[libvirt] [PATCH] Remove hard dependency on DMI
The udev node device driver tries to glean a few tidbits of information from /sys/devices/virtual/dmi/id. If either the BIOS or the kernel does not support DMI, libvirtd logs the error Failed to get udev device for syspath '/sys/devices/virtual/dmi/id' or '/sys/class/dmi/id' and refuses to start. This seems a bit extreme; information obtained from DMI, like the system board vendor and BIOS version, is useful but not critical to libvirt's operation. This patch eliminates the hard dependency on DMI. On systems without DMI, libvirtd logs a warning, and leaves the "computer" node device's properties empty. Signed-off-by: Ed Swierk <eswierk@aristanetworks.com> --- Index: libvirt-0.7.6/src/node_device/node_device_udev.c =================================================================== --- libvirt-0.7.6.orig/src/node_device/node_device_udev.c +++ libvirt-0.7.6/src/node_device/node_device_udev.c @@ -1471,9 +1471,9 @@ static int udevSetupSystemDev(void) if (device == NULL) { device = udev_device_new_from_syspath(udev, DMI_DEVPATH_FALLBACK); if (device == NULL) { - VIR_ERROR("Failed to get udev device for syspath '%s' or '%s'", - DMI_DEVPATH, DMI_DEVPATH_FALLBACK); - goto out; + VIR_WARN("Failed to get udev device for syspath '%s' or '%s'", + DMI_DEVPATH, DMI_DEVPATH_FALLBACK); + goto out2; } } @@ -1532,6 +1532,7 @@ static int udevSetupSystemDev(void) udev_device_unref(device); +out2: dev = virNodeDeviceAssignDef(&driverState->devs, def); if (dev == NULL) { VIR_ERROR("Failed to create device for '%s'", def->name);
On 03/03/2010 05:32 PM, Ed Swierk wrote:
The udev node device driver tries to glean a few tidbits of information from /sys/devices/virtual/dmi/id. If either the BIOS or the kernel does not support DMI, libvirtd logs the error
Failed to get udev device for syspath '/sys/devices/virtual/dmi/id' or '/sys/class/dmi/id'
and refuses to start.
This seems a bit extreme; information obtained from DMI, like the system board vendor and BIOS version, is useful but not critical to libvirt's operation.
Agreed.
This patch eliminates the hard dependency on DMI. On systems without DMI, libvirtd logs a warning, and leaves the "computer" node device's properties empty.
Although I use goto a lot, I generally try to avoid multiple labels within a function, just because I think it gets out of hand really quickly. Although it's a slightly more invasive patch, would you refactor the code to look something like what I've attached? I haven't even compile tested it as I'm running late, but that's the idea. Dave
Signed-off-by: Ed Swierk<eswierk@aristanetworks.com>
--- Index: libvirt-0.7.6/src/node_device/node_device_udev.c =================================================================== --- libvirt-0.7.6.orig/src/node_device/node_device_udev.c +++ libvirt-0.7.6/src/node_device/node_device_udev.c @@ -1471,9 +1471,9 @@ static int udevSetupSystemDev(void) if (device == NULL) { device = udev_device_new_from_syspath(udev, DMI_DEVPATH_FALLBACK); if (device == NULL) { - VIR_ERROR("Failed to get udev device for syspath '%s' or '%s'", - DMI_DEVPATH, DMI_DEVPATH_FALLBACK); - goto out; + VIR_WARN("Failed to get udev device for syspath '%s' or '%s'", + DMI_DEVPATH, DMI_DEVPATH_FALLBACK); + goto out2; } }
@@ -1532,6 +1532,7 @@ static int udevSetupSystemDev(void)
udev_device_unref(device);
+out2: dev = virNodeDeviceAssignDef(&driverState->devs, def); if (dev == NULL) { VIR_ERROR("Failed to create device for '%s'", def->name);
-- libvir-list mailing list libvir-list@redhat.com https://www.redhat.com/mailman/listinfo/libvir-list
On Wed, Mar 3, 2010 at 2:57 PM, Dave Allan <dallan@redhat.com> wrote:
Although I use goto a lot, I generally try to avoid multiple labels within a function, just because I think it gets out of hand really quickly. Although it's a slightly more invasive patch, would you refactor the code to look something like what I've attached? I haven't even compile tested it as I'm running late, but that's the idea.
