Pádraig Brady wrote: > Eric Sandeen wrote: >> Attached is a first stab at an fallocate utility, which will preallocate >> blocks to a file. This can be useful for virtual images, database >> files, testing, etc. Normally we'd hope that various tools will start >> using preallocation internally, but until then such a utility may be >> useful, and could be scripted as well. > > FYI here is how we're thinking of adding > fallocate() and posix_fallocate() to coreutils: > http://www.pixelbeat.org/patches/coreutils/inbox_may_2009.html Ah ok, sorry for being behind. :) Actually I guess I saw some of that fly by, too, and forgot... >> I put a fallback in the attached tool so that if fallocate() fails, it >> goes to posix_fallocate. On recent (not bleeding) glibc, >> posix_fallocate tries to call sys_fallocate, and resorts to writing 0s >> if that fails. On even older glibc, it goes straight to 0-writing. The >> user won't know which one happened, unless they allocate a few gigabytes >> and notice the delay. > > You'd get most of the functionality by just calling posix_fallocate(). > Calling fallocate() explicitly allows you to support the -n option. > I.E. this tool you're proposing would be equivalent to `truncate --allocate` > which we're going to add to coreutils, with the additional functionality > of the -n option. Is that worth a separate tool? Should we add -n > functionality to `truncate`? It's probably not that critical, though some people may wish it were there. They can always RFE if they need it I guess. Having the ability to allocate at a given offset may be nice too. For usefulness in test scripts, exposing all of fallocate's interfaces would be nice. TBH I think "truncate --allocate" sounds a little odd. (Now that I think back, I think I mentioned this before). truncate(1) and truncate(2) specifically refer to i_size, which to fs people like me, has nothing to do with the actual blocks allocated to a file. IOW i_size & i_blocks are independent. So adding "allocate" to "truncate" feels a little orthogonal, like "cp --mv" but I (as I think I said before ;) could live with it. Thanks, -Eric > cheers, > Pádraig. > -- To unsubscribe from this list: send the line "unsubscribe util-linux-ng" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
