On Thu, Aug 22, 2024 at 04:36:13PM -0700, Andrii Nakryiko wrote:

On Thu, Aug 22, 2024 at 2:24 AM Tony Ambardar tony.ambardar@gmail.com wrote:

...

From: Tony Ambardar tony.ambardar@gmail.com

Fix missing newlines and extraneous terminal spaces in messages.

Signed-off-by: Tony Ambardar tony.ambardar@gmail.com

tools/lib/bpf/btf.c | 7 ++++--- tools/lib/bpf/btf_dump.c | 2 +- tools/lib/bpf/btf_relocate.c | 2 +- tools/lib/bpf/libbpf.c | 4 ++-- tools/lib/bpf/relo_core.c | 2 +- 5 files changed, 9 insertions(+), 8 deletions(-)

diff --git a/tools/lib/bpf/btf.c b/tools/lib/bpf/btf.c index 32c00db3b91b..cf4f7bd7ff5c 100644 --- a/tools/lib/bpf/btf.c +++ b/tools/lib/bpf/btf.c @@ -2940,7 +2940,8 @@ static int btf_ext_setup_info(struct btf_ext *btf_ext,

    /* If no records, return failure now so .BTF.ext won't be used. */
    if (!info_left) {

•           pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
  

•           pr_debug("%s section in .BTF.ext has no records\n",
  

•                    ext_sec->desc);
  

if it fits in 100 characters, it stays in a single line, please don't touch that

OK, will do. Most of these are due to CI and local checkpatch, plus kernel docs on style/submission guidelines. It can be hard to decide which warnings/errors can be ignored and which apply, so it would be helpful at least if the CI flavour of checkpatch aligned with guidance above. Not sure what that involves, but sure that others besides me would benefit.

...

            return -EINVAL;
    }

@@ -3028,7 +3029,7 @@ static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)

    if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
        data_size < hdr->hdr_len) {

•           pr_debug("BTF.ext header not found");
  

•           pr_debug("BTF.ext header not found\n");
            return -EINVAL;
    }
  
  

@@ -3290,7 +3291,7 @@ int btf__dedup(struct btf *btf, const struct btf_dedup_opts *opts)

    d = btf_dedup_new(btf, opts);
    if (IS_ERR(d)) {

•           pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
  

•           pr_debug("btf_dedup_new failed: %ld\n", PTR_ERR(d));
            return libbpf_err(-EINVAL);
    }
  
  

diff --git a/tools/lib/bpf/btf_dump.c b/tools/lib/bpf/btf_dump.c index 894860111ddb..25e7c44d9f95 100644 --- a/tools/lib/bpf/btf_dump.c +++ b/tools/lib/bpf/btf_dump.c @@ -1304,7 +1304,7 @@ static void btf_dump_emit_type_decl(struct btf_dump *d, __u32 id, * chain, restore stack, emit warning, and try to * proceed nevertheless */

•                   pr_warn("not enough memory for decl stack:%d", err);
  

•                   pr_warn("not enough memory for decl stack:%d\n", err);
  

nit: add space after ':' while you are at it?

Right, done. Thanks!

...

                    d->decl_stack_cnt = stack_start;
                    return;
            }

[...]

I ran out of time looking through this, I'll try to get back to this patch set later today or tomorrow. So please don't repost, but see my comments below.

On Thu, Aug 22, 2024 at 2:25 AM Tony Ambardar tony.ambardar@gmail.com wrote:

From: Tony Ambardar tony.ambardar@gmail.com

Support for handling BTF data of either endianness was added in [1], but did not include BTF.ext data for lack of use cases. Later, support for static linking [2] provided a use case, but this feature and later ones were restricted to native-endian usage.

Add support for BTF.ext handling in either endianness. Convert BTF.ext data to native endianness when read into memory for further processing, and support raw data access that restores the original byte-order for output.

Add new API functions btf_ext__endianness() and btf_ext__set_endianness() for query and setting byte-order, as already exist for BTF data.

[1]:commit 3289959b97ca ("libbpf: Support BTF loading and raw data output in both endianness") [2]:commit 8fd27bf69b86 ("libbpf: Add BPF static linker BTF and BTF.ext support")

Is this some standard syntax to refer to commit? If not, maybe drop ":commit" parts?

