Skip to contentSkip to conversation

bpf: pack struct bpf_fib_lookup

PATCH Applied Anton Protopopov 6 participants 11 messages Apr 3, 2024 – Apr 4, 2024 BPF

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 bpf: pack struct bpf_fib_lookup

The struct bpf_fib_lookup is supposed to be of size 64. A recent commit 59b418c7063d ("bpf: Add a check for struct bpf_fib_lookup size") added a static assertion to check this property so that future changes to the structure will not accidentally break this assumption.

As it immediately turned out, on some 32-bit arm systems, when AEABI=n, the total size of the structure was equal to 68, see [1]. This happened because the bpf_fib_lookup structure contains a union of two 16-bit fields:

    union {
            __u16 tot_len;
            __u16 mtu_result;
    };

which was supposed to compile to a 16-bit-aligned 16-bit field. On the aforementioned setups it was instead both aligned and padded to 32-bits.

Declare this inner union as __attribute__((packed, aligned(2))) such that it always is of size 2 and is aligned to 16 bits.

[1] https://lore.kernel.org/all/CA+G9fYtsoP51f-oP_Sp5MOq-Ffv8La2RztNpwvE6+R1VtFiLrw@mail.gmail.com/#t

---
 include/uapi/linux/bpf.h       | 2 +-
 tools/include/uapi/linux/bpf.h | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
include/uapi/linux/bpf.h +1 -1
index 79c548276b6b..6fe9f11c8abe 100644
@@ -7157,7 +7157,7 @@ struct bpf_fib_lookup {
 
 		/* output: MTU value */
 		__u16	mtu_result;
-	};
+	} __attribute__((packed, aligned(2)));
 	/* input: L3 device index for lookup
 	 * output: device index from FIB lookup
 	 */
tools/include/uapi/linux/bpf.h +1 -1
index 79c548276b6b..6fe9f11c8abe 100644
@@ -7157,7 +7157,7 @@ struct bpf_fib_lookup {
 
 		/* output: MTU value */
 		__u16	mtu_result;
-	};
+	} __attribute__((packed, aligned(2)));
 	/* input: L3 device index for lookup
 	 * output: device index from FIB lookup
 	 */
Signature · 1 line
2.34.1
Reviewed-by on v1 1/1
#1 lore raw

From: Anton Protopopov <[email protected]>
Date: Wed, 3 Apr 2024 14:33:03 +0200

Quoted 18 lines

The struct bpf_fib_lookup is supposed to be of size 64. A recent commit 59b418c7063d ("bpf: Add a check for struct bpf_fib_lookup size") added a static assertion to check this property so that future changes to the structure will not accidentally break this assumption.

As it immediately turned out, on some 32-bit arm systems, when AEABI=n, the total size of the structure was equal to 68, see [1]. This happened because the bpf_fib_lookup structure contains a union of two 16-bit fields:

    union {
            __u16 tot_len;
            __u16 mtu_result;
    };

which was supposed to compile to a 16-bit-aligned 16-bit field. On the aforementioned setups it was instead both aligned and padded to 32-bits.

Quoted 6 lines

Declare this inner union as __attribute__((packed, aligned(2))) such that it always is of size 2 and is aligned to 16 bits.

[1] https://lore.kernel.org/all/CA+G9fYtsoP51f-oP_Sp5MOq-Ffv8La2RztNpwvE6+R1VtFiLrw@mail.gmail.com/#t

Quoted 23 lines
---
 include/uapi/linux/bpf.h       | 2 +-
 tools/include/uapi/linux/bpf.h | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
include/uapi/linux/bpf.h +1 -1
index 79c548276b6b..6fe9f11c8abe 100644
@@ -7157,7 +7157,7 @@ struct bpf_fib_lookup {
 
