Show HN: Host your own minimal file sharing site (now with expiring files) | Hacker News Reader

Show HN: Host your own minimal file sharing site (now with expiring files)

github.com

110 pointsnikosch8630 comments

Please investigate 'safe' php file upload handling, specifically: the `is_uploaded_file` and `move_uploaded_file` functions.

Also perhaps consider the `mkdir` function, and perhaps 'file_put_contents'. Your reliance on shell functions via `exec` is quite concerning, given what you're doing.

Lastly, I'd suggest either `random_int` or `random_bytes` (combined with e.g. bin2hex, or a hashing function to get an ascii string) instead of uniqid.

I'm not quite sure why there's a check on user agent, to carry out the same logic regardless?

I think OP's code is actually markedly better than your first two suggestions exactly because it doesn't use any of the advanced PHP APIs. It's all there, bare and easy to check, with no gotchas, ini-dependent caveats and what not.

I agree that uniqid(), as used, is not the best choice, but that's only because it may produce dupes if the machine's clock is rolled back (ie by ntpd). Regardless of the prng choice, the code could use a check for whether selected folder already exists.

What "gotchas" or caveats are you expecting from `is_uploaded_file` or `move_uploaded_file`?

Any use of `exec` should be considered dangerous, when there's native language/stdlib support for achieving the same thing.

Edit: corrected function name.

`is_uploaded_file` and `move_uploaded_file` are the band-aid API meant for broken code that may be mistakenly working with some random local files instead of uploaded ones.

This is akin to using "if (3 == x)" notation in C in a fear of accidentally writing "if (x = 3)".

For this particular case the use of both `rename` and `exec` would've been perfectly fine.

> Any use of `exec` should be considered dangerous

By whom? It may be a red flag if used by an inexperienced programmer or if it appears in an otherwise messy code base. But if I know what I'm doing, then it's just another API that's perfectly OK when used with due care.

> `is_uploaded_file` and `move_uploaded_file` are the band-aid

It's not a band-aid. When a file is uploaded (see RFC 1867) via HTTP POST, PHP internally allocates a hash table that tracks which files on the filesystem have been uploaded. When the process terminates, it uses that hashtable to remove any remaining uploaded files in the tmp space.

move_uploaded_file() confirms that the filename was indeed created as a result of an upload, and additionally removes that entry from the hash table.

> move_uploaded_file() confirms that the filename was indeed created as a result of an upload, and additionally removes that entry from the hash table.

In other words, move_uploaded_file() protects against moving a non-uploaded file or moving one more than once.

So it's for an upload processing code that, at some point, is becoming unsure of what the heck it is doing.

It is a band-aid.

> In other words, move_uploaded_file() protects against moving a non-uploaded file or moving one more than once.

Sure.

> So it's for an upload processing code that, at some point, is becoming unsure of what the heck it is doing.

How do you get to this conclusion from the above?

> It is a band-aid.

Nope. Uploaded file checking would seem to be necessary, though rather than requiring the developer to implement it, there's some lang-sec value in php doing the check automatically with any $_FILES references and instead allowing the developer to suppress it. That's an area where implementation can be refined, but it doesn't make the current solution a band-aid by any stretch.

Edit: as an example, I'd much rather see the example given in php docs actually implemented as part of the framework rather than delegated to the developer. But aside from just dumping all uploaded files in one quarantined location, this solution is a functional pattern. https://www.php.net/manual/en/function.is-uploaded-file.php

(I'm not a php dev, and docs on the relevant functions date back to 2003, so for all I know, this has been rectified in the intervening 17 years)

`exec` is and should be disabled for as many hosts as possible.

The OPs code might be totally fine, but that other script laying there since 1999 might not be. Or a third party library, or else.

By that logic, I guess you are fine with using `exec` to run Python code to rename the file? It's "just another API".

Part of knowing what you're doing involves knowing the standard library of the language you've chosen to use. There might be (very rare) times you need to use exec in a PHP application - a file upload script is not one of them.

good point with the mkdir, mv and touch, I just pushed an update to use php functions for that instead of shell. No reason to be concerned though, the user controlled input is sanitized and no shell command is/was executed based on unsanitized user input.

the user agent check was a relic from when it behaved differently, browser vs curl, it is not needed anymore and has been removed. Pull Request are always welcome

edit: I also just pushed an update to use bin2hex(random_bytes(16)) instead of uniquid to make sure we never have collisions, for arguments sake (even though I thought that risk was rather low to begin with)

There are already transfer.sh [1] and other established projects (which I can’t recall right now) that serve this purpose. With more fine-grained control too. A golang compiled executable is also arguably more lightweight than a php app, whether the feature set/code is simpler or not.

[1] https://github.com/dutchcoders/transfer.sh/

Edit: Since more than one person has replied with non-self-hosted solutions, I should note that transfer.sh can be self-hosted, although you can also use the website itself.

https://bashupload.com/

That’s not self-hosted.

https://0x0.st

Yeah, this & https://transfer.sh are my choice for a long time.

The goal was a minimum viable implementation of a file sharing service that can be self hosted. Simplicity to this extreme serves many purposes, it is easy to understand and debug, has less attack surface, less dependencies and is small (in terms of lines of Code, less than 100, and size, less than 50K).

If one needs more features or more control, there are certainly many options out there, I created this not to compete with other projects but to provide a super simple solution to a specific problem.

I like the idea of a really small MVP; I might give this a try later in Prolog to see how they compare.

Definitely not as concise as your PHP impl but meh https://gist.github.com/MagnificentPako/b67b081d5cac0d0185ef...

There is also Mozilla's Send.firefox.com, which could be self-hosted https://github.com/mozilla/send

If you want something with a bit more features, there is filebin[1]. It has clients for Linux, BSD and Android, runs on PHP with support for various databases, offers syntax highlighting for multiple formats and other stuff like plain view, multipaste, tar download and LDAP integration.

[1] http://git.server-speed.net/users/flo/filebin/

I built something similar (I call it IFS[0]). Uses no database, and x-plat via .NET Core.

[0]: https://github.com/Sebazzz/IFS/

There's also dl, which supports creating single use links for third parties to upload files: https://www.thregr.org/~wavexx/software/dl/

I've been hosting dl at work for many years and it's filled our needs admirably. Simple, zero maintenance, just works.

It even has a useful Thunderbird extension that auto-uploads any large attachment using the FileLink API when composing a new mail, adding the dl URL in the email's body. Unfortunately the extension stopped working after the migration to WebExtensions.

That's pretty handy actually

Awesome, thanks!

Reply on news.ycombinator.com