Skip to contentSkip to conversation

bpf: Fix Null-Pointer Dereference in kernel_clone() via BPF fmod_ret on security_task_alloc

PATCH In review Feng Yang 3 participants 8 messages Apr 11 – Apr 21

Subsystems (4)
lore
Share & tools
lore

Feng Yang

author
#0 lore raw

v3 cover bpf: Fix Null-Pointer Dereference in kernel_clone() via BPF fmod_ret on security_task_alloc

From: Feng Yang <[email protected]>

This patch set adds return value validation for fmod_ret to prevent system crashes caused by incorrect return values.

Changes in v3:
- Do not move the code, and make some formatting changes. Thanks, Leon Hwang.
- Error injection always ensures that 0 is a valid return value. Thanks, Menglong Dong.
- Link to v2: https://lore.kernel.org/all/[email protected]/
Changes in v2:
- Add validation for fault injection and include selftests. Thanks, Jiri Olsa, Jiayuan Chen.
- Link to v1: https://lore.kernel.org/all/[email protected]/

Feng Yang (2):
bpf: Fix Null-Pointer Dereference in kernel_clone() via BPF fmod_ret
on security_task_alloc
selftests/bpf: Add selftests for verifying return values of fmod_ret.

 kernel/bpf/verifier.c                         | 114 +++++++++++++++++-
 .../selftests/bpf/prog_tests/verifier.c       |   2 +
 .../bpf/progs/verifier_fmod_ret_return.c      |  59 +++++++++
 3 files changed, 173 insertions(+), 2 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/progs/verifier_fmod_ret_return.c
Signature · 1 line
2.43.0
author
#1 lore raw

PATCH v3 1/2 bpf: Fix Null-Pointer Dereference in kernel_clone() via BPF fmod_ret on security_task_alloc

From: Feng Yang <[email protected]>

Using the following BPF program will cause a kernel panic:
SEC("fmod_ret/security_task_alloc")
int fmod_task_alloc(void *ctx)