Signed-off-by: Tony Ambardar tony.ambardar@gmail.com

tools/lib/bpf/btf.c | 163 ++++++++++++++++++++++++++++++-- tools/lib/bpf/btf.h | 3 + tools/lib/bpf/libbpf.map | 2 + tools/lib/bpf/libbpf_internal.h | 2 + 4 files changed, 161 insertions(+), 9 deletions(-)

diff --git a/tools/lib/bpf/btf.c b/tools/lib/bpf/btf.c index cf4f7bd7ff5c..cb4ee4ed4b5c 100644 --- a/tools/lib/bpf/btf.c +++ b/tools/lib/bpf/btf.c @@ -2884,6 +2884,52 @@ int btf__add_decl_tag(struct btf *btf, const char *value, int ref_type_id, return btf_commit_type(btf, sz); }

+/*

• • Swap endianness of the info segment in a BTF.ext data section:
• • • requires BTF.ext header data in native byte order
• • • expects info record fields are 32-bit
• */

+static int btf_ext_bswap_info(void *data, const __u32 data_size)

pass btf_ext_header (in native endianness) as input argument

+{

•   const struct btf_ext_header *hdr = data;
  

•   __u32 info_size, sum_len, i, *p;
  

•   if (data_size < offsetofend(struct btf_ext_header, hdr_len)) {
  

•           pr_warn("BTF.ext initial header not found\n");
  

•           return -EINVAL;
  

•   }
  

•   if (data_size < hdr->hdr_len) {
  

•           pr_warn("BTF.ext header not found\n");
  

•           return -EINVAL;
  

•   }
  

and we already checked this, no?

•   if (hdr->hdr_len < offsetofend(struct btf_ext_header, line_info_len)) {
  

•           pr_warn("BTF.ext header missing func_info, line_info\n");
  

•           return -EINVAL;
  

•   }
  

we have this check in btf_ext__new() already, do the swap after we validated this and keep this function simpler and more focused. You are mixing sanity checks with byte swapping, it's confusing.

•   sum_len = hdr->func_info_len + hdr->line_info_len;
  

•   if (hdr->hdr_len >= offsetofend(struct btf_ext_header, core_relo_len))
  

•           sum_len += hdr->core_relo_len;
  

•   info_size = data_size - hdr->hdr_len;
  

•   if (info_size != sum_len) {
  

•           pr_warn("BTF.ext info size mismatch with header data\n");
  

•           return -EINVAL;
  

•   }
  

•   if (info_size && info_size % sizeof(__u32)) {
  

•           pr_warn("BTF.ext info size not 32-bit multiple\n");
  

•           return -EINVAL;
  

•   }
  

same as above, all this has nothing to do with endianness, it's all sanity checks. Let's extract them into a separate validation function that can happen before byte swap.

•   p = data + hdr->hdr_len;
  

•   for (i = 0; i < info_size / sizeof(__u32); i++, p++)
  

•           *p = bswap_32(*p);
  

this is very suspicious, it might work right now because all fields are u32, but I'm uncomfortable making these decision.

I think for byte swapping case we need to make sure that BTF version is exactly known to us (i.e., all record length is knowable for us), and then do more verbose, but correct byte swap based on bpf_func_info/bpf_line_info/etc records. Just treating all this as u32 is not great.

•   return 0;
  

+}

struct btf_ext_sec_setup_param { __u32 off; __u32 len; @@ -3023,24 +3069,56 @@ static int btf_ext_setup_core_relos(struct btf_ext *btf_ext) return btf_ext_setup_info(btf_ext, &param); }

-static int btf_ext_parse_hdr(__u8 *data, __u32 data_size) +/* Swap byte-order of BTF.ext header */ +static void btf_ext_bswap_hdr(struct btf_ext_header *h) {

•   const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
  

•   __u32 hdr_len = h->hdr_len; /* need native byte-order */
  
  

•   if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
  

•       data_size < hdr->hdr_len) {
  

•           pr_debug("BTF.ext header not found\n");
  

•   if (h->magic == bswap_16(BTF_MAGIC))
  

•           hdr_len = bswap_32(hdr_len);
  

why so complicated? just pass hdr_len as input argument into btf_ext_bswap_hdr(), and then uniformly swap everything, no double-checking of decisions we already made

•   h->magic = bswap_16(h->magic);
  

•   h->hdr_len = bswap_32(h->hdr_len);
  

•   h->func_info_off = bswap_32(h->func_info_off);
  

•   h->func_info_len = bswap_32(h->func_info_len);
  

•   h->line_info_off = bswap_32(h->line_info_off);
  

•   h->line_info_len = bswap_32(h->line_info_len);
  

•   if (hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
  

•           return;
  

•   h->core_relo_off = bswap_32(h->core_relo_off);
  

•   h->core_relo_len = bswap_32(h->core_relo_len);
  

+}

+static int btf_ext_parse_hdr(struct btf_ext *btf_ext) +{

•   struct btf_ext_header *hdr = btf_ext->hdr;
  

•   __u32 hdr_len, data_size = btf_ext->data_size;
  

•   if (data_size < offsetofend(struct btf_ext_header, hdr_len)) {
  

•           pr_debug("BTF.ext initial header not found\n");
  

"BTF.ext header too short" ?

            return -EINVAL;
    }

•   hdr_len = hdr->hdr_len;
    if (hdr->magic == bswap_16(BTF_MAGIC)) {
  

•           pr_warn("BTF.ext in non-native endianness is not supported\n");
  

•           return -ENOTSUP;
  

•           btf_ext->swapped_endian = true;
  

