capstan doesnt start qemu, invalid argument

[maci0]

capstan run cloudius/osv
Created instance: cloudius-osv
qemu-system-x86_64: -drive file=/home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2,if=none,id=hd0,aio=native,cache=none: could not open disk image /home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2: Invalid argument

exit status 1

[maci0]

FIY my /home is on zfs maybe thats the culprint ?

[dorlaor]

On Mon, Sep 8, 2014 at 10:22 AM, Marcel Wysocki notifications@github.com
wrote:

FIY my /home is on zfs maybe thats the culprint ?

It's not a local filesystem issue.
It's either a bug or something happened to this image.
Can you run 'capstan rmi cloudius/osv'
and run it again?

Is this file present? I check and it works for me.
What happens when you run 'qemu-system-x86_64' directly w/o any parameter?
I like to see it can run (expect a pxe prompt)

—
Reply to this email directly or view it on GitHub
#118 (comment)
.

[penberg]

@maci0 AFAICT, that error is coming from QEMU and actually might be related to ZFS. If removing the image and downloading it again like @dorlaor suggested doesn't fix it, can you please run capstan with strace -f and post the results here?

[maci0]

[pid 27835] rt_sigaction(SIGBUS, {0x7f94d88c9fb0, [], SA_RESTORER|SA_SIGINFO, 0x7f94d68d6720}, NULL, 8) = 0
[pid 27835] prctl(PR_MCE_KILL, 0x1, 0x1, 0, 0) = 0
[pid 27835] brk(0)                      = 0x7f94d97d6000
[pid 27835] brk(0x7f94d97f7000)         = 0x7f94d97f7000
[pid 27835] open("/home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2", O_RDONLY|O_NONBLOCK|O_CLOEXEC) = 9
[pid 27835] fstat(9, {st_mode=S_IFREG|0644, st_size=197120, ...}) = 0
[pid 27835] close(9)                    = 0
[pid 27835] open("/home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2", O_RDONLY|O_NONBLOCK|O_CLOEXEC) = 9
[pid 27835] fstat(9, {st_mode=S_IFREG|0644, st_size=197120, ...}) = 0
[pid 27835] close(9)                    = 0
[pid 27835] stat("/home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2", {st_mode=S_IFREG|0644, st_size=197120, ...}) = 0
[pid 27835] open("/home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2", O_RDWR|O_DIRECT|O_CLOEXEC) = -1 EINVAL (Invalid argument)
[pid 27835] rt_sigprocmask(SIG_BLOCK, NULL, [BUS USR1 ALRM IO], 8) = 0
[pid 27835] mmap(NULL, 1052672, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f94d84f1000
[pid 27835] rt_sigprocmask(SIG_SETMASK, [BUS USR1 ALRM IO], [BUS USR1 ALRM IO], 8) = 0
[pid 27835] brk(0)                      = 0x7f94d97f7000
[pid 27835] brk(0)                      = 0x7f94d97f7000
[pid 27835] brk(0x7f94d97f6000)         = 0x7f94d97f6000
[pid 27835] brk(0)                      = 0x7f94d97f6000
[pid 27835] write(2, "qemu-system-x86_64:", 19qemu-system-x86_64:) = 19
[pid 27835] write(2, " -drive", 7 -drive)      = 7
[pid 27835] write(2, " file=/home/maci/.capstan/instan"..., 101 file=/home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2,if=none,id=hd0,aio=native,cache=none) = 101
[pid 27835] write(2, ": ", 2: )           = 2
[pid 27835] write(2, "could not open disk image /home/"..., 102could not open disk image /home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2: Invalid argument) = 102
[pid 27835] write(2, "\n", 1
)           = 1

i think the problem is O_DIRECT openzfs/zfs#224
which comes from the cache=none option

[dorlaor]

On Tue, Sep 9, 2014 at 11:37 AM, Marcel Wysocki notifications@github.com
wrote:

