在 2019/9/25 21:08, Eryu Guan 写道:
On Wed, Sep 25, 2019 at 03:16:55PM +0800, Su Yanjun wrote:
On some linux distros(RHEL7, centos 7) copy_file_range uses
general implementation (splice interface). splice interace
uses pipe_to_file. pipe_to_file only work for different page.
The userspace cant's be aware of such error because copy_file_range
returns ok too.
This looks like a kernel bug to me.
Yes, it's RHEL kernel 3.10's bug. The new kernel is ok.
So for such case when copy_file_range return we read back data
then check it.
Yeah, I think we should make sure the data copied is correct.
Signed-off-by: Su Yanjun <suyj.fnst@xxxxxxxxxxxxxx>
---
ltp/fsx.c | 32 ++++++++++++++++++++++++++++++++
1 file changed, 32 insertions(+)
diff --git a/ltp/fsx.c b/ltp/fsx.c
index 06d08e4..0439430 100644
--- a/ltp/fsx.c
+++ b/ltp/fsx.c
@@ -1602,6 +1602,7 @@ do_copy_range(unsigned offset, unsigned length, unsigned dest)
size_t olen;
ssize_t nr;
int tries = 0;
+ int ret = 0;
if (length == 0) {
if (!quiet && testcalls > simulatedopcount)
@@ -1665,6 +1666,37 @@ do_copy_range(unsigned offset, unsigned length, unsigned dest)
memset(good_buf + file_size, '\0', dest - file_size);
if (dest + length > file_size)
file_size = dest + length;
+ /*
+ * Although copy_file_range returns ok
+ * for some linux distros that use general implementation
+ * (splice interface) such as RHEL 7, centos 7 that use
+ * pipe_to_file will cause test fail.
+ *
+ * Here we add a little more check here for copy_file_range
+ * We read copied data then check it. If check fail here
+ * then report it.
+ */
Above comments seem like not necessary, the content check is inspired by
Rthis HEL7 issue, but not the main purpose.
+ ret = lseek(fd, (off_t)dest, SEEK_SET);
+ if (ret == (off_t)-1) {
+ prterr("doread: lseek");
^^^^^^^ not doread
+ report_failure(140);
Use a different failure number than that in doread, i.e. don't copy the
code from doread.
+ }
+ ret = fsxread(fd, temp_buf, length, dest);
+ if (ret != length) {
+ if (ret == -1)
+ prterr("doread: read");
+ else
+ prt("short read: 0x%x bytes instead of 0x%x\n",
+ ret, length);
Same here, not doread, and weired indention above.
+ report_failure(141);
+ }
+ if (memcmp(good_buf+dest, temp_buf, length) != 0) {
+ prt("copy range: 0x%x to 0x%x at 0x%x\n", offset,
+ offset + length, dest);
+ prterr("do_copy_range:");
+ report_failure(161);
I think we could take use of check_buffers().
And similar check could be added to do_clone_range as well, I think.
Ok, i'll send patch v2 later.
Thanks
Thanks,
Eryu
+
+ }
}
#else
--
2.7.4