•           hdr_len = bswap_32(hdr_len);
    } else if (hdr->magic != BTF_MAGIC) {
            pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
            return -EINVAL;
    }
  
  

•   if (data_size < hdr_len) {
  

•           pr_debug("BTF.ext header not found\n");
  

•           return -EINVAL;
  

•   }
  

•   /* Maintain native byte-order in memory for introspection */
  

•   if (btf_ext->swapped_endian)
  

•           btf_ext_bswap_hdr(hdr);
  

•   if (hdr->version != BTF_VERSION) {
            pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
            return -ENOTSUP;
  

@@ -3067,6 +3145,7 @@ void btf_ext__free(struct btf_ext *btf_ext) free(btf_ext->line_info.sec_idxs); free(btf_ext->core_relo_info.sec_idxs); free(btf_ext->data);

•   free(btf_ext->data_swapped);
    free(btf_ext);
  

}

@@ -3087,7 +3166,12 @@ struct btf_ext *btf_ext__new(const __u8 *data, __u32 size) } memcpy(btf_ext->data, data, size);

•   err = btf_ext_parse_hdr(btf_ext->data, size);
  

•   err = btf_ext_parse_hdr(btf_ext);
  

•   if (err)
  

•           goto done;
  

•   if (btf_ext->swapped_endian)
  

•           err = btf_ext_bswap_info(btf_ext->data, btf_ext->data_size);
    if (err)
            goto done;
  

nit:

if (btf_ext->swapped_endian) { err = ...; if (err) goto done; }

i.e., keep error handling close and at the same nesting level

@@ -3120,15 +3204,76 @@ struct btf_ext *btf_ext__new(const __u8 *data, __u32 size) return btf_ext; }

+static void *btf_ext_raw_data(const struct btf_ext *btf_ext_ro, __u32 *size,

•                         bool swap_endian)
  

+{

•   struct btf_ext *btf_ext = (struct btf_ext *)btf_ext_ro;
  

•   const __u32 data_sz = btf_ext->data_size;
  

•   void *data;
  

•   int err;
  

•   data = swap_endian ? btf_ext->data_swapped : btf_ext->data;
  

•   if (data) {
  

•           *size = data_sz;
  

•           return data;
  

•   }
  

•   data = calloc(1, data_sz);
  

•   if (!data)
  

•           return NULL;
  

•   memcpy(data, btf_ext->data, data_sz);
  

•   if (swap_endian) {
  

•           err = btf_ext_bswap_info(data, data_sz);
  

•           if (err) {
  

•                   free(data);
  

•                   return NULL;
  

•           }
  

•           btf_ext_bswap_hdr(data);
  

•           btf_ext->data_swapped = data;
  

•   }
  

•   *size = data_sz;
  

•   return data;
  

+}

const void *btf_ext__raw_data(const struct btf_ext *btf_ext, __u32 *size) {

•   __u32 data_sz;
  

•   void *data;
  

•   data = btf_ext_raw_data(btf_ext, &data_sz, btf_ext->swapped_endian);
  

•   if (!data)
  

•           return errno = ENOMEM, NULL;
  

•   *size = btf_ext->data_size;
  

•   return btf_ext->data;
  

•   return data;
  

}

__attribute__((alias("btf_ext__raw_data"))) const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size);

+enum btf_endianness btf_ext__endianness(const struct btf_ext *btf_ext) +{

•   return (is_host_big_endian() != btf_ext->swapped_endian) ?
  

•           BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
  

+}

+int btf_ext__set_endianness(struct btf_ext *btf_ext, enum btf_endianness endian) +{

•   bool is_tgt_big_endian = (endian == BTF_BIG_ENDIAN);
  

•   if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
  

•           return libbpf_err(-EINVAL);
  

•   btf_ext->swapped_endian = (is_host_big_endian() != is_tgt_big_endian);
  

•   if (!btf_ext->swapped_endian) {
  

•           free(btf_ext->data_swapped);
  

•           btf_ext->data_swapped = NULL;
  

•   }
  

•   return 0;
  

+}

struct btf_dedup;

diff --git a/tools/lib/bpf/btf.h b/tools/lib/bpf/btf.h index b68d216837a9..e3cf91687c78 100644 --- a/tools/lib/bpf/btf.h +++ b/tools/lib/bpf/btf.h @@ -167,6 +167,9 @@ LIBBPF_API const char *btf__str_by_offset(const struct btf *btf, __u32 offset); LIBBPF_API struct btf_ext *btf_ext__new(const __u8 *data, __u32 size); LIBBPF_API void btf_ext__free(struct btf_ext *btf_ext); LIBBPF_API const void *btf_ext__raw_data(const struct btf_ext *btf_ext, __u32 *size); +LIBBPF_API enum btf_endianness btf_ext__endianness(const struct btf_ext *btf_ext); +LIBBPF_API int btf_ext__set_endianness(struct btf_ext *btf_ext,

•                                  enum btf_endianness endian);
  
  

