Skip to contentSkip to conversation

bpf: bpf_core_calc_relo_insn() should verify relocation type id

PATCH In review Eduard Zingerman 2 participants 8 messages Aug 21, 2024

Subsystems (2)
lore
Share & tools
lore

Eduard Zingerman

author maintainer
#0 lore raw

PATCH 1/1 bpf: bpf_core_calc_relo_insn() should verify relocation type id

In case of malformed relocation record of kind BPF_CORE_TYPE_ID_LOCAL referencing a non-existing BTF type, function bpf_core_calc_relo_insn would cause a null pointer deference.

Fix this by adding a proper check, as malformed relocation records could be passed from user space.

Simplest reproducer is a program:

    r0 = 0
    exit

With a single relocation record:

    .insn_off = 0,          /* patch first instruction */
    .type_id = 100500,      /* this type id does not exist */
    .access_str_off = 6,    /* offset of string "0" */

.kind = BPF_CORE_TYPE_ID_LOCAL,

See the link for original reproducer.

---
 tools/lib/bpf/relo_core.c | 5 +++++
 1 file changed, 5 insertions(+)
tools/lib/bpf/relo_core.c +5 -0
index 63a4d5ad12d1..a04724831ebc 100644
@@ -1297,6 +1297,11 @@ int bpf_core_calc_relo_insn(const char *prog_name,
 
 	local_id = relo->type_id;
 	local_type = btf_type_by_id(local_btf, local_id);
+	if (!local_type) {
+		pr_warn("prog '%s': relo #%d: bad type id %u\n",
+			prog_name, relo_idx, local_id);
+		return -EINVAL;
+	}
 	local_name = btf__name_by_offset(local_btf, local_type->name_off);
 	if (!local_name)
 		return -EINVAL;
Signature · 1 line
2.45.2
#1 lore raw
Quoted 35 lines · Eduard Zingerman

In case of malformed relocation record of kind BPF_CORE_TYPE_ID_LOCAL referencing a non-existing BTF type, function bpf_core_calc_relo_insn would cause a null pointer deference.

Fix this by adding a proper check, as malformed relocation records could be passed from user space.

Simplest reproducer is a program:

    r0 = 0
    exit

With a single relocation record:

    .insn_off = 0,          /* patch first instruction */
    .type_id = 100500,      /* this type id does not exist */
    .access_str_off = 6,    /* offset of string "0" */

.kind = BPF_CORE_TYPE_ID_LOCAL,

See the link for original reproducer.

---
 tools/lib/bpf/relo_core.c | 5 +++++
 1 file changed, 5 insertions(+)
tools/lib/bpf/relo_core.c +0 -0
index 63a4d5ad12d1..a04724831ebc 100644

tools/lib/bpf/relo_core.c int bpf_core_calc_relo_insn(const char *prog_name, · line 1300

Quoted 4 lines
tools/lib/bpf/relo_core.c +1 -0
@@ -1297,6 +1297,11 @@ int bpf_core_calc_relo_insn(const char *prog_name,
 
        local_id = relo->type_id;
        local_type = btf_type_by_id(local_btf, local_id);
+       if (!local_type) {

This is a meaningless check at least for libbpf's implementation of btf_type_by_id(), it never returns NULL. Commit you point to in Fixes tag clearly states the differences.

So you'd need to validate local_id directly against number of types in local_btf.

pw-bot: cr

tools/lib/bpf/relo_core.c int bpf_core_calc_relo_insn(const char *prog_name, · line 1301

Quoted 1 line
tools/lib/bpf/relo_core.c +1 -0
+               pr_warn("prog '%s': relo #%d: bad type id %u\n",

nit: this part of CO-RE-related code normally uses [%u] "syntax" to point to BTF type IDs, please adjust for consistency

Quoted 9 lines
tools/lib/bpf/relo_core.c +3 -1
+                       prog_name, relo_idx, local_id);
+               return -EINVAL;
+       }
        local_name = btf__name_by_offset(local_btf, local_type->name_off);
        if (!local_name)
                return -EINVAL;
--

2.45.2

author maintainer
#2 lore raw

On Wed, 2024-08-21 at 09:59 -0700, Andrii Nakryiko wrote:

[...]

