What keeps drivers/base/sys.c sysdev_show() from overrunning buffer?
The calls in drivers/base/sys.c to sysdev_show(), which seem to resolve to the routines node_read_cpumap() and node_read_meminfo() in node.c, do not take any buffer count (size). They used to, by Patrick removed the count parameter in Jan 2003, from here and other such places.
What's to keep the node_read_*() sprintf's from overrunning these buffers?
I am developing some changes to the cpumask_t print routines, which include using snprintf() instead of sprintf(), and watching buffer limits. These changes are motivated by the need to handle such things as 512 CPUs.
I couldn't plug my new routine into read_cpumap() to display the node_dev->cpumap (a cpumask_t), for want of a buffer count.
Signature · 3 lines
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <[email protected]> 1.650.933.1373Quoted 4 lines · Paul Jackson
The calls in
drivers/base/sys.ctosysdev_show(), which seem to resolve to the routinesnode_read_cpumap()andnode_read_meminfo()innode.c, do not take any buffer count (size). They used to, by Patrick removed the count parameter in Jan 2003, from here and other such places.
Quoted 8 lines
What's to keep the node_read_*() sprintf's from overrunning these buffers?
I am developing some changes to thecpumask_tprint routines, which include usingsnprintf()instead ofsprintf(), and watching buffer limits. These changes are motivated by the need to handle such things as 512 CPUs.
I couldn't plug my new routine intoread_cpumap()to display thenode_dev->cpumap(acpumask_t), for want of a buffer count.
There was some infrastructure erected for this at one point (seq_file); I wonder why it's not using that. But yes, this needs to get taken care of one way or another.
-- wli