Difference between revisions of "Talk:Homebrew ABI"

From Nintendo Switch Brew
Jump to navigation Jump to search
 
(2 intermediate revisions by 2 users not shown)
Line 73: Line 73:
  
 
--[[User:Misson20000|Misson20000]] ([[User talk:Misson20000|talk]]) 00:57, 10 November 2018 (UTC)
 
--[[User:Misson20000|Misson20000]] ([[User talk:Misson20000|talk]]) 00:57, 10 November 2018 (UTC)
 +
 +
Workarounds (which only exist for a debugger) don't belong in the HBABI. --[[User:Yellows8|Yellows8]] ([[User talk:Yellows8|talk]]) 19:17, 12 February 2019 (UTC)
 +
 +
Almost all of HBABI is workarounds to facilitate running homebrew applications in dirty environments:
 +
  - bad entrypoint arguments (MainThreadHandle)
 +
  - reusing the same process (OverrideHeap, Argv (can't use loader-args), AllocPages, LockRegion, RandomSeed)
 +
  - legacy exploits (OverrideService, AppletWorkaround)
 +
  - working around 0xffff8001 check in SVCs (ProcessHandle)
 +
 +
And although it's not specified in the HBABI, reading back argv[0] for ASET is also a way to work around not being installed as a real application and being able to access romfs like official applications do.
 +
 +
The only thing I'm working around here is the ASET workaround that needlessly ties homebrew to the SD card. Even if this is a workaround for a debugger, I don't think that's grounds for not belonging in the HBABI. It's not like Nintendo doesn't do anything similar ("Redirect*" commands in lr).
 +
--[[User:Misson20000|Misson20000]] ([[User talk:Misson20000|talk]]) 21:42, 12 February 2019 (UTC)
 +
 +
Reusing the same process is the approach selected for nx-hbloader which is the standard way to load homebrew, and I see no reason why they would be labeled as "workarounds". AllocPages and LockRegion were unilaterally added by you several months ago and nx-hbloader has no use for them. OverrideService and AppletWorkaround are not presently used by anything and they were mostly added for future-proofing in case we need to support a more restrictive environment where we need to pass stolen service handles, they don't exist for "legacy" exploits.
 +
 +
I see no reason to add a romfs override key. If we want to support non-SD card paths, we will add other device name prefixes to argv[0] (such as host:/) and add support for these devices in libnx.
 +
 +
--[[User:Fincs|Fincs]] ([[User talk:Fincs|talk]]) 22:12, 12 February 2019 (UTC)

Latest revision as of 22:13, 12 February 2019

StdioSockets

Currently, there's a broken link in the Loader Config Keys section on the StdioSockets key, caused by a recent removal of the section. I understand the temptation to remove StdioSockets, since it wasn't very well designed in the first place, and nobody uses it anymore, but I don't think that it makes sense to remove it from the ABI specification. The key shouldn't be erased entirely, because that would indicate that it is unreserved. A backwards-incompatible change like that defeats one of the purposes of the ABI. Removing the link and adding a label that the key is deprecated would be acceptable, but it'd leave no specification on the key. I think that, unless I'm misinterpreting the original intent of the edit, it should be reverted and labels should be added to the key's specification that it is deprecated and no longer in use.

Mentioning WinterMute.

--Misson20000 (talk) 17:50, 18 June 2018 (CDT)

Probably better to just reserve it for future use I guess. Had someone interrogate me about why nxlink didn't use the homebrew ABI "properly" so it seemed better to remove it rather than leave it there confusing people.

WinterMute (talk) 04:26, 19 June 2018 (CDT)

RomFS Override

The current strategy for providing RomFS to homebrew applications depends on the application being able to find and open its own NRO file via argv[0] and load RomFS from ASET. This is not practical for several use cases that I am encountering:

- The NRO file is not persisted to the SD card.

- The NRO file is not stored somewhere that the homebrew library knows how to access.

- The application does not exist as an NRO file at all.

I'd like to propose a new HBABI key to allow homebrew applications to access RomFS without needing to open their own NRO file via argv[0].

   ==== RomFSOverride ====
   This is used to override an application's RomFS. If this key is passed and recognized, the application shall make no attempt to open itself via fopen(argv[0]).
   
   * Key: 15
   * Value[0]: Handle to a session implementing Filesystem_services#IFileSystem for the application's RomFS.
   * Value[1]: Ignored.
   
   The given handle must respond to IFileSystem commands 0-12.

Possible alternatives include instead passing IFile/IStorage for the RomFS section, or IFile/IStorage for the entire ASET section.

Concerns include ownership of the session handle. Should the application be required to close the handle or leave it open? Personally, I'm inclined to require that the application borrow the session from the loader and leave it open in case closing the handle is not something desirable. This also simplifies the logic in the loader, since if the handle needs to be closed, the loader can always just close it instead of having to figure out whether it was recognized or not. Possible alternatives to this ownership scheme include either adding flags to HBABI entries for which words contain handles that should be closed if not recognized, or a flag that the application sets on each entry it recognizes so that the loader can take care of closing the handle if the application did not recognize it.

