| Message ID | 20220526194752.307192-1-javierm@redhat.com (mailing list archive) |
|---|---|
| State | Accepted, archived |
| Headers | show |
| Series | fbdev: vesafb: Fix a use-after-free due early fb_info cleanup | expand |
[2022-05-26 21:47:52] Javier Martinez Canillas: > Commit b3c9a924aab6 ("fbdev: vesafb: Cleanup fb_info in .fb_destroy rather > than .remove") fixed a use-after-free error due the vesafb driver freeing > the fb_info in the .remove handler instead of doing it in .fb_destroy. > > This can happen if the .fb_destroy callback is executed after the .remove > callback, since the former tries to access a pointer freed by the latter. > > But that change didn't take into account that another possible scenario is > that .fb_destroy is called before the .remove callback. For example, if no > process has the fbdev chardev opened by the time the driver is removed. > > If that's the case, fb_info will be freed when unregister_framebuffer() is > called, making the fb_info pointer accessed in vesafb_remove() after that > to no longer be valid. > > To prevent that, move the expression containing the info->par to happen > before the unregister_framebuffer() function call. > > Fixes: b3c9a924aab6 ("fbdev: vesafb: Cleanup fb_info in .fb_destroy rather than .remove") > Reported-by: Pascal Ernster <dri-devel@hardfalcon.net> > Signed-off-by: Javier Martinez Canillas <javierm@redhat.com> Tested on a bare metal machine and on a test VM that had both crashed when booting a kernel lacking this patch. Again, thanks a lot for the quick fix! :) Tested-by: Pascal Ernster <dri-devel@hardfalcon.net> > --- > > drivers/video/fbdev/vesafb.c | 5 +++-- > 1 file changed, 3 insertions(+), 2 deletions(-) > > diff --git a/drivers/video/fbdev/vesafb.c b/drivers/video/fbdev/vesafb.c > index e25e8de5ff67..929d4775cb4b 100644 > --- a/drivers/video/fbdev/vesafb.c > +++ b/drivers/video/fbdev/vesafb.c > @@ -490,11 +490,12 @@ static int vesafb_remove(struct platform_device *pdev) > { > struct fb_info *info = platform_get_drvdata(pdev); > > - /* vesafb_destroy takes care of info cleanup */ > - unregister_framebuffer(info); > if (((struct vesafb_par *)(info->par))->region) > release_region(0x3c0, 32); > > + /* vesafb_destroy takes care of info cleanup */ > + unregister_framebuffer(info); > + > return 0; > } >
On 5/26/22 21:47, Javier Martinez Canillas wrote: > Commit b3c9a924aab6 ("fbdev: vesafb: Cleanup fb_info in .fb_destroy rather > than .remove") fixed a use-after-free error due the vesafb driver freeing > the fb_info in the .remove handler instead of doing it in .fb_destroy. > > This can happen if the .fb_destroy callback is executed after the .remove > callback, since the former tries to access a pointer freed by the latter. > > But that change didn't take into account that another possible scenario is > that .fb_destroy is called before the .remove callback. For example, if no > process has the fbdev chardev opened by the time the driver is removed. > > If that's the case, fb_info will be freed when unregister_framebuffer() is > called, making the fb_info pointer accessed in vesafb_remove() after that > to no longer be valid. > > To prevent that, move the expression containing the info->par to happen > before the unregister_framebuffer() function call. > > Fixes: b3c9a924aab6 ("fbdev: vesafb: Cleanup fb_info in .fb_destroy rather than .remove") > Reported-by: Pascal Ernster <dri-devel@hardfalcon.net> > Signed-off-by: Javier Martinez Canillas <javierm@redhat.com> applied to the fbdev git tree. Thanks! Helge > --- > > drivers/video/fbdev/vesafb.c | 5 +++-- > 1 file changed, 3 insertions(+), 2 deletions(-) > > diff --git a/drivers/video/fbdev/vesafb.c b/drivers/video/fbdev/vesafb.c > index e25e8de5ff67..929d4775cb4b 100644 > --- a/drivers/video/fbdev/vesafb.c > +++ b/drivers/video/fbdev/vesafb.c > @@ -490,11 +490,12 @@ static int vesafb_remove(struct platform_device *pdev) > { > struct fb_info *info = platform_get_drvdata(pdev); > > - /* vesafb_destroy takes care of info cleanup */ > - unregister_framebuffer(info); > if (((struct vesafb_par *)(info->par))->region) > release_region(0x3c0, 32); > > + /* vesafb_destroy takes care of info cleanup */ > + unregister_framebuffer(info); > + > return 0; > } >
diff --git a/drivers/video/fbdev/vesafb.c b/drivers/video/fbdev/vesafb.c index e25e8de5ff67..929d4775cb4b 100644 --- a/drivers/video/fbdev/vesafb.c +++ b/drivers/video/fbdev/vesafb.c @@ -490,11 +490,12 @@ static int vesafb_remove(struct platform_device *pdev) { struct fb_info *info = platform_get_drvdata(pdev); - /* vesafb_destroy takes care of info cleanup */ - unregister_framebuffer(info); if (((struct vesafb_par *)(info->par))->region) release_region(0x3c0, 32); + /* vesafb_destroy takes care of info cleanup */ + unregister_framebuffer(info); + return 0; }
Commit b3c9a924aab6 ("fbdev: vesafb: Cleanup fb_info in .fb_destroy rather than .remove") fixed a use-after-free error due the vesafb driver freeing the fb_info in the .remove handler instead of doing it in .fb_destroy. This can happen if the .fb_destroy callback is executed after the .remove callback, since the former tries to access a pointer freed by the latter. But that change didn't take into account that another possible scenario is that .fb_destroy is called before the .remove callback. For example, if no process has the fbdev chardev opened by the time the driver is removed. If that's the case, fb_info will be freed when unregister_framebuffer() is called, making the fb_info pointer accessed in vesafb_remove() after that to no longer be valid. To prevent that, move the expression containing the info->par to happen before the unregister_framebuffer() function call. Fixes: b3c9a924aab6 ("fbdev: vesafb: Cleanup fb_info in .fb_destroy rather than .remove") Reported-by: Pascal Ernster <dri-devel@hardfalcon.net> Signed-off-by: Javier Martinez Canillas <javierm@redhat.com> --- drivers/video/fbdev/vesafb.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)