[pve-devel] [PATCH qemu-server 1/2] qmeventd: improve getting vmid from PID in presence of legacy cgroup entries

[pve-devel] [PATCH qemu-server 1/2] qmeventd: improve getting vmid from PID in presence of legacy cgroup entries

[Fiona Ebner]

On a hybrid cgroup system, the /proc/<PID>/cgroup file usually looks
like:

> 13:pids:/qemu.slice/110.scope
> 12:perf_event:/
> 11:devices:/qemu.slice
> 10:misc:/
> 9:hugetlb:/
> 8:freezer:/
> 7:cpu,cpuacct:/qemu.slice/110.scope
> 6:memory:/qemu.slice/110.scope
> 5:rdma:/
> 4:cpuset:/
> 3:blkio:/qemu.slice
> 2:net_cls,net_prio:/
> 1:name=systemd:/qemu.slice/110.scope
> 0::/qemu.slice/110.scope

but after 'umount /sys/fs/cgroup/pids' or for other reasons, it could
happen that an entry like '11:devices:/qemu.slice' is the first to
match the '/qemu.slice' part, which previously made the code expect to
find the VMID.

To improve detection, as a first step, match the trailing slash too.

Reported in the community forum:
https://forum.proxmox.com/threads/129320/post-571654

Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
---
 qmeventd/qmeventd.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/qmeventd/qmeventd.c b/qmeventd/qmeventd.c
index 921c0dca..503bce69 100644
--- a/qmeventd/qmeventd.c
+++ b/qmeventd/qmeventd.c
@@ -109,7 +109,7 @@ get_vmid_from_pid(pid_t pid)
 	}
 	cgroup_path++;
 
-	if (strncmp(cgroup_path, "/qemu.slice", 11)) {
+	if (strncmp(cgroup_path, "/qemu.slice/", 12)) {
 	    continue;
 	}
 
@@ -142,7 +142,7 @@ get_vmid_from_pid(pid_t pid)
     if (errno) {
 	fprintf(stderr, "error parsing vmid for %d: %s\n", pid, strerror(errno));
     } else {
-	fprintf(stderr, "error parsing vmid for %d: no qemu.slice cgroup entry\n", pid);
+	fprintf(stderr, "error parsing vmid for %d: no matching qemu.slice cgroup entry\n", pid);
     }
 
 ret:
-- 
2.39.2

[pve-devel] [PATCH qemu-server 2/2] qmeventd: further improve getting vmid from PID

[Fiona Ebner]

by also expecting the ".scope" part and trying the next entry if it is
not present instead of immediately failing.

Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
---
 qmeventd/qmeventd.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/qmeventd/qmeventd.c b/qmeventd/qmeventd.c
index 503bce69..7e9b2354 100644
--- a/qmeventd/qmeventd.c
+++ b/qmeventd/qmeventd.c
@@ -128,12 +128,12 @@ get_vmid_from_pid(pid_t pid)
 	errno = 0;
 	char *endptr = NULL;
 	vmid = strtoul(vmid_start, &endptr, 10);
+	if (!endptr || strncmp(endptr, ".scope", 6)) {
+	    continue;
+	}
 	if (errno != 0) {
 	    fprintf(stderr, "error parsing vmid for %d: %s\n", pid, strerror(errno));
 	    vmid = 0;
-	} else if (*endptr != '.') {
-	    fprintf(stderr, "unexpected cgroup entry %s\n", buf);
-	    vmid = 0;
 	}
 
 	goto ret;
-- 
2.39.2

Re: [pve-devel] [PATCH qemu-server 1/2] qmeventd: improve getting vmid from PID in presence of legacy cgroup entries

[Fiona Ebner]

Am 10.07.23 um 09:58 schrieb Fiona Ebner:
> On a hybrid cgroup system, the /proc/<PID>/cgroup file usually looks
> like:
> 
>> 13:pids:/qemu.slice/110.scope
>> 12:perf_event:/
>> 11:devices:/qemu.slice
>> 10:misc:/
>> 9:hugetlb:/
>> 8:freezer:/
>> 7:cpu,cpuacct:/qemu.slice/110.scope
>> 6:memory:/qemu.slice/110.scope
>> 5:rdma:/
>> 4:cpuset:/
>> 3:blkio:/qemu.slice
>> 2:net_cls,net_prio:/
>> 1:name=systemd:/qemu.slice/110.scope
>> 0::/qemu.slice/110.scope
> 
> but after 'umount /sys/fs/cgroup/pids' or for other reasons, it could
> happen that an entry like '11:devices:/qemu.slice' is the first to
> match the '/qemu.slice' part, which previously made the code expect to
> find the VMID.
> 
> To improve detection, as a first step, match the trailing slash too.
> 
> Reported in the community forum:
> https://forum.proxmox.com/threads/129320/post-571654
> 
> Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>

I'll send a v2 with a bit more. Sorry for the noise.