device: 0.690
kernel: 0.663
files: 0.652
permissions: 0.561
socket: 0.544
ppc: 0.524
semantic: 0.421
architecture: 0.411
network: 0.402
register: 0.387
graphic: 0.377
PID: 0.371
user-level: 0.293
performance: 0.267
risc-v: 0.229
boot: 0.226
mistranslation: 0.225
debug: 0.204
vnc: 0.195
arm: 0.190
virtual: 0.182
hypervisor: 0.150
peripherals: 0.145
x86: 0.137
VMM: 0.103
i386: 0.100
assembly: 0.088
TCG: 0.077
KVM: 0.076

local/9pfs: Too many levels of symbolic links

Version: 2.9.1

The primary symptom is resolving symlink fails w/ error "too many levels of symbolic links".

My analysis showed that local_readlink() uses local_open_nofollow() to open the file and then tries to read it. local_open_nofollow() then tries to open the file w/ O_NOFOLLOW, which obviously fails if the requested file is a symlink.

Turned out that security_model=mapped can't cope with symlinks in the host file system.

Instead, security_model=passthrough works as expected. OTOH, I'll have to check, whether this mode already provides a safe chroot, or guest can escape and damage the host system.

The wiki page needs some more documentation on that:
https://wiki.qemu.org/Documentation/9psetup

The QEMU project is currently considering to move its bug tracking to
another system. For this we need to know which bugs are still valid
and which could be closed already. Thus we are setting older bugs to
"Incomplete" now.

If you still think this bug report here is valid, then please switch
the state back to "New" within the next 60 days, otherwise this report
will be marked as "Expired". Or please mark it as "Fix Released" if
the problem has been solved with a newer version of QEMU already.

Thank you and sorry for the inconvenience.


Enrico, with security_model=mapped (a.k.a. security_model=mapped-xattr) 9p is not following symlinks on host. That's the expected behaviour.

There are 2 distinct models:

security_model=passthrough uses the ownership information, permissions and symlink info etc. directly from the host's file system. This model requires the qemu binary to run as high privileged user (i.e. root) though, because it needs to be able to change file ownership, permissions and so forth. For that reason this model is only appropriate for use cases where there is a certain trust to what the guest system is doing. E.g. this model is commonly used by kernel coders to build kernel/driver code on host and then test run it as guest.

For untrusted guests (i.e. most cases) it is recommended to use security_model=mapped-xattr instead. In this mode all ownership information, permissions, etc. are emulated (i.e. mapped) by storing them as extended attributes on top of files on host. This way the qemu binary can run with a regular (unprivileged) user and pretend to host it would have all rights to change ownership and permissions.

You can read more details about this in the 9p developer description about the 9p 'local' driver here:
https://wiki.qemu.org/Documentation/9p#9p_Filesystem_Drivers

[Expired for QEMU because there has been no activity for 60 days.]