Skip to contentSkip to conversation

3.6-rc4 audit_log_d_path oops.

PATCH Dave Jones 3 participants 10 messages Sep 6, 2012

Subsystems (5)
lore
Share & tools
lore

Dave Jones

author
#0 lore raw

Hit this in overnight fuzz testing..

BUG: unable to handle kernel NULL pointer dereference at 0000000000000020
IP: [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
PGD 12fded067 PUD 142c06067 PMD 0 
Oops: 0000 [#1] SMP 
Modules linked in: tun fuse ipt_ULOG binfmt_misc nfnetlink nfc caif_socket caif phonet can llc2 pppoe pppox ppp_generic slhc irda crc_ccitt rds af_key decnet rose x25 atm netrom appletalk ipx p8023 psnap p8022 llc ax25 lockd sunrpc bluetooth rfkill ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 ip6table_filter ip6_tables nf_conntrack_ipv4 nf_defrag_ipv4 xt_state nf_conntrack kvm_intel kvm crc32c_intel ghash_clmulni_intel microcode pcspkr i2c_i801 e1000e uinput i915 video i2c_algo_bit drm_kms_helper drm i2c_core
CPU 5 
Pid: 7007, comm: trinity-child5 Not tainted 3.6.0-rc4+ #36
RIP: 0010:[<ffffffff81103365>]  [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
RSP: 0018:ffff880116b33ec8  EFLAGS: 00010246
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000035
RDX: ffffffff819dbf5d RSI: ffffffff819da5bb RDI: 0000000000000000
RBP: ffff880116b33ee8 R08: ffff88001942533e R09: 0000000000000000
R10: 0000000000000001 R11: 0000000000000000 R12: ffff880116b33f38
R13: ffff880116b33f38 R14: ffffffff819fa1eb R15: 0000000000000005
FS:  00007f5742581740(0000) GS:ffff880148600000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000020 CR3: 00000001046d3000 CR4: 00000000001407e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Process trinity-child5 (pid: 7007, threadinfo ffff880116b32000, task ffff880019424d00)
Stack:
 ffff880116b33ee8 0000000000000000 ffff880116b33f38 ffff880019424d00
 ffff880116b33f18 ffffffff81103522 00000000000001d0 ffff880119ed6500
 00000000000001d0 0000000000000109 ffff880116b33f78 ffffffff811ebb95
Call Trace:
 [<ffffffff81103522>] audit_log_link_denied+0x92/0x100
 [<ffffffff811ebb95>] sys_linkat+0x195/0x1e0
 [<ffffffff813573de>] ? trace_hardirqs_on_thunk+0x3a/0x3f
 [<ffffffff816a50ed>] system_call_fastpath+0x1a/0x1f
Code: 5d e8 4c 89 65 f0 4c 89 6d f8 0f 1f 44 00 00 48 85 f6 48 89 fb 49 89 d5 74 11 48 89 f2 31 c0 48 c7 c6 bb a5 9d 81 e8 8b e1 ff ff <8b> 73 20 40 f6 c6 01 75 62 48 8b 3d 33 33 93 01 48 85 ff 74 4e 
RIP  [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
 RSP <ffff880116b33ec8>
CR2: 0000000000000020
---[ end trace 85b88c850143bb1c ]---

That's here in kernel/audit.c

1433
1434 /* We will allow 11 spaces for ' (deleted)' to be appended */
1435 pathname = kmalloc(PATH_MAX+11, ab->gfp_mask);

'ab' is NULL.

Looks like audit_log_link_denied needs to handle potential failure from audit_log_start and abort early. (oddly, it looks like every other function called there checks for !ab.)

Maybe additional code should be added here to printk the audit message to dmesg so that we don't lose it entirely, but for now, minimal fix.

kernel/audit.c +3 -0
index ea3b7b6..c3e85bb 100644
@@ -1466,6 +1466,9 @@ void audit_log_link_denied(const char *operation, struct path *link)
 
 	ab = audit_log_start(current->audit_context, GFP_KERNEL,
 			     AUDIT_ANOM_LINK);
+	if (!ab)
+		return;
+
 	audit_log_format(ab, "op=%s action=denied", operation);
 	audit_log_format(ab, " pid=%d comm=", current->pid);
 	audit_log_untrustedstring(ab, current->comm);
author
#1 lore raw

Check all returns from audit_log_start

Following on from the previous patch that fixed an oops, these are all the other similar code patterns in the tree with the same checks added. I never saw these causing problems, but checking this everywhere seems to make more sense than every subsequent routine that gets passed 'ab' having to check it.

