On Mon, Aug 18, 2025 at 08:17:14AM -0400, Kenneth Gober wrote: > On Sun, Aug 17, 2025 at 9:42 AM dan <d...@nnnne-o-o-o.com> wrote: > > > I suggested I'm just watching at the behavior of the mv on Open Indiana > > and it seems to get that it always behave like a sequential cp + rm despite > > the presence of two different fs. And that it sounds safer after all.. > > > > When the source and destination are on different file systems, cp + rm is > required, because the bits > must be physically copied to the new file system. > > When the source and destination are on the same file system, the file does > not actually get moved. > Instead, the directory entry (which points to the inode for the file) is > moved. The inode itself (which > holds all the information about the file other than its name(s)) does not > get moved or copied. It's > like ln + rm, instead of cp + rm. Just moving the directory entry is much > safer (and faster) than > actually moving the data to new disk sectors.
For the ln + rm case, a syscall exists: rename(2), which does that operation atomically. It's all there in the source code of mv. > > -ken
