Skip to contentSkip to conversation

libbpf: Fix perm errors for LDIMM_64_FULL_RANGE_OFF use

PATCH Applied Emil Tsalapatis 3 participants 8 messages Feb 14 – Feb 17

Subsystems (2)
lore
Share & tools
lore

Emil Tsalapatis

author reviewer
#0 lore raw

cover libbpf: Fix perm errors for LDIMM_64_FULL_RANGE_OFF use

Commit 728ff167910e ("libbpf: Add gating for arena globals relocation feature") adds a feature flag for testing whether the running kernel supports LDIMM64 instructions with large direct offsets. Fix two edge cases that can cause unexpected -EPERM errors in two ways:

1) The probe program used for the feature has type TRACEPOINT, but it's possible the caller does not have permission to load it, even if it is able to do so for generic BPF programs. Use the SOCKET_FILTER type instead that requires fewer permissions. This does not affect the check itself, which will always fail verification anyway.

2) The probe is triggered during bpf_object__collect_relos(), itself called in bpf_object_open(), to compute the arena relocation offsets of arena variables. However, the caller may not have permissions to load BPF programs. This is the case in some systems with the bpftool calls made by the BPF selftests during compilation, e.g., for skeleton generation. Move all uses of the feature check to bpf_object_prepare() time instead.

Fixes 728ff167910e ("libbpf: Add gating for arena globals relocation feature")

v1 -> v2: https://lore.kernel.org/bpf/[email protected]/

- Adjust the hash of the original commit post-tree rebase
- Ensure close() is not called on invalid prog_fd in feature probe (Coverity)

Emil Tsalapatis (2):
libbpf: Do not use PROG_TYPE_TRACEPOINT program for feature gating
libbpf: Delay feature gate check until object prepare time

 tools/lib/bpf/features.c |  6 +++---
 tools/lib/bpf/libbpf.c   | 31 +++++++++++++++++++++++--------
 2 files changed, 26 insertions(+), 11 deletions(-)
Signature · 1 line
2.49.0
author reviewer
#1 lore raw

PATCH v2 1/2 libbpf: Do not use PROG_TYPE_TRACEPOINT program for feature gating

Commit 728ff167910e uses a PROG_TYPE_TRACEPOINT BPF test program to check whether the running kernel supports large LDIMM64 offsets. The feature gate incorrectly assumes that the program will fail at verification time with one of two messages, depending on whether the feature is supported by the running kernel. However, PROG_TYPE_TRACEPOINT programs may fail to load before verification even starts, e.g., if the shell does not have the appropriate capabilities. Use a BPF_PROG_TYPE_SOCKET_FILTER program for the feature gate instead.

Also fix two minor issues. First, ensure the log buffer for the test is
initialized: Failing program load before verification led to libbpf dumping uninitialized data to stdout. Also, ensure that close() is only called for program_fd in the probe if the program load actually succeeded. The call was currently failing silently with -EBADF most of the time.

Fixes 728ff167910e ("libbpf: Add gating for arena globals relocation feature")
---
 tools/lib/bpf/features.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
tools/lib/bpf/features.c +3 -3
index b65ab109e3ff..cf3937e08520 100644
@@ -508,7 +508,7 @@ static int probe_kern_arg_ctx_tag(int token_fd)
 
 static int probe_ldimm64_full_range_off(int token_fd)
 {
-	char log_buf[1024];
+	char log_buf[1024] = {};
 	int prog_fd, map_fd;
 	int ret;
 	LIBBPF_OPTS(bpf_map_create_opts, map_opts,
@@ -536,14 +536,14 @@ static int probe_ldimm64_full_range_off(int token_fd)
 	}
 	insns[0].imm = map_fd;
 
-	prog_fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, "global_reloc", "GPL", insns, insn_cnt, &prog_opts);
+	prog_fd = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, "global_reloc", "GPL", insns, insn_cnt, &prog_opts);
 	ret = -errno;
 
 	close(map_fd);
-	close(prog_fd);
 
 	if (prog_fd >= 0) {
 		pr_warn("Error in %s(): Program loading unexpectedly succeeded.\n", __func__);
+		close(prog_fd);
 		return -EINVAL;
 	}
 
