LU-16510 build: include unsafe_memcpy definition

The original LU-16510 missed a key part of the
unsafe_memcpy code from the upstream kernel, and so we
weren't actually defining unsafe_memcpy() as intended.

Thanks to Aurelien Degremont <adegremont@nvidia.com> for
pointing this out.

Lustre-change: https://review.whamcloud.com/50573
Lustre-commit: 565b21bf65e385a9b4fd8ee31cabe7892345b783

Fixes: 919b93b9 ("LU-16510 build: fortified memcpy from linux 6.1")
Signed-off-by: Patrick Farrell <pfarrell@whamcloud.com>
Change-Id: Ib9e2d56ed0b3691f1ab9fcd25403fa86ac784b6d
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Aurelien Degremont <adegremont@nvidia.com>
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/50667
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/libcfs/include/libcfs/linux/linux-fortify-string.h b/libcfs/include/libcfs/linux/linux-fortify-string.h
index bf03e0e..aeed8c5 100644 (file)
--- a/libcfs/include/libcfs/linux/linux-fortify-string.h
+++ b/libcfs/include/libcfs/linux/linux-fortify-string.h
@@ -80,6 +80,22 @@ extern char *__underlying_strncpy(char *p, const char *q, __kernel_size_t size)
 #define __underlying_strncpy   __builtin_strncpy
 #endif
 
+/**
+ * unsafe_memcpy - memcpy implementation with no FORTIFY bounds checking
+ *
+ * @dst: Destination memory address to write to
+ * @src: Source memory address to read from
+ * @bytes: How many bytes to write to @dst from @src
+ * @justification: Free-form text or comment describing why the use is needed
+ *
+ * This should be used for corner cases where the compiler cannot do the
+ * right thing, or during transitions between APIs, etc. It should be used
+ * very rarely, and includes a place for justification detailing where bounds
+ * checking has happened, and why existing solutions cannot be employed.
+ */
+#define unsafe_memcpy(dst, src, bytes, justification)          \
+       __underlying_memcpy(dst, src, bytes)
+
 /*
  * Clang's use of __builtin_*object_size() within inlines needs hinting via
  * __pass_*object_size(). The preference is to only ever use type 1 (member