Arch Linux User Repository

Have you updated your kernel without rebooting? I believe Manjaro (like Arch) will delete the /lib/modules/<old_kernel_version> when doing an upgrade, so you can't install kernel modules after you've updated it.

Errors out

:: Processing package changes...
(1/1) installing netatop                                                                                                                                                 [#########################################################################################################] 100%
depmod: ERROR: could not open directory /lib/modules/5.15.6-2-MANJARO: No such file or directory
depmod: FATAL: could not search modules: No such file or directory
modprobe: FATAL: Module netatop not found in directory /lib/modules/5.15.6-2-MANJARO
error: command failed to execute correctly

I have submitted a patch to upstream by building on eggz' work, and my own grep fu :)

This time, it is related to a change in the timespec struct in the kernel, and a move to proc_ops from filesystem_operations. See commits 22b17db4ea05561c7c8e4d770f10751e22e339f9 and d56c0d45f0e27f814e87a1676b6bdccccbc252e9 of the kernel repo.

Here's the patch:

diff -aur src.pristine/netatop-3.0/module/netatop.c src.new/netatop-3.0/module/netatop.c
--- src.pristine/netatop-3.0/module/netatop.c   2020-04-27 08:28:05.649361867 -0400
+++ src.new/netatop-3.0/module/netatop.c    2020-04-27 09:00:29.473109963 -0400
@@ -270,7 +270,12 @@

 static struct task_struct  *knetatop_task;

-static struct timespec boottime;
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 6, 0)
+               static struct timespec64    boottime;
+#else
+               static struct timespec  boottime;
+#endif

 /*
 ** function prototypes
@@ -346,14 +351,22 @@
         .owner          = THIS_MODULE,
 };

+#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 6, 0)
+static struct proc_ops netatop_proc_fops = {
+   .proc_open          = netatop_open,
+   .proc_read          = seq_read,
+   .proc_lseek         = seq_lseek,
+   .proc_release       = single_release,
+};
+#else
 static struct file_operations netatop_proc_fops = {
-   .open           = netatop_open,
-   .read           = seq_read,
-   .llseek         = seq_lseek,
-   .release    = single_release,
-   .owner          = THIS_MODULE,
+   .open           = netatop_open,
+   .read           = seq_read,
+   .llseek         = seq_lseek,
+   .proc_release   = single_release,
+   .owner                  = THIS_MODULE
 };
-
+#endif

 /*
 ** hook function to be called for every incoming local packet
@@ -1745,7 +1758,11 @@
        spin_lock_init(&shash[i].lock);
    }

-   getboottime(&boottime);
+   #if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 6, 0)
+       getboottime64(&boottime);
+   #else
+       getboottime(&boottime);
+   #endif

    /*
    ** register getsockopt for user space communication

FYI, eggz provided some amendments to prepare() in the comments for the netatop-dkms package that worked for me.

Thanks for reporting. I noticed it too, but I've been very busy. If the developer got your message, it will probably take him some time to find the changes that were made in the kernel again. As long as you mentioned the same things you mentioned here, then he'll probably know what to look for.

It seems that this is broken again with kernel 5.6. I am not entirely sure why, but it seems to have something to do with getboottime or getboottime64, maybe involving changes in timekeeping.h or "struct timespc"... but to be honest, I'm nowhere near clever enough to figure out exactly what changed.

Edit: I have sent a message to the maintainer.

@bartus why are you flagging this package out of date?

Please Note: The maintainer of netatop on the atop website has responded to me. Netatop v3.0 is now available. I already tested it with this makepkg, & it works perfectly fine now. He also added a SystemD service file which I have not looked into yet.

Please test it all & see. This should be able to be implemented without any patches now.

Patch for kernel>=5.5

git am < <(curl -s http://ix.io/1DAW)

Hi,

I also did a bit of digging, and it seems that the real_start_time was renamed to start_boottime in v5.5.0 of the kernel. You can verify that either by running

git diff v5.4.15-arch1:include/linux/sched.h v5.5-arch1:include/linux/sched.h

on the Arch linux kernel repo, or just check the change on the kernel repo itself: https://github.com/torvalds/linux/commit/cf25e24db61cc9df42c47485a2ec2bff4e9a3692.

Here are some links to a fixed PKGBUILD with a patch.

Contents of PKGBUILD: https://hastebin.com/cakewikuqi.bash

Contents of the patch: https://hastebin.com/nomoxunati.diff

Cheers,