]> www.infradead.org Git - nvme.git/commitdiff
ima: Avoid blocking in RCU read-side critical section
authorGUO Zihua <guozihua@huawei.com>
Tue, 7 May 2024 01:25:41 +0000 (01:25 +0000)
committerPaul Moore <paul@paul-moore.com>
Thu, 13 Jun 2024 18:26:50 +0000 (14:26 -0400)
A panic happens in ima_match_policy:

BUG: unable to handle kernel NULL pointer dereference at 0000000000000010
PGD 42f873067 P4D 0
Oops: 0000 [#1] SMP NOPTI
CPU: 5 PID: 1286325 Comm: kubeletmonit.sh
Kdump: loaded Tainted: P
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
               BIOS 0.0.0 02/06/2015
RIP: 0010:ima_match_policy+0x84/0x450
Code: 49 89 fc 41 89 cf 31 ed 89 44 24 14 eb 1c 44 39
      7b 18 74 26 41 83 ff 05 74 20 48 8b 1b 48 3b 1d
      f2 b9 f4 00 0f 84 9c 01 00 00 <44> 85 73 10 74 ea
      44 8b 6b 14 41 f6 c5 01 75 d4 41 f6 c5 02 74 0f
RSP: 0018:ff71570009e07a80 EFLAGS: 00010207
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000200
RDX: ffffffffad8dc7c0 RSI: 0000000024924925 RDI: ff3e27850dea2000
RBP: 0000000000000000 R08: 0000000000000000 R09: ffffffffabfce739
R10: ff3e27810cc42400 R11: 0000000000000000 R12: ff3e2781825ef970
R13: 00000000ff3e2785 R14: 000000000000000c R15: 0000000000000001
FS:  00007f5195b51740(0000)
GS:ff3e278b12d40000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000010 CR3: 0000000626d24002 CR4: 0000000000361ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 ima_get_action+0x22/0x30
 process_measurement+0xb0/0x830
 ? page_add_file_rmap+0x15/0x170
 ? alloc_set_pte+0x269/0x4c0
 ? prep_new_page+0x81/0x140
 ? simple_xattr_get+0x75/0xa0
 ? selinux_file_open+0x9d/0xf0
 ima_file_check+0x64/0x90
 path_openat+0x571/0x1720
 do_filp_open+0x9b/0x110
 ? page_counter_try_charge+0x57/0xc0
 ? files_cgroup_alloc_fd+0x38/0x60
 ? __alloc_fd+0xd4/0x250
 ? do_sys_open+0x1bd/0x250
 do_sys_open+0x1bd/0x250
 do_syscall_64+0x5d/0x1d0
 entry_SYSCALL_64_after_hwframe+0x65/0xca

Commit c7423dbdbc9e ("ima: Handle -ESTALE returned by
ima_filter_rule_match()") introduced call to ima_lsm_copy_rule within a
RCU read-side critical section which contains kmalloc with GFP_KERNEL.
This implies a possible sleep and violates limitations of RCU read-side
critical sections on non-PREEMPT systems.

Sleeping within RCU read-side critical section might cause
synchronize_rcu() returning early and break RCU protection, allowing a
UAF to happen.

The root cause of this issue could be described as follows:
| Thread A | Thread B |
| |ima_match_policy |
| |  rcu_read_lock |
|ima_lsm_update_rule | |
|  synchronize_rcu | |
| |    kmalloc(GFP_KERNEL)|
| |      sleep |
==> synchronize_rcu returns early
|  kfree(entry) | |
| |    entry = entry->next|
==> UAF happens and entry now becomes NULL (or could be anything).
| |    entry->action |
==> Accessing entry might cause panic.

To fix this issue, we are converting all kmalloc that is called within
RCU read-side critical section to use GFP_ATOMIC.

