* Re: xl network-attach SEGV in 4.2 and 4.1 [and 1 more messages]
2013-04-18 16:01 ` Ian Jackson
@ 2013-04-18 16:46 ` Ian Jackson
2013-04-19 7:12 ` Jan Beulich
2013-04-18 18:58 ` xl network-attach SEGV in 4.2 and 4.1 Roger Pau Monné
2013-11-12 17:23 ` [PATCH RESEND] libxl: Avoid realloc(, 0) when libxl__xs_directory returns empty list Ian Jackson
2 siblings, 1 reply; 11+ messages in thread
From: Ian Jackson @ 2013-04-18 16:46 UTC (permalink / raw)
To: Marek Marczykowski; +Cc: Ian Campbell, Jan Beulich, xen-devel@lists.xen.org
Marek Marczykowski writes ("[Xen-devel] xl network-attach SEGV in 4.2 and 4.1"):
> 4.2 and 4.1 suffers from SEGV during xl network-attach in
> libxl__device_nic_add. In 4.3-unstable it is fixed by:
> 5420f2650 libxl: Set vfb and vkb devid if not done so by the caller
>
> So either the patch need to be backported to 4.1 and 4.2, or fixed by this one:
> - ------
> libxl: Fix SEGV in network-attach
I have applied this to 4.2 and 4.1. I had to fix up a conflict in 4.1
and also I fixed a whitespace bug. The patches as committed are below.
> When "device/vif" directory exists but is empty l!=NULL, but nb==0, so
> l[nb-1] is invalid. Add missing check.
...
> if (!(l = libxl__xs_directory(gc, XBT_NULL,
> - - libxl__sprintf(gc, "%s/device/vif",
> dompath), &nb))) {
> + libxl__sprintf(gc, "%s/device/vif",
> dompath), &nb)) ||
> + nb == 0) {
Something in the arrangements you used to prepare this patch
linewrapped it _before you signed it_.
Ian Jackson writes ("Re: [Xen-devel] xl network-attach SEGV in 4.2 and 4.1"):
> Author: Ian Jackson <ian.jackson@eu.citrix.com>
> Date: Thu Apr 18 16:27:46 2013 +0100
>
> libxl: Avoid realloc(,0) when libxl__xs_directory returns empty list
IWBNI someone would review this and ack it; I'd like it to go into
4.1, 4.2 and unstable.
Ian.
^ permalink raw reply [flat|nested] 11+ messages in thread* Re: xl network-attach SEGV in 4.2 and 4.1 [and 1 more messages]
2013-04-18 16:46 ` xl network-attach SEGV in 4.2 and 4.1 [and 1 more messages] Ian Jackson
@ 2013-04-19 7:12 ` Jan Beulich
2013-04-22 9:49 ` Ian Jackson
0 siblings, 1 reply; 11+ messages in thread
From: Jan Beulich @ 2013-04-19 7:12 UTC (permalink / raw)
To: Ian Jackson; +Cc: xen-devel@lists.xen.org, Ian Campbell, Marek Marczykowski
>>> On 18.04.13 at 18:46, Ian Jackson <Ian.Jackson@eu.citrix.com> wrote:
> Marek Marczykowski writes ("[Xen-devel] xl network-attach SEGV in 4.2 and
> 4.1"):
>> 4.2 and 4.1 suffers from SEGV during xl network-attach in
>> libxl__device_nic_add. In 4.3-unstable it is fixed by:
>> 5420f2650 libxl: Set vfb and vkb devid if not done so by the caller
>>
>> So either the patch need to be backported to 4.1 and 4.2, or fixed by this
> one:
>> - ------
>> libxl: Fix SEGV in network-attach
>
> I have applied this to 4.2 and 4.1. I had to fix up a conflict in 4.1
> and also I fixed a whitespace bug. The patches as committed are below.
This fixup on 4.1 went wrong - you dropped an & that's necessary.
I'll take the liberty to fix this right away.
