tl;dr version: If you can’t use Apps2SD, do adb shell, pm setInstallLocation 2, move any app to SD (ignoring possible “failed” errors at first try).
Our beloved Frozen Yoghurt came with many new features welcome to the community at large, and one feature which had a mixed reception: “Apps on external storage”, which allows the user to install applications to its phone’s external storage – mostly in order to free up internal disk space.
Many custom ROM distributions for Android already had this feature built in, going by the moniker “Apps to SD” (or “Apps2SD” or just “A2SD”).
The typical implementation of A2SD works by using an ext2/ext3 partition on the SD card of your device – and usually only works when it’s exactly the second partition. For the sake of argument, one such custom implementation of A2SD will be included at the end of this post.
What it then does is just completely move all the applications to the SD partition, leaving only the /data partition behind, and uses a bind mount to fool the system into believing that the files are still on the same file system. So, in essence, the a2sd patch “cheats” and pretends that nothing actually has happened while quietly siphoning the apps to the SD card.
This, of course, only works when you actually have root access to your device and are allowed to play around with all the interesting system data itself. If you’re working on an unrooted/stock handset and firmware, you don’t have the option of using this feature; and also if you’re too lazy or unknowing or prissy to set up an ext partition on your SD card.
Thus the «official» Apps to SD comes into play – if your device is running Android 2.2, that is.
An important thing to note about understanding the official implementation is that it assumes that the user has no direct access to the /system partition. Especially: the user is not able to access any installed Android application package in any way that allows copying files.
What Froyo does when installing an application to SD is pretty simple: it creates a file on the SD card and uses this as a container to store the application in. Said container is used with a crypted loop mount, that is the actual data on the SD card is encrypted, and will be decrypted at load time when accessing the application.
The idea behind this seemingly convoluted setup is simple: if you have paid for an application, you could just store it on SD and then copy it if it is not encrypted. If it is encrypted, you cannot access the application in a “simple” way to copy (i.e. pirate) it.
Additionally, the application (with the default settings) needs to allow Android to move it to the SD card – otherwise the system does not enable the functionality, probably to ensure that applications aren’t “broken” by SD storage.
Of course this is easily manhandled by using the USB debugging interface with adb shell: just issue pm setInstallLocation 2. This tells he package manager (hence «
/dev/block/dm-41 on /mnt/asec/de.hafas.android.db-1 type vfat [...]
And yes, that’s 41 device mapper crypto loops. At least they don’t produce that much overhead as to noticeably slow down the system.
One of the boons of the Froyo implementation is that with above command, it can easily be used even with an unrooted phone and without repartitioning your SD drive. The disadvantages are that Android requires a fair bit of time after booting to mount all the crypto loop devices, which will result in your applications being accessible rather late after booting. Also, you will not be able to use widgets of any app that is on SD.
Here come the advantages of the customized A2SD approach: you can still access widgets and applications on your SD card even when it is mounted to your computer – because Android will only mount away the root partition (the FAT one), and not your ext partition. And you’ll have less overhead due to the crypto business.
And, as promised, the code that enables A2SD on most current ROMs:
#!/system/bin/sh # # Apps2SD using symlinks and bind mounts # Originally by cyanogen (firstname.lastname@example.org) # Modified to use a cleaner /sd-ext implementation by IEF (email@example.com) # execute any postinstall script then kill it if [ -e /dev/block/mmcblk0p2 ]; then # mount and set perms busybox mkdir /sd-ext busybox mount -o noatime,nodiratime -t auto /dev/block/mmcblk0p2 /sd-ext; busybox chown 1000:1000 /sd-ext; busybox chmod 771 /sd-ext; # clean up any old symlinks, create data directories for i in data; do if [ -h /data/$i ]; then rm /data/$i; fi; if [ ! -d /data/$i ]; then mkdir /data/$i; busybox chown 1000:1000 /data/$i; busybox chmod 771 /data/$i; fi; done; # don't allow /data/data on sd because of upgrade issues - move it if possible if [ -d /sd-ext/data ]; then busybox cp -a /sd-ext/data/* /data/data/; busybox rm -rf /sd-ext/data; fi; # move apps from internal memory to sdcard for i in app app-private dalvik-cache; do if [ ! -d /sd-ext/$i ]; then mkdir /sd-ext/$i; fi busybox chown 1000:1000 /sd-ext/$i; busybox chmod 771 /sd-ext/$i if [ -d /data/$i ] && [ ! -h /data/$i ]; then busybox cp -a /data/$i/* /sd-ext/$i/; busybox rm -f /data/$i/*; fi; done; # symlink app dirs - they must be on the same filesystem for i in app app-private dalvik-cache; do if [ -d /data/$i ] && [ ! -h /data/$i ]; then busybox rm -rf /data/$i; busybox ln -s /sd-ext/$i /data/$i; fi; done; # clean up old whiteouts for i in local misc property system tombstones data; do if [ -h /sd-ext/$i ]; then rm -f /sd-ext/$i; fi done; # please don't put odex files in the app directory people! # it causes dexopt to crash when switching builds! busybox rm -f /sd-ext/app/*.odex setprop shadow.apps2sd.active 1; echo "+++ Apps-to-SD successfully enabled"; else # replace symlinks with directories so we can boot without sd for i in app app-private dalvik-cache; do if [ -h /data/$i ]; then rm -f /data/$i; mkdir /data/$i; busybox chown 1000:1000 /data/$i; busybox chmod 771 /data/$i; fi; done; setprop shadow.apps2sd.active 0; fi; sync;
This is run as an init script.