Is there a piece of code in libvirt that exemplifies the preferred error handling style? (http://libvirt.org/hacking.html doesn't cover this issue, as far as I can tell.) Just in the very small part of libvirt I've hacked on recently I've found a variety of styles, including - pair every allocation with a goto label that frees the allocation and all the earlier ones, and goto the appropriate label on error - don't use goto at all, and on error, do the necessary frees and return -1, with each error case having to do one more free - a combination of the above, with each error case doing the necessary frees, but using goto out more or less as an alias for return -1 - none of the above, not bothering to free anything when an allocation fails (see udevSetupSystemDev for an example) There are probably arguments to be made for each of these styles, but it would be helpful to know which of them is preferred when writing new code or refactoring existing code. That said, I'll gladly refactor my patch towards the preferred style. --Ed
On 03/03/2010 07:20 PM, Ed Swierk wrote:
On Wed, Mar 3, 2010 at 2:57 PM, Dave Allan<dallan@redhat.com> wrote:
Although I use goto a lot, I generally try to avoid multiple labels within a function, just because I think it gets out of hand really quickly. Although it's a slightly more invasive patch, would you refactor the code to look something like what I've attached? I haven't even compile tested it as I'm running late, but that's the idea.
Is there a piece of code in libvirt that exemplifies the preferred error handling style? (http://libvirt.org/hacking.html doesn't cover this issue, as far as I can tell.) Just in the very small part of libvirt I've hacked on recently I've found a variety of styles, including
Agreed that we should add a statement to the hacking guide. My preferences are as follows.
- pair every allocation with a goto label that frees the allocation and all the earlier ones, and goto the appropriate label on error
I like Robert Love's description of this style at the very end of the thread at: http://kerneltrap.org/node/553/2131 I like this style, but my impression is that generally the libvirt community prefers to have a single label that frees everything, perhaps conditionally on error, unless it's absolutely necessary to have multiple labels. I reworked udevSetupSystemDev into this style (which also fixes the bug you pointed out that it didn't properly free resources on error). The patch also makes failure to find DMI data non-fatal.
- don't use goto at all, and on error, do the necessary frees and return -1, with each error case having to do one more free
I find this style troublesome to maintain, as any additional allocations require modifications to each error case.
- a combination of the above, with each error case doing the necessary frees, but using goto out more or less as an alias for return -1
Again, I think duplicating the frees in each error case is less maintainable than having them in on place.
- none of the above, not bothering to free anything when an allocation fails (see udevSetupSystemDev for an example)
Failure to cleanup is a bug. Please send mail (and, even better, patches) about any other instances you find.
There are probably arguments to be made for each of these styles, but it would be helpful to know which of them is preferred when writing new code or refactoring existing code.
That said, I'll gladly refactor my patch towards the preferred style.
--Ed
On Thu, Mar 4, 2010 at 10:31 AM, Dave Allan <dallan@redhat.com> wrote:
I reworked udevSetupSystemDev into this style (which also fixes the bug you pointed out that it didn't properly free resources on error). The patch also makes failure to find DMI data non-fatal.
Your patch works fine. Thanks. Acked-by: Ed Swierk <eswierk@aristanetworks.com>
On 03/04/2010 07:49 PM, Ed Swierk wrote:
On Thu, Mar 4, 2010 at 10:31 AM, Dave Allan<dallan@redhat.com> wrote:
I reworked udevSetupSystemDev into this style (which also fixes the bug you pointed out that it didn't properly free resources on error). The patch also makes failure to find DMI data non-fatal.
Your patch works fine. Thanks.
