While testing a patch for RFE to add fallocate support to vfs_glusterfs in samba-rhgs, I hit a problem with the read cache created by quick-read not being invalidated in case of glfs_fallocate() or glfs_discard().

To reproduce:

1) untar the attached reproducer.
2) Build the test programs with make all
3) Edit fields in Makefile to set hostname, volume and/or filename/logfile name for test.
4) Run test with make test.

The command writes a 2Kb file with pattern 0x11 for each byte written to the file.
The second call to t_readv first calls discard on half the file. It then reads the file and checks the bytes on the file for 0 in the first half and 1s on the second half.

Expected:
# make test
./t_write vm140-111 gv0 test211 log
./t_readv vm140-111 gv0 test211 log
Success

Actual Result:
# make test
./t_write vm140-111 gv0 test211 log
./t_readv vm140-111 gv0 test211 log
char at position 0 not zeroed out
make: *** [test] Error 251


Attached patch invalidates the cache in quick-read when fallocate, discard or zerofill operations are called.