Implement a pipe between TWRP and MTP to allow TWRP to tell MTP
to remove storage partitions as they become unavailable (e.g.
during a wipe, unmount, etc) instead of disabling MTP completely.
This includes some fixes and improvements in destructors to
properly remove / delete various items. This also means that we
will not be toggling adb off and on quite as often.
I do not like that we had to add another thread, but we were
unable to use select() on the mtp_usb character device because
this device does not support polling. Select always returned
indicating that the mtp file descriptor was ready to be read and
the resulting read would block. The read block prevented us from
being able to include reading of the pipe between TWRP and MTP in
the main MTP thread.
We might want to add a return pipe letting TWRP know if the
removal of the storage device was successful, but I am not sure
how we want to implement this. It would invovle timeouts in both
TWRP and MTP to ensure that we returned a failure indicator in a
timely manner to TWRP and prevent deleting the storage device in
the case of a failure. Right now we make no attempt to ensure that
an MTP operation is underway like a large file transfer, but we
were not doing anything like this in the past. In some respects we
have limited control over what happens. If the user installs a
zip that unmounts a storage partition, we will not know about the
change in storage status anyway. Regular Android does not have
these troubles because partitions rarely get unmounted like in
recovery. At some point, we have to hold the user accountable for
performing actions that may remove a storage partition while they
are using MTP anyway.
Ideally we do not want to toggle the USB IDs and thus toggle adb
off and on during early boot, but I am not sure what the best way
to handle that at this time.
Change-Id: I9343e5396bf6023d3b994de1bf01ed91d129bc14
Originally the inotify_init() call was part of a separate thread
which would occasionally allow the readDirs function to run and
start trying to add watches before inotify_fd was init'ed properly
and result in the add watch failing. This patch set relocates the
inotify_init call to happen outside and before starting the
separate thread to ensure that inotify_fd is set properly.
Change-Id: I8748c23473b60b57887df9d692834f3d2b249802
- use std::map instead of linked list
- read directories on demand
- fix writing zip files to storage root
- fix creating directories
- lots of minor fixes
- simplify generation of storage IDs and make them spec compliant
Change-Id: I2137c27549ddbdc58466f2e3aeda464fac70a3c5
Do some error checking of lstat calls and better handle those
errors.
Use lstat instead of dirent to check if an item is a directory
because dirent does not return the correct item type when using
exfat-fuse.
Change-Id: I0d008aff55a0e0b4fbeae24cf8d0c04453b7b77e
