https://bugs.kde.org/show_bug.cgi?id=458001
Méven <[email protected]> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |[email protected] --- Comment #10 from Méven <[email protected]> --- Thanks for reopening. Dolphin does ask the server to copy: KIO tries FICLONE first, like cp, and then copies with copy_file_range(). The difference is the size of each copy_file_range() call. cp passes the whole file in one call. KIO copies in pieces: 512 KiB up to KF 6.30, and since 6.31 a size that adapts to the speed of the copy. ZFS only clones a range that starts and ends on the block size of the dataset (its recordsize). For any other range the server copies the data instead, so the copy stays on the server but takes the space twice. With recordsize=1M, 512 KiB pieces never line up with a block. To confirm this, could you provide: - your KDE Frameworks version - your ZFS version, and the output of: zfs get recordsize <dataset> - the calls Dolphin makes during a copy, from: strace -f -e trace=copy_file_range,ioctl -p $(pidof dolphin) (copy one large file in Dolphin while this runs, then stop it with Ctrl+C) If the cause is confirmed, the fix is to keep the boundaries of each copied piece aligned on a large block size. -- You are receiving this mail because: You are watching all bug changes.