LIBBPF_API int btf__find_str(struct btf *btf, const char *s); LIBBPF_API int btf__add_str(struct btf *btf, const char *s); diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map index 8f0d9ea3b1b4..5c17632807b6 100644 --- a/tools/lib/bpf/libbpf.map +++ b/tools/lib/bpf/libbpf.map @@ -421,6 +421,8 @@ LIBBPF_1.5.0 { global: btf__distill_base; btf__relocate;

•           btf_ext__endianness;
  

•           btf_ext__set_endianness;
            bpf_map__autoattach;
            bpf_map__set_autoattach;
            bpf_program__attach_sockmap;
  

diff --git a/tools/lib/bpf/libbpf_internal.h b/tools/lib/bpf/libbpf_internal.h index 8cda511a1982..6b0270c83537 100644 --- a/tools/lib/bpf/libbpf_internal.h +++ b/tools/lib/bpf/libbpf_internal.h @@ -484,6 +484,8 @@ struct btf_ext { struct btf_ext_header *hdr; void *data; };

•   void *data_swapped;
  

•   bool swapped_endian;
    struct btf_ext_info func_info;
    struct btf_ext_info line_info;
    struct btf_ext_info core_relo_info;
  

-- 2.34.1

On Fri, Aug 23, 2024 at 12:47:47PM -0700, Andrii Nakryiko wrote:

On Thu, Aug 22, 2024 at 2:25 AM Tony Ambardar tony.ambardar@gmail.com wrote:

...

From: Tony Ambardar tony.ambardar@gmail.com

Allow bpf_object__open() to access files of either endianness, and convert included BPF programs to native byte-order in-memory for introspection.

Signed-off-by: Tony Ambardar tony.ambardar@gmail.com

tools/lib/bpf/libbpf.c | 21 +++++++++++++++++++-- tools/lib/bpf/libbpf_internal.h | 11 +++++++++++ 2 files changed, 30 insertions(+), 2 deletions(-)

Instructions are not the only data that would need swapping. We have user's data sections and stuff like that, which, generally speaking, isn't that safe to just byteswap.

I do understand the appeal of being endianness-agnostic, but doesn't extend all the way to actually loading BPF programs. At least I wouldn't start there.

Yes, absolutely. I first planned to move the endianness check from "open" to "load" functions but got waylaid tracing skeleton code into the latter and left it to continue progress. Let me figure out the best place to put a check without breaking things.

We need to make open phase endianness agnostic, load should just fail for swapped endianness case. So let's record the fact that we are not in native endianness, and fail early in load step.

This will still allow us to generate skeletons and stuff like that, right?

...

diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 8a0a0c1e37e1..a542031f4f73 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -940,6 +940,21 @@ bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data, return 0; }

+static void bpf_object_bswap_progs(struct bpf_object *obj) +{

•   struct bpf_program *prog = obj->programs;
  

•   struct bpf_insn *insn;
  

•   int p, i;
  

•   for (p = 0; p < obj->nr_programs; p++, prog++) {
  

•           insn = prog->insns;
  

•           for (i = 0; i < prog->insns_cnt; i++, insn++)
  

•                   bpf_insn_bswap(insn);
  

•           pr_debug("prog '%s': converted %zu insns to native byteorder\n",
  

"byte order"?

Good catch. Fixed.

...

•                    prog->name, prog->insns_cnt);
  

•   }
  

+}

static const struct btf_member * find_member_by_offset(const struct btf_type *t, __u32 bit_offset) { @@ -1610,7 +1625,6 @@ static int bpf_object__check_endianness(struct bpf_object *obj) #else # error "Unrecognized __BYTE_ORDER__" #endif

•   pr_warn("elf: endianness mismatch in %s.\n", obj->path);
    return -LIBBPF_ERRNO__ENDIAN;
  

}

@@ -3953,6 +3967,10 @@ static int bpf_object__elf_collect(struct bpf_object *obj) return -LIBBPF_ERRNO__FORMAT; }

•   /* change BPF program insns to native endianness for introspection */
  

•   if (bpf_object__check_endianness(obj))
  

let's rename this to "is_native_endianness()" and return true/false. "check" makes sense as something that errors out, but now it's purely a query, so "check" naming is confusing.

Right, I mistook this as exported before and left it.

BTW, so libelf will transparently byte-swap relocations and stuff like that to native endianness, is that right?

Correct. Sections with types like ELF_T_REL (.rel) and ELF_T_SYM (.symtab) get translated automagically. See patch #3 for example.

...