Quoted 14 lines
Quoted 14 lines
---
 tools/lib/bpf/relo_core.c | 5 +++++
 1 file changed, 5 insertions(+)
tools/lib/bpf/relo_core.c +0 -0
index 63a4d5ad12d1..a04724831ebc 100644
@@ -1297,6 +1297,11 @@ int bpf_core_calc_relo_insn(const char *prog_name,
 
Quoted 7 lines

tools/lib/bpf/relo_core.c int bpf_core_calc_relo_insn(const char *prog_name, · line 1300

Quoted 3 lines
tools/lib/bpf/relo_core.c +1 -0
        local_id = relo->type_id;
        local_type = btf_type_by_id(local_btf, local_id);
+       if (!local_type) {

This is a meaningless check at least for libbpf's implementation of btf_type_by_id(), it never returns NULL. Commit you point to in Fixes tag clearly states the differences.

That is not true on kernel side. bpf_core_calc_relo_insn() is called from bpf_core_apply():

int bpf_core_apply(struct bpf_core_ctx *ctx, const struct bpf_core_relo *relo,
int relo_idx, void *insn)

{
        bool need_cands = relo->kind != BPF_CORE_TYPE_ID_LOCAL;
        ...
        if (need_cands) {
                ...
                // code below would report an error if relo->type_id is bogus
                cc = bpf_core_find_cands(ctx, relo->type_id);
                if (IS_ERR(cc)) {
                        bpf_log(ctx->log, "target candidate search failed for %d\n",
                                relo->type_id);
                        err = PTR_ERR(cc);
                        goto out;
                }
                ...
        }

        err = bpf_core_calc_relo_insn((void *)ctx->log, relo, relo_idx, ctx->btf, &cands, specs,
                                      &targ_res);
        ...
}

If need_cands is false the bogus type_id could reach into bpf_core_calc_relo_insn(). Which is exactly the case with repro submitted by Liu.
There is also a simplified repro here:
https://github.com/kernel-patches/bpf/commit/017a9dcf17e572f9b7c32aa62a81df8ef41cef17
But I can't submit it as a test yet.

Quoted 3 lines

So you'd need to validate local_id directly against number of types in local_btf.

How is this better than a null check?

Quoted 8 lines

pw-bot: cr

tools/lib/bpf/relo_core.c int bpf_core_calc_relo_insn(const char *prog_name, · line 1301

Quoted 1 line
tools/lib/bpf/relo_core.c +1 -0
+               pr_warn("prog '%s': relo #%d: bad type id %u\n",

nit: this part of CO-RE-related code normally uses [%u] "syntax" to point to BTF type IDs, please adjust for consistency

Ok

#3 lore raw
Quoted 60 lines · Eduard Zingerman

On Wed, 2024-08-21 at 09:59 -0700, Andrii Nakryiko wrote:

[...]

Quoted 14 lines
Quoted 14 lines
---
 tools/lib/bpf/relo_core.c | 5 +++++
 1 file changed, 5 insertions(+)
tools/lib/bpf/relo_core.c +0 -0
index 63a4d5ad12d1..a04724831ebc 100644
@@ -1297,6 +1297,11 @@ int bpf_core_calc_relo_insn(const char *prog_name,
 
Quoted 7 lines

tools/lib/bpf/relo_core.c int bpf_core_calc_relo_insn(const char *prog_name, · line 1300

Quoted 3 lines
tools/lib/bpf/relo_core.c +1 -0
        local_id = relo->type_id;
        local_type = btf_type_by_id(local_btf, local_id);
+       if (!local_type) {

This is a meaningless check at least for libbpf's implementation of btf_type_by_id(), it never returns NULL. Commit you point to in Fixes tag clearly states the differences.

That is not true on kernel side. bpf_core_calc_relo_insn() is called from bpf_core_apply():

int bpf_core_apply(struct bpf_core_ctx *ctx, const struct bpf_core_relo *relo,
int relo_idx, void *insn)

{
        bool need_cands = relo->kind != BPF_CORE_TYPE_ID_LOCAL;
        ...
        if (need_cands) {
                ...
                // code below would report an error if relo->type_id is bogus
                cc = bpf_core_find_cands(ctx, relo->type_id);
                if (IS_ERR(cc)) {
                        bpf_log(ctx->log, "target candidate search failed for %d\n",
                                relo->type_id);
                        err = PTR_ERR(cc);
                        goto out;
                }
                ...
        }

        err = bpf_core_calc_relo_insn((void *)ctx->log, relo, relo_idx, ctx->btf, &cands, specs,
                                      &targ_res);
        ...
}

If need_cands is false the bogus type_id could reach into bpf_core_calc_relo_insn(). Which is exactly the case with repro submitted by Liu.
There is also a simplified repro here:
https://github.com/kernel-patches/bpf/commit/017a9dcf17e572f9b7c32aa62a81df8ef41cef17
But I can't submit it as a test yet.

