3 ms·

> because it’s such a run-away-screaming scary idea for security I think that's the root of our disagreement. Few years ago I've been of the same mind about gi

> because it’s such a run-away-screaming scary idea for security I think that's the root of our disagreement. Few years ago I've been of the same mind about giving the web new powerful features, but after watching the space and seeing the existing alternatives I've come to change my mind. As mentioned in another thread below (https://news.ycombinator.com/item?id=30013287 https://news.ycombinator.com/item?id=30013287), vendors who need this sort of functionality, already find ways to implement it via other proprietary methods like local executables that, unlike implementations of Web APIs, are not reviewed by other teams and usually expose all sorts of critical stuff over local HTTP servers or in another insecure fashion. In the end, it's not a question of "if" we want to expose those features to the web apps, it's "how" we can do so with minimal risks to the users, and that's where web APIs with thought-out permission models, explicit requests and history of cross-origin checks can help. Of course, you can also say that vendors can continue to implement those things insecurely anyway even when new APIs exist, but 1) in practice when they're given a simpler way to do the same thing, they tend to go for that more often instead of inventing custom hacks and 2) that's where advocacy of the new APIs comes in, and what I'm trying to do by showing what the web can be if we let it. As for your other arguments about potential for breaking changes if/when those APIs get adopted by other browsers, I agree I might be overoptimistic and your prediction might very well turn out to be true. Those APIs map to basic USB concepts quite closely, so I can't imagine what changes would be necessary, but, of course, I don't have enough experience with WebUSB outside this project to say that it's 100% impossible :) In the end, it's a bit of a chicken-and-egg problem - in order for browsers to see the interest or get feedback, there have to be apps trying to build something with those APIs and reporting bugs / feature requests, and for those apps to use those APIs, there has to be at least one implementation first. This problem can be chewed from either end, and I'm trying to do my part by showing developers how those APIs can be used for porting interesting apps and libraries.

chrismorgan 5y ago
My main disagreement with you is not based on the soundness or otherwise of WebUSB functionally, but your position on it as a specification. Still, I’ll address the matter of WebUSB itself. WebUSB is so scary for security because neither USB devices nor desktop operating systems have been written with such a use case in mind. There are two key problems in WebUSB: ⓐ USB devices weren’t designed with public exposure in mind, so that they’re commonly insecure and brickable or hijackable if you get access to them; and ⓑ operating systems weren’t designed with malicious USB devices in mind. Combine these, and you’ve got potential total sandbox escape and remote code execution, guarded only by a flimsy permissions popup that doesn’t explain what can happen—can’t explain what can happen. The threat here is probably the most severe (using the definition in the severity/probability model of risk assessment) ever contemplated for the web. Yes, there are legitimate uses of such an API. Yes, this makes life much easier and even safer for such uses. But if the concept doesn’t also give you the heeby-jeebies, you haven’t thought the implications through enough, haven’t seen how bad programmers are at their jobs, and haven’t observed how hopelessly ineffective permissions popups are at protecting users. I watched the video in your article and was horrified that it’s not even trying to protect you or indicate that this is anything other than routine: you just get a “‹origin› wants to connect” popup with a list of routinely-cryptic device names (“ILCE-6600” rather than “Sony α6600”) that you can choose from. Yet this is a thing that, with the wrong USB device and a malicious attacker that knows about a vulnerability in it, could literally set fire to your computer, steal all your passwords, or silently install malware on the USB device or computer. For myself, I might prefer to have WebUSB, because I trust my judgement in what I execute and because it’s more likely to support my platform. But for the population at large, on the balance of things, no, I agree with Mozilla’s position that it’s harmful, and would regretfully condemn people to continue downloading and executing untrusted code from the manufacturer, because that’s more likely to be done deliberately and only for first-party code. WebUSB is problematic because there’s probably no safe subset of its functionality: it seems to be fundamentally dangerous. People have tried proposing a few subsets, from memory (I’ve occasionally paid a little attention to it), but so far they’ve all been rejected as either not safe enough or too crippling; and so Chromium has declined to be crippled and implemented the wildly unsafe with only mild mitigations to limit the class of devices exposed, and Gecko and WebKit have declined to play it dangerously. It’s similar to the question of exposing raw TCP, which was contemplated, but ended up being deemed too dangerous, and so it was wrapped up in HTTP as WebSockets: something not quite as useful, but safe. It’s possible someone could come up with something like that for WebUSB, something that constrains it to something safe. Such a constraint could easily require major API changes. That WebUSB currently matches the fundamental USB concepts closely is no saving grace in the stability of the protocol.
- RReverser 5y ago
  > It’s possible someone could come up with something like that for WebUSB, something that constrains it to something safe. That sounds like WebSerial and WebHID to me, which don't require similar shenanigans with Zadig on Windows or Linux permission changes, because they allow access to a limited subset of USB device functions. But that will never cover all the use-cases for USB apps. For those the only choices are not to do anything at all and let closed native executables to continue dominate this space (which gives them unlimited access without even those simplest permission prompts), or to work on WebUSB and encourage devs who do need to interact with such devices over USB to switch to that instead. I don't see 3rd option or how it would be possible in our non-ideal world, and among those two WebUSB still seems like a better idea.