{
        return 1;
}
[  383.899321] BUG: kernel NULL pointer dereference, address: 0000000000000b99
[  383.899327] #PF: supervisor read access in kernel mode
[  383.899330] #PF: error_code(0x0000) - not-present page
[  383.899332] PGD 8000000108a60067 P4D 8000000108a60067 PUD 104550067 PMD 0
[  383.899341] Oops: Oops: 0000 [#1] SMP PTI
[  383.899346] CPU: 1 UID: 0 PID: 12925 Comm: test Kdump: loaded Not tainted 7.0.0-rc6+ #1 PREEMPT(lazy)
[  383.899349] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[  383.899351] RIP: 0010:get_task_pid+0x20/0x80
[  383.899358] Code: 90 90 90 90 90 90 90 90 90 90 f3 0f 1e fa 0f 1f 44 00 00 55 89 f5 53 48 89 fb e8 0b 98 0a 00 85 ed 75 32 48 81 c3 98 0b 00 00 <48> 8b 1b 48 85 db 74 14 b8 01 00 00 00 f0 0f c1 03 85 c0 74 27 8d
[  383.899362] RSP: 0018:ffffd3ca0ab13b38 EFLAGS: 00010206
[  383.899367] RAX: 0000000000000001 RBX: 0000000000000b99 RCX: 0000000000000008
[  383.899371] RDX: ffff8b2c85860000 RSI: 0000000000000000 RDI: 0000000000000001
[  383.899374] RBP: 0000000000000000 R08: 0000010754cf50c8 R09: 0000010754cf50c8
[  383.899377] R10: ffffffff95a6c000 R11: 0000000000000024 R12: 0000000000000000
[  383.899380] R13: 0000000000000000 R14: 0000000000000001 R15: 0000000001200000
[  383.899384] FS:  00007f3eae907740(0000) GS:ffff8b2d05e5c000(0000) knlGS:0000000000000000
[  383.899388] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  383.899390] CR2: 0000000000000b99 CR3: 0000000106a94003 CR4: 00000000000706f0
[  383.899393] Call Trace:
[  383.899397]  <TASK>
[  383.899401]  kernel_clone+0xe8/0x480
[  383.899406]  __do_sys_clone+0x65/0x90
[  383.899410]  do_syscall_64+0xca/0x860
[  383.899421]  ? next_uptodate_folio+0x85/0x2a0
[  383.899427]  ? percpu_counter_add_batch+0x4c/0x90
[  383.899434]  ? filemap_map_pages+0x3b7/0x4d0
[  383.899437]  ? do_read_fault+0x107/0x210
[  383.899442]  ? do_fault+0x1b2/0x330
[  383.899445]  ? __handle_mm_fault+0x49b/0x7a0
[  383.899448]  ? count_memcg_events+0xc4/0x160
[  383.899453]  ? handle_mm_fault+0xbb/0x370
[  383.899456]  ? do_user_addr_fault+0x209/0x680
[  383.899459]  ? irqentry_exit+0x7a/0x660
[  383.899462]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
[  383.899466] RIP: 0033:0x7f3eaeb426e7
[  383.899470] Code: 5d c3 90 f3 0f 1e fa 64 48 8b 04 25 10 00 00 00 45 31 c0 31 d2 31 f6 bf 11 00 20 01 4c 8d 90 d0 02 00 00 b8 38 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 39 89 c2 85 c0 75 2c 64 48 8b 04 25 10 00 00
[  383.899472] RSP: 002b:00007fff23a2c838 EFLAGS: 00000246 ORIG_RAX: 0000000000000038
[  383.899477] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f3eaeb426e7
[  383.899479] RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000001200011
[  383.899481] RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000000
[  383.899483] R10: 00007f3eae907a10 R11: 0000000000000246 R12: 0000000000000001
[  383.899485] R13: 00007fff23a2dbe8 R14: 0000000000403de0 R15: 00007f3eaecf5000
[  383.899488]  </TASK>
[  383.899489] Modules linked in: nft_fib_inet nft_fib_ipv4 nft_fib_ipv6 nft_fib nft_reject_inet nf_reject_ipv4 nf_reject_ipv6 nft_reject nft_ct nft_chain_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 ip_set rfkill nf_tables sunrpc intel_rapl_msr intel_rapl_common intel_uncore_frequency_common intel_pmc_core pmt_telemetry snd_intel8x0 pmt_discovery pmt_class snd_ac97_codec intel_pmc_ssram_telemetry ac97_bus intel_vsec snd_pcm snd_timer snd rapl soundcore joydev sg i2c_piix4 pcspkr vboxguest i2c_smbus nfnetlink sr_mod cdrom sd_mod ata_generic vmwgfx ahci libahci ata_piix drm_ttm_helper ghash_clmulni_intel video ttm e1000 wmi libata serio_raw dm_mirror dm_region_hash dm_log dm_multipath dm_mod fuse i2c_dev autofs4
[  383.899624] CR2: 0000000000000b99

This is because
1. The BPF program is designed to override the normal behavior of security_task_alloc and forcibly return a non-zero error value (e.g., -ENOMEM or 1).
2. User space triggers a process creation syscall (like fork or clone). The kernel invokes copy_process to allocate and initialize a new task_struct for the child process.
3. copy_process invokes security_task_alloc. Due to the attached BPF program, it returns the positive value 1.
4. copy_process treats any non-zero return from security_task_alloc as a failure. It aborts initialization, cleans up, and returns the error code cast to a pointer via ERR_PTR(1) (which evaluates to the memory address 0x1).
5. In kernel_clone(), the kernel uses the IS_ERR() macro to check if the returned task_struct *p is an error pointer. However, IS_ERR() only checks for negative error codes in the range [-MAX_ERRNO, -1]. Since 0x1 is not in this range, the check IS_ERR(p) evaluates to false.
6. The kernel incorrectly assumes p is a valid task_struct pointer and proceeds to call get_task_pid(p, PIDTYPE_PID). This dereferences the invalid address 0x1 (plus the offset of the thread_pid field), triggering a general protection fault or null-pointer dereference.

Therefore, for security-related functions, similar to LSM, the validity of their return values is verified to fix the issue. Return value validation for fault injection is also added accordingly.