Signature · 1 line
2.49.0
#2 lore raw

Re: [PATCH bpf v2 1/2] libbpf: Do not use PROG_TYPE_TRACEPOINT program for feature gating

Quoted 28 lines · Emil Tsalapatis

Commit 728ff167910e uses a PROG_TYPE_TRACEPOINT BPF test program to check whether the running kernel supports large LDIMM64 offsets. The feature gate incorrectly assumes that the program will fail at verification time with one of two messages, depending on whether the feature is supported by the running kernel. However, PROG_TYPE_TRACEPOINT programs may fail to load before verification even starts, e.g., if the shell does not have the appropriate capabilities. Use a BPF_PROG_TYPE_SOCKET_FILTER program for the feature gate instead.

Also fix two minor issues. First, ensure the log buffer for the test is
initialized: Failing program load before verification led to libbpf dumping uninitialized data to stdout. Also, ensure that close() is only called for program_fd in the probe if the program load actually succeeded. The call was currently failing silently with -EBADF most of the time.

Fixes 728ff167910e ("libbpf: Add gating for arena globals relocation feature")
---
 tools/lib/bpf/features.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
tools/lib/bpf/features.c +0 -0
index b65ab109e3ff..cf3937e08520 100644

tools/lib/bpf/features.c static int probe_kern_arg_ctx_tag(int token_fd) · line 511

Quoted 5 lines
tools/lib/bpf/features.c +1 -1
@@ -508,7 +508,7 @@ static int probe_kern_arg_ctx_tag(int token_fd)
 
 static int probe_ldimm64_full_range_off(int token_fd)
 {
-       char log_buf[1024];
+       char log_buf[1024] = {};

minor nit: this will clear whole log_buf, but we really just need to set first character to zero, so instead I'd do

log_buf[0] = '\0';

before program load

minor, because this is not performance critical code, but it's a bit wasteful nevertheless

Quoted 24 lines
tools/lib/bpf/features.c +2 -3
        int prog_fd, map_fd;
        int ret;
        LIBBPF_OPTS(bpf_map_create_opts, map_opts,
@@ -536,14 +536,14 @@ static int probe_ldimm64_full_range_off(int token_fd)
        }
        insns[0].imm = map_fd;
 
-       prog_fd = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, "global_reloc", "GPL", insns, insn_cnt, &prog_opts);
+       prog_fd = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, "global_reloc", "GPL", insns, insn_cnt, &prog_opts);
        ret = -errno;
 
        close(map_fd);
-       close(prog_fd);
 
        if (prog_fd >= 0) {
                pr_warn("Error in %s(): Program loading unexpectedly succeeded.\n", __func__);
+               close(prog_fd);
                return -EINVAL;
        }
 
--

2.49.0

author reviewer
#3 lore raw

PATCH v2 2/2 libbpf: Delay feature gate check until object prepare time

In reply to Emil Tsalapatis

Commit 728ff167910e ("libbpf: Add gating for arena globals relocation feature") adds a feature gate check that loads a map and BPF program to
test the running kernel supports large direct offsets for LDIMM64 instructions. This check is currently used to calculate arena symbol offsets during bpf_object__collect_relos, itself called by bpf_object_open.

However, the program calling bpf_object_open may not have the permissions to load maps and programs. This is the case with the BPF selftests, where bpftool is invoked at compilation time during skeleton generation. This causes errors as the feature gate unexpectedly fails with -EPERM.

Avoid this by moving all uses of the FEAT_LDIMM64_FULL_RANGE_OFF feature gate to BPF object preparation time instead.

Fixes 728ff167910e ("libbpf: Add gating for arena globals relocation feature")
---
 tools/lib/bpf/libbpf.c | 31 +++++++++++++++++++++++--------
 1 file changed, 23 insertions(+), 8 deletions(-)
tools/lib/bpf/libbpf.c +23 -8
index 93e59ed8d9a1..ff38a9db1b13 100644
@@ -3009,11 +3009,11 @@ static int init_arena_map_data(struct bpf_object *obj, struct bpf_map *map,
 	memcpy(obj->arena_data, data, data_sz);
 	obj->arena_data_sz = data_sz;
 