Fixes: c7423dbdbc9e ("ima: Handle -ESTALE returned by ima_filter_rule_match()")
Cc: stable@vger.kernel.org
Signed-off-by: GUO Zihua <guozihua@huawei.com>
Acked-by: John Johansen <john.johansen@canonical.com>
Reviewed-by: Mimi Zohar <zohar@linux.ibm.com>
Reviewed-by: Casey Schaufler <casey@schaufler-ca.com>
[PM: fixed missing comment, long lines, !CONFIG_IMA_LSM_RULES case]
Signed-off-by: Paul Moore <paul@paul-moore.com>
include/linux/lsm_hook_defs.h
include/linux/security.h
kernel/auditfilter.c
security/apparmor/audit.c
security/apparmor/include/audit.h
security/integrity/ima/ima.h
security/integrity/ima/ima_policy.c
security/security.c
security/selinux/include/audit.h
security/selinux/ss/services.c
security/smack/smack_lsm.c

index f804b76cde44eeda37830fa80851ec9dd2843626..44488b1ab9a97684cc881f319594aee31ee28421 100644 (file)
@@ -413,7 +413,7 @@ LSM_HOOK(void, LSM_RET_VOID, key_post_create_or_update, struct key *keyring,
 
 #ifdef CONFIG_AUDIT
 LSM_HOOK(int, 0, audit_rule_init, u32 field, u32 op, char *rulestr,
-        void **lsmrule)
+        void **lsmrule, gfp_t gfp)
 LSM_HOOK(int, 0, audit_rule_known, struct audit_krule *krule)
 LSM_HOOK(int, 0, audit_rule_match, u32 secid, u32 field, u32 op, void *lsmrule)
 LSM_HOOK(void, LSM_RET_VOID, audit_rule_free, void *lsmrule)
index 21cf70346b330e11d305f6b2c8f85a6c1f6a0e4a..de3af33e6ff502c2d4cc0ecc50724dfbd83faf37 100644 (file)
@@ -2048,7 +2048,8 @@ static inline void security_key_post_create_or_update(struct key *keyring,
 
 #ifdef CONFIG_AUDIT
 #ifdef CONFIG_SECURITY
-int security_audit_rule_init(u32 field, u32 op, char *rulestr, void **lsmrule);
+int security_audit_rule_init(u32 field, u32 op, char *rulestr, void **lsmrule,
+                            gfp_t gfp);
 int security_audit_rule_known(struct audit_krule *krule);
 int security_audit_rule_match(u32 secid, u32 field, u32 op, void *lsmrule);
 void security_audit_rule_free(void *lsmrule);
@@ -2056,7 +2057,7 @@ void security_audit_rule_free(void *lsmrule);
 #else
 
 static inline int security_audit_rule_init(u32 field, u32 op, char *rulestr,
-                                          void **lsmrule)
+                                          void **lsmrule, gfp_t gfp)
 {
        return 0;
 }
index be8c680121e466efb71d487cef470eddc35ad574..d6ef4f4f9cba503e9dcd6c35f30e3a834729df5f 100644 (file)
@@ -529,7 +529,8 @@ static struct audit_entry *audit_data_to_entry(struct audit_rule_data *data,
                        entry->rule.buflen += f_val;
                        f->lsm_str = str;
                        err = security_audit_rule_init(f->type, f->op, str,
-                                                      (void **)&f->lsm_rule);
+                                                      (void **)&f->lsm_rule,
+                                                      GFP_KERNEL);
                        /* Keep currently invalid fields around in case they
                         * become valid after a policy reload. */
                        if (err == -EINVAL) {
@@ -799,7 +800,7 @@ static inline int audit_dupe_lsm_field(struct audit_field *df,
 
        /* our own (refreshed) copy of lsm_rule */
        ret = security_audit_rule_init(df->type, df->op, df->lsm_str,
-                                      (void **)&df->lsm_rule);
+                                      (void **)&df->lsm_rule, GFP_KERNEL);
        /* Keep currently invalid fields around in case they
         * become valid after a policy reload. */
        if (ret == -EINVAL) {
index 45beb1c5f747a8492cb84eec0e680103b54f1f9d..6b5181c668b5b7c1937e16ef703fbfe10e1fcb5f 100644 (file)
@@ -217,7 +217,7 @@ void aa_audit_rule_free(void *vrule)
        }
 }
 
-int aa_audit_rule_init(u32 field, u32 op, char *rulestr, void **vrule)
+int aa_audit_rule_init(u32 field, u32 op, char *rulestr, void **vrule, gfp_t gfp)
 {
        struct aa_audit_rule *rule;
 
@@ -230,14 +230,14 @@ int aa_audit_rule_init(u32 field, u32 op, char *rulestr, void **vrule)
                return -EINVAL;
        }
 