Fixes 6ba43b761c41 ("bpf: Attachment verification for BPF_MODIFY_RETURN")
---
 kernel/bpf/verifier.c | 114 +++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 112 insertions(+), 2 deletions(-)
kernel/bpf/verifier.c +112 -2
index 56fcc96dc780..8285f626919e 100644
@@ -18363,6 +18363,112 @@ static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
 	return 0;
 }
 
+#ifdef CONFIG_FUNCTION_ERROR_INJECTION
+
+/*
+ * The return values corresponding to the error injection
+ * types shall be interpreted as follows:
+ *
+ * EI_ETYPE_NULL
+ *   This function will return `NULL` if it fails. e.g. return an allocated
+ *   object address.
+ *
+ * EI_ETYPE_ERRNO
+ *   This function will return an `-errno` error code if it fails. e.g. return
+ *   -EINVAL if the input is wrong. This will include the functions which will
+ *   return an address which encodes `-errno` by ERR_PTR() macro.
+ *
+ * EI_ETYPE_ERRNO_NULL
+ *   This function will return an `-errno` or `NULL` if it fails. If the caller
+ *   of this function checks the return value with IS_ERR_OR_NULL() macro, this
+ *   type will be appropriate.
+ *
+ * EI_ETYPE_TRUE
+ *   This function will return `true` (non-zero positive value) if it fails.
+ *
+ * Since returning 0 from modify_return means not modifying the return value,
+ * 0 must always be included.
+ */
+static int modify_return_get_retval_range(const struct bpf_prog *prog,
+					  struct bpf_retval_range *range)
+{
+	unsigned long addr = (unsigned long)prog->aux->dst_trampoline->func.addr;
+
+	if (within_error_injection_list(addr)) {
+		switch (get_injectable_error_type(addr)) {
+		case EI_ETYPE_NULL:
+			range->minval = 0;
+			range->maxval = 0;
+			break;
+		case EI_ETYPE_ERRNO:
+		case EI_ETYPE_ERRNO_NULL:
+			range->minval = -MAX_ERRNO;
+			range->maxval = 0;
+			break;
+		case EI_ETYPE_TRUE:
+			range->minval = 0;
+			range->maxval = 1;
+			break;
+		}
+		range->return_32bit = true;
+
+		return 0;
+	}
+
+	return -EINVAL;
+}
+
+#else
+
+/* The system call return value is allowed to be an arbitrary value. */
+static int modify_return_get_retval_range(const struct bpf_prog *prog,
+					  struct bpf_retval_range *range)
+{
+	return -EINVAL;
+}
+
+#endif /* CONFIG_FUNCTION_ERROR_INJECTION */
+
+#define SECURITY_PREFIX "security_"
+/* hooks return 0 or 1 */
+BTF_SET_START(bool_security_hooks)
+#ifdef CONFIG_SECURITY_NETWORK_XFRM
+BTF_ID(func, security_xfrm_state_pol_flow_match)
+#endif
+#ifdef CONFIG_AUDIT
+BTF_ID(func, security_audit_rule_known)
+#endif
+/*
+ * The inode_xattr_skipcap hook is called from within
+ * security_inode_removexattr and security_inode_setxattr.
+ * Since these two functions can also return values from
+ * inode_removexattr and inode_setxattr respectively,
+ * there is no need to restrict their return values to 0 or 1.
+ */
+BTF_SET_END(bool_security_hooks)
+
+/*
+ * Similar to bpf_lsm_get_retval_range,
+ * ensure that the return values of fmod_ret are valid.
+ */
+static int bpf_security_get_retval_range(const struct bpf_prog *prog,
+					 struct bpf_retval_range *range)
+{
+	if (strncmp(SECURITY_PREFIX, prog->aux->attach_func_name,
+		    sizeof(SECURITY_PREFIX) - 1))
+		return -EINVAL;
+
+	if (btf_id_set_contains(&bool_security_hooks, prog->aux->attach_btf_id)) {
+		range->minval = 0;
+		range->maxval = 1;
+	} else {
+		range->minval = -MAX_ERRNO;
+		range->maxval = 0;
+	}
+	range->return_32bit = true;
+
+	return 0;
+}
 
 static bool return_retval_range(struct bpf_verifier_env *env, struct bpf_retval_range *range)
 {
@@ -18416,8 +18522,13 @@ static bool return_retval_range(struct bpf_verifier_env *env, struct bpf_retval_
 			*range = retval_range(0, 0);
 			break;
 		case BPF_TRACE_RAW_TP:
-		case BPF_MODIFY_RETURN:
 			return false;
+		case BPF_MODIFY_RETURN:
+			if (!bpf_security_get_retval_range(env->prog, range))
+				break;
+			if (modify_return_get_retval_range(env->prog, range))
+				return false;
+			break;
 		case BPF_TRACE_ITER:
 		default:
 			break;
@@ -25460,7 +25571,6 @@ static int check_struct_ops_btf_id(struct bpf_verifier_env *env)
 	return bpf_prog_ctx_arg_info_init(prog, st_ops_desc->arg_info[member_idx].info,
 					  st_ops_desc->arg_info[member_idx].cnt);
 }
-#define SECURITY_PREFIX "security_"
 
