xfs
[Top] [All Lists]

Re: [PATCH 2/2 V2] mkfs.xfs: don't call blkid_get_topology on regular fi

To: Brian Foster <bfoster@xxxxxxxxxx>
Subject: Re: [PATCH 2/2 V2] mkfs.xfs: don't call blkid_get_topology on regular files
From: Eric Sandeen <sandeen@xxxxxxxxxxx>
Date: Fri, 06 Jun 2014 11:47:56 -0500
Cc: Eric Sandeen <sandeen@xxxxxxxxxx>, xfs-oss <xfs@xxxxxxxxxxx>
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <5391DDE4.90007@xxxxxxxxxxx>
References: <53852A05.5040006@xxxxxxxxxx> <5390C1B7.7050106@xxxxxxxxxxx> <20140606152059.GB3048@xxxxxxxxxxxxxx> <5391DDE4.90007@xxxxxxxxxxx>
User-agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.9; rv:24.0) Gecko/20100101 Thunderbird/24.5.0
On 6/6/14, 10:27 AM, Eric Sandeen wrote:
> On 6/6/14, 10:21 AM, Brian Foster wrote:
>> On Thu, Jun 05, 2014 at 02:15:03PM -0500, Eric Sandeen wrote:
>>> If we encounter a target that's really a regular file,
>>> even without "-d file..." on the cmdline, call
>>> platform_findsizes() instead of blkid_get_topology to
>>> try to discover the "sector size" via the fsgeom() call.
>>>
>>> Signed-off-by: Eric Sandeen <sandeen@xxxxxxxxxx>
>>> ---
>>>
>>> V2: Lose local "isa_file" flag, just switch based on 
>>> (xi->disfile) or (stat works & S_ISREG)
>>>
>>> diff --git a/mkfs/xfs_mkfs.c b/mkfs/xfs_mkfs.c
>>> index 188b6b3..3b8bf67 100644
>>> --- a/mkfs/xfs_mkfs.c
>>> +++ b/mkfs/xfs_mkfs.c
>>> @@ -456,9 +456,25 @@ static void get_topology(
>>>     struct fs_topology      *ft,
>>>     int                     force_overwrite)
>>>  {
>>> -   if (!xi->disfile) {
>>> -           const char *dfile = xi->volname ? xi->volname : xi->dname;
>>> +   struct stat statbuf;
>>> +   char *dfile = xi->volname ? xi->volname : xi->dname;
>>>  
>>> +   /*
>>> +    * Don't call blkid for topology if this is a "-d file" target, or
>>> +    * if we've simply been pointed at a regular file.  platform_findsizes
>>> +    * will attempt to find the underlying sector size of the host fs.
>>> +    */
>>> +   if (xi->disfile ||
>>> +       (!stat(dfile, &statbuf) && S_ISREG(statbuf.st_mode))) {
>>> +           int fd;
>>> +           long long dummy;
>>> +
>>> +           fd = open(dfile, O_RDONLY);
>>> +           if (fd >= 0) {
>>> +                   platform_findsizes(dfile, fd, &dummy, &ft->lsectorsize);
>>> +                   close(fd);
>>> +           }
>>
>> The patch looks fine:
>>
>> Reviewed-by: Brian Foster <bfoster@xxxxxxxxxx>
>>
>> It does look like we'd still be susceptible to error in a situation
>> where the file hasn't been created yet (which would occur in
>> libxfs_init()).
> 
> Hm, let's see - the only way to have mkfs create the file is with the
> -d file,name=foo,size=bar invocation.
> 
> Sprinkling some printf's into the if/else above, I see:
> 
> # rm -f testfile
> # mkfs/mkfs.xfs -dfile,name=testfile,size=1g
> ** doing platform_findsizes **
> meta-data=testfile               isize=256    agcount=4, agsize=65536 blks
> ...
> 
> so it looks like the file does get created before we get here.

Oh, I see, sorry.  Crud, you're right.

What prompted all this was that making a fs-in-a-file on a filesystem hosted
on a hard 4k sector device fails, because it tries to do 512-byte DIO.

If we use the "-d file" invocation, it works, because it doesn't do direct IO.

Sigh.  It's complicated.  Let me think about the 2nd patch just a bit more.

Perhaps only doing the platform_findsizes if stat succeeds, not if xi->disfile.

-Eric

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