Later we could remove all those same checks from audit_log_format and friends. For now, this just prevents similar bugs being introduced as the one in my previous patch.

include/linux/audit.h +1 -1
index 36abf2a..2df27a7 100644
@@ -665,7 +665,7 @@ extern __printf(4, 5)
 void audit_log(struct audit_context *ctx, gfp_t gfp_mask, int type,
 	       const char *fmt, ...);
 
-extern struct audit_buffer *audit_log_start(struct audit_context *ctx, gfp_t gfp_mask, int type);
+extern struct audit_buffer __must_check *audit_log_start(struct audit_context *ctx, gfp_t gfp_mask, int type);
 extern __printf(2, 3)
 void audit_log_format(struct audit_buffer *ab, const char *fmt, ...);
 extern void		    audit_log_end(struct audit_buffer *ab);
kernel/audit.c +7 -1
index ea3b7b6..832c549 100644
@@ -271,6 +271,9 @@ static int audit_log_config_change(char *function_name, int new, int old,
 	int rc = 0;
 
 	ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
+	if (!ab)
+		return -EINVAL;
+
 	audit_log_format(ab, "%s=%d old=%d auid=%u ses=%u", function_name, new,
 			 old, loginuid, sessionid);
 	if (sid) {
@@ -632,6 +635,9 @@ static int audit_log_common_recv_msg(struct audit_buffer **ab, u16 msg_type,
 	}
 
 	*ab = audit_log_start(NULL, GFP_KERNEL, msg_type);
+	if (!*ab)
+		return -EINVAL;
+
 	audit_log_format(*ab, "pid=%d uid=%u auid=%u ses=%u",
 			 pid, uid, auid, ses);
 	if (sid) {
@@ -1145,7 +1151,7 @@ static inline void audit_get_stamp(struct audit_context *ctx,
  * will be written at syscall exit.  If there is no associated task, then
  * task context (ctx) should be NULL.
  */
-struct audit_buffer *audit_log_start(struct audit_context *ctx, gfp_t gfp_mask,
+struct audit_buffer __must_check *audit_log_start(struct audit_context *ctx, gfp_t gfp_mask,
 				     int type)
 {
 	struct audit_buffer	*ab	= NULL;
kernel/audit_tree.c +9 -7
index ed206fd..2580edf 100644
@@ -462,13 +462,15 @@ static void kill_rules(struct audit_tree *tree)
 		if (rule->tree) {
 			/* not a half-baked one */
 			ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
-			audit_log_format(ab, "op=");
-			audit_log_string(ab, "remove rule");
-			audit_log_format(ab, " dir=");
-			audit_log_untrustedstring(ab, rule->tree->pathname);
-			audit_log_key(ab, rule->filterkey);
-			audit_log_format(ab, " list=%d res=1", rule->listnr);
-			audit_log_end(ab);
+			if (ab) {
+				audit_log_format(ab, "op=");
+				audit_log_string(ab, "remove rule");
+				audit_log_format(ab, " dir=");
+				audit_log_untrustedstring(ab, rule->tree->pathname);
+				audit_log_key(ab, rule->filterkey);
+				audit_log_format(ab, " list=%d res=1", rule->listnr);
+				audit_log_end(ab);
+			}
 			rule->tree = NULL;
 			list_del_rcu(&entry->list);
 			list_del(&entry->rule.list);
kernel/auditsc.c +4 -0
index 4b96415..1f39dcf 100644
@@ -2707,6 +2707,8 @@ void audit_core_dumps(long signr)
 		return;
 
 	ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_ANOM_ABEND);
+	if (!ab)
+		return;
 	audit_log_abend(ab, "memory violation", signr);
 	audit_log_end(ab);
 }
@@ -2716,6 +2718,8 @@ void __audit_seccomp(unsigned long syscall, long signr, int code)
 	struct audit_buffer *ab;
 