chrismorgan 5y ago

My main disagreement with you is not based on the soundness or otherwise of WebUSB functionally, but your position on it as a specification. Still, I’ll address the matter of WebUSB itself. WebUSB is so scary for security because neither USB devices nor desktop operating systems have been written with such a use case in mind. There are two key problems in WebUSB: ⓐ USB devices weren’t designed with public exposure in mind, so that they’re commonly insecure and brickable or hijackable if you get access to them; and ⓑ operating systems weren’t designed with malicious USB devices in mind. Combine these, and you’ve got potential total sandbox escape and remote code execution, guarded only by a flimsy permissions popup that doesn’t explain what can happen—can’t explain what can happen. The threat here is probably the most severe (using the definition in the severity/probability model of risk assessment) ever contemplated for the web. Yes, there are legitimate uses of such an API. Yes, this makes life much easier and even safer for such uses. But if the concept doesn’t also give you the heeby-jeebies, you haven’t thought the implications through enough, haven’t seen how bad programmers are at their jobs, and haven’t observed how hopelessly ineffective permissions popups are at protecting users. I watched the video in your article and was horrified that it’s not even trying to protect you or indicate that this is anything other than routine: you just get a “‹origin› wants to connect” popup with a list of routinely-cryptic device names (“ILCE-6600” rather than “Sony α6600”) that you can choose from. Yet this is a thing that, with the wrong USB device and a malicious attacker that knows about a vulnerability in it, could literally set fire to your computer, steal all your passwords, or silently install malware on the USB device or computer. For myself, I might prefer to have WebUSB, because I trust my judgement in what I execute and because it’s more likely to support my platform. But for the population at large, on the balance of things, no, I agree with Mozilla’s position that it’s harmful, and would regretfully condemn people to continue downloading and executing untrusted code from the manufacturer, because that’s more likely to be done deliberately and only for first-party code. WebUSB is problematic because there’s probably no safe subset of its functionality: it seems to be fundamentally dangerous. People have tried proposing a few subsets, from memory (I’ve occasionally paid a little attention to it), but so far they’ve all been rejected as either not safe enough or too crippling; and so Chromium has declined to be crippled and implemented the wildly unsafe with only mild mitigations to limit the class of devices exposed, and Gecko and WebKit have declined to play it dangerously. It’s similar to the question of exposing raw TCP, which was contemplated, but ended up being deemed too dangerous, and so it was wrapped up in HTTP as WebSockets: something not quite as useful, but safe. It’s possible someone could come up with something like that for WebUSB, something that constrains it to something safe. Such a constraint could easily require major API changes. That WebUSB currently matches the fundamental USB concepts closely is no saving grace in the stability of the protocol.

RReverser 5y ago
> It’s possible someone could come up with something like that for WebUSB, something that constrains it to something safe. That sounds like WebSerial and WebHID to me, which don't require similar shenanigans with Zadig on Windows or Linux permission changes, because they allow access to a limited subset of USB device functions. But that will never cover all the use-cases for USB apps. For those the only choices are not to do anything at all and let closed native executables to continue dominate this space (which gives them unlimited access without even those simplest permission prompts), or to work on WebUSB and encourage devs who do need to interact with such devices over USB to switch to that instead. I don't see 3rd option or how it would be possible in our non-ideal world, and among those two WebUSB still seems like a better idea.