Skip to contentSkip to conversation

ch: Do not read past the end of vendor_labels[]

PATCH In review Bart Van Assche 2 participants 4 messages Jun 29, 2020 SCSI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ch: Do not read past the end of vendor_labels[]

This patch fixes the following gcc 10 compiler error:

In function 'memcpy',

    inlined from 'ch_ioctl' at drivers/scsi/ch.c:666:4:
./include/linux/string.h:377:4: error: call to '__read_overflow2' declared with attribute error: detected read beyond size of object passed as 2nd parameter
  377 |    __read_overflow2();
      |    ^~~~~~~~~~~~~~~~~~
Cc Hannes Reinecke <[email protected]>
Cc Arnd Bergmann <[email protected]>
---
 drivers/scsi/ch.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
drivers/scsi/ch.c +8 -4
index b81b397366db..b675a01380eb 100644
@@ -651,19 +651,23 @@ static long ch_ioctl(struct file *file,
 		memset(&vparams,0,sizeof(vparams));
 		if (ch->counts[CHET_V1]) {
 			vparams.cvp_n1  = ch->counts[CHET_V1];
-			memcpy(vparams.cvp_label1,vendor_labels[0],16);
+			strncpy(vparams.cvp_label1, vendor_labels[0],
+				ARRAY_SIZE(vparams.cvp_label1));
 		}
 		if (ch->counts[CHET_V2]) {
 			vparams.cvp_n2  = ch->counts[CHET_V2];
-			memcpy(vparams.cvp_label2,vendor_labels[1],16);
+			strncpy(vparams.cvp_label2, vendor_labels[1],
+				ARRAY_SIZE(vparams.cvp_label2));
 		}
 		if (ch->counts[CHET_V3]) {
 			vparams.cvp_n3  = ch->counts[CHET_V3];
-			memcpy(vparams.cvp_label3,vendor_labels[2],16);
+			strncpy(vparams.cvp_label3, vendor_labels[2],
+				ARRAY_SIZE(vparams.cvp_label3));
 		}
 		if (ch->counts[CHET_V4]) {
 			vparams.cvp_n4  = ch->counts[CHET_V4];
-			memcpy(vparams.cvp_label4,vendor_labels[3],16);
+			strncpy(vparams.cvp_label4, vendor_labels[3],
+				ARRAY_SIZE(vparams.cvp_label4));
 		}
 		if (copy_to_user(argp, &vparams, sizeof(vparams)))
 			return -EFAULT;
#1 lore raw
Quoted 21 lines · Bart Van Assche

This patch fixes the following gcc 10 compiler error:

In function 'memcpy',

    inlined from 'ch_ioctl' at drivers/scsi/ch.c:666:4:
./include/linux/string.h:377:4: error: call to '__read_overflow2' declared with attribute error: detected read beyond size of object passed as 2nd parameter
  377 |    __read_overflow2();
      |    ^~~~~~~~~~~~~~~~~~
Cc Hannes Reinecke <[email protected]>
Cc Arnd Bergmann <[email protected]>
---
 drivers/scsi/ch.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
drivers/scsi/ch.c +0 -0
index b81b397366db..b675a01380eb 100644

drivers/scsi/ch.c static long ch_ioctl(struct file *file, · line 656

Quoted 7 lines
drivers/scsi/ch.c +2 -1
@@ -651,19 +651,23 @@ static long ch_ioctl(struct file *file,
                memset(&vparams,0,sizeof(vparams));
                if (ch->counts[CHET_V1]) {
                        vparams.cvp_n1  = ch->counts[CHET_V1];
-                       memcpy(vparams.cvp_label1,vendor_labels[0],16);
+                       strncpy(vparams.cvp_label1, vendor_labels[0],
+                               ARRAY_SIZE(vparams.cvp_label1));
                }

Against which tree is this? I see in mainline the correct

      strncpy(vparams.cvp_label1,vendor_labels[0],16);

rather than the broken memcpy. If this was changed recently to the broken version, maybe send a revert, or add a "Fixes" tag?

Arnd

#2 lore raw
Quoted 16 lines · Arnd Bergmann

drivers/scsi/ch.c static long ch_ioctl(struct file *file, · line 656

Quoted 12 lines · Bart Van Assche
drivers/scsi/ch.c +2 -1
index b81b397366db..b675a01380eb 100644
@@ -651,19 +651,23 @@ static long ch_ioctl(struct file *file,
                memset(&vparams,0,sizeof(vparams));
                if (ch->counts[CHET_V1]) {
                        vparams.cvp_n1  = ch->counts[CHET_V1];
-                       memcpy(vparams.cvp_label1,vendor_labels[0],16);
+                       strncpy(vparams.cvp_label1, vendor_labels[0],
+                               ARRAY_SIZE(vparams.cvp_label1));
                }

Against which tree is this? I see in mainline the correct

Quoted 4 lines
      strncpy(vparams.cvp_label1,vendor_labels[0],16);

rather than the broken memcpy. If this was changed recently to the broken version, maybe send a revert, or add a "Fixes" tag?

Hi Arnd,

Thanks for having taken a look. This patch applies to Martin's for-next branch. The most recent ch patch I found in Linus' master branch is "ch:
remove ch_mutex()" from February 2020. I haven't found any more recent ch patches in the linux-next/master branch either. Have I perhaps been looking at the wrong repository or the wrong branch?

Thanks,

Bart.

#3 lore raw
Quoted 22 lines · Bart Van Assche
Quoted 16 lines · Arnd Bergmann

drivers/scsi/ch.c static long ch_ioctl(struct file *file, · line 656

Quoted 12 lines · Bart Van Assche
drivers/scsi/ch.c +2 -1
index b81b397366db..b675a01380eb 100644
@@ -651,19 +651,23 @@ static long ch_ioctl(struct file *file,
                memset(&vparams,0,sizeof(vparams));
                if (ch->counts[CHET_V1]) {
                        vparams.cvp_n1  = ch->counts[CHET_V1];
-                       memcpy(vparams.cvp_label1,vendor_labels[0],16);
+                       strncpy(vparams.cvp_label1, vendor_labels[0],
+                               ARRAY_SIZE(vparams.cvp_label1));
                }

Against which tree is this? I see in mainline the correct

Quoted 4 lines
      strncpy(vparams.cvp_label1,vendor_labels[0],16);

rather than the broken memcpy. If this was changed recently to the broken version, maybe send a revert, or add a "Fixes" tag?

Quoted 7 lines

Hi Arnd,

Thanks for having taken a look. This patch applies to Martin's for-next branch. The most recent ch patch I found in Linus' master branch is "ch:
remove ch_mutex()" from February 2020. I haven't found any more recent ch patches in the linux-next/master branch either. Have I perhaps been looking at the wrong repository or the wrong branch?

That is the right branch, and I don't see any later changes to the file after Feb 2020 in there or in mainline either, but I also clearly see it using strncpy(). See also:

https://elixir.bootlin.com/linux/v5.8-rc3/source/drivers/scsi/ch.c#L647

I think there were some patches under discussion about replacing a lot of strncpy() calls with the more intuitive memcpy(), strnlcpy() or strscpy() alternatives, but in this case strncpy() in in fact the correct one (as you also concluded) and I don't see any patches to this file getting applied to that effect.

Arnd