-	/* place globals at the end of the arena (if supported) */
-	if (kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF))
-		obj->arena_data_off = mmap_sz - data_alloc_sz;
-	else
-		obj->arena_data_off = 0;
+	/*
+	 * possible offset for arena globals into the arena,
+	 * if supported at load time
+	 */
+	obj->arena_data_off = mmap_sz - data_alloc_sz;
 
 	/* make bpf_map__init_value() work for ARENA maps */
 	map->mmaped = obj->arena_data;
@@ -4672,7 +4672,7 @@ static int bpf_program__record_reloc(struct bpf_program *prog,
 		reloc_desc->type = RELO_DATA;
 		reloc_desc->insn_idx = insn_idx;
 		reloc_desc->map_idx = obj->arena_map_idx;
-		reloc_desc->sym_off = sym->st_value + obj->arena_data_off;
+		reloc_desc->sym_off = sym->st_value;
 
 		map = &obj->maps[obj->arena_map_idx];
 		pr_debug("prog '%s': found arena map %d (%s, sec %d, off %zu) for insn %u\n",
@@ -5556,6 +5556,7 @@ bpf_object__create_maps(struct bpf_object *obj)
 {
 	struct bpf_map *map;
 	unsigned int i, j;
+	char *arena_globals;
 	int err;
 	bool retried;
 
@@ -5633,7 +5634,15 @@ bpf_object__create_maps(struct bpf_object *obj)
 					return err;
 				}
 				if (obj->arena_data) {
-					memcpy(map->mmaped + obj->arena_data_off, obj->arena_data,
+					/*
+					 * Only relocate the globals to the end of the arena
+					 * if the kernel supports it.
+					 */
+					arena_globals = map->mmaped;
+					if (kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF))
+						arena_globals += obj->arena_data_off;
+
+					memcpy(arena_globals, obj->arena_data,
 						obj->arena_data_sz);
 					zfree(&obj->arena_data);
 				}
@@ -6386,6 +6395,12 @@ bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
 		case RELO_DATA:
 			map = &obj->maps[relo->map_idx];
 			insn[1].imm = insn[0].imm + relo->sym_off;
+
+			if (relo->map_idx == obj->arena_map_idx &&
+			    kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF)) {
+				insn[1].imm += obj->arena_data_off;
+			}
+
 			if (obj->gen_loader) {
 				insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
 				insn[0].imm = relo->map_idx;
@@ -14456,7 +14471,7 @@ int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
 		if (!map_skel->mmaped)
 			continue;
 
-		if (map->def.type == BPF_MAP_TYPE_ARENA)
+		if (map->def.type == BPF_MAP_TYPE_ARENA && kernel_supports(*s->obj, FEAT_LDIMM64_FULL_RANGE_OFF))
 			*map_skel->mmaped = map->mmaped + map->obj->arena_data_off;
 		else
 			*map_skel->mmaped = map->mmaped;
Signature · 1 line
2.49.0
#4 lore raw

Re: [PATCH bpf v2 2/2] libbpf: Delay feature gate check until object prepare time

Quoted 13 lines · Emil Tsalapatis

Commit 728ff167910e ("libbpf: Add gating for arena globals relocation feature") adds a feature gate check that loads a map and BPF program to
test the running kernel supports large direct offsets for LDIMM64 instructions. This check is currently used to calculate arena symbol offsets during bpf_object__collect_relos, itself called by bpf_object_open.

However, the program calling bpf_object_open may not have the permissions to load maps and programs. This is the case with the BPF selftests, where bpftool is invoked at compilation time during skeleton generation. This causes errors as the feature gate unexpectedly fails with -EPERM.

Quoted 9 lines

Avoid this by moving all uses of the FEAT_LDIMM64_FULL_RANGE_OFF feature gate to BPF object preparation time instead.