 		/* output: MTU value */
 		__u16	mtu_result;
-	};
+	} __attribute__((packed, aligned(2)));
 	/* input: L3 device index for lookup
 	 * output: device index from FIB lookup
 	 */
tools/include/uapi/linux/bpf.h +0 -0
index 79c548276b6b..6fe9f11c8abe 100644

tools/include/uapi/linux/bpf.h struct bpf_fib_lookup { · line 7163

Quoted 8 lines
tools/include/uapi/linux/bpf.h +1 -1
@@ -7157,7 +7157,7 @@ struct bpf_fib_lookup {
 
 		/* output: MTU value */
 		__u16	mtu_result;
-	};
+	} __attribute__((packed, aligned(2)));
 	/* input: L3 device index for lookup
 	 * output: device index from FIB lookup
 	 */

Thanks,
Olek

maintainer Acked-by on v1 1/1
#2 lore raw
In reply to Anton Protopopov
Quoted 18 lines · Anton Protopopov

The struct bpf_fib_lookup is supposed to be of size 64. A recent commit 59b418c7063d ("bpf: Add a check for struct bpf_fib_lookup size") added a static assertion to check this property so that future changes to the structure will not accidentally break this assumption.

As it immediately turned out, on some 32-bit arm systems, when AEABI=n, the total size of the structure was equal to 68, see [1]. This happened because the bpf_fib_lookup structure contains a union of two 16-bit fields:

     union {
             __u16 tot_len;
             __u16 mtu_result;
     };

which was supposed to compile to a 16-bit-aligned 16-bit field. On the aforementioned setups it was instead both aligned and padded to 32-bits.

Quoted 6 lines

Declare this inner union as __attribute__((packed, aligned(2))) such that it always is of size 2 and is aligned to 16 bits.

   [1] https://lore.kernel.org/all/CA+G9fYtsoP51f-oP_Sp5MOq-Ffv8La2RztNpwvE6+R1VtFiLrw@mail.gmail.com/#t
#3 lore raw
Quoted 23 lines · Daniel Borkmann
Quoted 21 lines · Anton Protopopov

The struct bpf_fib_lookup is supposed to be of size 64. A recent commit 59b418c7063d ("bpf: Add a check for struct bpf_fib_lookup size") added a static assertion to check this property so that future changes to the structure will not accidentally break this assumption.

As it immediately turned out, on some 32-bit arm systems, when AEABI=n, the total size of the structure was equal to 68, see [1]. This happened because the bpf_fib_lookup structure contains a union of two 16-bit fields:

     union {
             __u16 tot_len;
             __u16 mtu_result;
     };

which was supposed to compile to a 16-bit-aligned 16-bit field. On the aforementioned setups it was instead both aligned and padded to 32-bits.

Declare this inner union as __attribute__((packed, aligned(2))) such that it always is of size 2 and is aligned to 16 bits.

Quoted 6 lines
Quoted 3 lines
   [1] https://lore.kernel.org/all/CA+G9fYtsoP51f-oP_Sp5MOq-Ffv8La2RztNpwvE6+R1VtFiLrw@mail.gmail.com/#t

Let's continue the tag fest :) Also fixes tag:

Fixes e1850ea9bd9e ("bpf: bpf_fib_lookup return MTU value as output

when looked up")

Should this also go through the bpf tree instead of bpf-next?

#4 lore raw
In reply to Anton Protopopov
Quoted 10 lines · Anton Protopopov

The struct bpf_fib_lookup is supposed to be of size 64. A recent commit 59b418c7063d ("bpf: Add a check for struct bpf_fib_lookup size") added a static assertion to check this property so that future changes to the structure will not accidentally break this assumption.

As it immediately turned out, on some 32-bit arm systems, when AEABI=n, the total size of the structure was equal to 68, see [1]. This happened because the bpf_fib_lookup structure contains a union of two 16-bit fields:

Quoted 4 lines
    union {
            __u16 tot_len;
            __u16 mtu_result;
    };

This union was introduced three years ago, so fixing it now means another incompatible ABI change. While you clearly change it back to what it should have been the entire time, it's not obvious that this is the correct thing to do after three years.

