fsinfo: include access bits for cockpit-files

[jelly]

In cockpit-files in some places we now check for file read/write or directory read/write with test -w, however this is something fsinfo could provide!

When watching a directory fsinfo already reacts on changes to files and the directory itself when permissions change except for ACL's. (Calling setfacl on a file gives no change event).

It seems we do watch for ATTRIB changes but they aren't send as updates.

We could provide if a file is readable / writeable and also for cwdInfo (the watched directory) with:

{
  'writeable': os.access(filename, os.W_OK)
  'readable': os.access(filename, os.R_OK)
}

Sadly os.access returns a boolean and not a bitmask so we would need 2 extra syscalls per file, unless I am missing something. I have tested ACL's and os.access takes those into account.

ACL Demo

# as root
echo 'phone' > /tmp/banana
chmod 640 /tmp/banana

# as user
$ cat /tmp/banana
permissions error

# as root
setfactl -m u:admin:r /tmp/banana

# as user
cat /tmp/banana
phone

As the access man page mentions something about ACL's not working on older glib'cs I tested on rhel-8-10 and there ACL's are accounted for.

[martinpitt]

Sadly os.access returns a boolean and not a bitmask so we would need 2 extra syscalls per file, unless I am missing something

This would be opt-in like everything else, in the info attribute, no? If you really would ask for both readable and writable, it would be two syscalls. But (1) that seems rare, and (2) unavoidable. os.access() cannot be derived from looking at the permission, as there's so much more going into it (user, auxiliary groups, ACLs, SELinux, etc.)

So this seems fine to me.

[allisonkarlitskaya]

I think this makes sense, subject to:

• it absolutely must be conditional (because of the syscall overhead)
• I prefer the names r-ok and w-ok, and maybe also add x-ok while you're at it
• please do it via the already-open fd (faccessat(fd, "", AT_EMPTY_PATH)) instead of via the filename

[allisonkarlitskaya]

One more: please make sure to use AT_SYMLINK_NOFOLLOW. We don't want to accidentally report information about symlink targets.

[allisonkarlitskaya]

It doesn't look like Python passes AT_EMPTY_PATH and it doesn't look like there's a flag to get us there either, so probably you have to do this name-based using the dir_fd. That's not excellent, but it's already what we do for stat, so eh...

[jelly]

Just checked the Python bug tracker and there is no issue yet for supporting passing AT_EMPTY_PATH to faccessat at all and it would need to support a new flags argument for that.

[jelly]

Just checked the Python bug tracker and there is no issue yet for supporting passing AT_EMPTY_PATH to faccessat at all and it would need to support a new flags argument for that.

Honestly, this is not the worst:

AT_EMPTY_PATH = 0x1000
if libc.faccessat(dir_fd, "", os.R_OK, AT_EMPTY_PATH) == 0:
    print('can read dir')

Also note that AT_EMPTY_PATH is Linux specific so unsure if Python wants to support that /o\