2010/12/3 Eric Blake <eblake@redhat.com>:
Making this change will allow the future patches to use virFileWriteStr to create a file, rather than its current limitation of only working on pre-existing files.
* src/util/util.h (virFileWriteStr): Alter signature. * src/util/util.c (virFileWriteStr): Allow file creation. * src/network/bridge_driver.c (networkEnableIpForwarding) (networkDisableIPV6): Adjust clients. * src/node_device/node_device_driver.c (nodeDeviceVportCreateDelete): Likewise. * src/util/cgroup.c (virCgroupSetValueStr): Likewise. * src/util/pci.c (pciBindDeviceToStub, pciUnBindDeviceFromStub): Likewise. Based on a report from Jean-Baptiste Rouault. ---
Alternatively, I only counted 16 existing users of virFileWriteStr; and this is an internal API. We could easily rewrite all clients to always pass a third parameter, and change the signature of virFileWriteStr to require a mode_t argument. Hmm; some of those clients are writing to kernel files that should always exist (/proc/sys/net/ipv4/ip_forward, for example), where it's tough to justify what we would pass as a mode_t argument. So maybe pass mode==0 as a sentinel to require a pre-existing file
How does this look? Admittedly, all existing uses were okay with a mode parameter of 0; and I haven't yet seen your patch that would use a non-zero mode, but this still makes more sense to me.
Oh, and VIR_FORCE_CLOSE preserves errno, so I was able to simplify virFileWriteStr in the process.
src/network/bridge_driver.c | 8 ++++---- src/node_device/node_device_driver.c | 2 +- src/util/cgroup.c | 2 +- src/util/pci.c | 16 ++++++++-------- src/util/util.c | 13 ++++++++----- src/util/util.h | 3 ++- 6 files changed, 24 insertions(+), 20 deletions(-)
ACK. Matthias