 #ifdef CONFIG_FUNCTION_ERROR_INJECTION
 
Signature · 1 line
2.43.0
#2 lore raw
Quoted 157 lines · Feng Yang

From: Feng Yang <[email protected]>

Using the following BPF program will cause a kernel panic:
SEC("fmod_ret/security_task_alloc")
int fmod_task_alloc(void *ctx)

{
        return 1;
}
[  383.899321] BUG: kernel NULL pointer dereference, address: 0000000000000b99
[  383.899327] #PF: supervisor read access in kernel mode
[  383.899330] #PF: error_code(0x0000) - not-present page
[  383.899332] PGD 8000000108a60067 P4D 8000000108a60067 PUD 104550067 PMD 0
[  383.899341] Oops: Oops: 0000 [#1] SMP PTI
[  383.899346] CPU: 1 UID: 0 PID: 12925 Comm: test Kdump: loaded Not tainted 7.0.0-rc6+ #1 PREEMPT(lazy)
[  383.899349] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[  383.899351] RIP: 0010:get_task_pid+0x20/0x80
[  383.899358] Code: 90 90 90 90 90 90 90 90 90 90 f3 0f 1e fa 0f 1f 44 00 00 55 89 f5 53 48 89 fb e8 0b 98 0a 00 85 ed 75 32 48 81 c3 98 0b 00 00 <48> 8b 1b 48 85 db 74 14 b8 01 00 00 00 f0 0f c1 03 85 c0 74 27 8d
[  383.899362] RSP: 0018:ffffd3ca0ab13b38 EFLAGS: 00010206
[  383.899367] RAX: 0000000000000001 RBX: 0000000000000b99 RCX: 0000000000000008
[  383.899371] RDX: ffff8b2c85860000 RSI: 0000000000000000 RDI: 0000000000000001
[  383.899374] RBP: 0000000000000000 R08: 0000010754cf50c8 R09: 0000010754cf50c8
[  383.899377] R10: ffffffff95a6c000 R11: 0000000000000024 R12: 0000000000000000
[  383.899380] R13: 0000000000000000 R14: 0000000000000001 R15: 0000000001200000
[  383.899384] FS:  00007f3eae907740(0000) GS:ffff8b2d05e5c000(0000) knlGS:0000000000000000
[  383.899388] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  383.899390] CR2: 0000000000000b99 CR3: 0000000106a94003 CR4: 00000000000706f0
[  383.899393] Call Trace:
[  383.899397]  <TASK>
[  383.899401]  kernel_clone+0xe8/0x480
[  383.899406]  __do_sys_clone+0x65/0x90
[  383.899410]  do_syscall_64+0xca/0x860
[  383.899421]  ? next_uptodate_folio+0x85/0x2a0
[  383.899427]  ? percpu_counter_add_batch+0x4c/0x90
[  383.899434]  ? filemap_map_pages+0x3b7/0x4d0
[  383.899437]  ? do_read_fault+0x107/0x210
[  383.899442]  ? do_fault+0x1b2/0x330
[  383.899445]  ? __handle_mm_fault+0x49b/0x7a0
[  383.899448]  ? count_memcg_events+0xc4/0x160
[  383.899453]  ? handle_mm_fault+0xbb/0x370
[  383.899456]  ? do_user_addr_fault+0x209/0x680
[  383.899459]  ? irqentry_exit+0x7a/0x660
[  383.899462]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
[  383.899466] RIP: 0033:0x7f3eaeb426e7
[  383.899470] Code: 5d c3 90 f3 0f 1e fa 64 48 8b 04 25 10 00 00 00 45 31 c0 31 d2 31 f6 bf 11 00 20 01 4c 8d 90 d0 02 00 00 b8 38 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 39 89 c2 85 c0 75 2c 64 48 8b 04 25 10 00 00
[  383.899472] RSP: 002b:00007fff23a2c838 EFLAGS: 00000246 ORIG_RAX: 0000000000000038
[  383.899477] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f3eaeb426e7
[  383.899479] RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000001200011
[  383.899481] RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000000
[  383.899483] R10: 00007f3eae907a10 R11: 0000000000000246 R12: 0000000000000001
[  383.899485] R13: 00007fff23a2dbe8 R14: 0000000000403de0 R15: 00007f3eaecf5000
[  383.899488]  </TASK>
[  383.899489] Modules linked in: nft_fib_inet nft_fib_ipv4 nft_fib_ipv6 nft_fib nft_reject_inet nf_reject_ipv4 nf_reject_ipv6 nft_reject nft_ct nft_chain_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 ip_set rfkill nf_tables sunrpc intel_rapl_msr intel_rapl_common intel_uncore_frequency_common intel_pmc_core pmt_telemetry snd_intel8x0 pmt_discovery pmt_class snd_ac97_codec intel_pmc_ssram_telemetry ac97_bus intel_vsec snd_pcm snd_timer snd rapl soundcore joydev sg i2c_piix4 pcspkr vboxguest i2c_smbus nfnetlink sr_mod cdrom sd_mod ata_generic vmwgfx ahci libahci ata_piix drm_ttm_helper ghash_clmulni_intel video ttm e1000 wmi libata serio_raw dm_mirror dm_region_hash dm_log dm_multipath dm_mod fuse i2c_dev autofs4
[  383.899624] CR2: 0000000000000b99

