[CRIU] I can't dump mysqld

[Andrew Vagin]

On Mon, Aug 01, 2016 at 01:16:33PM +0900, Yohei Kamitsukasa wrote:
> On 7/30/16 11:50, Andrew Vagin wrote:
> 
> > On Sat, Jul 30, 2016 at 11:16:45AM +0900, Yohei Kamitsukasa wrote:
> > > On 7/28/16 15:17, Andrew Vagin wrote:
> > > 
> > > > On Thu, Jul 28, 2016 at 09:58:39AM +0900, Yohei Kamitsukasa wrote:
> > > > > rt_sigaction(SIGCHLD, {0x45b940, [CHLD], SA_RESTORER|SA_RESTART|SA_SIGINFO, 0x7f6c58d163d0}, NULL, 8) = 0
> > > > > ptrace(PTRACE_SETSIGMASK, 8367, 0x8, 0x7ffe72948b30) = 0
> > > > > ptrace(PTRACE_SETREGSET, 8367, NT_PRSTATUS, [{0x7ffe72948b90, 216}]) = 0
> > > > > ptrace(PTRACE_CONT, 8367, NULL, SIG_0)  = 0
> > > > > accept(6,
> > > > Could you check that all processes live in the same network namespaces?
> > > > 
> > > > "readlink /proc/pid/ns/net" should return the same value for all
> > > > processes.
> > > Thank you for teaching how to check to me:)
> > > I checked the network namespaces of mysqld's processes and they were in same
> > > network namespace all. (A process of CRIU dumping was in the network
> > > namespace.)
> > > 
> > Could you apply the following patch [1] and try again?
> > [1]: https://patchwork.criu.org/patch/1136/mbox/
> > 
> > This patch will not fix the problem, but it will show more info about an
> > error in a log file.
> > 
> > Thanks,
> > Andrew
> Thank you for writing the patch.
> I applied the patch and show the result.
> 
> (00.074960) Dump private signals of 1191
> (00.075032) Dump private signals of 1192
> (00.075103) Dump shared signals of 1093
> (00.075308) Parasite syscall_ip at 0x400000
> (00.076765) Set up parasite blob using memfd
> (00.076806) Putting parasite blob into 0x7f0174cd9000->0x7fead3688000
> (00.076866) Dumping GP/FPU registers for 1093
> (00.076905) Warn  (arch/x86/crtools.c:133): Will restore 1093 with
> interrupted system call
> (00.076979) xsave runtime structure
> (00.077017) -----------------------
> (00.077061) cwd:37f swd:0 twd:0 fop:0 mxcsr:1fa0 mxcsr_mask:ffff
> (00.077101) magic1:0 extended_size:0 xstate_bv:0 xstate_size:0
> (00.077139) xstate_bv: 0
> (00.077176) -----------------------
> (00.077215) Putting tsock into pid 1093
> (00.077604) Error (parasite-syscall.c:528): Unable to connect a transport
> socket: Permission denied

Do you use a kernel security module (selinux, apparmor)? I think this
error may be reported due to their rules. Could you try to disable the
kernel security module and try to dump the mysql service? We need to
proof that the problem is really about security modules.

Thanks,
Andrew

> (00.077693) Error (parasite-syscall.c:134): Can't block signals for 1093: No
> such process
> (00.077737) Error (cr-dump.c:1237): Can't infect (pid: 1093) with parasite
> (00.078966) Unlock network
> (00.079009) Unfreezing tasks into 1
> (00.079048)     Unseizing 1093 into 1
> (00.079118) Error (ptrace.c:54): Unable to detach from 1093: No such process
> (00.080127) Error (cr-dump.c:1614): Dumping FAILED.
> 
> It showed 'Permission denied'. I executed the command by root user.
> Is the problem related with either MySQL setting or MySQL ran by systemd?
> (https://criu.org/Integration#Systemd)
> 
> Thanks,
> Yohei Kamitsukasa