Fixes 728ff167910e ("libbpf: Add gating for arena globals relocation feature")
---
 tools/lib/bpf/libbpf.c | 31 +++++++++++++++++++++++--------
 1 file changed, 23 insertions(+), 8 deletions(-)

This works, but see some suggestions for a follow up to contain feature-checking a bit better

Quoted 8 lines
tools/lib/bpf/libbpf.c +0 -0
index 93e59ed8d9a1..ff38a9db1b13 100644
@@ -3009,11 +3009,11 @@ static int init_arena_map_data(struct bpf_object *obj, struct bpf_map *map,
        memcpy(obj->arena_data, data, data_sz);
        obj->arena_data_sz = data_sz;
 

tools/lib/bpf/libbpf.c static int init_arena_map_data(struct bpf_object *obj, struct bpf_map *map, · line 3016

Quoted 10 lines
tools/lib/bpf/libbpf.c +5 -5
-       /* place globals at the end of the arena (if supported) */
-       if (kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF))
-               obj->arena_data_off = mmap_sz - data_alloc_sz;
-       else
-               obj->arena_data_off = 0;
+       /*
+        * possible offset for arena globals into the arena,
+        * if supported at load time
+        */
+       obj->arena_data_off = mmap_sz - data_alloc_sz;

I'd keep this obj->arena_data_off at zero at this point (we don't use it until prepare step, skeleton also just uses obj->arena_data directly without offset during open phase)...

tools/lib/bpf/libbpf.c static int bpf_program__record_reloc(struct bpf_program *prog, · line 4675

Quoted 9 lines
tools/lib/bpf/libbpf.c +1 -1
        /* make bpf_map__init_value() work for ARENA maps */
        map->mmaped = obj->arena_data;
@@ -4672,7 +4672,7 @@ static int bpf_program__record_reloc(struct bpf_program *prog,
                reloc_desc->type = RELO_DATA;
                reloc_desc->insn_idx = insn_idx;
                reloc_desc->map_idx = obj->arena_map_idx;
-               reloc_desc->sym_off = sym->st_value + obj->arena_data_off;
+               reloc_desc->sym_off = sym->st_value;

this is good, relos don't need to record this offset...

Quoted 13 lines
tools/lib/bpf/libbpf.c +1 -0
                map = &obj->maps[obj->arena_map_idx];
                pr_debug("prog '%s': found arena map %d (%s, sec %d, off %zu) for insn %u\n",
@@ -5556,6 +5556,7 @@ bpf_object__create_maps(struct bpf_object *obj)
 {
        struct bpf_map *map;
        unsigned int i, j;
+       char *arena_globals;
        int err;
        bool retried;
 
@@ -5633,7 +5634,15 @@ bpf_object__create_maps(struct bpf_object *obj)
                                        return err;

tools/lib/bpf/libbpf.c bpf_object__create_maps(struct bpf_object *obj) · line 5643

Quoted 10 lines
tools/lib/bpf/libbpf.c +7 -1
                                }
                                if (obj->arena_data) {
-                                       memcpy(map->mmaped + obj->arena_data_off, obj->arena_data,
+                                       /*
+                                        * Only relocate the globals to the end of the arena
+                                        * if the kernel supports it.
+                                        */
+                                       arena_globals = map->mmaped;
+                                       if (kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF))
+                                               arena_globals += obj->arena_data_off;

and here I'd instead set arena_data_off to its final value (zero or proper offset), something like

if (kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF)

    obj->arena_data_off = bpf_map_mmap_sz(map) -

roundup(sysconf(_SC_PAGE_SIZE), sysconf(_SC_PAGE_SIZE));

(note, we did all the correctness checks early in init_arena_map_data, we don't have to do any more here)

and then the rest of the code stays exactly the same, down to the original memcpy(map->mmaped + obj->arena_data_off, ...), and no extra variables

Quoted 6 lines
tools/lib/bpf/libbpf.c +2 -0
+
+                                       memcpy(arena_globals, obj->arena_data,
                                                obj->arena_data_sz);
                                        zfree(&obj->arena_data);
                                }

tools/lib/bpf/libbpf.c bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog) · line 6403

Quoted 9 lines
tools/lib/bpf/libbpf.c +6 -0
@@ -6386,6 +6395,12 @@ bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
                case RELO_DATA:
                        map = &obj->maps[relo->map_idx];
                        insn[1].imm = insn[0].imm + relo->sym_off;
+
+                       if (relo->map_idx == obj->arena_map_idx &&
+                           kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF)) {
+                               insn[1].imm += obj->arena_data_off;
+                       }
+

here too we don't need feature checking, our obj->arena_data_off should have a valid value (might be zero, that's fine)

tools/lib/bpf/libbpf.c int bpf_object__load_skeleton(struct bpf_object_skeleton *s) · line 14475

