xfs
[Top] [All Lists]

Re: Questions about XFS discard and xfs_free_extent() code (newbie)

To: "xfs@xxxxxxxxxxx" <xfs@xxxxxxxxxxx>
Subject: Re: Questions about XFS discard and xfs_free_extent() code (newbie)
From: Chris Murphy <lists@xxxxxxxxxxxxxxxxx>
Date: Sat, 21 Dec 2013 10:03:47 -0700
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <31B8E1DCCA4F45918621D6461ED19B19@alyakaslap>
References: <AD612A564BB84E75B010AE37687DFC8E@alyakaslap> <20131218230615.GQ31386@dastard> <78FC295EC7FF48C987266DC48B183930@alyakaslap> <20131219105513.GZ31386@dastard> <31B8E1DCCA4F45918621D6461ED19B19@alyakaslap>
On Dec 19, 2013, at 12:24 PM, Alex Lyakas <alex@xxxxxxxxxxxxxxxxx> wrote:

> Hi Dave,
> It makes sense. I agree it might break some guarantees. Although if the user 
> deleted some blocks in the file or the whole file, maybe it's ok to not have 
> a clear promise what he sees after the crash. 

User perspective: I disagree. Sounds like a possible zombie file invasion, with 
no clear way of reversion. The file either needs to be gone, as in not 
accessible in user space, or it needs to be present and intact.  There isn't a 
reasonable expectation for a file to be resurrected from the dead that's also 
corrupted.

If the file name isn't also corrupted, the problem is worse. It looks like a 
legitimate file, yet it's useless. The zombie files will be subject to backup 
and restore, just like their valid predecessors. All I need is to stumble upon 
a handful of these files, which I won't necessarily remember were deleted 
files, to start assuming I have some sort of weird file system corruption, at 
which point at best I'll become really confused not knowing what to do next. At 
worse, I may start throwing hammers that end up causing worse problems.


Chris Murphy
<Prev in Thread] Current Thread [Next in Thread>