This is because
1. The BPF program is designed to override the normal behavior of security_task_alloc and forcibly return a non-zero error value (e.g., -ENOMEM or 1).
2. User space triggers a process creation syscall (like fork or clone). The kernel invokes copy_process to allocate and initialize a new task_struct for the child process.
3. copy_process invokes security_task_alloc. Due to the attached BPF program, it returns the positive value 1.
4. copy_process treats any non-zero return from security_task_alloc as a failure. It aborts initialization, cleans up, and returns the error code cast to a pointer via ERR_PTR(1) (which evaluates to the memory address 0x1).
5. In kernel_clone(), the kernel uses the IS_ERR() macro to check if the returned task_struct *p is an error pointer. However, IS_ERR() only checks for negative error codes in the range [-MAX_ERRNO, -1]. Since 0x1 is not in this range, the check IS_ERR(p) evaluates to false.
6. The kernel incorrectly assumes p is a valid task_struct pointer and proceeds to call get_task_pid(p, PIDTYPE_PID). This dereferences the invalid address 0x1 (plus the offset of the thread_pid field), triggering a general protection fault or null-pointer dereference.

Therefore, for security-related functions, similar to LSM, the validity of their return values is verified to fix the issue. Return value validation for fault injection is also added accordingly.

Fixes 6ba43b761c41 ("bpf: Attachment verification for BPF_MODIFY_RETURN")
---
 kernel/bpf/verifier.c | 114 +++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 112 insertions(+), 2 deletions(-)
kernel/bpf/verifier.c +55 -0
index 56fcc96dc780..8285f626919e 100644
@@ -18363,6 +18363,112 @@ static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn)
        return 0;
 }
 
+#ifdef CONFIG_FUNCTION_ERROR_INJECTION
+
+/*
+ * The return values corresponding to the error injection
+ * types shall be interpreted as follows:
+ *
+ * EI_ETYPE_NULL
+ *   This function will return `NULL` if it fails. e.g. return an allocated
+ *   object address.
+ *
+ * EI_ETYPE_ERRNO
+ *   This function will return an `-errno` error code if it fails. e.g. return
+ *   -EINVAL if the input is wrong. This will include the functions which will
+ *   return an address which encodes `-errno` by ERR_PTR() macro.
+ *
+ * EI_ETYPE_ERRNO_NULL
+ *   This function will return an `-errno` or `NULL` if it fails. If the caller
+ *   of this function checks the return value with IS_ERR_OR_NULL() macro, this
+ *   type will be appropriate.
+ *
+ * EI_ETYPE_TRUE
+ *   This function will return `true` (non-zero positive value) if it fails.
+ *
+ * Since returning 0 from modify_return means not modifying the return value,
+ * 0 must always be included.
+ */
+static int modify_return_get_retval_range(const struct bpf_prog *prog,
+                                         struct bpf_retval_range *range)
+{
+       unsigned long addr = (unsigned long)prog->aux->dst_trampoline->func.addr;
+
+       if (within_error_injection_list(addr)) {
+               switch (get_injectable_error_type(addr)) {
+               case EI_ETYPE_NULL:
+                       range->minval = 0;
+                       range->maxval = 0;
+                       break;
+               case EI_ETYPE_ERRNO:
+               case EI_ETYPE_ERRNO_NULL:
+                       range->minval = -MAX_ERRNO;
+                       range->maxval = 0;
+                       break;
+               case EI_ETYPE_TRUE:
+                       range->minval = 0;
+                       range->maxval = 1;
+                       break;
+               }
+               range->return_32bit = true;
+
+               return 0;
+       }
+
+       return -EINVAL;
+}
+

