procfs uglyness caused by "cat"
hello list,
when doing "cat /proc/uptime" or similar files, the routine which prepares the buffer gets called twice.
this is because "cat" reads two times from the file, as a system-call-trace shows:
# strace -f cat /proc/uptime
...
read(3, "5241.09 5082.74\n", 1024) = 16
...
read(3, "", 1024) = 0
...this leads to uptime_read_proc() being called two times, the second time with parameter off=16, but since this is ignored, the work is done twice: clock_...gettime(), cputime_to_timespec(), sprintf(), proc_calc_metrics() and so on.
insert a printk-statement if you don't beleive this.
btw, there's no wrong information produced because of this, because *page points to the same location both calls.
a simple way to get rid of this:
static int uptime_read_proc(char *page, char **start, off_t off,
int count, int *eof, void *data)
{
struct timespec uptime;
struct timespec idle;
int len;
cputime_t idletime;+ if (off)
+ return 0;
cputime_add(init_task.utime, init_task.stime);
do_posix_clock_monotonic_gettime(&uptime);
cputime_to_timespec(idletime, &idle);
len = sprintf(page,"%lu.%02lu %lu.%02lu\n",...
and so on.
this affects possibly all /proc files which ignore the offset parameter and are evaluated(?) with "cat".
regards,
h.rosmanith
Robert Hancock explained what's wrong with your proposal of:
+ if (off)
+ return 0;
For an alternative that seems to work better, see the processing behind the /proc/<pid>/cpuset file, by grep'ing for seq_file in kernel/cpuset.c.
Essentially, it composes the result string on the open, and then lets user code read and seek over that string at will.
Signature · 3 lines
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <[email protected]> 1.925.600.0401