Quoted 10 lines
tools/lib/bpf/libbpf.c +1 -1
                        if (obj->gen_loader) {
                                insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
                                insn[0].imm = relo->map_idx;
@@ -14456,7 +14471,7 @@ int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
                if (!map_skel->mmaped)
                        continue;
 
-               if (map->def.type == BPF_MAP_TYPE_ARENA)
+               if (map->def.type == BPF_MAP_TYPE_ARENA && kernel_supports(*s->obj, FEAT_LDIMM64_FULL_RANGE_OFF))
                        *map_skel->mmaped = map->mmaped + map->obj->arena_data_off;

and here as well, no need to expose feature check here, we have map->obj->arena_data_off set correctly

Quoted 5 lines
tools/lib/bpf/libbpf.c +0 -1
                else
                        *map_skel->mmaped = map->mmaped;
--

2.49.0

author reviewer
#5 lore raw

Re: [PATCH bpf v2 2/2] libbpf: Delay feature gate check until object prepare time

Quoted 17 lines · Andrii Nakryiko
Quoted 16 lines · Emil Tsalapatis

Commit 728ff167910e ("libbpf: Add gating for arena globals relocation feature") adds a feature gate check that loads a map and BPF program to
test the running kernel supports large direct offsets for LDIMM64 instructions. This check is currently used to calculate arena symbol offsets during bpf_object__collect_relos, itself called by bpf_object_open.

However, the program calling bpf_object_open may not have the permissions to load maps and programs. This is the case with the BPF selftests, where bpftool is invoked at compilation time during skeleton generation. This causes errors as the feature gate unexpectedly fails with -EPERM.

Avoid this by moving all uses of the FEAT_LDIMM64_FULL_RANGE_OFF feature gate to BPF object preparation time instead.

Quoted 6 lines
Quoted 6 lines
Fixes 728ff167910e ("libbpf: Add gating for arena globals relocation feature")
---
 tools/lib/bpf/libbpf.c | 31 +++++++++++++++++++++++--------
 1 file changed, 23 insertions(+), 8 deletions(-)

Hi Andrii,

ack to this and the point for the first patch, two minor followups below.

Quoted 12 lines

This works, but see some suggestions for a follow up to contain feature-checking a bit better

Quoted 8 lines
tools/lib/bpf/libbpf.c +0 -0
index 93e59ed8d9a1..ff38a9db1b13 100644
@@ -3009,11 +3009,11 @@ static int init_arena_map_data(struct bpf_object *obj, struct bpf_map *map,
        memcpy(obj->arena_data, data, data_sz);
        obj->arena_data_sz = data_sz;
 
Quoted 15 lines

tools/lib/bpf/libbpf.c static int init_arena_map_data(struct bpf_object *obj, struct bpf_map *map, · line 3016

Quoted 10 lines
tools/lib/bpf/libbpf.c +5 -5
-       /* place globals at the end of the arena (if supported) */
-       if (kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF))
-               obj->arena_data_off = mmap_sz - data_alloc_sz;
-       else
-               obj->arena_data_off = 0;
+       /*
+        * possible offset for arena globals into the arena,
+        * if supported at load time
+        */
+       obj->arena_data_off = mmap_sz - data_alloc_sz;

I'd keep this obj->arena_data_off at zero at this point (we don't use it until prepare step, skeleton also just uses obj->arena_data directly without offset during open phase)...

Do we even want to keep the assignment? The field should be 0 by default since we're calloc()'ing the object.

Quoted 39 lines

tools/lib/bpf/libbpf.c static int bpf_program__record_reloc(struct bpf_program *prog, · line 4675

Quoted 9 lines
tools/lib/bpf/libbpf.c +1 -1
        /* make bpf_map__init_value() work for ARENA maps */
        map->mmaped = obj->arena_data;
@@ -4672,7 +4672,7 @@ static int bpf_program__record_reloc(struct bpf_program *prog,
                reloc_desc->type = RELO_DATA;
                reloc_desc->insn_idx = insn_idx;
                reloc_desc->map_idx = obj->arena_map_idx;
-               reloc_desc->sym_off = sym->st_value + obj->arena_data_off;
+               reloc_desc->sym_off = sym->st_value;

this is good, relos don't need to record this offset...