•           bpf_object_bswap_progs(obj);
  

•   /* sort BPF programs by section name and in-section instruction offset
     * for faster search
     */
  

@@ -7993,7 +8011,6 @@ static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, }

    err = bpf_object__elf_init(obj);

•   err = err ? : bpf_object__check_endianness(obj);
    err = err ? : bpf_object__elf_collect(obj);
    err = err ? : bpf_object__collect_externs(obj);
    err = err ? : bpf_object_fixup_btf(obj);
  

diff --git a/tools/lib/bpf/libbpf_internal.h b/tools/lib/bpf/libbpf_internal.h index 6b0270c83537..f53daa601c6f 100644 --- a/tools/lib/bpf/libbpf_internal.h +++ b/tools/lib/bpf/libbpf_internal.h @@ -11,6 +11,7 @@

#include <stdlib.h> #include <limits.h> +#include <byteswap.h> #include <errno.h> #include <linux/err.h> #include <fcntl.h> @@ -590,6 +591,16 @@ static inline bool is_ldimm64_insn(struct bpf_insn *insn) return insn->code == (BPF_LD | BPF_IMM | BPF_DW); }

+static inline void bpf_insn_bswap(struct bpf_insn *insn) +{

•   /* dst_reg & src_reg nibbles */
  

•   __u8 *regs = (__u8 *)insn + offsetofend(struct bpf_insn, code);
  

•   *regs = (*regs >> 4) | (*regs << 4);
  

hm... we have fields, just do a brain-dead swap instead of all this mucking with offsetofend(

__u8 tmp_reg = insn->dst_reg;

insn->dst_reg = insn->src_reg; insn->src_reg = tmp_reg;

?

Main reason for this is most compilers recognize the shift/or statement pattern and emit a rotate op as I recall. And the offsetofend() seemed clearest at documenting "the byte after opcode" while not obscuring these are nibble fields. So would prefer to leave it unless you have strong objections or I'm off the mark somehow. Let me know either way? Thanks!

...

•   insn->off = bswap_16(insn->off);
  

•   insn->imm = bswap_32(insn->imm);
  

+}

/* Unconditionally dup FD, ensuring it doesn't use [0, 2] range.

• Original FD is not closed or altered in any other way.
• Preserves original FD value, if it's invalid (negative).

-- 2.34.1

On Thu, Aug 22, 2024 at 2:25 AM Tony Ambardar tony.ambardar@gmail.com wrote:

From: Tony Ambardar tony.ambardar@gmail.com

Allow static linking object files of either endianness, checking that input files have consistent byte-order, and setting output endianness from input.

Linking requires in-memory processing of programs, relocations, sections, etc. in native endianness, and output conversion to target byte-order. This is enabled by built-in ELF translation and recent BTF/BTF.ext endianness functions. Further add local functions for swapping byte-order of sections containing BPF insns.

Signed-off-by: Tony Ambardar tony.ambardar@gmail.com

tools/lib/bpf/linker.c | 106 ++++++++++++++++++++++++++++++++++------- 1 file changed, 90 insertions(+), 16 deletions(-)

Mostly just stylistic and code organization nits, the change overall looks good.

diff --git a/tools/lib/bpf/linker.c b/tools/lib/bpf/linker.c index 7489306cd6f7..9bf218db443e 100644 --- a/tools/lib/bpf/linker.c +++ b/tools/lib/bpf/linker.c @@ -135,6 +135,7 @@ struct bpf_linker { int fd; Elf *elf; Elf64_Ehdr *elf_hdr;

•   bool swapped_endian;
  
    /* Output sections metadata */
    struct dst_sec *secs;
  

@@ -324,13 +325,8 @@ static int init_output_elf(struct bpf_linker *linker, const char *file)

    linker->elf_hdr->e_machine = EM_BPF;
    linker->elf_hdr->e_type = ET_REL;

-#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__

•   linker->elf_hdr->e_ident[EI_DATA] = ELFDATA2LSB;
  

-#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__

•   linker->elf_hdr->e_ident[EI_DATA] = ELFDATA2MSB;
  

-#else -#error "Unknown __BYTE_ORDER__" -#endif

•   /* Set unknown ELF endianness, assign later from input files */
  

•   linker->elf_hdr->e_ident[EI_DATA] = ELFDATANONE;
  
    /* STRTAB */
    /* initialize strset with an empty string to conform to ELF */
  

@@ -541,19 +537,21 @@ static int linker_load_obj_file(struct bpf_linker *linker, const char *filename, const struct bpf_linker_file_opts *opts, struct src_obj *obj) { -#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__

•   const int host_endianness = ELFDATA2LSB;
  

-#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__

•   const int host_endianness = ELFDATA2MSB;
  

-#else -#error "Unknown __BYTE_ORDER__" -#endif int err = 0; Elf_Scn *scn; Elf_Data *data; Elf64_Ehdr *ehdr; Elf64_Shdr *shdr; struct src_sec *sec;

•   unsigned char obj_byteorder;
  

•   unsigned char *link_byteorder = &linker->elf_hdr->e_ident[EI_DATA];
  

nit: not a fan of pointer into e_ident, just read local value of byte order, and then assign it directly below (it's only in one of the branches, no duplication, really)

+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__

•   const unsigned char host_byteorder = ELFDATA2LSB;
  

+#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__

•   const unsigned char host_byteorder = ELFDATA2MSB;
  

+#else +#error "Unknown __BYTE_ORDER__" +#endif

    pr_debug("linker: adding object file '%s'...\n", filename);

