sys_sync_file_range()

Hello Andrew Morton,

The patch f79e2abb9bd4: "[PATCH] sys_sync_file_range()" from Mar 31, 2006, leads to the following static checker warning:

	fs/sync.c:298 ksys_sync_file_range()
	warn: signed overflow undefined. 'offset + nbytes < offset'

fs/sync.c
279 int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes,
280 unsigned int flags)
281 {

   282          int ret;
   283          struct fd f;
   284          struct address_space *mapping;
   285          loff_t endbyte;                 /* inclusive */
   286          umode_t i_mode;
   287  
   288          ret = -EINVAL;
   289          if (flags & ~VALID_FLAGS)
   290                  goto out;
   291  
   292          endbyte = offset + nbytes;
   293  
   294          if ((s64)offset < 0)
   295                  goto out;
   296          if ((s64)endbyte < 0)
   297                  goto out;
   298          if (endbyte < offset)
                    ^^^^^^^^^^^^^^^^

This is undefined but I think it works with the kernel build options?

   299                  goto out;
   300  
   301          if (sizeof(pgoff_t) == 4) {
   302                  if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
   303                          /*
   304                           * The range starts outside a 32 bit machine's
   305                           * pagecache addressing capabilities.  Let it "succeed"
   306                           */
   307                          ret = 0;
   308                          goto out;
   309                  }
   310                  if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
   311                          /*
   312                           * Out to EOF
   313                           */
   314                          nbytes = 0;
   315                  }
   316          }

regards,
dan carpenter

Yup. -fno-strict-overflow. Please see Dmitry's explanation at
https://bugzilla.kernel.org/show_bug.cgi?id=199099.