-       rule = kzalloc(sizeof(struct aa_audit_rule), GFP_KERNEL);
+       rule = kzalloc(sizeof(struct aa_audit_rule), gfp);
 
        if (!rule)
                return -ENOMEM;
 
        /* Currently rules are treated as coming from the root ns */
        rule->label = aa_label_parse(&root_ns->unconfined->label, rulestr,
-                                    GFP_KERNEL, true, false);
+                                    gfp, true, false);
        if (IS_ERR(rule->label)) {
                int err = PTR_ERR(rule->label);
                aa_audit_rule_free(rule);
index acbb03b9bd25cce3fd314f2b8a8db9936259fa2c..0c8cc86b417b563f644d6d06e7e7c250750372fb 100644 (file)
@@ -200,7 +200,7 @@ static inline int complain_error(int error)
 }
 
 void aa_audit_rule_free(void *vrule);
-int aa_audit_rule_init(u32 field, u32 op, char *rulestr, void **vrule);
+int aa_audit_rule_init(u32 field, u32 op, char *rulestr, void **vrule, gfp_t gfp);
 int aa_audit_rule_known(struct audit_krule *rule);
 int aa_audit_rule_match(u32 sid, u32 field, u32 op, void *vrule);
 
index 3e568126cd48164ed851eee8166dc17aa64ba4ad..c51e24d24d1e9ec64d0713016213a8ba02c8da45 100644 (file)
@@ -546,7 +546,7 @@ static inline void ima_free_modsig(struct modsig *modsig)
 #else
 
 static inline int ima_filter_rule_init(u32 field, u32 op, char *rulestr,
-                                      void **lsmrule)
+                                      void **lsmrule, gfp_t gfp)
 {
        return -EINVAL;
 }
index c0556907c2e67a9307810eb56cd66052b7a53cc9..09da8e63923956855897795df8cb3748e39b2fd3 100644 (file)
@@ -401,7 +401,8 @@ static void ima_free_rule(struct ima_rule_entry *entry)
        kfree(entry);
 }
 
-static struct ima_rule_entry *ima_lsm_copy_rule(struct ima_rule_entry *entry)
+static struct ima_rule_entry *ima_lsm_copy_rule(struct ima_rule_entry *entry,
+                                               gfp_t gfp)
 {
        struct ima_rule_entry *nentry;
        int i;
@@ -410,7 +411,7 @@ static struct ima_rule_entry *ima_lsm_copy_rule(struct ima_rule_entry *entry)
         * Immutable elements are copied over as pointers and data; only
         * lsm rules can change
         */
-       nentry = kmemdup(entry, sizeof(*nentry), GFP_KERNEL);
+       nentry = kmemdup(entry, sizeof(*nentry), gfp);
        if (!nentry)
                return NULL;
 
@@ -425,7 +426,8 @@ static struct ima_rule_entry *ima_lsm_copy_rule(struct ima_rule_entry *entry)
 
                ima_filter_rule_init(nentry->lsm[i].type, Audit_equal,
                                     nentry->lsm[i].args_p,
-                                    &nentry->lsm[i].rule);
+                                    &nentry->lsm[i].rule,
+                                    gfp);
                if (!nentry->lsm[i].rule)
                        pr_warn("rule for LSM \'%s\' is undefined\n",
                                nentry->lsm[i].args_p);
@@ -438,7 +440,7 @@ static int ima_lsm_update_rule(struct ima_rule_entry *entry)
        int i;
        struct ima_rule_entry *nentry;
 
-       nentry = ima_lsm_copy_rule(entry);
+       nentry = ima_lsm_copy_rule(entry, GFP_KERNEL);
        if (!nentry)
                return -ENOMEM;
 