Quoted 13 lines
tools/lib/bpf/libbpf.c +1 -0
                map = &obj->maps[obj->arena_map_idx];
                pr_debug("prog '%s': found arena map %d (%s, sec %d, off %zu) for insn %u\n",
@@ -5556,6 +5556,7 @@ bpf_object__create_maps(struct bpf_object *obj)
 {
        struct bpf_map *map;
        unsigned int i, j;
+       char *arena_globals;
        int err;
        bool retried;
 
@@ -5633,7 +5634,15 @@ bpf_object__create_maps(struct bpf_object *obj)
                                        return err;

tools/lib/bpf/libbpf.c bpf_object__create_maps(struct bpf_object *obj) · line 5643

Quoted 10 lines
tools/lib/bpf/libbpf.c +7 -1
                                }
                                if (obj->arena_data) {
-                                       memcpy(map->mmaped + obj->arena_data_off, obj->arena_data,
+                                       /*
+                                        * Only relocate the globals to the end of the arena
+                                        * if the kernel supports it.
+                                        */
+                                       arena_globals = map->mmaped;
+                                       if (kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF))
+                                               arena_globals += obj->arena_data_off;

and here I'd instead set arena_data_off to its final value (zero or proper offset), something like

Quoted 4 lines

if (kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF)

    obj->arena_data_off = bpf_map_mmap_sz(map) -

roundup(sysconf(_SC_PAGE_SIZE), sysconf(_SC_PAGE_SIZE));

(I assume roundup is roundup(arena->arena_data_sz, sysconf(_SC_PAGE_SIZE))

I assume the trivial duplication of recomputing the rounded up value here and in init_arena_map_data is fine - rounding up data_sz to the nearest page is verbose but easy to understand imo.

Quoted 46 lines

(note, we did all the correctness checks early in init_arena_map_data, we don't have to do any more here)

and then the rest of the code stays exactly the same, down to the original memcpy(map->mmaped + obj->arena_data_off, ...), and no extra variables

Quoted 6 lines
tools/lib/bpf/libbpf.c +2 -0
+
+                                       memcpy(arena_globals, obj->arena_data,
                                                obj->arena_data_sz);
                                        zfree(&obj->arena_data);
                                }

tools/lib/bpf/libbpf.c bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog) · line 6403

Quoted 9 lines
tools/lib/bpf/libbpf.c +6 -0
@@ -6386,6 +6395,12 @@ bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
                case RELO_DATA:
                        map = &obj->maps[relo->map_idx];
                        insn[1].imm = insn[0].imm + relo->sym_off;
+
+                       if (relo->map_idx == obj->arena_map_idx &&
+                           kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF)) {
+                               insn[1].imm += obj->arena_data_off;
+                       }
+

here too we don't need feature checking, our obj->arena_data_off should have a valid value (might be zero, that's fine)

tools/lib/bpf/libbpf.c int bpf_object__load_skeleton(struct bpf_object_skeleton *s) · line 14475

Quoted 10 lines
tools/lib/bpf/libbpf.c +1 -1
                        if (obj->gen_loader) {
                                insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
                                insn[0].imm = relo->map_idx;
@@ -14456,7 +14471,7 @@ int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
                if (!map_skel->mmaped)
                        continue;
 
-               if (map->def.type == BPF_MAP_TYPE_ARENA)
+               if (map->def.type == BPF_MAP_TYPE_ARENA && kernel_supports(*s->obj, FEAT_LDIMM64_FULL_RANGE_OFF))
                        *map_skel->mmaped = map->mmaped + map->obj->arena_data_off;

and here as well, no need to expose feature check here, we have map->obj->arena_data_off set correctly

Quoted 5 lines
tools/lib/bpf/libbpf.c +0 -1
                else
                        *map_skel->mmaped = map->mmaped;
--

2.49.0

author reviewer
#6 lore raw

Re: [PATCH bpf v2 2/2] libbpf: Delay feature gate check until object prepare time

Quoted 104 lines · Emil Tsalapatis
Quoted 17 lines · Andrii Nakryiko
Quoted 16 lines · Emil Tsalapatis

Commit 728ff167910e ("libbpf: Add gating for arena globals relocation feature") adds a feature gate check that loads a map and BPF program to
test the running kernel supports large direct offsets for LDIMM64 instructions. This check is currently used to calculate arena symbol offsets during bpf_object__collect_relos, itself called by bpf_object_open.