kernel/bpf/verifier.c static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) · line 18430

Quoted 10 lines
kernel/bpf/verifier.c +10 -0
+#else
+
+/* The system call return value is allowed to be an arbitrary value. */
+static int modify_return_get_retval_range(const struct bpf_prog *prog,
+                                         struct bpf_retval_range *range)
+{
+       return -EINVAL;
+}
+
+#endif /* CONFIG_FUNCTION_ERROR_INJECTION */

I recall people already explained that this is no go. We cannot break all fmod_ret because error injection is disabled.

Also see sashiko reports.

Quoted 31 lines
kernel/bpf/verifier.c +31 -0
+
+#define SECURITY_PREFIX "security_"
+/* hooks return 0 or 1 */
+BTF_SET_START(bool_security_hooks)
+#ifdef CONFIG_SECURITY_NETWORK_XFRM
+BTF_ID(func, security_xfrm_state_pol_flow_match)
+#endif
+#ifdef CONFIG_AUDIT
+BTF_ID(func, security_audit_rule_known)
+#endif
+/*
+ * The inode_xattr_skipcap hook is called from within
+ * security_inode_removexattr and security_inode_setxattr.
+ * Since these two functions can also return values from
+ * inode_removexattr and inode_setxattr respectively,
+ * there is no need to restrict their return values to 0 or 1.
+ */
+BTF_SET_END(bool_security_hooks)
+
+/*
+ * Similar to bpf_lsm_get_retval_range,
+ * ensure that the return values of fmod_ret are valid.
+ */
+static int bpf_security_get_retval_range(const struct bpf_prog *prog,
+                                        struct bpf_retval_range *range)
+{
+       if (strncmp(SECURITY_PREFIX, prog->aux->attach_func_name,
+                   sizeof(SECURITY_PREFIX) - 1))
+               return -EINVAL;
+
+       if (btf_id_set_contains(&bool_security_hooks, prog->aux->attach_btf_id)) {

kernel/bpf/verifier.c static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) · line 18471

Quoted 10 lines
kernel/bpf/verifier.c +10 -0
+               range->minval = 0;
+               range->maxval = 1;
+       } else {
+               range->minval = -MAX_ERRNO;
+               range->maxval = 0;
+       }
+       range->return_32bit = true;
+
+       return 0;
+}

let's copy paste bpf_lsm_get_retval_range() for no good reason?!

please don't send such poor quality patches.

pw-bot: cr

author
#3 lore raw
Quoted 11 lines · On Sun, 12 Apr 2026 00:35:55 +0800 Alexei Starovoitov wrote:

kernel/bpf/verifier.c static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) · line 18430

Quoted 8 lines
kernel/bpf/verifier.c +8 -0
+/* The system call return value is allowed to be an arbitrary value. */
+static int modify_return_get_retval_range(const struct bpf_prog *prog,
+                                         struct bpf_retval_range *range)
+{
+       return -EINVAL;
+}
+
+#endif /* CONFIG_FUNCTION_ERROR_INJECTION */

I recall people already explained that this is no go. We cannot break all fmod_ret because error injection is disabled.

When error injection is disabled, the returned -EINVAL will result in a false return. This shouldn't change the previous logic.

case BPF_TRACE_RAW_TP:
- case BPF_MODIFY_RETURN:
return false;
+ case BPF_MODIFY_RETURN:
+ if (!bpf_security_get_retval_range(env->prog, range))
+ break;
+ if (modify_return_get_retval_range(env->prog, range))
+ return false;
+ break;

Quoted 2 lines

Also see sashiko reports.

