mirror of
https://github.com/jart/cosmopolitan.git
synced 2025-07-26 12:30:30 +00:00
Make improvements
- Introduce path module to redbean - Fix glitch with linenoise printing extra line on eof - Introduce closefrom() and close_range() system calls - Make file descriptor closing more secure in pledge.com
This commit is contained in:
Justine Tunney
parent
439ad21b12
commit
1837dc2e85
31 changed files with 806 additions and 75 deletions
|
|
@ -1959,6 +1959,96 @@ RE MODULE
|
|||
and re.Regex:search.
|
||||
|
||||
|
||||
────────────────────────────────────────────────────────────────────────────────
|
||||
PATH MODULE
|
||||
|
||||
The path module may be used to manipulate unix paths.
|
||||
|
||||
Note that we use unix paths on Windows. For example, if you have a
|
||||
path like C:\foo\bar then it should be /c/foo/bar with redbean. It
|
||||
should also be noted the unix module is more permissive when using
|
||||
Windows paths, where translation to win32 is very light.
|
||||
|
||||
path.dirname(str)
|
||||
└─→ str
|
||||
|
||||
Strips final component of path, e.g.
|
||||
|
||||
path │ dirname
|
||||
───────────────────
|
||||
. │ .
|
||||
.. │ .
|
||||
/ │ /
|
||||
usr │ .
|
||||
/usr/ │ /
|
||||
/usr/lib │ /usr
|
||||
/usr/lib/ │ /usr
|
||||
|
||||
path.basename(path:str)
|
||||
└─→ str
|
||||
|
||||
Returns final component of path, e.g.
|
||||
|
||||
path │ basename
|
||||
─────────────────────
|
||||
. │ .
|
||||
.. │ ..
|
||||
/ │ /
|
||||
usr │ usr
|
||||
/usr/ │ usr
|
||||
/usr/lib │ lib
|
||||
/usr/lib/ │ lib
|
||||
|
||||
path.join(str, ...)
|
||||
└─→ str
|
||||
|
||||
Concatenates path components, e.g.
|
||||
|
||||
x │ y │ joined
|
||||
─────────────────────────────────
|
||||
/ │ / │ /
|
||||
/usr │ lib │ /usr/lib
|
||||
/usr/ │ lib │ /usr/lib
|
||||
/usr/lib │ /lib │ /lib
|
||||
|
||||
You may specify 1+ arguments.
|
||||
|
||||
Specifying no arguments will raise an error. If nil arguments are
|
||||
specified, then they're skipped over. If exclusively nil arguments
|
||||
are passed, then nil is returned. Empty strings behave similarly to
|
||||
nil, but unlike nil may coerce a trailing slash.
|
||||
|
||||
path.exists(path:str)
|
||||
└─→ bool
|
||||
|
||||
Returns true if path exists.
|
||||
|
||||
This function is inclusive of regular files, directories, and
|
||||
special files. Symbolic links are followed are resolved. On error,
|
||||
false is returned.
|
||||
|
||||
path.isfile(path:str)
|
||||
└─→ bool
|
||||
|
||||
Returns true if path exists and is regular file.
|
||||
|
||||
Symbolic links are not followed. On error, false is returned.
|
||||
|
||||
path.isdir(path:str)
|
||||
└─→ bool
|
||||
|
||||
Returns true if path exists and is directory.
|
||||
|
||||
Symbolic links are not followed. On error, false is returned.
|
||||
|
||||
path.islink(path:str)
|
||||
└─→ bool
|
||||
|
||||
Returns true if path exists and is symbolic link.
|
||||
|
||||
Symbolic links are not followed. On error, false is returned.
|
||||
|
||||
|
||||
────────────────────────────────────────────────────────────────────────────────
|
||||
MAXMIND MODULE
|
||||
|
||||
|
|
@ -2220,12 +2310,36 @@ UNIX MODULE
|
|||
|
||||
Closes file descriptor.
|
||||
|
||||
This function should never be called twice for the same file
|
||||
descriptor, regardless of whether or not an error happened. The file
|
||||
descriptor is always gone after close is called. So it technically
|
||||
always succeeds, but that doesn't mean an error should be ignored.
|
||||
For example, on NFS a close failure could indicate data loss.
|
||||
|
||||
Closing does not mean that scheduled i/o operations have been
|
||||
completed. You'd need to use fsync() or fdatasync() beforehand to
|
||||
ensure that. You shouldn't need to do that normally, because our
|
||||
close implementation guarantees a consistent view, since on systems
|
||||
where it isn't guaranteed (like Windows) close will implicitly sync.