@@ -579,11 +577,25 @@ static int linker_load_obj_file(struct bpf_linker *linker, const char *filename, pr_warn_elf("failed to get ELF header for %s", filename); return err; }

•   if (ehdr->e_ident[EI_DATA] != host_endianness) {
  

•   /* Linker output endianness set by first input object */
  

•   obj_byteorder = ehdr->e_ident[EI_DATA];
  

•   if (obj_byteorder != ELFDATA2LSB && obj_byteorder != ELFDATA2MSB) {
            err = -EOPNOTSUPP;
  

•           pr_warn_elf("unsupported byte order of ELF file %s", filename);
  

•           pr_warn("linker: unknown byte order of ELF file %s\n", filename);
            return err;
    }
  

•   if (*link_byteorder == ELFDATANONE) {
  

•           *link_byteorder = obj_byteorder;
  

see above, I'd prefer:

linker->elf_hdr->e_ident[EI_DATA] = obj_byteorder;

•           linker->swapped_endian = obj_byteorder != host_byteorder;
  

•           pr_debug("linker: set %s-endian output byte order\n",
  

•                    obj_byteorder == ELFDATA2MSB ? "big" : "little");
  

•   } else if (*link_byteorder != obj_byteorder) {
  

•           err = -EOPNOTSUPP;
  

•           pr_warn("linker: byte order mismatch with ELF file %s\n", filename);
  

•           return err;
  

•   }
  

•   if (ehdr->e_type != ET_REL
        || ehdr->e_machine != EM_BPF
        || ehdr->e_ident[EI_CLASS] != ELFCLASS64) {
  

@@ -1111,6 +1123,27 @@ static bool sec_content_is_same(struct dst_sec *dst_sec, struct src_sec *src_sec return true; }

+static bool is_exec_sec(struct dst_sec *sec) +{

•   if (!sec || sec->ephemeral)
  

•           return false;
  

•   return (sec->shdr->sh_type == SHT_PROGBITS) &&
  

•          (sec->shdr->sh_flags & SHF_EXECINSTR);
  

+}

+static int exec_sec_bswap(void *raw_data, int size) +{

•   const int insn_cnt = size / sizeof(struct bpf_insn);
  

•   struct bpf_insn *insn = raw_data;
  

•   int i;
  

•   if (size % sizeof(struct bpf_insn))
  

•           return -EINVAL;
  

•   for (i = 0; i < insn_cnt; i++, insn++)
  

•           bpf_insn_bswap(insn);
  

•   return 0;
  

+}

static int extend_sec(struct bpf_linker *linker, struct dst_sec *dst, struct src_sec *src) { void *tmp; @@ -1170,6 +1203,16 @@ static int extend_sec(struct bpf_linker *linker, struct dst_sec *dst, struct src memset(dst->raw_data + dst->sec_sz, 0, dst_align_sz - dst->sec_sz); /* now copy src data at a properly aligned offset */ memcpy(dst->raw_data + dst_align_sz, src->data->d_buf, src->shdr->sh_size);

•           /* convert added bpf insns to native byte-order */
  

•           if (linker->swapped_endian && is_exec_sec(dst)) {
  

•                   err = exec_sec_bswap(dst->raw_data + dst_align_sz,
  

•                                        src->shdr->sh_size);
  

nit: I think exec_sec_bswap() shouldn't ever fail, so given we have is_exec_sec() now, let's do the size alignment check early on (and regardless of swapped_endian), and then just proceed with byte swap that can't fail

•                   if (err) {
  

•                           pr_warn("%s: error changing insns endianness\n", __func__);
  

•                           return err;
  

•                   }
  

•           }
    }
  
    dst->sec_sz = dst_final_sz;
  

@@ -2630,6 +2673,14 @@ int bpf_linker__finalize(struct bpf_linker *linker) if (!sec->scn) continue;

•           /* restore sections with bpf insns to target byte-order */
  

•           if (linker->swapped_endian && is_exec_sec(sec)) {
  

•                   err = exec_sec_bswap(sec->raw_data, sec->sec_sz);
  

and here we'll know that size is validly aligned anyways, so no checks required

•                   if (err) {
  

•                           pr_warn("error finalizing insns endianness\n");
  

•                           return libbpf_err(err);
  

•                   }
  

•           }
            sec->data->d_buf = sec->raw_data;
    }
  
  

@@ -2696,6 +2747,13 @@ static int emit_elf_data_sec(struct bpf_linker *linker, const char *sec_name, return 0; }

+static enum btf_endianness +linker_btf_endianness(const struct bpf_linker *linker) +{

•   unsigned char byteorder = linker->elf_hdr->e_ident[EI_DATA];
  

empty line between variable declaration and the rest of the code

•   return byteorder == ELFDATA2MSB ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
  

+}

but actually, this whole helper function seems unnecessary, just do everything inside finalize_btf, it's a pretty trivial piece of logic

static int finalize_btf(struct bpf_linker *linker) { LIBBPF_OPTS(btf_dedup_opts, opts); @@ -2742,6 +2800,22 @@ static int finalize_btf(struct bpf_linker *linker) return err; }

•   /* Set .BTF and .BTF.ext output byte order */
  

•   err = btf__set_endianness(linker->btf,
  

•                             linker_btf_endianness(linker));
  

•   if (err) {
  

•           pr_warn("failed to set .BTF output endianness: %d\n", err);
  

nit: you used "linker: " prefix for messages like this, stay consistent?

