Bug 219007
| Summary: | opening and closing /dev/dri/card0 in a QEMU KVM instance will shutdown system, 6.10.0-rc6+ | ||
|---|---|---|---|
| Product: | Drivers | Reporter: | Colin Ian King (colin.i.king) |
| Component: | Video(DRI - non Intel) | Assignee: | drivers_video-dri |
| Status: | NEW --- | ||
| Severity: | high | CC: | colin.i.king, regressions |
| Priority: | P3 | ||
| Hardware: | Intel | ||
| OS: | Linux | ||
| Kernel Version: | Subsystem: | ||
| Regression: | No | Bisected commit-id: | |
| Attachments: | dmesg 6.10 log | ||
Note this also can reproduce when running *without* root privileged too, so this is a user space DoS attack vector. Colin, quick reminder: many bugs reported here are never forwarded to any developer. For some of the details see https://lwn.net/Articles/910740/ and the links in there. So the people you are trying to reach most likely won't even see this. Mail is usually the best for reporting, as mentioned in https://docs.kernel.org/admin-guide/reporting-issues.html (which also tells people to avoid bugzilla in most cases) Forgot: I'll forward this report by mail, as it's a regression, so no need to do anything for you. But it's not the first bug I see from you here, so I thought a reminder might be wise. And one more comment: could you please share a dmesg from the VM so we know what drm driver is used. Created attachment 306610 [details]
dmesg 6.10 log
Attached, dmesg log. Apologies for the delayed reply.
|
The following code when run as root on a Debian sid amd64 server running in virt-manager (KVM QEMU) will shut the system down with 6.10.0-rc6. The fork() is required to cause racing on the open/close on /dev/dri/card0 #include <fcntl.h> #include <unistd.h> int main(void) { pid_t pid = fork(); while (1) { int fd; fd = openat(AT_FDCWD, "/dev/dri/card0", O_WRONLY|O_NONBLOCK|O_SYNC); close(fd); } } This was originally found using: while true; do sudo ./stress-ng --dev 4 --dev-file /dev/dri/card0 -t 5; done and narrowed down to the above reproducer. (cf: https://github.com/ColinIanKing/stress-ng/issues/407 ) This does not occur on pre 6.10 kernels, so it looks like a 6.10 regression.