|
||||
|
||||
File descriptors are automatically closed on exit().
|
||||
|
||||
Returns `EBADF` if `fd` wasn't valid.
|
||||
|
||||
Returns `EINTR` possibly maybe.
|
||||
|
||||
Returns `EIO` if an i/o error occurred.
|
||||
|
||||
unix.read(fd:int[, bufsiz:str[, offset:int]])
|
||||
├─→ data:str
|
||||
└─→ nil, unix.Errno
|
||||
|
||||
Reads from file descriptor.
|
||||
|
||||
This function returns empty string on end of file. The exception is
|
||||
if `bufsiz` is zero, in which case an empty returned string means
|
||||
the file descriptor works.
|
||||
|
||||
unix.write(fd:int, data:str[, offset:int])
|
||||
├─→ wrotebytes:int
|
||||
└─→ nil, unix.Errno
|
||||
|
|
@ -3778,14 +3892,66 @@ UNIX MODULE
|
|||
├─→ true
|
||||
└─→ nil, unix.Errno
|
||||
|
||||
Unveil parts of a restricted filesystem view, e.g.
|
||||
Restricts filesystem operations, e.g.
|
||||
|
||||
unix.unveil(".", "r")
|
||||
unix.unveil(nil, nil)
|
||||
unix.unveil(".", "r"); -- current dir + children visible
|
||||
unix.unveil("/etc", "r"); -- make /etc readable too
|
||||
unix.unveil(0, 0); -- commit and lock policy
|
||||
|
||||
This can be used for sandboxing file system access.
|
||||
Unveiling restricts a thread's view of the filesystem to a set of
|
||||
allowed paths with specific privileges.
|
||||
|
||||
Unveil support is a work in progress.
|
||||
Once you start using unveil(), the entire file system is considered
|
||||
hidden. You then specify, by repeatedly calling unveil(), which paths
|
||||
should become unhidden. When you're finished, you call `unveil(0,0)`
|
||||
which commits your policy, after which further use is forbidden, in
|
||||
the current thread, as well as any threads or processes it spawns.
|
||||
|
||||
There are some differences between unveil() on Linux versus OpenBSD.
|
||||
|
||||
1. Build your policy and lock it in one go. On OpenBSD, policies take
|
||||
effect immediately and may evolve as you continue to call unveil()
|
||||
but only in a more restrictive direction. On Linux, nothing will
|
||||
happen until you call `unveil(0,0)` which commits and locks.
|
||||
|
||||
2. Try not to overlap directory trees. On OpenBSD, if directory trees
|
||||
overlap, then the most restrictive policy will be used for a given
|
||||
file. On Linux overlapping may result in a less restrictive policy
|
||||
and possibly even undefined behavior.
|
||||
|
||||
3. OpenBSD and Linux disagree on error codes. On OpenBSD, accessing
|
||||
paths outside of the allowed set raises ENOENT, and accessing ones
|
||||
with incorrect permissions raises EACCES. On Linux, both these
|
||||
cases raise EACCES.
|
||||
|
||||
4. Unlike OpenBSD, Linux does nothing to conceal the existence of
|
||||
paths. Even with an unveil() policy in place, it's still possible
|
||||
to access the metadata of all files using functions like stat()
|
||||
and open(O_PATH), provided you know the path. A sandboxed process
|
||||
can always, for example, determine how many bytes of data are in
|
||||
/etc/passwd, even if the file isn't readable. But it's still not
|
||||
possible to use opendir() and go fishing for paths which weren't
|
||||
previously known.
|
||||
|
||||
This system call is supported natively on OpenBSD and polyfilled on
|
||||
Linux using the Landlock LSM[1].
|
||||
|
||||
`path` is the file or directory to unveil
|
||||
|
||||
`permissions` is a string consisting of zero or more of the
|
||||
following characters:
|
||||
|
||||
- 'r' makes `path` available for read-only path operations,
|
||||
corresponding to the pledge promise "rpath".
|
||||
|
||||
- `w` makes `path` available for write operations, corresponding
|
||||
to the pledge promise "wpath".
|
||||
|
||||
- `x` makes `path` available for execute operations,
|
||||
corresponding to the pledge promises "exec" and "execnative".
|
||||
|
||||
- `c` allows `path` to be created and removed, corresponding to
|
||||
the pledge promise "cpath".
|
||||
|
||||
unix.gmtime(unixts:int)
|
||||
├─→ year,mon,mday,hour,min,sec,gmtoffsec,wday,yday,dst:int,zone:str
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue