On Sat, Dec 31, 2016 at 06:04:32PM +0100, Veronika Kabatova wrote: > Currently, it is not possible to view assignment tree in a different way than > debug output, and there are no possibilities to change the settings during > runtime. This patchset adds the ability for socket communication, creating an > interface for users to change some settings during runtime (sleep interval, > CPU and IRQ banning). Strings needed for retrieving data and setting changes > are described in the first patch's commit message. > > The main purpose of this patchset is to create a simple user interface (using > curses library). The user interface is added into configuration script and > makefile and builds as a `irqbalance-ui` command. It uses aforementioned socket > interface to communicate with irqbalance and allows user to view assignment > tree, sleep interval, CPUs and IRQs that irqbalance is working with, and to > setup irqbalance's settings (which was yet possible only with restarting). > > The user interface makes it possible for users to check how the IRQs are > balanced in given moment (refresh interval of assignment tree is set to 5s) and > to easily change irqbalance's settings in case the requirements change later. > The socket interface iteslf can also be used for scripting (eg automatic banning > depending on load), which can be useful for system administrators for various > optimizations. > > Current plans for future development include unicode / ASCII signs in tree based > on locale settings, lspci and /proc/interrupts integration for IRQ source > information (eg device or kernel driver names) and displaying generic load of > CPUs. Ideas for other enhancements are welcome. > > The patches are also available as `ui_v2` branch at GitHub: > https://github.com/veruu/irqbalance/tree/ui_v2 > > ** Changes since v1 ** > 1) Addressing Neil's comments: > > I like the idea of what you are adding, but some initial comments: > > > > 1) Setting a fixed socket address limits people to a single irqbalance instance. > > I know of a few cases in which multiple irqbalances may be run on different cpu > > sets, and this would be in conflict with a single socket address. Suggest that > > you use abstract namespace mode of AF_LOCAL using the irqbalance pid and a > > standard string to create a unique path for each instance. > > > > Fixed. I think having a unique path with regular UNIX sockets (based on PID as > you said) should be enough, but I did implement abstract sockets as you > suggested because of their advantages. Thanks for the idea! > > > 2) Security is an issue here. You've not restricted socket access at all, > > meaning any user on the system can modify irqbalance behavior. Suggest that you > > use SCM_CREDENTIALS socket option to limit access to root. > > AFAIK (and based on my testing), this is not an issue with normal UNIX sockets > because non-root users don't have priviledges to connect to socket created by > root. This doesn't work with abstract sockets and additional Security can never > hurt, so a check for valid root credentials is done before accepting requests. > > 2) Document socket API in man page > > Veronika Kabatova (5): > Add ability for socket communication > Add helper functions for user interface > Add user interface functionality > Add main user interface files > Add user interface to configuration and build, document socket API in > man page > > Makefile.am | 8 +- > configure.ac | 4 +- > cputree.c | 9 +- > irqbalance.1 | 29 +++ > irqbalance.c | 286 +++++++++++++++++---- > irqbalance.h | 4 + > ui/helpers.c | 169 +++++++++++++ > ui/helpers.h | 31 +++ > ui/irqbalance-ui.c | 400 +++++++++++++++++++++++++++++ > ui/irqbalance-ui.h | 86 +++++++ > ui/ui.c | 729 +++++++++++++++++++++++++++++++++++++++++++++++++++++ > ui/ui.h | 53 ++++ > 12 files changed, 1748 insertions(+), 60 deletions(-) > create mode 100644 ui/helpers.c > create mode 100644 ui/helpers.h > create mode 100644 ui/irqbalance-ui.c > create mode 100644 ui/irqbalance-ui.h > create mode 100644 ui/ui.c > create mode 100644 ui/ui.h > > -- > 2.7.4 > > It seems you didn't throughly test this: gcc -DHAVE_CONFIG_H -I. -I. -W -Wall -Wshadow -Wformat -Wundef -D_GNU_SOURCE -I/usr/include/glib-2.0 -I/usr/lib64/glib-2.0/include -g -O2 -c -o classify.o classify.c classify.c:38:15: error: static declaration of ?cl_banned_irqs? follows non-static declaration static GList *cl_banned_irqs = NULL; ^~~~~~~~~~~~~~ In file included from classify.c:10:0: irqbalance.h:66:15: note: previous declaration of ?cl_banned_irqs? was here extern GList *cl_banned_irqs; ^~~~~~~~~~~~~~ Makefile:572: recipe for target 'classify.o' failed make[2]: *** [classify.o] Error 1 make[2]: Leaving directory '/home/nhorman/git/irqbalance' Makefile:639: recipe for target 'all-recursive' failed make[1]: *** [all-recursive] Error 1 make[1]: Leaving directory '/home/nhorman/git/irqbalance' Makefile:448: recipe for target 'all' failed make: *** [all] Error 2
