Caveat: yes, the real answer is buying more disk because it's safest, but I have a couple of outrageous suggestions that I haven't seen floated: 1) custom FUSE

Caveat: yes, the real answer is buying more disk because it's safest, but I have a couple of outrageous suggestions that I haven't seen floated: 1) custom FUSE filesystem that just skips the first N bytes of a file with a given name. Calculate N ahead of time by reading the file (you could calculate it on-demand but this will introduce a massive perf delay in some early file op that could cause a timeout for whatever software is reading the file) 2) (warning: this sounds insane but I've actually seen embedded devices that do this type of gymnastics in the name of not buying more storage space) if it's acceptable to temporarily have the file not be a text file, you could write custom code for sliding the data forward in the file by N bytes, starting with the N + (2*blocksize)th byte. The first two blocks are left until the end because you encode the progress of this operation into a state block, which includes a SHA of the state for validation purposes and which is written atomically, alternating between block 0 and 1 of the file. Your actual move of data must be designed in such a way that each data write is nondestructive if its state write is lost (i.e. it can only write to scratch space). Resume an interrupted move by figuring out which state validates and has the higher serial number. This is not something I'd attempt as a one-off. I'd want to be damn sure this code was well-tested.