| Message ID | 20240507093714.1031820-1-guozihua@huawei.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | ima: fix deadlock when traversing "ima_default_rules". | expand |
On Tue, 2024-05-07 at 09:37 +0000, GUO Zihua wrote: > From: liqiong <liqiong@nfschina.com> > > [ Upstream commit eb0782bbdfd0d7c4786216659277c3fd585afc0e ] > > The current IMA ruleset is identified by the variable "ima_rules" > that default to "&ima_default_rules". When loading a custom policy > for the first time, the variable is updated to "&ima_policy_rules" > instead. That update isn't RCU-safe, and deadlocks are possible. > Indeed, some functions like ima_match_policy() may loop indefinitely > when traversing "ima_default_rules" with list_for_each_entry_rcu(). > > When iterating over the default ruleset back to head, if the list > head is "ima_default_rules", and "ima_rules" have been updated to > "&ima_policy_rules", the loop condition (&entry->list != ima_rules) > stays always true, traversing won't terminate, causing a soft lockup > and RCU stalls. > > Introduce a temporary value for "ima_rules" when iterating over > the ruleset to avoid the deadlocks. > > Addition: > > A rcu_read_lock pair is added within ima_update_policy_flag to avoid > suspicious RCU usage warning. This pair of RCU lock was added with > commit 4f2946aa0c45 ("IMA: introduce a new policy option > func=SETXATTR_CHECK") on mainstream. > > Signed-off-by: liqiong <liqiong@nfschina.com> > Reviewed-by: THOBY Simon <Simon.THOBY@viveris.fr> > Fixes: 38d859f991f3 ("IMA: policy can now be updated multiple times") > Reported-by: kernel test robot <lkp@intel.com> (Fix sparse: incompatible types in comparison expression.) > Signed-off-by: Mimi Zohar <zohar@linux.ibm.com> > Sig=ned-off-by: GUO Zihua <guozihua@huawei.com> Hi Scott, I'm confused by this patch. Is it meant for upstream? thanks, Mimi
On 2024/5/7 19:54, Mimi Zohar wrote: > On Tue, 2024-05-07 at 09:37 +0000, GUO Zihua wrote: >> From: liqiong <liqiong@nfschina.com> >> >> [ Upstream commit eb0782bbdfd0d7c4786216659277c3fd585afc0e ] >> >> The current IMA ruleset is identified by the variable "ima_rules" >> that default to "&ima_default_rules". When loading a custom policy >> for the first time, the variable is updated to "&ima_policy_rules" >> instead. That update isn't RCU-safe, and deadlocks are possible. >> Indeed, some functions like ima_match_policy() may loop indefinitely >> when traversing "ima_default_rules" with list_for_each_entry_rcu(). >> >> When iterating over the default ruleset back to head, if the list >> head is "ima_default_rules", and "ima_rules" have been updated to >> "&ima_policy_rules", the loop condition (&entry->list != ima_rules) >> stays always true, traversing won't terminate, causing a soft lockup >> and RCU stalls. >> >> Introduce a temporary value for "ima_rules" when iterating over >> the ruleset to avoid the deadlocks. >> >> Addition: >> >> A rcu_read_lock pair is added within ima_update_policy_flag to avoid >> suspicious RCU usage warning. This pair of RCU lock was added with >> commit 4f2946aa0c45 ("IMA: introduce a new policy option >> func=SETXATTR_CHECK") on mainstream. >> >> Signed-off-by: liqiong <liqiong@nfschina.com> >> Reviewed-by: THOBY Simon <Simon.THOBY@viveris.fr> >> Fixes: 38d859f991f3 ("IMA: policy can now be updated multiple times") >> Reported-by: kernel test robot <lkp@intel.com> (Fix sparse: incompatible types in comparison expression.) >> Signed-off-by: Mimi Zohar <zohar@linux.ibm.com> >> Sig=ned-off-by: GUO Zihua <guozihua@huawei.com> > > Hi Scott, > > I'm confused by this patch. Is it meant for upstream? > > thanks, > > Mimi > It's a backport from upstream.
On 2024/5/8 10:06, Guozihua (Scott) wrote: > On 2024/5/7 19:54, Mimi Zohar wrote: >> On Tue, 2024-05-07 at 09:37 +0000, GUO Zihua wrote: >>> From: liqiong <liqiong@nfschina.com> >>> >>> [ Upstream commit eb0782bbdfd0d7c4786216659277c3fd585afc0e ] >>> >>> The current IMA ruleset is identified by the variable "ima_rules" >>> that default to "&ima_default_rules". When loading a custom policy >>> for the first time, the variable is updated to "&ima_policy_rules" >>> instead. That update isn't RCU-safe, and deadlocks are possible. >>> Indeed, some functions like ima_match_policy() may loop indefinitely >>> when traversing "ima_default_rules" with list_for_each_entry_rcu(). >>> >>> When iterating over the default ruleset back to head, if the list >>> head is "ima_default_rules", and "ima_rules" have been updated to >>> "&ima_policy_rules", the loop condition (&entry->list != ima_rules) >>> stays always true, traversing won't terminate, causing a soft lockup >>> and RCU stalls. >>> >>> Introduce a temporary value for "ima_rules" when iterating over >>> the ruleset to avoid the deadlocks. >>> >>> Addition: >>> >>> A rcu_read_lock pair is added within ima_update_policy_flag to avoid >>> suspicious RCU usage warning. This pair of RCU lock was added with >>> commit 4f2946aa0c45 ("IMA: introduce a new policy option >>> func=SETXATTR_CHECK") on mainstream. >>> >>> Signed-off-by: liqiong <liqiong@nfschina.com> >>> Reviewed-by: THOBY Simon <Simon.THOBY@viveris.fr> >>> Fixes: 38d859f991f3 ("IMA: policy can now be updated multiple times") >>> Reported-by: kernel test robot <lkp@intel.com> (Fix sparse: incompatible types in comparison expression.) >>> Signed-off-by: Mimi Zohar <zohar@linux.ibm.com> >>> Sig=ned-off-by: GUO Zihua <guozihua@huawei.com> >> >> Hi Scott, >> >> I'm confused by this patch. Is it meant for upstream? >> >> thanks, >> >> Mimi >> > It's a backport from upstream. > To clarify, it's meant for Linux-5.10.y.