Okay.
Thansks.

Quoted 1 line

let's copy paste bpf_lsm_get_retval_range() for no good reason?!

So should we modify this part according to Jiayuan Chen's logic?

Jiayuan's previous reply:
Also, I think a whitelist approach would be better here. The known danger is specifically those security hooks whose return values get fed into ERR_PTR() by callers, such as:
- security_task_alloc
- security_inode_readlink
- security_task_movememory
- security_inode_follow_link
- security_fs_context_submount
- security_dentry_create_files_as
- security_perf_event_alloc
- security_inode_get_acl

Quoted 1 line

please don't send such poor quality patches.

#4 lore raw
In reply to Feng Yang
Quoted 2 lines · Feng Yang

From: Feng Yang <[email protected]>

[...]

Quoted 5 lines

static bool return_retval_range(struct bpf_verifier_env *env, struct bpf_retval_range *range) {

kernel/bpf/verifier.c +0 -0
@@ -18416,8 +18522,13 @@ static bool return_retval_range(struct bpf_verifier_env *env, struct bpf_retval_
 			*range = retval_range(0, 0);

kernel/bpf/verifier.c static bool return_retval_range(struct bpf_verifier_env *env, struct bpf_retval_ · line 18531

Quoted 10 lines
kernel/bpf/verifier.c +6 -1
 			break;
 		case BPF_TRACE_RAW_TP:
-		case BPF_MODIFY_RETURN:
 			return false;
+		case BPF_MODIFY_RETURN:
+			if (!bpf_security_get_retval_range(env->prog, range))
+				break;
+			if (modify_return_get_retval_range(env->prog, range))
+				return false;
+			break;

return false by default, as what we did in the previous logic?

+ case BPF_MODIFY_RETURN:
+ if (!bpf_security_get_retval_range(env->prog, range))
+ break;
+ if (!modify_return_get_retval_range(env->prog, range))
+ break;
+ return false;

Quoted 12 lines
kernel/bpf/verifier.c +0 -1
 		case BPF_TRACE_ITER:
 		default:
 			break;
@@ -25460,7 +25571,6 @@ static int check_struct_ops_btf_id(struct bpf_verifier_env *env)
 	return bpf_prog_ctx_arg_info_init(prog, st_ops_desc->arg_info[member_idx].info,
 					  st_ops_desc->arg_info[member_idx].cnt);
 }
-#define SECURITY_PREFIX "security_"
 
 #ifdef CONFIG_FUNCTION_ERROR_INJECTION
 
author
#5 lore raw
Quoted 18 lines · Menglong Dong
Quoted 4 lines

static bool return_retval_range(struct bpf_verifier_env *env, struct bpf_retval_range *range) {

kernel/bpf/verifier.c +0 -0
@@ -18416,8 +18522,13 @@ static bool return_retval_range(struct bpf_verifier_env *env, struct bpf_retval_
 			*range = retval_range(0, 0);

kernel/bpf/verifier.c static bool return_retval_range(struct bpf_verifier_env *env, struct bpf_retval_ · line 18531

Quoted 10 lines
kernel/bpf/verifier.c +6 -1
 			break;
 		case BPF_TRACE_RAW_TP:
-		case BPF_MODIFY_RETURN:
 			return false;
+		case BPF_MODIFY_RETURN:
+			if (!bpf_security_get_retval_range(env->prog, range))
+				break;
+			if (modify_return_get_retval_range(env->prog, range))
+				return false;
+			break;

return false by default, as what we did in the previous logic?

kernel/bpf/verifier.c static bool return_retval_range(struct bpf_verifier_env *env, struct bpf_retval_

Quoted 6 lines
kernel/bpf/verifier.c +6 -0
+		case BPF_MODIFY_RETURN:
+			if (!bpf_security_get_retval_range(env->prog, range))
+				break;
+			if (!modify_return_get_retval_range(env->prog, range))
+				break;
+			return false;

Okay, thank you very much.

Quoted 1 line
Quoted 1 line
kernel/bpf/verifier.c +0 -0
 		case BPF_TRACE_ITER:
author
#6 lore raw
In reply to Feng Yang

[...]

