on 2017/3/30 16:52, Amir Goldstein wrote: > On Thu, Mar 30, 2017 at 11:22 AM, zhangyi (F) <yi.zhang@xxxxxxxxxx> wrote: >> On 2017/3/30 15:03, Amir Goldstein wrote: >>>> Hi: >>>> Current overlayfs does not support underlying filesystem which have "weird" d_flags (like DCACHE_OP_HASH， >>>> DCACHE_OP_COMPARE, DCACHE_OP_REVALIDATE, etc). So we cannot support fs have it's own d_hash, d_compare and >>>> d_revalidate functions. >>> >>> Not accurate. d_revalidate() is allowed for lower, but not for upper >>> (ovl_dentry_remote()). >>> >> >> Thank you for the corrections, I see it in the ovl_lower_dir(). >> >>>> >>>> Do you have a TODO plan to support it in a future version ? >>>> >>> >>> Which of the constrains is relevant to your use case? (which of the flags) >>> Do you need it for lower / upper or both? >>> Is it a proprietary file system of a known one you need to support? >>> >> >> We want to use d_hash(), d_compare() and d_revalidate() to our own proprietary file system >> (not a remote fs, not widely used, and it's file name is case-insensitive) for lower >> and upper both, and we also need to use d_revalidate() to do some special handling. >> >> I see there were many other rarely used（?）file systems also use d_hash(), d_compare(), >> d_revalidate() in kernel(such as hfs, jfs, fat, etc.), they are also not supportted by ovlfs. >> So I would like to know are these file systems need to support in future ? >> > > I see you are not the first that requested this feature: > https://lkml.org/lkml/2014/10/28/1049 > > The solution for the general case is (too?) complex. > > However, for the case of all layers on the same fs, where d_hash() > d_compare() are consistent > across all layers, I guess this is doable and even not that hard to do. > > The 'samefs' case is an interesting case, because it is commonly used > by container > managers (e.g. dockerd). > > I've just posted the first overlayfs feature that depends on 'samefs' > https://marc.info/?l=linux-unionfs&m=149079818122019&w=2 > and I have plans to develop more features for 'samefs' case (e.g. NFS export). > > I don't mind writing a POC patch for you to test your use case, but I > will ask you > to first write a test which mixes case insensitive names in > lower/upper, so this patch > can be properly tested. > An xfstest overlay test is greatly preferred. > I can test it with overlayfs over vfat. > Thank you for your support. Now we are not make a decision whether to modify the underlying file system or the overlayfs. If so, we can plan to write and test it. Thanks, ZhangYi. -- To unsubscribe from this list: send the line "unsubscribe linux-unionfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