I do wonder to what degree we still want to care about OABI at all. As I mentioned, I stopped testing it myself because it is no longer relevant to most people, and there are probably a number of other ABI issues.

Quoted 5 lines

which was supposed to compile to a 16-bit-aligned 16-bit field. On the aforementioned setups it was instead both aligned and padded to 32-bits.

Declare this inner union as __attribute__((packed, aligned(2))) such that it always is of size 2 and is aligned to 16 bits.

I think you probably want 32-bit alignment for the structure, to keep the ABI unchanged on all other architectures.

Arnd

maintainer
#5 lore raw
Quoted 26 lines · Arnd Bergmann
Quoted 10 lines · Anton Protopopov

The struct bpf_fib_lookup is supposed to be of size 64. A recent commit 59b418c7063d ("bpf: Add a check for struct bpf_fib_lookup size") added a static assertion to check this property so that future changes to the structure will not accidentally break this assumption.

As it immediately turned out, on some 32-bit arm systems, when AEABI=n, the total size of the structure was equal to 68, see [1]. This happened because the bpf_fib_lookup structure contains a union of two 16-bit fields:

Quoted 4 lines
     union {
             __u16 tot_len;
             __u16 mtu_result;
     };

This union was introduced three years ago, so fixing it now means another incompatible ABI change. While you clearly change it back to what it should have been the entire time, it's not obvious that this is the correct thing to do after three years.

I do wonder to what degree we still want to care about OABI at all. As I mentioned, I stopped testing it myself because it is no longer relevant to most people, and there are probably a number of other ABI issues.

Quoted 8 lines
Quoted 5 lines

which was supposed to compile to a 16-bit-aligned 16-bit field. On the aforementioned setups it was instead both aligned and padded to 32-bits.

Declare this inner union as __attribute__((packed, aligned(2))) such that it always is of size 2 and is aligned to 16 bits.

I think you probably want 32-bit alignment for the structure, to keep the ABI unchanged on all other architectures.

Fwiw, on x86 nothing should change on this regard, see below pahole dump before/after. I think similar might be true for other archs as otherwise we should have seen a kbuild bot complaint on hitting the size assert.

So it seems at this point only for the reported case of AEABI=n ... if it is no longer relevant to most people, I'd think the likelihood of such users and users who utilize this specific helper is even smaller. But we could queue it to bpf-next to give it some time.

Before patch:

struct bpf_fib_lookup {
        __u8                       family;               /*     0     1 */
        __u8                       l4_protocol;          /*     1     1 */
        __be16                     sport;                /*     2     2 */
        __be16                     dport;                /*     4     2 */
        union {
                __u16              tot_len;              /*     6     2 */
                __u16              mtu_result;           /*     6     2 */
        };                                               /*     6     2 */
        __u32                      ifindex;              /*     8     4 */
        union {
                __u8               tos;                  /*    12     1 */
                __be32             flowinfo;             /*    12     4 */
                __u32              rt_metric;            /*    12     4 */
        };                                               /*    12     4 */
        union {
                __be32             ipv4_src;             /*    16     4 */
                __u32              ipv6_src[4];          /*    16    16 */
        };                                               /*    16    16 */
        union {
                __be32             ipv4_dst;             /*    32     4 */
                __u32              ipv6_dst[4];          /*    32    16 */
        };                                               /*    32    16 */
        union {
                struct {
                        __be16     h_vlan_proto;         /*    48     2 */
                        __be16     h_vlan_TCI;           /*    50     2 */
                };                                       /*    48     4 */
                __u32              tbid;                 /*    48     4 */
        };                                               /*    48     4 */
        __u8                       smac[6];              /*    52     6 */
        __u8                       dmac[6];              /*    58     6 */

        /* size: 64, cachelines: 1, members: 12 */
};

