nvram: Fix write beyond end condition; prove to gcc copy is safe

In nvram_write, first of all, correctly handle the case where the file
pointer is already beyond the end; we should return EOF in that case.

Second, make the logic a bit more explicit so that gcc can statically
prove that the copy_from_user() is safe.  Once the condition of the
beyond-end filepointer is eliminated, the copy is safe but gcc can't
prove it, causing build failures for i386 allyesconfig.

Third, eliminate the entirely superfluous variable "len", and just use
the passed-in variable "count" instead.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Cc: Arjan van de Ven <arjan@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Wim Van Sebroeck <wim@iguana.be>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <tip-*@git.kernel.org>

1 files changed, 10 insertions, 4 deletions

diff --git a/drivers/char/nvram.c b/drivers/char/nvram.c
index 4008e2c..fdbcc9f 100644
--- a/drivers/char/nvram.c
+++ b/drivers/char/nvram.c
@@ -264,10 +264,16 @@ static ssize_t nvram_write(struct file *file, const char __user *buf,
 	unsigned char contents[NVRAM_BYTES];
 	unsigned i = *ppos;
 	unsigned char *tmp;
-	int len;
 
-	len = (NVRAM_BYTES - i) < count ? (NVRAM_BYTES - i) : count;
-	if (copy_from_user(contents, buf, len))
+	if (i >= NVRAM_BYTES)
+		return 0;	/* Past EOF */
+
+	if (count > NVRAM_BYTES - i)
+		count = NVRAM_BYTES - i;
+	if (count > NVRAM_BYTES)
+		return -EFAULT;	/* Can't happen, but prove it to gcc */
+
+	if (copy_from_user(contents, buf, count))
 		return -EFAULT;
 
 	spin_lock_irq(&rtc_lock);
@@ -275,7 +281,7 @@ static ssize_t nvram_write(struct file *file, const char __user *buf,
 	if (!__nvram_check_checksum())
 		goto checksum_err;
 
-	for (tmp = contents; count-- > 0 && i < NVRAM_BYTES; ++i, ++tmp)
+	for (tmp = contents; count--; ++i, ++tmp)
 		__nvram_write_byte(*tmp, i);
 
 	__nvram_set_checksum();