Jan
>> When "device/vif" directory exists but is empty l!=NULL, but nb==0, so
>> l[nb-1] is invalid. Add missing check.
> ...
>> if (!(l = libxl__xs_directory(gc, XBT_NULL,
>> - - libxl__sprintf(gc, "%s/device/vif",
>> dompath), &nb))) {
>> + libxl__sprintf(gc, "%s/device/vif",
>> dompath), &nb)) ||
>> + nb == 0) {
>
> Something in the arrangements you used to prepare this patch
> linewrapped it _before you signed it_.
>
> Ian Jackson writes ("Re: [Xen-devel] xl network-attach SEGV in 4.2 and 4.1"):
>> Author: Ian Jackson <ian.jackson@eu.citrix.com>
>> Date: Thu Apr 18 16:27:46 2013 +0100
>>
>> libxl: Avoid realloc(,0) when libxl__xs_directory returns empty list
>
> IWBNI someone would review this and ack it; I'd like it to go into
> 4.1, 4.2 and unstable.
>
> Ian.
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: xl network-attach SEGV in 4.2 and 4.1
2013-04-18 16:01 ` Ian Jackson
2013-04-18 16:46 ` xl network-attach SEGV in 4.2 and 4.1 [and 1 more messages] Ian Jackson
@ 2013-04-18 18:58 ` Roger Pau Monné
2013-11-12 17:23 ` [PATCH RESEND] libxl: Avoid realloc(, 0) when libxl__xs_directory returns empty list Ian Jackson
2 siblings, 0 replies; 11+ messages in thread
From: Roger Pau Monné @ 2013-04-18 18:58 UTC (permalink / raw)
To: Ian Jackson; +Cc: xen-devel@lists.xen.org, Ian Campbell, Marek Marczykowski
On 18/04/13 18:01, Ian Jackson wrote:
> Ian Campbell writes ("Re: [Xen-devel] xl network-attach SEGV in 4.2 and 4.1"):
>> On Wed, 2013-04-17 at 03:49 +0100, Marek Marczykowski wrote:
>>> Hi all,
>>> When "device/vif" directory exists but is empty l!=NULL, but nb==0, so
>>> l[nb-1] is invalid. Add missing check.
>
> Thanks for pointing this out. The root cause is the expectation that
> libxl__xs_directory won't return a non-null list but set *nb to 0.
>
> I found some occurrences of this bug in (xen-unstable) staging.
> See patch below, which should probably go into 4.1 and 4.2.
>
>>> Signed-Off-by: Marek Marczykowski <marmarek@invisiblethingslab.com>
>>>
>>> diff --git a/tools/libxl/libxl.c b/tools/libxl/libxl.c
>>> index 5783cd2..9e06a7d 100644
>>> --- a/tools/libxl/libxl.c
>>> +++ b/tools/libxl/libxl.c
>>> @@ -2569,7 +2569,8 @@ void libxl__device_nic_add(libxl__egc *egc, uint32_t
>>> domid,
>>> goto out_free;
>>> }
>>> if (!(l = libxl__xs_directory(gc, XBT_NULL,
>>> - libxl__sprintf(gc, "%s/device/vif",
>>> dompath), &nb))) {
>>> + libxl__sprintf(gc, "%s/device/vif",
>>> dompath), &nb)) ||
>>> + nb == 0) {
>
> I think this is the right way to fix this in the stable trees. I
> looked at the backport of "5420f2650 libxl: Set vfb and vkb devid if
> not done so by the caller" and it's nontrivial, so although it's a
> bugfix I'd rather leave it.
>
> Ian.
>
>
> commit d1f52920ee96c91665813fb875c0c76e5eb7a312
> Author: Ian Jackson <ian.jackson@eu.citrix.com>
> Date: Thu Apr 18 16:27:46 2013 +0100
>
> libxl: Avoid realloc(,0) when libxl__xs_directory returns empty list
>
> If the named path is a leaf node, libxl__xs_directory can succeed,
> returning non-null, but set *nb to 0.