•           return err;
  

•   }
  

•   if (linker->btf_ext) {
  

•           err = btf_ext__set_endianness(linker->btf_ext,
  

•                                         linker_btf_endianness(linker));
  

•           if (err) {
  

•                   pr_warn("failed to set .BTF.ext output endianness: %d\n", err);
  

•                   return err;
  

•           }
  

•   }
  

•   /* Emit .BTF section */
    raw_data = btf__raw_data(linker->btf, &raw_sz);
    if (!raw_data)
  

-- 2.34.1

On Fri, Aug 23, 2024 at 12:47:56PM -0700, Andrii Nakryiko wrote:

On Thu, Aug 22, 2024 at 2:25 AM Tony Ambardar tony.ambardar@gmail.com wrote:

...

From: Tony Ambardar tony.ambardar@gmail.com

Track target endianness in 'struct bpf_gen' and process in-memory data in native byte-order, but on finalization convert the embedded loader BPF insns to target endianness.

The light skeleton also includes a target-accessed data blob which is heterogeneous and thus difficult to convert to target byte-order on finalization. Add support functions to convert data to target endianness as it is added to the blob.

Also add additional debug logging for data blob structure details and skeleton loading.

Signed-off-by: Tony Ambardar tony.ambardar@gmail.com

tools/lib/bpf/bpf_gen_internal.h | 1 + tools/lib/bpf/gen_loader.c | 187 +++++++++++++++++++++++-------- tools/lib/bpf/libbpf.c | 1 + tools/lib/bpf/skel_internal.h | 3 +- 4 files changed, 147 insertions(+), 45 deletions(-)

[...]

...

+/*

• • Fields of bpf_attr are set to values in native byte-order before being
• • written to the target-bound data blob, and may need endian conversion.
• • This macro allows setting the correct value in situ and is simpler than
• • writing a separate converter for *all fields* of *all records* included
• • in union bpf_attr.
• */

+#define move_tgt_endian(lval, rval) { \

•   if (!gen->swapped_endian)                               \
  

•           lval = (rval);                                  \
  

add {} here and for else branch, please

...

•   else                                                    \
  

•           switch (sizeof(lval)) {                         \
  

•           case 2:                                         \
  

•                   lval = bswap_16(rval);                  \
  

•                   break;                                  \
  

•           case 4:                                         \
  

•                   lval = bswap_32(rval);                  \
  

•                   break;                                  \
  

•           case 8:                                         \
  

•                   lval = bswap_64(rval);                  \
  

•                   break;                                  \
  

I'd also go for more compact:

case 2: lval = bswap_16(rval); break; case 4: lval = bswap_32(rval); break;

The irony is that I had this originally but checkpatch complained loudly.

...

•           default:                                        \
  

•                   lval = (rval);                          \
  

•                   pr_warn("unsupported bswap size!\n");   \
  

case 1: is unsupported? It just doesn't need a byte swap, so please add it explicitly and avoid unnecessary warnings

Good point.

...

•           }                                               \
  

•   }
  

void bpf_gen__load_btf(struct bpf_gen *gen, const void *btf_raw_data, __u32 btf_raw_size) {

[...]

...

    emit(gen, BPF_LDX_MEM(BPF_B, BPF_REG_9, BPF_REG_8, offsetofend(struct bpf_insn, code)));
    emit(gen, BPF_ALU32_IMM(BPF_AND, BPF_REG_9, reg_mask));
    emit(gen, BPF_STX_MEM(BPF_B, BPF_REG_8, BPF_REG_9, offsetofend(struct bpf_insn, code)));

@@ -931,11 +971,34 @@ static void cleanup_relos(struct bpf_gen *gen, int insns) cleanup_core_relo(gen); }

+/* Covert func, line, and core relo info records to target endianness,

typo: convert

Fixed, thanks.

...

• • checking the blob size is consistent with 32-bit fields.
• */

+static void info_blob_bswap(struct bpf_gen *gen, int info_off, __u32 size) +{

•   __u32 *field = gen->data_start + info_off;
  

•   int i, cnt = size / sizeof(__u32);
  

•   if (size && size % sizeof(__u32)) {
  

nit: () around mod operation

Done.

...