 	ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_ANOM_ABEND);
+	if (!ab)
+		return;
 	audit_log_abend(ab, "seccomp", signr);
 	audit_log_format(ab, " syscall=%ld", syscall);
 	audit_log_format(ab, " compat=%d", is_compat_task());
security/integrity/ima/ima_audit.c +3 -0
index 7a57f67..6fa60ff 100644
@@ -38,6 +38,9 @@ void integrity_audit_msg(int audit_msgno, struct inode *inode,
 		return;
 
 	ab = audit_log_start(current->audit_context, GFP_KERNEL, audit_msgno);
+	if (!ab)
+		return;
+
 	audit_log_format(ab, "pid=%d uid=%u auid=%u ses=%u",
 			 current->pid, current_cred()->uid,
 			 audit_get_loginuid(current),
security/integrity/ima/ima_policy.c +2 -0
index 1a95830..2c39e18 100644
@@ -276,6 +276,8 @@ static int ima_parse_rule(char *rule, struct ima_measure_rule_entry *entry)
 	int result = 0;
 
 	ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_INTEGRITY_RULE);
+	if (!ab)
+		return -EINVAL;
 
 	entry->uid = -1;
 	entry->action = UNKNOWN;
security/selinux/hooks.c +4 -0
index 6c77f63..59c5929 100644
@@ -2802,6 +2802,8 @@ static int selinux_inode_setxattr(struct dentry *dentry, const char *name,
 				audit_size = 0;
 			}
 			ab = audit_log_start(current->audit_context, GFP_ATOMIC, AUDIT_SELINUX_ERR);
+			if (!ab)
+				return -EINVAL;
 			audit_log_format(ab, "op=setxattr invalid_context=");
 			audit_log_n_untrustedstring(ab, value, audit_size);
 			audit_log_end(ab);
@@ -5318,6 +5320,8 @@ static int selinux_setprocattr(struct task_struct *p,
 				else
 					audit_size = size;
 				ab = audit_log_start(current->audit_context, GFP_ATOMIC, AUDIT_SELINUX_ERR);
+				if (!ab)
+					return -EINVAL;
 				audit_log_format(ab, "op=fscreate invalid_context=");
 				audit_log_n_untrustedstring(ab, value, audit_size);
 				audit_log_end(ab);
#2 lore raw

Re: Check all returns from audit_log_start

Quoted 11 lines · Dave Jones

Following on from the previous patch that fixed an oops, these are all the other similar code patterns in the tree with the same checks added. I never saw these causing problems, but checking this everywhere seems to make more sense than every subsequent routine that gets passed 'ab' having to check it.

Later we could remove all those same checks from audit_log_format and friends. For now, this just prevents similar bugs being introduced as the one in my previous patch.

Not certain because I haven't looked at what happens with the error code, but I think this might not be right. auditd can be explictly told not to audit certain events, in which case it is normal and expected that ab would come back NULL....

author
#3 lore raw

Re: Check all returns from audit_log_start

Quoted 11 lines · Eric Paris
Quoted 10 lines · Dave Jones

Following on from the previous patch that fixed an oops, these are all the other similar code patterns in the tree with the same checks added. I never saw these causing problems, but checking this everywhere seems to make more sense than every subsequent routine that gets passed 'ab' having to check it.

Later we could remove all those same checks from audit_log_format and friends. For now, this just prevents similar bugs being introduced as the one in my previous patch.

Quoted 6 lines
Quoted 1 line

Not certain because I haven't looked at what happens with the error code, but I think this might not be right. auditd can be explictly told not to audit certain events, in which case it is normal and expected that ab would come back NULL....

Ugh, that's a lot messier to have to audit every function that gets passed 'ab' to make sure it has a NULL check, but ok I'll go look at it.

hopefully audit_log_link_denied was a one off.

Dave

author
#4 lore raw

Re: Check all returns from audit_log_start

Quoted 9 lines · Dave Jones
Quoted 4 lines

Not certain because I haven't looked at what happens with the error code, but I think this might not be right. auditd can be explictly told not to audit certain events, in which case it is normal and expected that ab would come back NULL....