Acked-by: Ed Swierk<eswierk@aristanetworks.com>
Thanks for the ACK; anybody else willing to provide the second? Dave
On 03/05/2010 10:41 AM, Dave Allan wrote:
On 03/04/2010 07:49 PM, Ed Swierk wrote:
On Thu, Mar 4, 2010 at 10:31 AM, Dave Allan<dallan@redhat.com> wrote:
I reworked udevSetupSystemDev into this style (which also fixes the bug you pointed out that it didn't properly free resources on error). The patch also makes failure to find DMI data non-fatal.
Your patch works fine. Thanks.
Acked-by: Ed Swierk<eswierk@aristanetworks.com>
Thanks for the ACK; anybody else willing to provide the second?
From 23526641083527139548c68b4637bae8350d2f98 Mon Sep 17 00:00:00 2001 From: David Allan <dallan@redhat.com> Date: Thu, 4 Mar 2010 13:17:24 -0500 Subject: [PATCH 1/1] Free resources on error in udev startup
* The udev driver didn't properly free resources that it allocates when setting up the 'computer' device in the error case.
I didn't see this one in the repository yet; but if it helps to have another review via code inspection, I've read through your patch and it looks like a sane split in functionality. Therefore: ACK. -- Eric Blake eblake@redhat.com +1-801-349-2682 Libvirt virtualization library http://libvirt.org
On 03/09/2010 06:04 PM, Eric Blake wrote:
On 03/05/2010 10:41 AM, Dave Allan wrote:
On 03/04/2010 07:49 PM, Ed Swierk wrote:
On Thu, Mar 4, 2010 at 10:31 AM, Dave Allan<dallan@redhat.com> wrote:
I reworked udevSetupSystemDev into this style (which also fixes the bug you pointed out that it didn't properly free resources on error). The patch also makes failure to find DMI data non-fatal.
Your patch works fine. Thanks.
Acked-by: Ed Swierk<eswierk@aristanetworks.com>
Thanks for the ACK; anybody else willing to provide the second?
From 23526641083527139548c68b4637bae8350d2f98 Mon Sep 17 00:00:00 2001 From: David Allan<dallan@redhat.com> Date: Thu, 4 Mar 2010 13:17:24 -0500 Subject: [PATCH 1/1] Free resources on error in udev startup
* The udev driver didn't properly free resources that it allocates when setting up the 'computer' device in the error case.
I didn't see this one in the repository yet; but if it helps to have another review via code inspection, I've read through your patch and it looks like a sane split in functionality. Therefore:
ACK.
Ok, thanks, pushed. Dave
On Thu, Mar 04, 2010 at 01:31:33PM -0500, Dave Allan wrote:
On 03/03/2010 07:20 PM, Ed Swierk wrote:
On Wed, Mar 3, 2010 at 2:57 PM, Dave Allan<dallan@redhat.com> wrote:
Although I use goto a lot, I generally try to avoid multiple labels within a function, just because I think it gets out of hand really quickly. Although it's a slightly more invasive patch, would you refactor the code to look something like what I've attached? I haven't even compile tested it as I'm running late, but that's the idea.
Is there a piece of code in libvirt that exemplifies the preferred error handling style? (http://libvirt.org/hacking.html doesn't cover this issue, as far as I can tell.) Just in the very small part of libvirt I've hacked on recently I've found a variety of styles, including
Agreed that we should add a statement to the hacking guide. My preferences are as follows.
- pair every allocation with a goto label that frees the allocation and all the earlier ones, and goto the appropriate label on error
I like Robert Love's description of this style at the very end of the thread at:
http://kerneltrap.org/node/553/2131
I like this style, but my impression is that generally the libvirt community prefers to have a single label that frees everything, perhaps conditionally on error, unless it's absolutely necessary to have multiple labels.
If the cleanup code only involves free'ing memory, then having multiple labels is complete overkill. VIR_FREE() and every function named XXXFree() in libvirt is required to handle NULL as its arg. Thus you can safely call free on all the variables even if they wre not yet allocated (yes they have to have been initialized to NULL). This is much simpler & clearer than having multiple gotos Daniel -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://deltacloud.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|
participants (4)
-
Daniel P. Berrange -
Dave Allan -
Ed Swierk -
Eric Blake