>
> In three places in libxl this may result in a zero size argument being
> passed to malloc() or realloc(), which is not adviseable.
>
> Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
Acked-by: Roger Pau Monné <roger.pau@citrix.com>
>
> diff --git a/tools/libxl/libxl.c b/tools/libxl/libxl.c
> index 0f936c0..1c04d37 100644
> --- a/tools/libxl/libxl.c
> +++ b/tools/libxl/libxl.c
> @@ -1840,7 +1840,7 @@ libxl_device_vtpm *libxl_device_vtpm_list(libxl_ctx *ctx, uint32_t domid, int *n
>
> fe_path = libxl__sprintf(gc, "%s/device/vtpm", libxl__xs_get_dompath(gc, domid));
> dir = libxl__xs_directory(gc, XBT_NULL, fe_path, &ndirs);
> - if(dir) {
> + if (dir && ndirs) {
> vtpms = malloc(sizeof(*vtpms) * ndirs);
> libxl_device_vtpm* vtpm;
> libxl_device_vtpm* end = vtpms + ndirs;
> @@ -2314,7 +2314,7 @@ static int libxl__append_disk_list_of_type(libxl__gc *gc,
> be_path = libxl__sprintf(gc, "%s/backend/%s/%d",
> libxl__xs_get_dompath(gc, 0), type, domid);
> dir = libxl__xs_directory(gc, XBT_NULL, be_path, &n);
> - if (dir) {
> + if (dir && n) {
> libxl_device_disk *tmp;
> tmp = realloc(*disks, sizeof (libxl_device_disk) * (*ndisks + n));
> if (tmp == NULL)
> @@ -3006,7 +3006,7 @@ static int libxl__append_nic_list_of_type(libxl__gc *gc,
> be_path = libxl__sprintf(gc, "%s/backend/%s/%d",
> libxl__xs_get_dompath(gc, 0), type, domid);
> dir = libxl__xs_directory(gc, XBT_NULL, be_path, &n);
> - if (dir) {
> + if (dir && n) {
> libxl_device_nic *tmp;
> tmp = realloc(*nics, sizeof (libxl_device_nic) * (*nnics + n));
> if (tmp == NULL)
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
>
^ permalink raw reply [flat|nested] 11+ messages in thread* [PATCH RESEND] libxl: Avoid realloc(, 0) when libxl__xs_directory returns empty list
2013-04-18 16:01 ` Ian Jackson
2013-04-18 16:46 ` xl network-attach SEGV in 4.2 and 4.1 [and 1 more messages] Ian Jackson
2013-04-18 18:58 ` xl network-attach SEGV in 4.2 and 4.1 Roger Pau Monné
@ 2013-11-12 17:23 ` Ian Jackson
2013-11-12 17:31 ` Ian Campbell
2 siblings, 1 reply; 11+ messages in thread
From: Ian Jackson @ 2013-11-12 17:23 UTC (permalink / raw)
To: Ian Campbell, Marek Marczykowski, xen-devel@lists.xen.org; +Cc: Roger Pau Monne
Ian Jackson writes ("Re: [Xen-devel] xl network-attach SEGV in 4.2 and 4.1"):
> Thanks for pointing this out. The root cause is the expectation that
> libxl__xs_directory won't return a non-null list but set *nb to 0.
>
> I found some occurrences of this bug in (xen-unstable) staging.
> See patch below, which should probably go into 4.1 and 4.2.
Ping.
Ian.
>From ea275c0e29299a2399aa7814116d68bfdaeb21c1 Mon Sep 17 00:00:00 2001
From: Ian Jackson <ian.jackson@eu.citrix.com>
Date: Thu, 18 Apr 2013 16:27:46 +0100
Subject: [PATCH] libxl: Avoid realloc(,0) when libxl__xs_directory returns
empty list
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
If the named path is a leaf node, libxl__xs_directory can succeed,
returning non-null, but set *nb to 0.
In three places in libxl this may result in a zero size argument being
passed to malloc() or realloc(), which is not adviseable.
Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
Acked-by: Roger Pau Monné <roger.pau@citrix.com>
---
tools/libxl/libxl.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/tools/libxl/libxl.c b/tools/libxl/libxl.c
index 0f0f56c..887e0ca 100644
--- a/tools/libxl/libxl.c
+++ b/tools/libxl/libxl.c
@@ -1864,7 +1864,7 @@ libxl_device_vtpm *libxl_device_vtpm_list(libxl_ctx *ctx, uint32_t domid, int *n
fe_path = libxl__sprintf(gc, "%s/device/vtpm", libxl__xs_get_dompath(gc, domid));
dir = libxl__xs_directory(gc, XBT_NULL, fe_path, &ndirs);
- if(dir) {
+ if (dir && ndirs) {
vtpms = malloc(sizeof(*vtpms) * ndirs);
libxl_device_vtpm* vtpm;
libxl_device_vtpm* end = vtpms + ndirs;
@@ -2371,7 +2371,7 @@ static int libxl__append_disk_list_of_type(libxl__gc *gc,
be_path = libxl__sprintf(gc, "%s/backend/%s/%d",
libxl__xs_get_dompath(gc, 0), type, domid);
dir = libxl__xs_directory(gc, XBT_NULL, be_path, &n);
- if (dir) {
+ if (dir && n) {
libxl_device_disk *tmp;
tmp = realloc(*disks, sizeof (libxl_device_disk) * (*ndisks + n));
if (tmp == NULL)
@@ -3060,7 +3060,7 @@ static int libxl__append_nic_list_of_type(libxl__gc *gc,
be_path = libxl__sprintf(gc, "%s/backend/%s/%d",
libxl__xs_get_dompath(gc, 0), type, domid);
dir = libxl__xs_directory(gc, XBT_NULL, be_path, &n);
- if (dir) {
+ if (dir && n) {
libxl_device_nic *tmp;
tmp = realloc(*nics, sizeof (libxl_device_nic) * (*nnics + n));
if (tmp == NULL)
--
1.7.10.4
^ permalink raw reply related [flat|nested] 11+ messages in thread* Re: [PATCH RESEND] libxl: Avoid realloc(, 0) when libxl__xs_directory returns empty list
2013-11-12 17:23 ` [PATCH RESEND] libxl: Avoid realloc(, 0) when libxl__xs_directory returns empty list Ian Jackson
@ 2013-11-12 17:31 ` Ian Campbell
2013-11-12 17:57 ` Ian Jackson
0 siblings, 1 reply; 11+ messages in thread
From: Ian Campbell @ 2013-11-12 17:31 UTC (permalink / raw)
To: Ian Jackson; +Cc: Roger Pau Monne, Marek Marczykowski, xen-devel@lists.xen.org
On Tue, 2013-11-12 at 17:23 +0000, Ian Jackson wrote:
> Ian Jackson writes ("Re: [Xen-devel] xl network-attach SEGV in 4.2 and 4.1"):
> > Thanks for pointing this out. The root cause is the expectation that
> > libxl__xs_directory won't return a non-null list but set *nb to 0.
> >
> > I found some occurrences of this bug in (xen-unstable) staging.
> > See patch below, which should probably go into 4.1 and 4.2.
>
> Ping.
I could have sworn this went in...
>
> Ian.
>
> From ea275c0e29299a2399aa7814116d68bfdaeb21c1 Mon Sep 17 00:00:00 2001
> From: Ian Jackson <ian.jackson@eu.citrix.com>
> Date: Thu, 18 Apr 2013 16:27:46 +0100
> Subject: [PATCH] libxl: Avoid realloc(,0) when libxl__xs_directory returns
> empty list
> MIME-Version: 1.0
> Content-Type: text/plain; charset=UTF-8
> Content-Transfer-Encoding: 8bit
>
> If the named path is a leaf node, libxl__xs_directory can succeed,
> returning non-null, but set *nb to 0.
>
> In three places in libxl this may result in a zero size argument being
> passed to malloc() or realloc(), which is not adviseable.