Quoted 1 line
Quoted 5 lines
Quoted 2 lines

So you'd need to validate local_id directly against number of types in local_btf.

How is this better than a null check?

because id check will be useful for both kernel and libbpf sides?..

Quoted 11 lines
Quoted 8 lines

pw-bot: cr

tools/lib/bpf/relo_core.c int bpf_core_calc_relo_insn(const char *prog_name, · line 1301

Quoted 1 line
tools/lib/bpf/relo_core.c +1 -0
+               pr_warn("prog '%s': relo #%d: bad type id %u\n",

nit: this part of CO-RE-related code normally uses [%u] "syntax" to point to BTF type IDs, please adjust for consistency

Ok

author maintainer
#4 lore raw
Quoted 61 lines · Andrii Nakryiko
Quoted 60 lines · Eduard Zingerman

On Wed, 2024-08-21 at 09:59 -0700, Andrii Nakryiko wrote:

[...]

Quoted 14 lines
Quoted 14 lines
---
 tools/lib/bpf/relo_core.c | 5 +++++
 1 file changed, 5 insertions(+)
tools/lib/bpf/relo_core.c +0 -0
index 63a4d5ad12d1..a04724831ebc 100644
@@ -1297,6 +1297,11 @@ int bpf_core_calc_relo_insn(const char *prog_name,
 
Quoted 7 lines

tools/lib/bpf/relo_core.c int bpf_core_calc_relo_insn(const char *prog_name, · line 1300

Quoted 3 lines
tools/lib/bpf/relo_core.c +1 -0
        local_id = relo->type_id;
        local_type = btf_type_by_id(local_btf, local_id);
+       if (!local_type) {

This is a meaningless check at least for libbpf's implementation of btf_type_by_id(), it never returns NULL. Commit you point to in Fixes tag clearly states the differences.

That is not true on kernel side. bpf_core_calc_relo_insn() is called from bpf_core_apply():

int bpf_core_apply(struct bpf_core_ctx *ctx, const struct bpf_core_relo *relo,
int relo_idx, void *insn)

{
        bool need_cands = relo->kind != BPF_CORE_TYPE_ID_LOCAL;
        ...
        if (need_cands) {
                ...
                // code below would report an error if relo->type_id is bogus
                cc = bpf_core_find_cands(ctx, relo->type_id);
                if (IS_ERR(cc)) {
                        bpf_log(ctx->log, "target candidate search failed for %d\n",
                                relo->type_id);
                        err = PTR_ERR(cc);
                        goto out;
                }
                ...
        }

        err = bpf_core_calc_relo_insn((void *)ctx->log, relo, relo_idx, ctx->btf, &cands, specs,
                                      &targ_res);
        ...
}

If need_cands is false the bogus type_id could reach into bpf_core_calc_relo_insn(). Which is exactly the case with repro submitted by Liu.
There is also a simplified repro here:
https://github.com/kernel-patches/bpf/commit/017a9dcf17e572f9b7c32aa62a81df8ef41cef17
But I can't submit it as a test yet.

Quoted 1 line
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

So you'd need to validate local_id directly against number of types in local_btf.

How is this better than a null check?

because id check will be useful for both kernel and libbpf sides?..

This would require a special case for BPF_CORE_TYPE_ID_LOCAL in bpf_core_calc_relo_insn(). If you don't like this null check I'll modify bpf_core_apply() instead to always check relo->type_id and report and error.

This would also be in the line with what bpf_core_resolve_relo() does on libbpf side.

[...]

#5 lore raw
Quoted 71 lines · Eduard Zingerman
Quoted 61 lines · Andrii Nakryiko
Quoted 60 lines · Eduard Zingerman

On Wed, 2024-08-21 at 09:59 -0700, Andrii Nakryiko wrote:

[...]

