On Fri, Jan 07, 2022 at 05:52:27AM +0000, Al Viro wrote: > > Looks good, assuming Al is ok with the re-factoring. > > Reviewed-by: Ian Kent <raven@xxxxxxxxxx> > > Ummm.... Mind resending that? I'm still digging myself from under > the huge pile of mail, and this seems to have been lost in process... Non-obvious part is that current code only does this forgetting the root when we are certain that we won't look at it later in pathwalk. IOW, it's guaranteed to be the same through the entire thing. This patch changes that; the final component may very well be e.g. an absolute symlink. We won't know that until we unlazy, so we can't make forgetting conditional upon that. I _think_ it's not going to lead to any problems, but I'll need to take a good look at the entire thing after I get some sleep - I'm about to fall down right now. Other problems here (aside of whitespace damage - was that a cut'n'paste of some kind? Looks like 8859-1 NBSP for each leading space...) are * misleading name of the new helper - it sounds like "non-RCU side of complete_walk()" and that's not what it does * LOOKUP_CACHED needs to be mentioned in commit message (it's incompatible with O_CREAT, so it couldn't occur on that codepath - the transformation is an equivalent one, but that ought to be mentioned) * the change I mentioned above needs to be in commit message - this one is a lot more subtle. Anyway, I'll look into that tomorrow - too sleepy right now to do any serious analysis.
