On 07/05/25 05:08, Jan Kara wrote:
On Wed 07-05-25 07:56:21, Amir Goldstein wrote:
On Wed, May 7, 2025 at 1:39 AM Gustavo A. R. Silva
<gustavoars@xxxxxxxxxx> wrote:
-Wflex-array-member-not-at-end was introduced in GCC-14, and we are
getting ready to enable it, globally.
Modify FANOTIFY_INLINE_FH() macro, which defines a struct containing a
flexible-array member in the middle (struct fanotify_fh::buf), to use
struct_size_t() to pre-allocate space for both struct fanotify_fh and
its flexible-array member. Replace the struct with a union and relocate
the flexible structure (struct fanotify_fh) to the end.
See the memory layout of struct fanotify_fid_event before and after
changes below.
pahole -C fanotify_fid_event fs/notify/fanotify/fanotify.o
BEFORE:
struct fanotify_fid_event {
struct fanotify_event fae; /* 0 48 */
__kernel_fsid_t fsid; /* 48 8 */
struct {
struct fanotify_fh object_fh; /* 56 4 */
unsigned char _inline_fh_buf[12]; /* 60 12 */
}; /* 56 16 */
/* size: 72, cachelines: 2, members: 3 */
/* last cacheline: 8 bytes */
};
AFTER:
struct fanotify_fid_event {
struct fanotify_event fae; /* 0 48 */
__kernel_fsid_t fsid; /* 48 8 */
union {
unsigned char _inline_fh_buf[16]; /* 56 16 */
struct fanotify_fh object_fh __attribute__((__aligned__(1))); /* 56 4 */
I'm not that familiar with pahole, but I find it surprising to see this member
aligned(1), when struct fanotify_fh is defined as __aligned(4).
Yeah.
Yep, gotcha.
} __attribute__((__aligned__(1))); /* 56 16 */
/* size: 72, cachelines: 2, members: 3 */
/* forced alignments: 1 */
/* last cacheline: 8 bytes */
} __attribute__((__aligned__(8)));
So, with these changes, fix the following warnings:
fs/notify/fanotify/fanotify.h:317:28: warning: structure containing a flexible array member is not at the end of another structure [-Wflex-array-member-not-at-end]
fs/notify/fanotify/fanotify.h:289:28: warning: structure containing a flexible array member is not at the end of another structure [-Wflex-array-member-not-at-end]
Signed-off-by: Gustavo A. R. Silva <gustavoars@xxxxxxxxxx>
---
fs/notify/fanotify/fanotify.h | 12 ++++++------
1 file changed, 6 insertions(+), 6 deletions(-)
diff --git a/fs/notify/fanotify/fanotify.h b/fs/notify/fanotify/fanotify.h
index b44e70e44be6..91c26b1c1d32 100644
--- a/fs/notify/fanotify/fanotify.h
+++ b/fs/notify/fanotify/fanotify.h
@@ -275,12 +275,12 @@ static inline void fanotify_init_event(struct fanotify_event *event,
event->pid = NULL;
}
-#define FANOTIFY_INLINE_FH(name, size) \
-struct { \
- struct fanotify_fh name; \
- /* Space for object_fh.buf[] - access with fanotify_fh_buf() */ \
- unsigned char _inline_fh_buf[size]; \
-}
+#define FANOTIFY_INLINE_FH(name, size) \
+union { \
+ /* Space for object_fh and object_fh.buf[] - access with fanotify_fh_buf() */ \
+ unsigned char _inline_fh_buf[struct_size_t(struct fanotify_fh, buf, size)]; \
The name _inline_fh_buf is confusing in this setting
better use bytes[] as in DEFINE_FLEX() or maybe even consider
a generic helper DEFINE_FLEX_MEMBER() to use instead of
FANOTIFY_INLINE_FH(), because this is not fanotify specific,
except maybe for alignment (see below).
Yes, I guess a generic helper for this would be nice but if fanotify is the
only place that plays these tricks, we can keep it specific for now. I
agree naming the "space-buffer" field "bytes" would be less confusing.
I can send v2 with this change.
+ struct fanotify_fh name; \
+} __packed
Why added __packed?
The fact that struct fanotify_fh is 4 bytes aligned could end up with less
bytes reserved for the inline buffer if the union is not also 4 bytes aligned.
So maybe something like this:
#define FANOTIFY_INLINE_FH(name, size) \
DEFINE_FLEX_MEMBER(struct fanotify_fh, name, size) __aligned(4)
I guess you need to provide the "member" information to
DEFINE_FLEX_MEMBER() somewhere as well.
Yeah, I can write something like that as well.
I'll come back with v2, shortly.
Thanks for the feedback, folks! :)
--
Gustavo
