[musl] [BUG] hasmntopt weirdness

[musl] [BUG] hasmntopt weirdness

[Matthias Görgens]

Musl's `hasmntopt` is a bit weird.

For example, my `/proc/mounts` contains the following line:

shm /dev/shm tmpfs rw,nosuid,nodev,noexec,relatime,size=65536k,inode64 0 0

If I run `hasmntopt(entp, "suid")` on that entry with glibc, I get
NULL. In musl I get: "suid,nodev,noexec,relatime,size=65536k,inode64".

Do we have a spec for what should happen? I can find multiple
descriptions online, some of them would allow what musl does, but not
all. I don't know what's authoritative.

Arguably the musl behaviour is a bit silly by itself, and would
actively cause bugs for someone porting programs from glibc.

Re: [musl] [BUG] hasmntopt weirdness

[Rich Felker]

On Thu, Mar 30, 2023 at 06:02:43PM +0800, Matthias Görgens wrote:
> Musl's `hasmntopt` is a bit weird.
> 
> For example, my `/proc/mounts` contains the following line:
> 
> shm /dev/shm tmpfs rw,nosuid,nodev,noexec,relatime,size=65536k,inode64 0 0
> 
> If I run `hasmntopt(entp, "suid")` on that entry with glibc, I get
> NULL. In musl I get: "suid,nodev,noexec,relatime,size=65536k,inode64".
> 
> Do we have a spec for what should happen? I can find multiple
> descriptions online, some of them would allow what musl does, but not
> all. I don't know what's authoritative.
> 
> Arguably the musl behaviour is a bit silly by itself, and would
> actively cause bugs for someone porting programs from glibc.

I'm pretty sure it was agreed upon at one point that this was a bug,
but never pursued further.

From what I remember, I think hasmntopt needs to check that the match
found is at the start or immediately follows a comma (unless there is
some awful way to embed literal commas) and that it's followed by an
equals sign, comma, or nul terminator. Does this sound correct?

There's not really a need to use strstr to do this, but it might still
be the simplest implementation.

Rich

Re: [musl] [BUG] hasmntopt weirdness

[Matthias Görgens]

On Thu, 30 Mar 2023 at 21:57, Rich Felker <dalias@libc.org> wrote:
> I'm pretty sure it was agreed upon at one point that this was a bug,
> but never pursued further.

OK, that's good to know!

> From what I remember, I think hasmntopt needs to check that the match
> found is at the start or immediately follows a comma (unless there is
> some awful way to embed literal commas) and that it's followed by an
> equals sign, comma, or nul terminator. Does this sound correct?

For what it's worth, that's close to what glibc does.

Sorry, I'm new to this: Is there a spec to these things, or do we just copy what
the other C standard libraries are doing?

Re: [musl] [BUG] hasmntopt weirdness

[Rich Felker]

On Thu, Mar 30, 2023 at 10:47:27PM +0800, Matthias Görgens wrote:
> On Thu, 30 Mar 2023 at 21:57, Rich Felker <dalias@libc.org> wrote:
> > I'm pretty sure it was agreed upon at one point that this was a bug,
> > but never pursued further.
> 
> OK, that's good to know!
> 
> > From what I remember, I think hasmntopt needs to check that the match
> > found is at the start or immediately follows a comma (unless there is
> > some awful way to embed literal commas) and that it's followed by an
> > equals sign, comma, or nul terminator. Does this sound correct?
> 
> For what it's worth, that's close to what glibc does.
> 
> Sorry, I'm new to this: Is there a spec to these things, or do we just copy what
> the other C standard libraries are doing?

For standard functions, there is a spec.

For nonstandard ones, generally no, and that's why musl frowns on
them. We have to basically do what the standardization process would
have done: looking at historical practice and whether the interface
and behavior was sufficiently consistent across different historical
systems with the nonstandard extension that we can work out a behavior
that's not conflicting with what applications might expect.

When this is difficult or impossible, that's a strong indication that
we shouldn't adopt the interface at all. But due to historical
mistakes (not doing sufficient analysis to apply these principles), we
let a few things in we shouldn't have.