diff --git a/security/integrity/ima/ima_policy.c b/security/integrity/ima/ima_policy.c index 1c403e8a8044..4f5d44037081 100644 --- a/security/integrity/ima/ima_policy.c +++ b/security/integrity/ima/ima_policy.c @@ -210,7 +210,7 @@ static struct ima_rule_entry *arch_policy_entry __ro_after_init; static LIST_HEAD(ima_default_rules); static LIST_HEAD(ima_policy_rules); static LIST_HEAD(ima_temp_rules); -static struct list_head *ima_rules = &ima_default_rules; +static struct list_head __rcu *ima_rules = (struct list_head __rcu *)(&ima_default_rules); static int ima_policy __initdata; @@ -648,12 +648,14 @@ int ima_match_policy(struct inode *inode, const struct cred *cred, u32 secid, { struct ima_rule_entry *entry; int action = 0, actmask = flags | (flags << 1); + struct list_head *ima_rules_tmp; if (template_desc) *template_desc = ima_template_desc_current(); rcu_read_lock(); - list_for_each_entry_rcu(entry, ima_rules, list) { + ima_rules_tmp = rcu_dereference(ima_rules); + list_for_each_entry_rcu(entry, ima_rules_tmp, list) { if (!(entry->action & actmask)) continue; @@ -701,11 +703,15 @@ int ima_match_policy(struct inode *inode, const struct cred *cred, u32 secid, void ima_update_policy_flag(void) { struct ima_rule_entry *entry; + struct list_head *ima_rules_tmp; - list_for_each_entry(entry, ima_rules, list) { + rcu_read_lock(); + ima_rules_tmp = rcu_dereference(ima_rules); + list_for_each_entry_rcu(entry, ima_rules_tmp, list) { if (entry->action & IMA_DO_MASK) ima_policy_flag |= entry->action; } + rcu_read_unlock(); ima_appraise |= (build_ima_appraise | temp_ima_appraise); if (!ima_appraise) @@ -898,10 +904,10 @@ void ima_update_policy(void) list_splice_tail_init_rcu(&ima_temp_rules, policy, synchronize_rcu); - if (ima_rules != policy) { + if (ima_rules != (struct list_head __rcu *)policy) { ima_policy_flag = 0; - ima_rules = policy; + rcu_assign_pointer(ima_rules, policy); /* * IMA architecture specific policy rules are specified * as strings and converted to an array of ima_entry_rules @@ -989,7 +995,7 @@ static int ima_lsm_rule_init(struct ima_rule_entry *entry, pr_warn("rule for LSM \'%s\' is undefined\n", entry->lsm[lsm_rule].args_p); - if (ima_rules == &ima_default_rules) { + if (ima_rules == (struct list_head __rcu *)(&ima_default_rules)) { kfree(entry->lsm[lsm_rule].args_p); entry->lsm[lsm_rule].args_p = NULL; result = -EINVAL; @@ -1598,9 +1604,11 @@ void *ima_policy_start(struct seq_file *m, loff_t *pos) { loff_t l = *pos; struct ima_rule_entry *entry; + struct list_head *ima_rules_tmp; rcu_read_lock(); - list_for_each_entry_rcu(entry, ima_rules, list) { + ima_rules_tmp = rcu_dereference(ima_rules); + list_for_each_entry_rcu(entry, ima_rules_tmp, list) { if (!l--) { rcu_read_unlock(); return entry; @@ -1619,7 +1627,8 @@ void *ima_policy_next(struct seq_file *m, void *v, loff_t *pos) rcu_read_unlock(); (*pos)++; - return (&entry->list == ima_rules) ? NULL : entry; + return (&entry->list == &ima_default_rules || + &entry->list == &ima_policy_rules) ? NULL : entry; } void ima_policy_stop(struct seq_file *m, void *v) @@ -1823,6 +1832,7 @@ bool ima_appraise_signature(enum kernel_read_file_id id) struct ima_rule_entry *entry; bool found = false; enum ima_hooks func; + struct list_head *ima_rules_tmp; if (id >= READING_MAX_ID) return false; @@ -1834,7 +1844,8 @@ bool ima_appraise_signature(enum kernel_read_file_id id) func = read_idmap[id] ?: FILE_CHECK; rcu_read_lock(); - list_for_each_entry_rcu(entry, ima_rules, list) { + ima_rules_tmp = rcu_dereference(ima_rules); + list_for_each_entry_rcu(entry, ima_rules_tmp, list) { if (entry->action != APPRAISE) continue;