Cdiffs Udiffs Wdiffs Sdiffs Frames Old New Patch Raw usr/src/uts/common/fs/zfs/dmu_tx.c

1862 incremental zfs receive fails for sparse file > 8PB
dmu_tx_count_free is doing a horrible over-estimation of used memory. It
assumes that the file is fully non-sparse and calculates a worst-case estimate
of how much memory is needed to hold all metadata for the file. If a large
hole needs to be freed, the estimation goes into the TB-range, which obviously
fails later on.
This patch tries to calculate a more realistic estimate by counting the l1
blocks (the loop for this is already present) and assumes a worst-case
distribution of those blocks over the full length given.
Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com>
Reviewed by: Simon Klinkert <klinkert@webgods.de>

42 lines changed: 27 ins; 15 del; 0 mod; 1370 unchg

This code review page was prepared using /opt/onbld/bin/webrev. Webrev is maintained by the OpenSolaris project. The latest version may be obtained here.