[pid 27835] rt_sigaction(SIGBUS, {0x7f94d88c9fb0, [], SA_RESTORER|SA_SIGINFO, 0x7f94d68d6720}, NULL, 8) = 0
[pid 27835] prctl(PR_MCE_KILL, 0x1, 0x1, 0, 0) = 0
[pid 27835] brk(0) = 0x7f94d97d6000
[pid 27835] brk(0x7f94d97f7000) = 0x7f94d97f7000
[pid 27835] open("/home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2", O_RDONLY|O_NONBLOCK|O_CLOEXEC) = 9
[pid 27835] fstat(9, {st_mode=S_IFREG|0644, st_size=197120, ...}) = 0
[pid 27835] close(9) = 0
[pid 27835] open("/home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2", O_RDONLY|O_NONBLOCK|O_CLOEXEC) = 9
[pid 27835] fstat(9, {st_mode=S_IFREG|0644, st_size=197120, ...}) = 0
[pid 27835] close(9) = 0
[pid 27835] stat("/home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2", {st_mode=S_IFREG|0644, st_size=197120, ...}) = 0
[pid 27835] open("/home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2", O_RDWR|O_DIRECT|O_CLOEXEC) = -1 EINVAL (Invalid argument)
[pid 27835] rt_sigprocmask(SIG_BLOCK, NULL, [BUS USR1 ALRM IO], 8) = 0
[pid 27835] mmap(NULL, 1052672, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f94d84f1000
[pid 27835] rt_sigprocmask(SIG_SETMASK, [BUS USR1 ALRM IO], [BUS USR1 ALRM IO], 8) = 0
[pid 27835] brk(0) = 0x7f94d97f7000
[pid 27835] brk(0) = 0x7f94d97f7000
[pid 27835] brk(0x7f94d97f6000) = 0x7f94d97f6000
[pid 27835] brk(0) = 0x7f94d97f6000
[pid 27835] write(2, "qemu-system-x86_64:", 19qemu-system-x86_64:) = 19
[pid 27835] write(2, " -drive", 7 -drive) = 7
[pid 27835] write(2, " file=/home/maci/.capstan/instan"..., 101 file=/home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2,if=none,id=hd0,aio=native,cache=none) = 101
[pid 27835] write(2, ": ", 2: ) = 2
[pid 27835] write(2, "could not open disk image /home/"..., 102could not open disk image /home/maci/.capstan/instances/qemu/cloudius-osv/disk.qcow2: Invalid argument) = 102
[pid 27835] write(2, "\n", 1
) = 1

i think the problem is O_DIRECT openzfs/zfs#224
openzfs/zfs#224
which comes from the cache=none option

You are correct. Too bad...
@penberg, how about adding an option to use cache=writethrough|none ?
In the mean time, just patch Capstan manually.

—
Reply to this email directly or view it on GitHub
#118 (comment)
.

[tgrabiec]

@dOr I think this decision could be automated in the same way as OSv's run.py automates it since commit cloudius-systems/osv@1989d9e

[dorlaor]

On Tue, Sep 9, 2014 at 5:59 PM, Tomasz Grabiec notifications@github.com
wrote:

@dOr https://github.com/dor I think this decision could be automated in
the same way as OSv's run.py automates it since commit
cloudius-systems/osv@1989d9e
cloudius-systems/osv@1989d9e

It's possible.
The error report is misleading: "raise Exception('Path not found: ' + path)"
Also, one shouldn't use cache=unsafe w/o being verbose about it.
—

Reply to this email directly or view it on GitHub
#118 (comment)
.

[tgrabiec]

2014-09-09 17:02 GMT+02:00 dorlaor notifications@github.com:

On Tue, Sep 9, 2014 at 5:59 PM, Tomasz Grabiec notifications@github.com
wrote:

@dOr https://github.com/dor I think this decision could be automated
in
the same way as OSv's run.py automates it since commit
cloudius-systems/osv@1989d9e
<
cloudius-systems/osv@1989d9eabc97d68c3a57181ff584f1b830705d98>

It's possible.
The error report is misleading: "raise Exception('Path not found: ' +
path)"

This error is thrown only when the path does not exist, it's unrelated to
the O_DIRECT thing.

Also, one shouldn't use cache=unsafe w/o being verbose about it.

[penberg]

@maci0 I opened pull request #119 that should fix it. If you have the time, can you please test it on your machine?

[jpeach]

FYI, this change breaks the OS X build because sys call.O_DIRECT is not available ...

$ GOPATH=$HOME/go ./install
# github.com/cloudius-systems/capstan/hypervisor/qemu
hypervisor/qemu/qemu.go:231: undefined: syscall.O_DIRECT
# github.com/cloudius-systems/capstan/hypervisor/qemu
hypervisor/qemu/qemu.go:231: undefined: syscall.O_DIRECT

[penberg]

@jpeach I opened pull request #121 to fix that up. Thanks for reporting!