Ugh, that's a lot messier to have to audit every function that gets passed 'ab' to make sure it has a NULL check, but ok I'll go look at it.

hopefully audit_log_link_denied was a one off.

ok, from a quick look-over, that does seem to be the case.
This still feels like a nasty trap waiting for someone to walk into again though.

Dave

author
#5 lore raw
In reply to Dave Jones
Quoted 59 lines · Dave Jones

Hit this in overnight fuzz testing..

BUG: unable to handle kernel NULL pointer dereference at 0000000000000020
IP: [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
PGD 12fded067 PUD 142c06067 PMD 0 
Oops: 0000 [#1] SMP 
Modules linked in: tun fuse ipt_ULOG binfmt_misc nfnetlink nfc caif_socket caif phonet can llc2 pppoe pppox ppp_generic slhc irda crc_ccitt rds af_key decnet rose x25 atm netrom appletalk ipx p8023 psnap p8022 llc ax25 lockd sunrpc bluetooth rfkill ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 ip6table_filter ip6_tables nf_conntrack_ipv4 nf_defrag_ipv4 xt_state nf_conntrack kvm_intel kvm crc32c_intel ghash_clmulni_intel microcode pcspkr i2c_i801 e1000e uinput i915 video i2c_algo_bit drm_kms_helper drm i2c_core
CPU 5 
Pid: 7007, comm: trinity-child5 Not tainted 3.6.0-rc4+ #36
RIP: 0010:[<ffffffff81103365>]  [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
RSP: 0018:ffff880116b33ec8  EFLAGS: 00010246
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000035
RDX: ffffffff819dbf5d RSI: ffffffff819da5bb RDI: 0000000000000000
RBP: ffff880116b33ee8 R08: ffff88001942533e R09: 0000000000000000
R10: 0000000000000001 R11: 0000000000000000 R12: ffff880116b33f38
R13: ffff880116b33f38 R14: ffffffff819fa1eb R15: 0000000000000005
FS:  00007f5742581740(0000) GS:ffff880148600000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000020 CR3: 00000001046d3000 CR4: 00000000001407e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Process trinity-child5 (pid: 7007, threadinfo ffff880116b32000, task ffff880019424d00)
Stack:
 ffff880116b33ee8 0000000000000000 ffff880116b33f38 ffff880019424d00
 ffff880116b33f18 ffffffff81103522 00000000000001d0 ffff880119ed6500
 00000000000001d0 0000000000000109 ffff880116b33f78 ffffffff811ebb95
Call Trace:
 [<ffffffff81103522>] audit_log_link_denied+0x92/0x100
 [<ffffffff811ebb95>] sys_linkat+0x195/0x1e0
 [<ffffffff813573de>] ? trace_hardirqs_on_thunk+0x3a/0x3f
 [<ffffffff816a50ed>] system_call_fastpath+0x1a/0x1f
Code: 5d e8 4c 89 65 f0 4c 89 6d f8 0f 1f 44 00 00 48 85 f6 48 89 fb 49 89 d5 74 11 48 89 f2 31 c0 48 c7 c6 bb a5 9d 81 e8 8b e1 ff ff <8b> 73 20 40 f6 c6 01 75 62 48 8b 3d 33 33 93 01 48 85 ff 74 4e 
RIP  [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
 RSP <ffff880116b33ec8>
CR2: 0000000000000020
---[ end trace 85b88c850143bb1c ]---

That's here in kernel/audit.c

1433
1434 /* We will allow 11 spaces for ' (deleted)' to be appended */
1435 pathname = kmalloc(PATH_MAX+11, ab->gfp_mask);

'ab' is NULL.

Looks like audit_log_link_denied needs to handle potential failure from audit_log_start and abort early. (oddly, it looks like every other function called there checks for !ab.)