--Misson20000 (talk) 00:36, 6 November 2018 (UTC)

"- The NRO file is not persisted to the SD card." "- The NRO file is not stored somewhere that the homebrew library knows how to access." "- The application does not exist as an NRO file at all." The ABI requires the NRO to located on SD in the first place... --Yellows8 (talk) 19:34, 8 November 2018 (UTC)
Like yellows8 said, the homebrew ABI is designed around NRO files stored on SD card. If we ever support other filesystems (such as hostio access), we'd just add support for host:/ paths being passed through argv (as well as support code in libnx which would be trivial) - homebrew apps would need a recompile anyway with or without new ABI keys. Also, for that use case (hostio), romfs doesn't really make sense because you could just use assets in a folder inside the host filesystem instead of having to go through the trouble of building romfs (which can get pretty large). Finally, homebrew by definition is NRO formatted. Other formats should not be supported or promoted. --Fincs (talk) 19:47, 8 November 2018 (UTC)

yellows8: It says nowhere in the homebrew ABI document that the NRO is required to be located on the SD card. If you are absolutely convinced that that should be an HBABI requirement, I would encourage you to add it to the document. I very much think that should not be an HBABI requirement though. The only things keeping homebrew on the SD card are libnx's behavior of reading its own executable combined with the small set of default-mounted filesystems that libnx understands. Aside from this behavior (which I'm trying to get changed), there is no reason to require that homebrew applications be located on the SD card.

fincs: Apologies if I was unclear. When I said RomFS, I meant the filesystem that is visible to libnx applications under romfs:/* paths. For the hostio use case, if hostio were to be mounted under hostio:, the application would need to be written to tell whether it's on hostio or not and adjust its asset loading paths accordingly, whereas it would be significantly more convenient if the homebrew loader could override the romfs: mountpoint to use a hostio IFileSystem. I also disagree that homebrew is "by definition" NRO formatted. By convention, sure, and I'd even accept that this ABI does not apply to non-NRO formatted homebrew. I invite you to ignore that use case.

Here's a different idea for you.

   ==== Mount ====
   This is used to request that a filesystem be mounted by the homebrew application.
   
   * Key: 15
   * Value[0]: Pointer to a NULL-terminated string representing the desired mountpoint.
   * Value[1]: Handle to a session implementing Filesystem_services#IFileSystem.
   
   The length of the mountpoint shall be no longer than 32 characters, including the NULL terminator.
   This key may be combined with the Argv key to specify the location of an application's executable.
   

This way, we can still override romfs:/ with Mount["romfs", <IFileSystem>], or we can mount an arbitrary filesystem (Mount["myfs", <IFileSystem>]) and set argv[0] to a path on that new mountpoint (myfs:/application.nro) so that the application can still find its NRO file and read it back to provide the romfs:/ mount even if the NRO is not located on the SD card. I've even specified a length limit to match fsdev requirements.

--Misson20000 (talk) 23:42, 9 November 2018 (UTC)

Homebrew_ABI#NextLoadPath "NRO" "should start with "sdmc:/". Of course nx-hbloader doesn't enforce the latter besides only having sdmc mounted. --Yellows8 (talk) 00:05, 10 November 2018 (UTC)

For NextLoadPath? Sure, if you want to use NextLoadPath, it's best to give it a path to an NRO starting with "sdmc:/" because that's all that nx-hbloader understands and all that you want it to understand, which makes sense. But HBABI doesn't require that homebrew needs to be loaded by passing NextLoadPath back to the loader. After all, nx-hbmenu isn't loaded that way. That's besides the point, anyway. The point is that whether or not this restriction is actually documented on the HBABI page, I want you to evaluate why this restriction exists and whether it's necessary or not.

--Misson20000 (talk) 00:27, 10 November 2018 (UTC)

Where does this "IFilesystem" come from? --Yellows8 (talk) 01:47, 10 November 2018 (CET)

Doesn't much matter as far as the application is concerned. Could be some save data that the loader opened from FS, could be some BIS filesystem, or, in my case, a custom IFileSystem implementation that came from a custom service. Just as long as it implements the IFileSystem interface.

--Misson20000 (talk) 00:57, 10 November 2018 (UTC)

Workarounds (which only exist for a debugger) don't belong in the HBABI. --Yellows8 (talk) 19:17, 12 February 2019 (UTC)

Almost all of HBABI is workarounds to facilitate running homebrew applications in dirty environments:

 - bad entrypoint arguments (MainThreadHandle)
 - reusing the same process (OverrideHeap, Argv (can't use loader-args), AllocPages, LockRegion, RandomSeed)
 - legacy exploits (OverrideService, AppletWorkaround)
 - working around 0xffff8001 check in SVCs (ProcessHandle)

And although it's not specified in the HBABI, reading back argv[0] for ASET is also a way to work around not being installed as a real application and being able to access romfs like official applications do.

The only thing I'm working around here is the ASET workaround that needlessly ties homebrew to the SD card. Even if this is a workaround for a debugger, I don't think that's grounds for not belonging in the HBABI. It's not like Nintendo doesn't do anything similar ("Redirect*" commands in lr). --Misson20000 (talk) 21:42, 12 February 2019 (UTC)

Reusing the same process is the approach selected for nx-hbloader which is the standard way to load homebrew, and I see no reason why they would be labeled as "workarounds". AllocPages and LockRegion were unilaterally added by you several months ago and nx-hbloader has no use for them. OverrideService and AppletWorkaround are not presently used by anything and they were mostly added for future-proofing in case we need to support a more restrictive environment where we need to pass stolen service handles, they don't exist for "legacy" exploits.

I see no reason to add a romfs override key. If we want to support non-SD card paths, we will add other device name prefixes to argv[0] (such as host:/) and add support for these devices in libnx.

--Fincs (talk) 22:12, 12 February 2019 (UTC)