@@ -664,7 +666,7 @@ retry:
                }
 
                if (rc == -ESTALE && !rule_reinitialized) {
-                       lsm_rule = ima_lsm_copy_rule(rule);
+                       lsm_rule = ima_lsm_copy_rule(rule, GFP_ATOMIC);
                        if (lsm_rule) {
                                rule_reinitialized = true;
                                goto retry;
@@ -1140,7 +1142,8 @@ static int ima_lsm_rule_init(struct ima_rule_entry *entry,
        entry->lsm[lsm_rule].type = audit_type;
        result = ima_filter_rule_init(entry->lsm[lsm_rule].type, Audit_equal,
                                      entry->lsm[lsm_rule].args_p,
-                                     &entry->lsm[lsm_rule].rule);
+                                     &entry->lsm[lsm_rule].rule,
+                                     GFP_KERNEL);
        if (!entry->lsm[lsm_rule].rule) {
                pr_warn("rule for LSM \'%s\' is undefined\n",
                        entry->lsm[lsm_rule].args_p);
index e5da848c50b9156c4eab30fa3cc56b2694085309..e5ca08789f74183dbe6dacc68dde6bafa6a9b0f0 100644 (file)
@@ -5332,15 +5332,17 @@ void security_key_post_create_or_update(struct key *keyring, struct key *key,
  * @op: rule operator
  * @rulestr: rule context
  * @lsmrule: receive buffer for audit rule struct
+ * @gfp: GFP flag used for kmalloc
  *
  * Allocate and initialize an LSM audit rule structure.
  *
  * Return: Return 0 if @lsmrule has been successfully set, -EINVAL in case of
  *         an invalid rule.
  */
-int security_audit_rule_init(u32 field, u32 op, char *rulestr, void **lsmrule)
+int security_audit_rule_init(u32 field, u32 op, char *rulestr, void **lsmrule,
+                            gfp_t gfp)
 {
-       return call_int_hook(audit_rule_init, field, op, rulestr, lsmrule);
+       return call_int_hook(audit_rule_init, field, op, rulestr, lsmrule, gfp);
 }
 
 /**
index 52aca71210b476be5353d8b7332baf73eca8b361..29c7d4c86f6d57381cb852abb3277dc356afaf5f 100644 (file)
  *     @op: the operator the rule uses
  *     @rulestr: the text "target" of the rule
  *     @rule: pointer to the new rule structure returned via this
+ *     @gfp: GFP flag used for kmalloc
  *
  *     Returns 0 if successful, -errno if not.  On success, the rule structure
  *     will be allocated internally.  The caller must free this structure with
  *     selinux_audit_rule_free() after use.
  */
-int selinux_audit_rule_init(u32 field, u32 op, char *rulestr, void **rule);
+int selinux_audit_rule_init(u32 field, u32 op, char *rulestr, void **rule,
+                           gfp_t gfp);
 
 /**
  *     selinux_audit_rule_free - free an selinux audit rule structure.
index f20e1968b7f7a848661e79652d453d12e5e63dbf..e33e55384b75aaf1468af0209a3189db1f0e0dd2 100644 (file)
@@ -3507,7 +3507,8 @@ void selinux_audit_rule_free(void *vrule)
        }
 }
 
-int selinux_audit_rule_init(u32 field, u32 op, char *rulestr, void **vrule)
+int selinux_audit_rule_init(u32 field, u32 op, char *rulestr, void **vrule,
+                           gfp_t gfp)
 {
        struct selinux_state *state = &selinux_state;
        struct selinux_policy *policy;
@@ -3548,7 +3549,7 @@ int selinux_audit_rule_init(u32 field, u32 op, char *rulestr, void **vrule)
                return -EINVAL;
        }
 
-       tmprule = kzalloc(sizeof(struct selinux_audit_rule), GFP_KERNEL);
+       tmprule = kzalloc(sizeof(struct selinux_audit_rule), gfp);
        if (!tmprule)
                return -ENOMEM;
        context_init(&tmprule->au_ctxt);
index 70ba2841e181d3374eb31eb7b1d52038744f0a9b..f5cbec1e6a92301ed8701ccc82d4d5d59df93878 100644 (file)
@@ -4693,11 +4693,13 @@ static int smack_post_notification(const struct cred *w_cred,
  * @op: required testing operator (=, !=, >, <, ...)
  * @rulestr: smack label to be audited
  * @vrule: pointer to save our own audit rule representation
+ * @gfp: type of the memory for the allocation
  *
  * Prepare to audit cases where (@field @op @rulestr) is true.
  * The label to be audited is created if necessay.
  */
-static int smack_audit_rule_init(u32 field, u32 op, char *rulestr, void **vrule)
+static int smack_audit_rule_init(u32 field, u32 op, char *rulestr, void **vrule,
+                                gfp_t gfp)
 {
        struct smack_known *skp;
        char **rule = (char **)vrule;