After patch:

struct bpf_fib_lookup {
        __u8                       family;               /*     0     1 */
        __u8                       l4_protocol;          /*     1     1 */
        __be16                     sport;                /*     2     2 */
        __be16                     dport;                /*     4     2 */
        union {
                __u16              tot_len;              /*     6     2 */
                __u16              mtu_result;           /*     6     2 */
        } __attribute__((__aligned__(2)));               /*     6     2 */
        __u32                      ifindex;              /*     8     4 */
        union {
                __u8               tos;                  /*    12     1 */
                __be32             flowinfo;             /*    12     4 */
                __u32              rt_metric;            /*    12     4 */
        };                                               /*    12     4 */
        union {
                __be32             ipv4_src;             /*    16     4 */
                __u32              ipv6_src[4];          /*    16    16 */
        };                                               /*    16    16 */
        union {
                __be32             ipv4_dst;             /*    32     4 */
                __u32              ipv6_dst[4];          /*    32    16 */
        };                                               /*    32    16 */
        union {
                struct {
                        __be16     h_vlan_proto;         /*    48     2 */
                        __be16     h_vlan_TCI;           /*    50     2 */
                };                                       /*    48     4 */
                __u32              tbid;                 /*    48     4 */
        };                                               /*    48     4 */
        __u8                       smac[6];              /*    52     6 */
        __u8                       dmac[6];              /*    58     6 */

        /* size: 64, cachelines: 1, members: 12 */
        /* forced alignments: 1 */
} __attribute__((__aligned__(4)));

Thanks,
Daniel

#6 lore raw
Quoted 6 lines · Daniel Borkmann
Quoted 5 lines · Arnd Bergmann
Quoted 3 lines · Anton Protopopov

Declare this inner union as __attribute__((packed, aligned(2))) such that it always is of size 2 and is aligned to 16 bits.

Quoted 6 lines
Quoted 2 lines

I think you probably want 32-bit alignment for the structure, to keep the ABI unchanged on all other architectures.

Fwiw, on x86 nothing should change on this regard, see below pahole dump before/after. I think similar might be true for other archs as otherwise we should have seen a kbuild bot complaint on hitting the size assert.

It's not the structure layout that changes, just its alignment. Of course this is unlikely to cause actual bugs, but if there there is no real need to change it, I would leave the alignment the same as before.

Arnd

#7 lore raw
Quoted 10 lines · Arnd Bergmann
Quoted 9 lines · Daniel Borkmann
Quoted 7 lines · Arnd Bergmann
Quoted 3 lines · Anton Protopopov

Declare this inner union as __attribute__((packed, aligned(2))) such that it always is of size 2 and is aligned to 16 bits.

I think you probably want 32-bit alignment for the structure, to keep the ABI unchanged on all other architectures.

Quoted 8 lines
Quoted 3 lines

Fwiw, on x86 nothing should change on this regard, see below pahole dump before/after. I think similar might be true for other archs as otherwise we should have seen a kbuild bot complaint on hitting the size assert.

It's not the structure layout that changes, just its alignment. Of course this is unlikely to cause actual bugs, but if there there is no real need to change it, I would leave the alignment the same as before.

I think the struct will now be automatically 4-byte aligned, as it has the following layout:

    struct {
            u8 a;
            u8 b;
            u16 c;
            u16 d;
            union { u16 e; u16 f; } __aligned__(2);
            ...
    };

So if the union is 2-byte aligned, then the struct is automatically 4-byte aligned, because its address is 6 bytes less than the address of the union. In fact, as Daniel posted above, pahole shows that the struct actually has __aligned__(4) attribute in the patched version. I can add explicit __aligned__(4) to make this clear.

Quoted 1 line

Arnd

#8 lore raw

From: Anton Protopopov <[email protected]>
Date: Thu, 4 Apr 2024 09:56:51 +0200

Quoted 32 lines · Arnd Bergmann
Quoted 10 lines · Daniel Borkmann
Quoted 9 lines · Arnd Bergmann
Quoted 7 lines · Anton Protopopov
Quoted 3 lines

Declare this inner union as __attribute__((packed, aligned(2))) such that it always is of size 2 and is aligned to 16 bits.

