Skip to contentSkip to conversation

: GeForce4 440 Go 64M overflows fb_fix_screeninfo.id

2.6.2-rc2, rivafb Guido Guenther 3 participants 4 messages Jan 29, 2004

lore
Share & tools
lore
#0 lore raw

Hi,
fb_fix_screeninfo has space for 15 characters but riva/fbdev.c tries to copy more into it in case of the above card (and therefore corrupts memory). The result looks like:
rivafb: PCI nVidia NV20 framebuffer ver 0.9.5b (nVidiaGeForce4-4\224, 32MB @ 0x94000000)

                                                               ^^^^^

Possible fix attached. This also overwrites the initial "nVidia" in rivafb_fix.id making the output the same as in 2.4 (and using strlcpy makes sure we don't overflow again). With this patch:

rivafb: PCI nVidia NV20 framebuffer ver 0.9.5b (GeForce4-440-GO-M64, 32MB @ 0x94000000)

Can this go in?
-- Guido

linux/fb.h renamed from benh-rsync-2.6-clean/include/linux/fb.h +1 -1
@@ -114,7 +114,7 @@
 
 
 struct fb_fix_screeninfo {
-	char id[16];			/* identification string eg "TT Builtin" */
+	char id[32];			/* identification string eg "TT Builtin" */
 	unsigned long smem_start;	/* Start of frame buffer mem */
 					/* (physical address) */
 	__u32 smem_len;			/* Length of frame buffer mem */
video/riva/fbdev.c renamed from benh-rsync-2.6-clean/drivers/video/riva/fbdev.c +1 -1
@@ -1867,7 +1867,7 @@
 		goto err_out_kfree1;
 	memset(info->pixmap.addr, 0, 64 * 1024);
 
-	strcat(rivafb_fix.id, rci->name);
+	strlcpy(rivafb_fix.id, rci->name, sizeof(rivafb_fix.id));
 	default_par->riva.Architecture = rci->arch_rev;
 
 	default_par->Chipset = (pd->vendor << 16) | pd->device;
#1 lore raw
Quoted 5 lines · Guido Guenther

Hi,
fb_fix_screeninfo has space for 15 characters but riva/fbdev.c tries to copy more into it in case of the above card (and therefore corrupts memory). The result looks like:
rivafb: PCI nVidia NV20 framebuffer ver 0.9.5b (nVidiaGeForce4-4\224, 32MB @ 0x94000000)

Quoted 8 lines
                                                               ^^^^^

Possible fix attached. This also overwrites the initial "nVidia" in rivafb_fix.id making the output the same as in 2.4 (and using strlcpy makes sure we don't overflow again). With this patch:

rivafb: PCI nVidia NV20 framebuffer ver 0.9.5b (GeForce4-440-GO-M64, 32MB @ 0x94000000)

Can this go in?

No way. Second part (riva/fbdev.c) is OK, but first part is wrong. fb_fix_screeninfo is part of ABI, and as such cannot be changed. No fb application is going to work on your system - try 'fbset -i' (unless you rebuilt fbset binary after doing this change).

You'll have to live with 'GeForce4-440-GO' name.
Petr Vandrovec

Quoted 24 lines

-- Guido

linux/fb.h renamed from benh-rsync-2.6-clean/include/linux/fb.h +1 -1
@@ -114,7 +114,7 @@
 
 
 struct fb_fix_screeninfo {
-	char id[16];			/* identification string eg "TT Builtin" */
+	char id[32];			/* identification string eg "TT Builtin" */
 	unsigned long smem_start;	/* Start of frame buffer mem */
 					/* (physical address) */
 	__u32 smem_len;			/* Length of frame buffer mem */
video/riva/fbdev.c renamed from benh-rsync-2.6-clean/drivers/video/riva/fbdev.c +1 -1
@@ -1867,7 +1867,7 @@
 		goto err_out_kfree1;
 	memset(info->pixmap.addr, 0, 64 * 1024);
 
-	strcat(rivafb_fix.id, rci->name);
+	strlcpy(rivafb_fix.id, rci->name, sizeof(rivafb_fix.id));
 	default_par->riva.Architecture = rci->arch_rev;
 
 	default_par->Chipset = (pd->vendor << 16) | pd->device;
#2 lore raw
Quoted 6 lines

No way. Second part (riva/fbdev.c) is OK, but first part is wrong. fb_fix_screeninfo is part of ABI, and as such cannot be changed. No fb application is going to work on your system - try 'fbset -i' (unless you rebuilt fbset binary after doing this change).

You'll have to live with 'GeForce4-440-GO' name.

I applied the fbdev.c part but Petr is right. It would break userland.

#3 lore raw
Quoted 1 line · James Simmons

I applied the fbdev.c part but Petr is right. It would break userland.

That's fine. Thanks,
-- Guido