Skip to contentSkip to conversation

vt/fbcon: deinitialize resources in visual_init() after failed memory allocation

PATCH In review Grzegorz Halat 3 participants 6 messages Apr 26, 2019 – May 24, 2019

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 vt/fbcon: deinitialize resources in visual_init() after failed memory allocation

After memory allocation failure vc_allocate() doesn't clean up data which has been initialized in visual_init(). In case of fbcon this leads to divide-by-0 in fbcon_init() on next open of the same tty.

memory allocation in vc_allocate() may fail here:
1097: vc->vc_screenbuf = kzalloc(vc->vc_screenbuf_size, GFP_KERNEL);

on next open() fbcon_init() skips vc_font.data initialization:
1088: if (!p->fontdata) {

division by zero in fbcon_init() happens here:
1149: new_cols /= vc->vc_font.width;

Additional check is needed in fbcon_deinit() to prevent usage of uninitialized vc_screenbuf:

1251: if (vc->vc_hi_font_mask && vc->vc_screenbuf)
1252: set_vc_hi_font(vc, false);

Crash:

#6 [ffffc90001eafa60] divide_error at ffffffff81a00be4

    [exception RIP: fbcon_init+463]
    RIP: ffffffff814b860f  RSP: ffffc90001eafb18  RFLAGS: 00010246

...
#7 [ffffc90001eafb60] visual_init at ffffffff8154c36e #8 [ffffc90001eafb80] vc_allocate at ffffffff8154f53c #9 [ffffc90001eafbc8] con_install at ffffffff8154f624
...

---
 drivers/tty/vt/vt.c              | 11 +++++++++--
 drivers/video/fbdev/core/fbcon.c |  2 +-
 2 files changed, 10 insertions(+), 3 deletions(-)
drivers/tty/vt/vt.c +9 -2
index 650c66886c80..ec85d195678f 100644
@@ -1056,6 +1056,13 @@ static void visual_init(struct vc_data *vc, int num, int init)
 	vc->vc_screenbuf_size = vc->vc_rows * vc->vc_size_row;
 }
 
+
+static void visual_deinit(struct vc_data *vc)
+{
+	vc->vc_sw->con_deinit(vc);
+	module_put(vc->vc_sw->owner);
+}
+
 int vc_allocate(unsigned int currcons)	/* return 0 on success */
 {
 	struct vt_notifier_param param;
@@ -1103,6 +1110,7 @@ int vc_allocate(unsigned int currcons)	/* return 0 on success */
 
 	return 0;
 err_free:
+	visual_deinit(vc);
 	kfree(vc);
 	vc_cons[currcons].d = NULL;
 	return -ENOMEM;
@@ -1331,9 +1339,8 @@ struct vc_data *vc_deallocate(unsigned int currcons)
 		param.vc = vc = vc_cons[currcons].d;
 		atomic_notifier_call_chain(&vt_notifier_list, VT_DEALLOCATE, &param);
 		vcs_remove_sysfs(currcons);
-		vc->vc_sw->con_deinit(vc);
+		visual_deinit(vc);
 		put_pid(vc->vt_pid);
-		module_put(vc->vc_sw->owner);
 		vc_uniscr_set(vc, NULL);
 		kfree(vc->vc_screenbuf);
 		vc_cons[currcons].d = NULL;
drivers/video/fbdev/core/fbcon.c +1 -1
index cd059a801662..c59b23f6e9ba 100644
@@ -1248,7 +1248,7 @@ static void fbcon_deinit(struct vc_data *vc)
 	if (free_font)
 		vc->vc_font.data = NULL;
 
-	if (vc->vc_hi_font_mask)
+	if (vc->vc_hi_font_mask && vc->vc_screenbuf)
 		set_vc_hi_font(vc, false);
 
 	if (!con_is_bound(&fb_con))
Signature · 1 line
2.20.1
#1 lore raw
Quoted 4 lines · Grzegorz Halat

After memory allocation failure vc_allocate() doesn't clean up data which has been initialized in visual_init(). In case of fbcon this leads to divide-by-0 in fbcon_init() on next open of the same tty.

Hi,
A gentle reminder. Could you please review my patch? I've seen two crashes caused by this bug.

--
Grzegorz Halat

#2 lore raw
Quoted 9 lines · Grzegorz Halat
Quoted 4 lines · Grzegorz Halat

After memory allocation failure vc_allocate() doesn't clean up data which has been initialized in visual_init(). In case of fbcon this leads to divide-by-0 in fbcon_init() on next open of the same tty.

Hi,
A gentle reminder. Could you please review my patch? I've seen two crashes caused by this bug.

How? How are you triggering a memory allocation failure in a "normal" system?

Anyway, I'll queue this up, but it really does not seem like anything anyone would see "in the wild".

thanks,

greg k-h

#3 lore raw
Quoted 4 lines · Greg Kroah-Hartman

How? How are you triggering a memory allocation failure in a "normal" system?
Anyway, I'll queue this up, but it really does not seem like anything anyone would see "in the wild"

I've seen this crash twice in ours customer environment under low memory conditions.
There is a report in Debian bug tracker:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug€4443
and LKML bug report:
https://lkml.org/lkml/2017/12/18/591

--
Grzegorz

#4 lore raw
Quoted 7 lines · Grzegorz Halat
Quoted 4 lines · Greg Kroah-Hartman

How? How are you triggering a memory allocation failure in a "normal" system?
Anyway, I'll queue this up, but it really does not seem like anything anyone would see "in the wild"

Quoted 6 lines

I've seen this crash twice in ours customer environment under low memory conditions.
There is a report in Debian bug tracker:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug€4443
and LKML bug report:
https://lkml.org/lkml/2017/12/18/591

Ok, now queued up to go to Linus for 5.2-final.

thanks,

greg k-h

Acked-by on v1 1/1
#5 lore raw
In reply to Grzegorz Halat
Quoted 24 lines · Grzegorz Halat

After memory allocation failure vc_allocate() doesn't clean up data which has been initialized in visual_init(). In case of fbcon this leads to divide-by-0 in fbcon_init() on next open of the same tty.

memory allocation in vc_allocate() may fail here:
1097: vc->vc_screenbuf = kzalloc(vc->vc_screenbuf_size, GFP_KERNEL);

on next open() fbcon_init() skips vc_font.data initialization:
1088: if (!p->fontdata) {

division by zero in fbcon_init() happens here:
1149: new_cols /= vc->vc_font.width;

Additional check is needed in fbcon_deinit() to prevent usage of uninitialized vc_screenbuf:

1251: if (vc->vc_hi_font_mask && vc->vc_screenbuf)
1252: set_vc_hi_font(vc, false);

Crash:

#6 [ffffc90001eafa60] divide_error at ffffffff81a00be4

    [exception RIP: fbcon_init+463]
    RIP: ffffffff814b860f  RSP: ffffc90001eafb18  RFLAGS: 00010246
Quoted 7 lines

...
#7 [ffffc90001eafb60] visual_init at ffffffff8154c36e #8 [ffffc90001eafb80] vc_allocate at ffffffff8154f53c #9 [ffffc90001eafbc8] con_install at ffffffff8154f624
...

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics