193	Does the op arg need to be long ?

This revision is now accepted and ready to land.Apr 4 2020, 2:36 PM

markj accepted this revision.Apr 4 2020, 5:21 PM

markj added inline comments.

sys/fs/fuse/fuse_vnops.c
984–986	I would make wantparent and islastcn into bools as well: bool wantparent = (flags & (LOCKPARENT \| WANTPARENT)) != 0; bool islastcn = (flags & ISLASTCN) != 0; just to explicitly avoid narrowing.
sys/fs/smbfs/smbfs_vnops.c
1199	Same as above regarding the use of a bool.

sigsys_gmail.com added inline comments.Apr 4 2020, 6:23 PM

sys/fs/fuse/fuse_vnops.c
984–986	Oops missed those ones. I'll update the diff.
sys/sys/namei.h
193	I can't see any reason it would need to be. It gets saved in int variables too sometimes. But I just noticed an assert in vfs_lookup.c that checks if cn_nameiop gets "contaminated" with flags, so if the op is larger it allows to detect more errors like this.