•           pr_warn("info records not using 32-bit fields!\n");
  

•           return;
  

•   }
  

•   if (gen->swapped_endian)
  

•           for (i = 0; i < cnt; i++, field++)
  

•                   *field = bswap_32(*field);
  

+}

void bpf_gen__prog_load(struct bpf_gen *gen, enum bpf_prog_type prog_type, const char *prog_name, const char *license, struct bpf_insn *insns, size_t insn_cnt, struct bpf_prog_load_opts *load_attr, int prog_idx) {

•   int func_info_tot_sz = load_attr->func_info_cnt *
  

•                          load_attr->func_info_rec_size;
  

•   int line_info_tot_sz = load_attr->line_info_cnt *
  

•                          load_attr->line_info_rec_size;
  

•   int core_relo_tot_sz = gen->core_relo_cnt *
  

•                          sizeof(struct bpf_core_relo);
    int prog_load_attr, license_off, insns_off, func_info, line_info, core_relos;
    int attr_size = offsetofend(union bpf_attr, core_relo_rec_size);
    union bpf_attr attr;
  

@@ -947,32 +1010,60 @@ void bpf_gen__prog_load(struct bpf_gen *gen, license_off = add_data(gen, license, strlen(license) + 1); /* add insns to blob of bytes */ insns_off = add_data(gen, insns, insn_cnt * sizeof(struct bpf_insn));

•   pr_debug("gen: prog_load: license off %d insn off %d\n",
  

•            license_off, insns_off);
  
  

•   attr.prog_type = prog_type;
  

•   attr.expected_attach_type = load_attr->expected_attach_type;
  

•   attr.attach_btf_id = load_attr->attach_btf_id;
  

•   attr.prog_ifindex = load_attr->prog_ifindex;
  

•   attr.kern_version = 0;
  

•   attr.insn_cnt = (__u32)insn_cnt;
  

•   attr.prog_flags = load_attr->prog_flags;
  

•   attr.func_info_rec_size = load_attr->func_info_rec_size;
  

•   attr.func_info_cnt = load_attr->func_info_cnt;
  

•   func_info = add_data(gen, load_attr->func_info,
  

•                        attr.func_info_cnt * attr.func_info_rec_size);
  

•   /* convert blob insns to target endianness */
  

•   if (gen->swapped_endian) {
  

•           struct bpf_insn *insn = gen->data_start + insns_off;
  

•           int i;
  
  

•   attr.line_info_rec_size = load_attr->line_info_rec_size;
  

•   attr.line_info_cnt = load_attr->line_info_cnt;
  

•   line_info = add_data(gen, load_attr->line_info,
  

•                        attr.line_info_cnt * attr.line_info_rec_size);
  

•           for (i = 0; i < insn_cnt; i++, insn++)
  

•                   bpf_insn_bswap(insn);
  

•   }
  
  

•   attr.core_relo_rec_size = sizeof(struct bpf_core_relo);
  

•   attr.core_relo_cnt = gen->core_relo_cnt;
  

•   core_relos = add_data(gen, gen->core_relos,
  

•                        attr.core_relo_cnt * attr.core_relo_rec_size);
  

•   move_tgt_endian(attr.prog_type, prog_type);
  

•   move_tgt_endian(attr.expected_attach_type, load_attr->expected_attach_type);
  

•   move_tgt_endian(attr.attach_btf_id, load_attr->attach_btf_id);
  

•   move_tgt_endian(attr.prog_ifindex, load_attr->prog_ifindex);
  

•   attr.kern_version = 0;
  

•   move_tgt_endian(attr.insn_cnt, (__u32)insn_cnt);
  

•   move_tgt_endian(attr.prog_flags, load_attr->prog_flags);
  

•   move_tgt_endian(attr.func_info_rec_size, load_attr->func_info_rec_size);
  

•   move_tgt_endian(attr.func_info_cnt, load_attr->func_info_cnt);
  

this is quite intrusive, maybe instead of imperative move_tgt_endian() macro, develop the one that could be used as

attr.func_info_cnt = tgt_endian(load_attr->func_info_cnt);

I had considered this but it's not reliable since the source var size may not match the dest and the bswap will be improperly sized e.g. note above that move_tgt_endian() uses the _dest_ var to size the bswap.

While I completely agree this is intrusive, it's still safe and correct. The other idea I played with is to leave the assignments alone and fix up struct fields' endianness afterwards via macro. Something like:

attr.map_type = map_type; attr.key_size = key_size; attr.value_size = value_size; attr.map_flags = map_attr->map_flags; attr.map_extra = map_attr->map_extra;

BSWAP_FIELDS(attr, map_type, key_size, value_size, map_flags, map_extra);

But this would require some funky macro magic, possibly in a small header. What do you think? Does something similar exist already in kernel sources?

? I.e., working as an expression, taking into account the need to swap and byte size of the argument. Should be doable.

...

•   func_info = add_data(gen, load_attr->func_info, func_info_tot_sz);
  

•   pr_debug("gen: prog_load: func_info: off %d cnt %d rec size %d\n",
  

•            func_info, load_attr->func_info_cnt,
  

•            load_attr->func_info_rec_size);
  

[...]