On 08/08/2014 07:07 AM, Ján Tomko wrote:
On 08/05/2014 04:38 PM, John Ferlan wrote:
https://bugzilla.redhat.com/show_bug.cgi?id=1077068
Check for the NFS FS type being true for a "local" stat of the file to force usage of the 'st_size' value rather than calculating the size using the 'st_blocks' and 'st_blksize'. As described in the stat(2) man page:
"Use of the st_blocks and st_blksize fields may be less portable."
experimentation shows a 10M file could get the following output from stat:
st_size=10485760 st_blocks=88 st_blksize=1048576
resulting in a "44 KiB" value being displayed as the allocation value. While this value does match the "du -s" value of the same file, the "du -b" value shows the "st_size" field. Similarly a long listing of the file shows the 10M size.
Capacity should be the apparent size (what du -b shows, or st_size), while allocation should track the on-disk usage (du, st_blocks * 512).
It looks to me that the values are correct, it's just that posix_fallocate does neither work nor fail on NFS.
Ah yes - this is exactly where I went back and forth on... Digging through 'old' google results on posix_fallocate() and wondering whether it was incorrectly returning success or whether stat() was as pointed out in its man page not getting a reliable st_blocks value. However, if posix_fallocate() isn't working as specified for NFS and not producing any error message, then how does one really determine that? I also had some code that reworked the two callers/users in order to force the "allocation" paths to go through the slower lseek/safewrite calls. Is it worth resurrecting that and going with it? John