However, the program calling bpf_object_open may not have the permissions to load maps and programs. This is the case with the BPF selftests, where bpftool is invoked at compilation time during skeleton generation. This causes errors as the feature gate unexpectedly fails with -EPERM.

Avoid this by moving all uses of the FEAT_LDIMM64_FULL_RANGE_OFF feature gate to BPF object preparation time instead.

Quoted 6 lines
Quoted 6 lines
Fixes 728ff167910e ("libbpf: Add gating for arena globals relocation feature")
---
 tools/lib/bpf/libbpf.c | 31 +++++++++++++++++++++++--------
 1 file changed, 23 insertions(+), 8 deletions(-)

Hi Andrii,

ack to this and the point for the first patch, two minor followups below.

Quoted 12 lines

This works, but see some suggestions for a follow up to contain feature-checking a bit better

Quoted 8 lines
tools/lib/bpf/libbpf.c +0 -0
index 93e59ed8d9a1..ff38a9db1b13 100644
@@ -3009,11 +3009,11 @@ static int init_arena_map_data(struct bpf_object *obj, struct bpf_map *map,
        memcpy(obj->arena_data, data, data_sz);
        obj->arena_data_sz = data_sz;
 
Quoted 15 lines

tools/lib/bpf/libbpf.c static int init_arena_map_data(struct bpf_object *obj, struct bpf_map *map, · line 3016

Quoted 10 lines
tools/lib/bpf/libbpf.c +5 -5
-       /* place globals at the end of the arena (if supported) */
-       if (kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF))
-               obj->arena_data_off = mmap_sz - data_alloc_sz;
-       else
-               obj->arena_data_off = 0;
+       /*
+        * possible offset for arena globals into the arena,
+        * if supported at load time
+        */
+       obj->arena_data_off = mmap_sz - data_alloc_sz;

I'd keep this obj->arena_data_off at zero at this point (we don't use it until prepare step, skeleton also just uses obj->arena_data directly without offset during open phase)...

Do we even want to keep the assignment? The field should be 0 by default since we're calloc()'ing the object.

Quoted 39 lines

tools/lib/bpf/libbpf.c static int bpf_program__record_reloc(struct bpf_program *prog, · line 4675

Quoted 9 lines
tools/lib/bpf/libbpf.c +1 -1
        /* make bpf_map__init_value() work for ARENA maps */
        map->mmaped = obj->arena_data;
@@ -4672,7 +4672,7 @@ static int bpf_program__record_reloc(struct bpf_program *prog,
                reloc_desc->type = RELO_DATA;
                reloc_desc->insn_idx = insn_idx;
                reloc_desc->map_idx = obj->arena_map_idx;
-               reloc_desc->sym_off = sym->st_value + obj->arena_data_off;
+               reloc_desc->sym_off = sym->st_value;

this is good, relos don't need to record this offset...

Quoted 13 lines
tools/lib/bpf/libbpf.c +1 -0
                map = &obj->maps[obj->arena_map_idx];
                pr_debug("prog '%s': found arena map %d (%s, sec %d, off %zu) for insn %u\n",
@@ -5556,6 +5556,7 @@ bpf_object__create_maps(struct bpf_object *obj)
 {
        struct bpf_map *map;
        unsigned int i, j;
+       char *arena_globals;
        int err;
        bool retried;
 
@@ -5633,7 +5634,15 @@ bpf_object__create_maps(struct bpf_object *obj)
                                        return err;

tools/lib/bpf/libbpf.c bpf_object__create_maps(struct bpf_object *obj) · line 5643

Quoted 10 lines
tools/lib/bpf/libbpf.c +7 -1
                                }
                                if (obj->arena_data) {
-                                       memcpy(map->mmaped + obj->arena_data_off, obj->arena_data,
+                                       /*
+                                        * Only relocate the globals to the end of the arena
+                                        * if the kernel supports it.
+                                        */
+                                       arena_globals = map->mmaped;
+                                       if (kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF))
+                                               arena_globals += obj->arena_data_off;

and here I'd instead set arena_data_off to its final value (zero or proper offset), something like

Quoted 4 lines

if (kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF)

    obj->arena_data_off = bpf_map_mmap_sz(map) -

roundup(sysconf(_SC_PAGE_SIZE), sysconf(_SC_PAGE_SIZE));

Quoted 7 lines

(I assume roundup is roundup(arena->arena_data_sz, sysconf(_SC_PAGE_SIZE))

I assume the trivial duplication of recomputing the rounded up value here and in init_arena_map_data is fine - rounding up data_sz to the nearest page is verbose but easy to understand imo.