Maybe additional code should be added here to printk the audit message to dmesg so that we don't lose it entirely, but for now, minimal fix.

kernel/audit.c +0 -0
index ea3b7b6..c3e85bb 100644

kernel/audit.c void audit_log_link_denied(const char *operation, struct path *link) · line 1474

Quoted 9 lines
kernel/audit.c +3 -0
@@ -1466,6 +1466,9 @@ void audit_log_link_denied(const char *operation, struct path *link)
 
 	ab = audit_log_start(current->audit_context, GFP_KERNEL,
 			     AUDIT_ANOM_LINK);
+	if (!ab)
+		return;
+
 	audit_log_format(ab, "op=%s action=denied", operation);
 	audit_log_format(ab, " pid=%d comm=", current->pid);
 	audit_log_untrustedstring(ab, current->comm);

Added Kees, as this was introduced in a51d9eaa41866ab6b4b6ecad7b621f8b66ece0dc

I just realised, the funny thing about this is that the machine running that test had selinux/audit disabled. And yet here we are, screwing around with audit buffers.

Should there be a test on audit_enable=0 in audit_log_link_denied() ?

I'm now curious how much more of the audit code is getting run through similar lack of tests

Dave

#6 lore raw
Quoted 72 lines · Dave Jones
Quoted 59 lines · Dave Jones

Hit this in overnight fuzz testing..

BUG: unable to handle kernel NULL pointer dereference at 0000000000000020
IP: [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
PGD 12fded067 PUD 142c06067 PMD 0
Oops: 0000 [#1] SMP
Modules linked in: tun fuse ipt_ULOG binfmt_misc nfnetlink nfc caif_socket caif phonet can llc2 pppoe pppox ppp_generic slhc irda crc_ccitt rds af_key decnet rose x25 atm netrom appletalk ipx p8023 psnap p8022 llc ax25 lockd sunrpc bluetooth rfkill ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 ip6table_filter ip6_tables nf_conntrack_ipv4 nf_defrag_ipv4 xt_state nf_conntrack kvm_intel kvm crc32c_intel ghash_clmulni_intel microcode pcspkr i2c_i801 e1000e uinput i915 video i2c_algo_bit drm_kms_helper drm i2c_core
CPU 5
Pid: 7007, comm: trinity-child5 Not tainted 3.6.0-rc4+ #36
RIP: 0010:[<ffffffff81103365>]  [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
RSP: 0018:ffff880116b33ec8  EFLAGS: 00010246
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000035
RDX: ffffffff819dbf5d RSI: ffffffff819da5bb RDI: 0000000000000000
RBP: ffff880116b33ee8 R08: ffff88001942533e R09: 0000000000000000
R10: 0000000000000001 R11: 0000000000000000 R12: ffff880116b33f38
R13: ffff880116b33f38 R14: ffffffff819fa1eb R15: 0000000000000005
FS:  00007f5742581740(0000) GS:ffff880148600000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000020 CR3: 00000001046d3000 CR4: 00000000001407e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Process trinity-child5 (pid: 7007, threadinfo ffff880116b32000, task ffff880019424d00)
Stack:
 ffff880116b33ee8 0000000000000000 ffff880116b33f38 ffff880019424d00
 ffff880116b33f18 ffffffff81103522 00000000000001d0 ffff880119ed6500
 00000000000001d0 0000000000000109 ffff880116b33f78 ffffffff811ebb95
Call Trace:
 [<ffffffff81103522>] audit_log_link_denied+0x92/0x100
 [<ffffffff811ebb95>] sys_linkat+0x195/0x1e0
 [<ffffffff813573de>] ? trace_hardirqs_on_thunk+0x3a/0x3f
 [<ffffffff816a50ed>] system_call_fastpath+0x1a/0x1f
Code: 5d e8 4c 89 65 f0 4c 89 6d f8 0f 1f 44 00 00 48 85 f6 48 89 fb 49 89 d5 74 11 48 89 f2 31 c0 48 c7 c6 bb a5 9d 81 e8 8b e1 ff ff <8b> 73 20 40 f6 c6 01 75 62 48 8b 3d 33 33 93 01 48 85 ff 74 4e
RIP  [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
 RSP <ffff880116b33ec8>
CR2: 0000000000000020
---[ end trace 85b88c850143bb1c ]---

That's here in kernel/audit.c

1433
1434 /* We will allow 11 spaces for ' (deleted)' to be appended */
1435 pathname = kmalloc(PATH_MAX+11, ab->gfp_mask);

'ab' is NULL.