Issue Regarding BPF_PROG_TYPE_EXT Programs Bypassing Validation Logic:

__attribute__ ((noinline)) int B()
{
return 0; /* Valid return value */
}

SEC("fmod_ret/security_task_alloc")
int A()

{
        return B();
}

SEC("freplace/B")
int replace()
{
return 1; /* Invalid return value */
}

This method can verify whether the return value of freplace is legitimate by using env->prog->aux->dst_prog->aux->attach_btf_id.

However, this approach will reject legitimate programs if the return value of the original program does not come from the replaced function (as shown below).

__attribute__ ((noinline)) int B()

{
        return 0; 
}

SEC("fmod_ret/security_task_alloc")
int A()

{
        B();
        return 0;
}

SEC("freplace/B")
int replace()

{
        return 1;
}

I have not found a proper solution to this issue.

target.bpf.c:
__attribute__ ((noinline)) int B()

{
	bpf_printk("--\n");
	return -2;	
}

SEC("fmod_ret/security_task_alloc")
int A()

{
	int i = B();
	bpf_printk("return: %d\n", i);

return i;
}

freplace.bpf.c:
SEC("freplace/B")
int replace()

{
	bpf_printk("freplace\n");
	return -1;
}
In addition, the log output from testing the above program is as follows:
cat /sys/kernel/debug/tracing/trace_pipe 
          attack-5370    [002] ...11  4379.086626: bpf_trace_printk: freplace

          attack-5370    [002] ...11  4379.086656: bpf_trace_printk: return: -2

Strangely, the replacement program is executed, but the return value still originates from the original program.

author
#7 lore raw

PATCH v3 2/2 selftests/bpf: Add selftests for verifying return values of fmod_ret.

In reply to Feng Yang

From: Feng Yang <[email protected]>

Add selftests for verifying return values of fmod_ret. test_progs -t verifier_fmod_ret_return

569/1   verifier_fmod_ret_return/fmod_task_alloc_1:OK
569/2   verifier_fmod_ret_return/fmod_task_alloc_2:OK
569/3   verifier_fmod_ret_return/fmod_ret_sys_open_1:OK
569/4   verifier_fmod_ret_return/fmod_ret_sys_open_2:OK
569/5   verifier_fmod_ret_return/fmod_ret_sys_open_3:OK
569/6   verifier_fmod_ret_return/fmod_page_pool_alloc_netmems_1:OK
569/7   verifier_fmod_ret_return/fmod_page_pool_alloc_netmems_2:OK
569     verifier_fmod_ret_return:OK

Summary: 1/7 PASSED, 0 SKIPPED, 0 FAILED

---
 .../selftests/bpf/prog_tests/verifier.c       |  2 +
 .../bpf/progs/verifier_fmod_ret_return.c      | 59 +++++++++++++++++++
 2 files changed, 61 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/progs/verifier_fmod_ret_return.c
tools/testing/selftests/bpf/prog_tests/verifier.c +2 -0
index a96b25ebff23..5a8d14e3980e 100644
@@ -37,6 +37,7 @@
 #include "verifier_div0.skel.h"
 #include "verifier_div_mod_bounds.skel.h"
 #include "verifier_div_overflow.skel.h"
+#include "verifier_fmod_ret_return.skel.h"
 #include "verifier_global_subprogs.skel.h"
 #include "verifier_global_ptr_args.skel.h"
 #include "verifier_gotol.skel.h"
@@ -185,6 +186,7 @@ void test_verifier_direct_stack_access_wraparound(void) { RUN(verifier_direct_st
 void test_verifier_div0(void)                 { RUN(verifier_div0); }
 void test_verifier_div_mod_bounds(void)       { RUN(verifier_div_mod_bounds); }
 void test_verifier_div_overflow(void)         { RUN(verifier_div_overflow); }
+void test_verifier_fmod_ret_return(void)      { RUN(verifier_fmod_ret_return); }
 void test_verifier_global_subprogs(void)      { RUN(verifier_global_subprogs); }
 void test_verifier_global_ptr_args(void)      { RUN(verifier_global_ptr_args); }
 void test_verifier_gotol(void)                { RUN(verifier_gotol); }
tools/testing/selftests/bpf/progs/verifier_fmod_ret_return.c new file +59 -0

59 lines not on this page · raw message

Signature · 1 line
2.43.0