I think you probably want 32-bit alignment for the structure, to keep the ABI unchanged on all other architectures.

Quoted 8 lines
Quoted 3 lines

Fwiw, on x86 nothing should change on this regard, see below pahole dump before/after. I think similar might be true for other archs as otherwise we should have seen a kbuild bot complaint on hitting the size assert.

It's not the structure layout that changes, just its alignment. Of course this is unlikely to cause actual bugs, but if there there is no real need to change it, I would leave the alignment the same as before.

I think the struct will now be automatically 4-byte aligned, as it has the following layout:

    struct {
            u8 a;
            u8 b;
            u16 c;
            u16 d;
            union { u16 e; u16 f; } __aligned__(2);
            ...
    };
Quoted 5 lines

So if the union is 2-byte aligned, then the struct is automatically 4-byte aligned, because its address is 6 bytes less than the address of the union. In fact, as Daniel posted above, pahole shows that the struct actually has __aligned__(4) attribute in the patched version. I can add explicit __aligned__(4) to make this clear.

I think it would be fine to not introduce an explicit attribute as long as pahole shows the structure is already 4-byte aligned.

Regarding aligning it to 32 or 64 bytes -- the actual benefit can be checked via scripts/bloat-o-meter. Usually alignment bigger than 8 bytes don't change anything in the object code, at least on x86_64.

Quoted 2 lines
Quoted 2 lines

Arnd

Thanks,
Olek

#9 lore raw
In reply to Anton Protopopov
Quoted 32 lines · Anton Protopopov
Quoted 10 lines · Arnd Bergmann
Quoted 9 lines · Daniel Borkmann
Quoted 7 lines · Arnd Bergmann
Quoted 3 lines · Anton Protopopov

Declare this inner union as __attribute__((packed, aligned(2))) such that it always is of size 2 and is aligned to 16 bits.

I think you probably want 32-bit alignment for the structure, to keep the ABI unchanged on all other architectures.

Quoted 8 lines
Quoted 3 lines

Fwiw, on x86 nothing should change on this regard, see below pahole dump before/after. I think similar might be true for other archs as otherwise we should have seen a kbuild bot complaint on hitting the size assert.

It's not the structure layout that changes, just its alignment. Of course this is unlikely to cause actual bugs, but if there there is no real need to change it, I would leave the alignment the same as before.

I think the struct will now be automatically 4-byte aligned, as it has the following layout:

    struct {
            u8 a;
            u8 b;
            u16 c;
            u16 d;
            union { u16 e; u16 f; } __aligned__(2);
            ...
    };
Quoted 5 lines

So if the union is 2-byte aligned, then the struct is automatically 4-byte aligned, because its address is 6 bytes less than the address of the union. In fact, as Daniel posted above, pahole shows that the struct actually has __aligned__(4) attribute in the patched version. I can add explicit __aligned__(4) to make this clear.

Ah right. I misread your patch and assumed that you gave the outer structure 2-byte alignment as well, but it's only the inner union that needs it, which you did correctly.

Arnd

#10 lore raw
In reply to Anton Protopopov
This patch was applied to bpf/bpf-next.git (master) by Andrii Nakryiko <[email protected]>:

Hello:

This patch was applied to bpf/bpf-next.git (master) by Andrii Nakryiko <[email protected]>:

Quoted 11 lines · On Wed, 3 Apr 2024 14:33:03 +0200 you wrote:

The struct bpf_fib_lookup is supposed to be of size 64. A recent commit 59b418c7063d ("bpf: Add a check for struct bpf_fib_lookup size") added a static assertion to check this property so that future changes to the structure will not accidentally break this assumption.

As it immediately turned out, on some 32-bit arm systems, when AEABI=n, the total size of the structure was equal to 68, see [1]. This happened because the bpf_fib_lookup structure contains a union of two 16-bit fields:

[...]

Here is the summary with links:
- [bpf-next] bpf: pack struct bpf_fib_lookup

    https://git.kernel.org/bpf/bpf-next/c/f91717007217

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html