Backing up via cp causes blocks to be consumed on the source machine??

I need to replace the re-badged LSI 9220-8i (probably really a Chinese counterfeit) that handles the zfs mirrors on my server-of-all-work since it has shed a port and a second one is flakey. But I'm out of sata ports on the server, and so must do the pre-replace backup across my lan. I'm using cp because I've never used dump and don't want to experiment on this task.

So I nfs-mounted the source filesystems and started the copy. I'm watching progress using df and notice that the free block count is going down both on the destination drive AND on the source, which scared me. Why would a copy be consuming blocks on the nfs-mounted source?
 
It would appear that the blocks are being used for buffering, though I would have thought buffers would be allocated in swap and not counted. Apparently I happened to look at it while it was doing a long series of allocations without freeing anything up. [sigh] I'll purge the thread. hmm...apparently I can't purge the thread.
 
Back
Top