I thought it was strictly speaking fine, but I can see why it would be
best to avoid.
> Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
> Acked-by: Roger Pau Monné <roger.pau@citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
> ---
> tools/libxl/libxl.c | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/tools/libxl/libxl.c b/tools/libxl/libxl.c
> index 0f0f56c..887e0ca 100644
> --- a/tools/libxl/libxl.c
> +++ b/tools/libxl/libxl.c
> @@ -1864,7 +1864,7 @@ libxl_device_vtpm *libxl_device_vtpm_list(libxl_ctx *ctx, uint32_t domid, int *n
>
> fe_path = libxl__sprintf(gc, "%s/device/vtpm", libxl__xs_get_dompath(gc, domid));
> dir = libxl__xs_directory(gc, XBT_NULL, fe_path, &ndirs);
> - if(dir) {
> + if (dir && ndirs) {
> vtpms = malloc(sizeof(*vtpms) * ndirs);
> libxl_device_vtpm* vtpm;
> libxl_device_vtpm* end = vtpms + ndirs;
> @@ -2371,7 +2371,7 @@ static int libxl__append_disk_list_of_type(libxl__gc *gc,
> be_path = libxl__sprintf(gc, "%s/backend/%s/%d",
> libxl__xs_get_dompath(gc, 0), type, domid);
> dir = libxl__xs_directory(gc, XBT_NULL, be_path, &n);
> - if (dir) {
> + if (dir && n) {
> libxl_device_disk *tmp;
> tmp = realloc(*disks, sizeof (libxl_device_disk) * (*ndisks + n));
> if (tmp == NULL)
> @@ -3060,7 +3060,7 @@ static int libxl__append_nic_list_of_type(libxl__gc *gc,
> be_path = libxl__sprintf(gc, "%s/backend/%s/%d",
> libxl__xs_get_dompath(gc, 0), type, domid);
> dir = libxl__xs_directory(gc, XBT_NULL, be_path, &n);
> - if (dir) {
> + if (dir && n) {
> libxl_device_nic *tmp;
> tmp = realloc(*nics, sizeof (libxl_device_nic) * (*nnics + n));
> if (tmp == NULL)
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel
^ permalink raw reply [flat|nested] 11+ messages in thread* Re: [PATCH RESEND] libxl: Avoid realloc(, 0) when libxl__xs_directory returns empty list
2013-11-12 17:31 ` Ian Campbell
@ 2013-11-12 17:57 ` Ian Jackson
2013-11-25 13:55 ` Ian Jackson
0 siblings, 1 reply; 11+ messages in thread
From: Ian Jackson @ 2013-11-12 17:57 UTC (permalink / raw)
To: Ian Campbell; +Cc: Roger Pau Monne, Marek Marczykowski, xen-devel@lists.xen.org
Ian Campbell writes ("Re: [PATCH RESEND] libxl: Avoid realloc(,0) when libxl__xs_directory returns empty list"):
> On Tue, 2013-11-12 at 17:23 +0000, Ian Jackson wrote:
> > Ping.
>
> I could have sworn this went in...
Yes. I have found a bunch of these lying around in branches in my
main tree.
> > In three places in libxl this may result in a zero size argument being
> > passed to malloc() or realloc(), which is not adviseable.
>
> I thought it was strictly speaking fine, but I can see why it would be
> best to avoid.
Calling malloc(0) merely produces a fiddly-to-deal-with corner case,
where it may succeed by returning 0.
realloc(something,0) is actually wrong. If it returns 0 you don't
know whether that means "I tried to allocate you a new 0-length block,
but failed, so your original block is still there" or "I have
successfully freed your old block; have this null pointer as an
answer".
> > Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
> > Acked-by: Roger Pau Monné <roger.pau@citrix.com>
>
> Acked-by: Ian Campbell <ian.campbell@citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
And put on the backport list.
Thanks,
Ian.
^ permalink raw reply [flat|nested] 11+ messages in thread