Actually there is an issue: If we move the flag check here, then we haven't properly calculated arena_data_off by the time we do the relocations in bpf_object__relocate_data(). I'll just put the single feature in bpf_object__realocate instead.

Quoted 46 lines
Quoted 46 lines

(note, we did all the correctness checks early in init_arena_map_data, we don't have to do any more here)

and then the rest of the code stays exactly the same, down to the original memcpy(map->mmaped + obj->arena_data_off, ...), and no extra variables

Quoted 6 lines
tools/lib/bpf/libbpf.c +2 -0
+
+                                       memcpy(arena_globals, obj->arena_data,
                                                obj->arena_data_sz);
                                        zfree(&obj->arena_data);
                                }

tools/lib/bpf/libbpf.c bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog) · line 6403

Quoted 9 lines
tools/lib/bpf/libbpf.c +6 -0
@@ -6386,6 +6395,12 @@ bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
                case RELO_DATA:
                        map = &obj->maps[relo->map_idx];
                        insn[1].imm = insn[0].imm + relo->sym_off;
+
+                       if (relo->map_idx == obj->arena_map_idx &&
+                           kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF)) {
+                               insn[1].imm += obj->arena_data_off;
+                       }
+

here too we don't need feature checking, our obj->arena_data_off should have a valid value (might be zero, that's fine)

tools/lib/bpf/libbpf.c int bpf_object__load_skeleton(struct bpf_object_skeleton *s) · line 14475

Quoted 10 lines
tools/lib/bpf/libbpf.c +1 -1
                        if (obj->gen_loader) {
                                insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
                                insn[0].imm = relo->map_idx;
@@ -14456,7 +14471,7 @@ int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
                if (!map_skel->mmaped)
                        continue;
 
-               if (map->def.type == BPF_MAP_TYPE_ARENA)
+               if (map->def.type == BPF_MAP_TYPE_ARENA && kernel_supports(*s->obj, FEAT_LDIMM64_FULL_RANGE_OFF))
                        *map_skel->mmaped = map->mmaped + map->obj->arena_data_off;

and here as well, no need to expose feature check here, we have map->obj->arena_data_off set correctly

Quoted 5 lines
tools/lib/bpf/libbpf.c +0 -1
                else
                        *map_skel->mmaped = map->mmaped;
--

2.49.0

#7 lore raw
In reply to Emil Tsalapatis
This series was applied to bpf/bpf.git (master) by Alexei Starovoitov <[email protected]>:

Hello:

This series was applied to bpf/bpf.git (master)
by Alexei Starovoitov <[email protected]>:

Quoted 5 lines · On Fri, 13 Feb 2026 21:10:11 -0500 you wrote:

Commit 728ff167910e ("libbpf: Add gating for arena globals relocation feature") adds a feature flag for testing whether the running kernel supports LDIMM64 instructions with large direct offsets. Fix two edge cases that can cause unexpected -EPERM errors in two ways:

Quoted 7 lines

1) The probe program used for the feature has type TRACEPOINT, but it's possible the caller does not have permission to load it, even if it is able to do so for generic BPF programs. Use the SOCKET_FILTER type instead that requires fewer permissions. This does not affect the check itself, which will always fail verification anyway.

[...]

Here is the summary with links:
- [bpf,v2,1/2] libbpf: Do not use PROG_TYPE_TRACEPOINT program for feature gating

    https://git.kernel.org/bpf/bpf/c/623e7b898a0b
  - [bpf,v2,2/2] libbpf: Delay feature gate check until object prepare time
    https://git.kernel.org/bpf/bpf/c/9e07e5f57d93

You are awesome, thank you!

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