This is an idea for substantially reducing the number of syscalls needed
by monitoring tools whilst mostly re-using the existing API.
The proposed files in this proof-of-concept patch set are:
* /proc/all/stat
A stat line for each process in the existing format.
* /proc/all/statm
statm lines but starting with a PID column.
* /proc/all/status
status info for all processes in the existing format.
* /proc/all/io
The existing /proc/pid/io data but formatted as a single line for
each process, similarly to stat/statm, with a PID column added.
* /proc/all/statx
Gathers info from stat, statm and io; the purpose is actually
not so much to reduce syscalls but to help userspace be more
efficient by not having to store data in e.g. hashtables in order
to gather it from separate /proc/all/ files.
The format proposed here starts with the unchanged stat line
and begins the other info with a few characters, repeating for
each process:
...
25 (cat) R 1 1 0 0 -1 4194304 185 0 16 0 2 0 0 0 20 ...
m 662 188 167 5 0 112 0
io 4292 0 12 0 0 0 0
...
There has been a proposal with some overlapping goals: /proc/task-diag
(https://github.com/avagin/linux-task-diag), but I'm not sure about
its current status.
Best Wishes,
Eugene
Eugene Lubarsky (5):
fs/proc: Introduce /proc/all/stat
fs/proc: Introduce /proc/all/statm
fs/proc: Introduce /proc/all/status
fs/proc: Introduce /proc/all/io
fs/proc: Introduce /proc/all/statx
fs/proc/base.c | 215 +++++++++++++++++++++++++++++++++++++++++++--
fs/proc/internal.h | 1 +
fs/proc/root.c | 1 +
3 files changed, 210 insertions(+), 7 deletions(-)
--
2.25.1