Looks like audit_log_link_denied needs to handle potential failure from audit_log_start and abort early. (oddly, it looks like every other function called there checks for !ab.)

Maybe additional code should be added here to printk the audit message to dmesg so that we don't lose it entirely, but for now, minimal fix.

kernel/audit.c +0 -0
index ea3b7b6..c3e85bb 100644

kernel/audit.c void audit_log_link_denied(const char *operation, struct path *link) · line 1474

Quoted 9 lines
kernel/audit.c +3 -0
@@ -1466,6 +1466,9 @@ void audit_log_link_denied(const char *operation, struct path *link)
 
     ab = audit_log_start(current->audit_context, GFP_KERNEL,
                          AUDIT_ANOM_LINK);
+    if (!ab)
+            return;
+
     audit_log_format(ab, "op=%s action=denied", operation);
     audit_log_format(ab, " pid=%d comm=", current->pid);
     audit_log_untrustedstring(ab, current->comm);

Added Kees, as this was introduced in a51d9eaa41866ab6b4b6ecad7b621f8b66ece0dc

Quoted 6 lines

I just realised, the funny thing about this is that the machine running that test had selinux/audit disabled. And yet here we are, screwing around with audit buffers.

Should there be a test on audit_enable=0 in audit_log_link_denied() ?

I'm now curious how much more of the audit code is getting run through similar lack of tests

Possible there are other places. The usual idium is to put a a static inline audit_[blah] in the header file which just does if (!audit_dummy_context()) return __audit_[blah]; return 0;

Thus if audit is off or compiled out nothing gets run.

#7 lore raw
In reply to Dave Jones
Quoted 61 lines · Dave Jones
Quoted 60 lines · Dave Jones

Hit this in overnight fuzz testing..

BUG: unable to handle kernel NULL pointer dereference at 0000000000000020
IP: [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
PGD 12fded067 PUD 142c06067 PMD 0
Oops: 0000 [#1] SMP
Modules linked in: tun fuse ipt_ULOG binfmt_misc nfnetlink nfc caif_socket caif phonet can llc2 pppoe pppox ppp_generic slhc irda crc_ccitt rds af_key decnet rose x25 atm netrom appletalk ipx p8023 psnap p8022 llc ax25 lockd sunrpc bluetooth rfkill ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 ip6table_filter ip6_tables nf_conntrack_ipv4 nf_defrag_ipv4 xt_state nf_conntrack kvm_intel kvm crc32c_intel ghash_clmulni_intel microcode pcspkr i2c_i801 e1000e uinput i915 video i2c_algo_bit drm_kms_helper drm i2c_core
CPU 5
Pid: 7007, comm: trinity-child5 Not tainted 3.6.0-rc4+ #36
RIP: 0010:[<ffffffff81103365>]  [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
RSP: 0018:ffff880116b33ec8  EFLAGS: 00010246
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000035
RDX: ffffffff819dbf5d RSI: ffffffff819da5bb RDI: 0000000000000000
RBP: ffff880116b33ee8 R08: ffff88001942533e R09: 0000000000000000
R10: 0000000000000001 R11: 0000000000000000 R12: ffff880116b33f38
R13: ffff880116b33f38 R14: ffffffff819fa1eb R15: 0000000000000005
FS:  00007f5742581740(0000) GS:ffff880148600000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000020 CR3: 00000001046d3000 CR4: 00000000001407e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Process trinity-child5 (pid: 7007, threadinfo ffff880116b32000, task ffff880019424d00)
Stack:
 ffff880116b33ee8 0000000000000000 ffff880116b33f38 ffff880019424d00
 ffff880116b33f18 ffffffff81103522 00000000000001d0 ffff880119ed6500
 00000000000001d0 0000000000000109 ffff880116b33f78 ffffffff811ebb95
Call Trace:
 [<ffffffff81103522>] audit_log_link_denied+0x92/0x100
 [<ffffffff811ebb95>] sys_linkat+0x195/0x1e0
 [<ffffffff813573de>] ? trace_hardirqs_on_thunk+0x3a/0x3f
 [<ffffffff816a50ed>] system_call_fastpath+0x1a/0x1f
Code: 5d e8 4c 89 65 f0 4c 89 6d f8 0f 1f 44 00 00 48 85 f6 48 89 fb 49 89 d5 74 11 48 89 f2 31 c0 48 c7 c6 bb a5 9d 81 e8 8b e1 ff ff <8b> 73 20 40 f6 c6 01 75 62 48 8b 3d 33 33 93 01 48 85 ff 74 4e
RIP  [<ffffffff81103365>] audit_log_d_path+0x35/0xf0
 RSP <ffff880116b33ec8>
CR2: 0000000000000020
---[ end trace 85b88c850143bb1c ]---

