On Wed, Mar 27, 2024 at 11:19:06AM -0600, Jens Axboe wrote: > On 3/27/24 10:45 AM, Christian Brauner wrote: > > There's a bunch of flags that are purely based on what the file > > operations support while also never being conditionally set or unset. > > IOW, they're not subject to change for individual file opens. Imho, such > > flags don't need to live in f_mode they might as well live in the fops > > structs itself. And the fops struct already has that lonely > > mmap_supported_flags member. We might as well turn that into a generic > > fops_flags member and move a few flags from FMODE_* space into FOP_* > > space. That gets us four FMODE_* bits back and the ability for new > > static flags that are about file ops to not have to live in FMODE_* > > space but in their own FOP_* space. It's not the most beautiful thing > > ever but it gets the job done. Yes, there'll be an additional pointer > > chase but hopefully that won't matter for these flags. > > Not doing that extra dereference is kind of the point of the FMODE_* > flags, at least the ones that I care about. Probably not a huge deal for > these cases though, as we're generally going to call one of the f_op > handlers shortly anyway. The cases where we don't, at least for > io_uring, we already cache the state separately. > > Hence more of a general observation than an objection to the patch. I do > like freeing up FMODE space, as it's (pretty) full. Yes, I'm actuely aware that for some flags having them FMODE_* bits might be performance sensitive. Moving FMODE_PATH would probably be very noticeable due to it's use in __fget_files_rcu() and there might be other cases like FMODE_NOWAIT and so on. So I was delibaretely moving flags that really remain static and that are unlikely to be in a hot path. And we certainly need to be careful about this. I think in the long-run it can bring us benefits for new flag proposals that we wouldn't be willing to accept if they had to live in FMODE_* space. Either because they're a one-off (the MAP_SYNC flag comes to mind) or because they're not performance sensitive and of course only if they're static. We should still push back on unnecessary FMODE_* bit additions but we may be able to be a little less stingy with FOP_* bits for a bit.