Quoted 14 lines
Quoted 14 lines
---
 tools/lib/bpf/relo_core.c | 5 +++++
 1 file changed, 5 insertions(+)
tools/lib/bpf/relo_core.c +0 -0
index 63a4d5ad12d1..a04724831ebc 100644
@@ -1297,6 +1297,11 @@ int bpf_core_calc_relo_insn(const char *prog_name,
 
Quoted 7 lines

tools/lib/bpf/relo_core.c int bpf_core_calc_relo_insn(const char *prog_name, · line 1300

Quoted 3 lines
tools/lib/bpf/relo_core.c +1 -0
        local_id = relo->type_id;
        local_type = btf_type_by_id(local_btf, local_id);
+       if (!local_type) {

This is a meaningless check at least for libbpf's implementation of btf_type_by_id(), it never returns NULL. Commit you point to in Fixes tag clearly states the differences.

That is not true on kernel side. bpf_core_calc_relo_insn() is called from bpf_core_apply():

int bpf_core_apply(struct bpf_core_ctx *ctx, const struct bpf_core_relo *relo,
int relo_idx, void *insn)

{
        bool need_cands = relo->kind != BPF_CORE_TYPE_ID_LOCAL;
        ...
        if (need_cands) {
                ...
                // code below would report an error if relo->type_id is bogus
                cc = bpf_core_find_cands(ctx, relo->type_id);
                if (IS_ERR(cc)) {
                        bpf_log(ctx->log, "target candidate search failed for %d\n",
                                relo->type_id);
                        err = PTR_ERR(cc);
                        goto out;
                }
                ...
        }

        err = bpf_core_calc_relo_insn((void *)ctx->log, relo, relo_idx, ctx->btf, &cands, specs,
                                      &targ_res);
        ...
}

If need_cands is false the bogus type_id could reach into bpf_core_calc_relo_insn(). Which is exactly the case with repro submitted by Liu.
There is also a simplified repro here:
https://github.com/kernel-patches/bpf/commit/017a9dcf17e572f9b7c32aa62a81df8ef41cef17
But I can't submit it as a test yet.

Quoted 1 line
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

So you'd need to validate local_id directly against number of types in local_btf.

How is this better than a null check?

because id check will be useful for both kernel and libbpf sides?..

Quoted 7 lines

This would require a special case for BPF_CORE_TYPE_ID_LOCAL in bpf_core_calc_relo_insn(). If you don't like this null check I'll modify bpf_core_apply() instead to always check relo->type_id and report and error.

This would also be in the line with what bpf_core_resolve_relo() does on libbpf side.

Ok, then let's do that. I don't want static analysers complaining about this when checking libbpf code base.

Quoted 3 lines

[...]

author maintainer
#6 lore raw

On Wed, 2024-08-21 at 13:08 -0700, Andrii Nakryiko wrote:

[...]

Quoted 2 lines

Ok, then let's do that. I don't want static analysers complaining about this when checking libbpf code base.

Just want to rant a bit.
Here is a footgun in the relo_core.c:

    #ifdef __KERNEL__
    ...
    #undef pr_warn
    #undef pr_info
    #undef pr_debug
    #define pr_warn(fmt, log, ...)      bpf_log((void *)log, fmt, "", ##__VA_ARGS__)
    #define pr_info(fmt, log, ...)      bpf_log((void *)log, fmt, "", ##__VA_ARGS__)
    #define pr_debug(fmt, log, ...)     bpf_log((void *)log, fmt, "", ##__VA_ARGS__)
                          ^^^                   ^^^^^^^^^^^       ^^
                     first format param,        prog_name         replacement for
                     usually prog_name          cast to           first param
                                                verifier log
    ...
    #else
    ...
    #endif

    int bpf_core_calc_relo_insn(const char *prog_name, ...)
    {
        ...
        pr_warn("prog '%s': relo #%d: bad type id %u\n",
                prog_name, relo_idx, local_id);
        ...
    }

And in the verifier.c:

    err = bpf_core_calc_relo_insn((void *)ctx->log, relo, ...);
                                  ^^^^^^^^^^^^^^^^
                                  This is a prog_name parameter

Just spent more than an hour trying to figure out why passing real program name (char *) does not work.
I'll think on a refactoring, but that is for another series.

author maintainer
#7 lore raw

On Wed, 2024-08-21 at 15:32 -0700, Eduard Zingerman wrote:

[...]

Quoted 3 lines

Just spent more than an hour trying to figure out why passing real program name (char *) does not work.
I'll think on a refactoring, but that is for another series.

I passed env->prog->aux->name there, and pr_warn caused corruption of the aux structure. Sigh.