vhd-util check is very slow because of the double loop. We can create an array of sector ranges occupied by each extent and then sort it. We can then check for overlaps, so the whole process should take O(NlogN) time instead of O(N^2).

Pause requests are idempotent. If this is the expected behavior, it should be documented, otherwise an error should be returned.

Essentially I'm getting following compile time errors

td-req.c: In function 'guest_copy2':
td-req.c:359:18: error: 'struct gntdev_grant_copy_segment' has no member named 'iov'
gcopy_seg->iov.iov_base = tapreq->vma + (i << PAGE_SHIFT) + (blkif_seg->first_sect << SECTOR_SHIFT);
^
td-req.c:360:18: error: 'struct gntdev_grant_copy_segment' has no member named 'iov'
gcopy_seg->iov.iov_len = (blkif_seg->last_sect - blkif_seg->first_sect + 1) << SECTOR_SHIFT;
^
td-req.c:361:18: error: 'struct gntdev_grant_copy_segment' has no member named 'ref'
gcopy_seg->ref = blkif_seg->gref;
^
td-req.c:362:18: error: 'struct gntdev_grant_copy_segment' has no member named 'offset'
gcopy_seg->offset = blkif_seg->first_sect << SECTOR_SHIFT;
^
td-req.c:365:10: error: 'struct ioctl_gntdev_grant_copy' has no member named 'dir'
gcopy.dir = blkif_rq_wr(&tapreq->msg);
^
td-req.c:366:10: error: 'struct ioctl_gntdev_grant_copy' has no member named 'domid'
gcopy.domid = blkif->domid;

Environment -
[root@localhost blktap-3.2.0.xs1086]# rpm -qa | grep kernel | grep 4.16
kernel-devel-4.16.8-1.x86_64
kernel-headers-4.16.8-1.x86_64
kernel-4.16.8-1.x86_64

[root@localhost blktap-3.2.0.xs1086]# rpm -qa | grep xen
xe-guest-utilities-xenstore-7.0.0-24.x86_64
xen-dom0-libs-devel-4.6.1-2.1.0.851.31985.x86_64
xenserver-ddk-files-7.0.0-16.noarch
xen-hypervisor-4.6.1-2.1.0.851.31985.x86_64
xen-dom0-libs-4.6.1-2.1.0.851.31985.x86_64
xen-libs-devel-4.6.1-2.1.0.851.31985.x86_64
xenserver-release-7.0.0-125770c.96.x86_64
xen-devel-4.6.1-2.1.0.851.31985.x86_64
xen-libs-4.6.1-2.1.0.851.31985.x86_64

Kindly suggest a way ahead to make it compatible.

blktap does not build with dkms on debian sid with 3.9.1 kernal

DKMS make.log for blktap-2.0.91 for kernel 3.9-1-amd64 (x86_64)
Fr 7. Jun 21:08:00 UTC 2013
make: Entering directory /usr/src/linux-headers-3.9-1-amd64' CC [M] /var/lib/dkms/blktap/2.0.91/build/control.o In file included from /var/lib/dkms/blktap/2.0.91/build/control.c:30:0: /var/lib/dkms/blktap/2.0.91/build/blktap.h:75:41: warning: variably modified 'pending' at file scope [enabled by default] CC [M] /var/lib/dkms/blktap/2.0.91/build/ring.o In file included from /var/lib/dkms/blktap/2.0.91/build/ring.c:30:0: /var/lib/dkms/blktap/2.0.91/build/blktap.h:75:41: warning: variably modified 'pending' at file scope [enabled by default] /var/lib/dkms/blktap/2.0.91/build/ring.c: In function 'blktap_ring_map_request': /var/lib/dkms/blktap/2.0.91/build/ring.c:206:2: error: too few arguments to function 'do_mmap_pgoff' In file included from /usr/src/linux-headers-3.9-1- common/include/linux/pagemap.h:7:0, from /usr/src/linux-headers-3.9-1-common/include/linux/blkdev.h:13, from /var/lib/dkms/blktap/2.0.91/build/ring.c:27: /usr/src/linux-headers-3.9-1-common/include/linux/mm.h:1454:22: note: declared here /var/lib/dkms/blktap/2.0.91/build/ring.c: In function 'blktap_ring_mmap_request': /var/lib/dkms/blktap/2.0.91/build/ring.c:438:19: error: 'VM_RESERVED' undeclared (first use in this function) /var/lib/dkms/blktap/2.0.91/build/ring.c:438:19: note: each undeclared identifier is reported only once for each function it appears in /var/lib/dkms/blktap/2.0.91/build/ring.c: In function 'blktap_ring_mmap_sring': /var/lib/dkms/blktap/2.0.91/build/ring.c:474:19: error: 'VM_RESERVED' undeclared (first use in this function) make[3]: *** [/var/lib/dkms/blktap/2.0.91/build/ring.o] Error 1 make[2]: *** [_module_/var/lib/dkms/blktap/2.0.91/build] Error 2 make[1]: *** [sub-make] Fehler 2 make: *** [all] Fehler 2 make: Leaving directory/usr/src/linux-headers-3.9-1-amd64'

