| commit | e3fc69eb7932705c4ac3ff97665b59d2d8106992 | |
| author | Matthew Dillon <dillon@apollo.backplane.com> | |
| Sat, 27 Dec 2014 22:44:14 +0000 (27 14:44 -0800) | ||
| committer | Matthew Dillon <dillon@apollo.backplane.com> | |
| Sun, 28 Dec 2014 00:44:36 +0000 (27 16:44 -0800) | ||
| tree | 84b87816a03d67219c0a2fb21c12700787227e55 | treesnapshot (tar.gz zip) |
| parent | e86ad61a2ea73db6f3071d794ea3a02ea5ba5c12 | commitdiff |
kernel - Fix incorrect assertion in ffs_reallocblks()
* When softupdates is not enabled reallocblks should not try to free
an unallocated block. This appears to be possible if the reallocation
range covers part of a file with a hole in it. FSX appears to reveal
this case.
* It is also possible that this case is due to some interaction between
the DragonFly buffer cache and UFS that breaks b_offset assignment
assumptions made by UFS, but I couldn't find any particular vector
for the possibility.
* When softupdates is not enabled reallocblks should not try to free
an unallocated block. This appears to be possible if the reallocation
range covers part of a file with a hole in it. FSX appears to reveal
this case.
* It is also possible that this case is due to some interaction between
the DragonFly buffer cache and UFS that breaks b_offset assignment
assumptions made by UFS, but I couldn't find any particular vector
for the possibility.
| sys/vfs/ufs/ffs_alloc.c | diffblobblamehistory |