That's here in kernel/audit.c

1433
1434 /* We will allow 11 spaces for ' (deleted)' to be appended */
1435 pathname = kmalloc(PATH_MAX+11, ab->gfp_mask);

'ab' is NULL.

Looks like audit_log_link_denied needs to handle potential failure from audit_log_start and abort early. (oddly, it looks like every other function called there checks for !ab.)

Maybe additional code should be added here to printk the audit message to dmesg so that we don't lose it entirely, but for now, minimal fix.

kernel/audit.c +0 -0
index ea3b7b6..c3e85bb 100644
@@ -1466,6 +1466,9 @@ void audit_log_link_denied(const char *operation, struct path *link)
 
Quoted 8 lines

kernel/audit.c void audit_log_link_denied(const char *operation, struct path *link) · line 1474

Quoted 8 lines
kernel/audit.c +3 -0
     ab = audit_log_start(current->audit_context, GFP_KERNEL,
                          AUDIT_ANOM_LINK);
+    if (!ab)
+            return;
+
     audit_log_format(ab, "op=%s action=denied", operation);
     audit_log_format(ab, " pid=%d comm=", current->pid);
     audit_log_untrustedstring(ab, current->comm);

This fix looks fine to me.

Quoted 4 lines

Added Kees, as this was introduced in a51d9eaa41866ab6b4b6ecad7b621f8b66ece0dc

I just realised, the funny thing about this is that the machine running that test had selinux/audit disabled. And yet here we are, screwing around with audit buffers.

The intent was to have this message show up in dmesg even if auditd wasn't running, and even if the specific process wasn't being explicitly audited.

Quoted 3 lines

Should there be a test on audit_enable=0 in audit_log_link_denied() ?

I'm now curious how much more of the audit code is getting run through similar lack of tests

What is the condition in which audit_log_start fails?

-Kees

Signature · 2 lines
Kees Cook
Chrome OS Security
author
#8 lore raw
Quoted 7 lines · Kees Cook
Quoted 2 lines

I just realised, the funny thing about this is that the machine running that test had selinux/audit disabled. And yet here we are, screwing around with audit buffers.

The intent was to have this message show up in dmesg even if auditd wasn't running, and even if the specific process wasn't being explicitly audited.

Quoted 5 lines
Quoted 3 lines

Should there be a test on audit_enable=0 in audit_log_link_denied() ?

I'm now curious how much more of the audit code is getting run through similar lack of tests

What is the condition in which audit_log_start fails?

in the case of that oops, given I had booted with audit=0, I suspect it was hitting the first check...

1157 if (audit_initialized != AUDIT_INITIALIZED)
1158 return NULL;

Dave

#9 lore raw
Quoted 10 lines · Dave Jones
Quoted 9 lines · Kees Cook
Quoted 2 lines

I just realised, the funny thing about this is that the machine running that test had selinux/audit disabled. And yet here we are, screwing around with audit buffers.

The intent was to have this message show up in dmesg even if auditd wasn't running, and even if the specific process wasn't being explicitly audited.

Quoted 2 lines

Should there be a test on audit_enable=0 in audit_log_link_denied() ?

Quoted 8 lines
Quoted 3 lines
Quoted 1 line

I'm now curious how much more of the audit code is getting run through similar lack of tests

What is the condition in which audit_log_start fails?

in the case of that oops, given I had booted with audit=0, I suspect it was hitting the first check...

1157 if (audit_initialized != AUDIT_INITIALIZED)
1158 return NULL;

Ah-ha, okay. Yeah, I'm fine with the fix you had. If _start fails, just return.

-Kees

Signature · 2 lines
Kees Cook
Chrome OS Security