Could you please tag revision ec6e15f as '0.9.0'. This is the revision we use in xenserver-core, and tagging it here will allow us to use the upstream repository instead of our own branch.

Hi there!

Would it be possible to have an iops/rate limits and then expose it to the VBD object of XenServer? This is often a no-go for multi-tenant environnements because of that.

Any idea @chandrikas ?

Hi,

I just took over this package in Fedora, previous it's 2.0.90 maintained by @ke4qqq.

I just checked the VERSION file, saying that it's 3.0.0 now, but the git tags confuse me again, why it's 0.9.2?

Also libvhdio-2.0.90.so is still the present library in the tree. Shouldn't it be renamed to libvhdio-3.0.0.so?

I hope I can get an answer.

Thanks.

How to reproduce :

[root@localhost ~]# ./run.py -p blktap
Launching docker with args ['docker', 'run', '-i', '-t', '-u', 'builder', '-e', 'PACKAGES=blktap', 'xenserver/xenserver-build-env', '/usr/local/bin/init-container.sh']
.
.
.

sh-4.2$ ls
local-SRPMs
sh-4.2$ cd local-SRPMs/
sh-4.2$ ls
blktap-3.7.0-1.0.src.rpm
sh-4.2$ rpmbuild --rebuild blktap-3.7.0-1.0.src.rpm
Installing blktap-3.7.0-1.0.src.rpm
.
.
/bin/sh ../libtool --tag=CC --mode=compile gcc -DHAVE_CONFIG_H -I. -I.. -D_GNU_SOURCE -I../include -Wall -Werror -DXC_WANT_COMPAT_EVTCHN_API -DXC_WANT_COMPAT_GNTTAB_API -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -m64 -mtune=generic -c -o tapdisk-control.lo tapdisk-control.c
libtool: compile: gcc -DHAVE_CONFIG_H -I. -I.. -D_GNU_SOURCE -I../include -Wall -Werror -DXC_WANT_COMPAT_EVTCHN_API -DXC_WANT_COMPAT_GNTTAB_API -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -m64 -mtune=generic -c tapdisk-control.c -fPIC -DPIC -o .libs/tapdisk-control.o
In file included from td-blkif.h:43:0,
from tapdisk-vbd.h:40,
from tapdisk-control.c:53:
td-req.h:37:24: fatal error: xen/gntdev.h: No such file or directory
#include <xen/gntdev.h>
^
compilation terminated.

###################

gntdev.h is present at path /usr/include/xen/sys/gntdev.h

Indirect descriptors[1][2] feature dramatically improves throughput on advanced block device.

Upstream Linux accepted this feature long time ago, in both blkback and blkfront. And late Linux distributions already come with xen blkfront driver that support this feature.

