On Sat, Aug 30, 2025 at 12:40:32PM -0700, Linus Torvalds wrote: > On Sat, 30 Aug 2025 at 00:33, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote: > > > > So... Siddhesh, could you clarify the claim about breaking getmntent(3)? > > Does it or does it not happen on every system that has readonly AFS > > volumes mounted? > > Hmm. Looking at various source trees using Debian code search, at > least dietlibc doesn't treat '#' specially at all. > > And glibc seems to treat only a line that *starts* with a '#' > (possibly preceded by space/tab combinations) as an empty line. > > klibc checks for '#' at the beginning of the file (without any > potential space skipping before) > > Busybox seems to do the same "skip whitespace, then skip lines > starting with '#'" that glibc does. > > So I think the '#'-escaping logic is wrong. We should only escape '#' > marks at the beginning of a line (since we already escape spaces and > tabs, the "preceded by whitespace" doesn't matter). > > And that means that we shouldn't do it in 'mangle()' at all - because > it's irrelevant for any field but the first. > > And the first field in /proc/mounts is that 'r->mnt_devname' (or > show_devname), and again, that should only trigger on the first > character, not every character. *nod* Amusingly enough, glibc addmntent(3) does *not* consider # for an octal escape. BTW, another amuzing bogosity: seq_escape(m, "blah", "X") => "blah" seq_escape(m, "blah", "b") => "\142lah" seq_escape(m, "blah", "") => "\142\154\141\150" IOW, about 10 years ago an empty string switched meaning from "escape nothing" to "escape everything"...