[1] https://xenproject.org/2013/08/07/indirect-descriptors-for-xen-pv-disks/
[2] https://lwn.net/Articles/544614/

Hi there!

We started to think about various changes on blktap, especially to achieve these goals:

1. IO_uring support for faster IOPS (we already have a PoC)
2. Faster integration/merge in recent Linux kernels
3. Ideally, having it upstream Linux if possible at some point

Obviously, doing this together and going into the right direction after some architecture decision might be faster for everyone, deliver better perfs for both Citrix Hypervisor and XCP-ng. What do you think? Where you'd like to start?

Function returns success when vhd_get_header returns failure.

Hi, I have a issue concerning the ring size, I would like to modify the default value of MAX_RING_PAGE_ORDER to 4, in theory it's ok it's the max authorized value by the blkfront module.

But I don't understand why I can't boot correctly:

xenbus_probe_frontend: Waiting for devices to initialise: 25s...20s...15s...10s...5s...0s...

Should I change something else? Or is there a bug in blktap when MAX_RING_PAGE_ORDER is equal to 4? Thank you!

In vhd-util-check.c, all calls to memcmp() check the return value without any comparison operator (eg. if (memcmp(...)) ). I'm not a C guru, but from what I've read memcmp() will return 0 if equal, and potentially any other value if not. Shouldn't these checks be something like if (memcmp(...) != 0)?

To support my thoughts, I'm testing this on a VHD which I have no reason to believe is corrupt and I'm getting an invalid cookie message.

libvhd.c compares the value to 0 everywhere except in vhd_validate_batmap_header(). That appears to be a bug too.

Am I missing something here?

It would be great to update the documentation to explain how to enable debugging for tapback.

Thanks,
Flavien

closing a paused vbd is now legitimate

It'd be great to have a configure script check whether the grant-copy interface is available or not so we don't need to have a branch for buildroot.

Please consider creating man pages for binaries such as vhd-util lvm-util, etc.

This is a comment I got on the Archlinux package for this git repo, reporting it here just in case:

changes in gcc 4.8 bring a warning for pointer access:
http://gcc.gnu.org/gcc-4.8/porting_to.html

This is triggered in the current setup; also another error indicating that optimization needs to be on when compiling.

• I patched around this by using CFLAGS="-Wno-sizeof-pointer-memaccess -O" ./configure --prefix=/usr

But that is probably not going to be that neccesary in the coming weeks as there's a Debian bugreport about it, so someone has probably started talking to the git folks about it and they will either patch the source or ... similar. (I haven't actually loaded the kernel module yet, so I'm not going to recommend this solution just yet.)

I am happy to actually document this and submit a patch - feel free to tell me whether you wish it sent via mailing list or a pull request.

I am making this request as a function of packaging blktap for Fedora (see review request here https://bugzilla.redhat.com/show_bug.cgi?id=768545)

Hi,

I've seen failures to close blktap disks:

Sep 7 17:51:29.235595 xl: tap-err:tap_ctl_read_raw: failure reading data 0/544
Sep 7 17:51:29.235606 xl: tap-err:tap_ctl_send_and_receive: failed to receive 'close' message
Sep 7 17:51:29.235626 xl: [4274] libxl_blktap3.c:175:libxl__device_destroy_tapdisk:Failed to destroy tap device id 4163 minor 4
...
Sep 7 17:51:29.288949 tapdisk[4163]: ERROR: errno -32 at tapdisk_ctl_conn_send_event: close: failure sending message at offset 0/544

tap-ctl list shows processes lingering in state 0x2.

I was unable to repro under strace. This commit b6ef33c looks like what I was preparing to write. Are any details for CA-356508 available? Were close failures seen with lingering tapdisk processes?

Thanks.

nbd prints "timeout N", this is the event ID, not an actual time out being hit, fix this misleading print

I compile xen with cenos 7.2 then has some error ( fatal error: xen/xen.h: No such file or directory),I can download source code of xen, or I can download rpm of xen to solve it? but I don't know xen version,can you help me？