audit: fix recursive locking deadlock and dangling rule removal
cover audit: fix recursive locking deadlock and dangling rule removal
This patch series fixes a recursive locking deadlock in audit_dupe_exe() code path, as well as, a bug that prevents the deletion of dangling audit rules.
The first patch fixes the deadlock in audit_dupe_exe() by introducing the audit_watch_ctx struct to pass the fsnotify event context down the call chain. This safely bypasses the need for kern_path_parent() during move events, avoiding a double acquisition of the I_MUTEX_PARENT lock.
The second one, simply removes the d_really_is_negative() check in audit_alloc_mark() to allow dummy mark allocation, fixing the -ENOENT failure when deleting dangling rules for unlinked executables.
---
Ricardo Robaina (2):
audit: fix recursive locking deadlock in audit_dupe_exe()
audit: fix removal of dangling executable rules
kernel/audit.h | 13 ++++++++++--- kernel/audit_fsnotify.c | 29 ++++++++++++++++++----------- kernel/audit_watch.c | 25 +++++++++++++++++-------- kernel/auditfilter.c | 9 +++++---- 4 files changed, 50 insertions(+), 26 deletions(-)
Signature · 1 line
2.53.0
PATCH 1/2 audit: fix recursive locking deadlock in audit_dupe_exe()
A deadlock occurs in the audit subsystem when duplicating executable-related rules.
When a file is moved (e.g., via do_renameat2()), the VFS layer locks the parent directory (I_MUTEX_PARENT), which synchronously triggers an fsnotify_move event. If an existing executable audit rule matches the file being moved, the audit subsystem catches this event and calls audit_dupe_exe() to duplicate the watch and update the rule. Then, audit_alloc_mark() would call kern_path_parent() to resolve the path, leading to a blind attempt to acquire the exact same I_MUTEX_PARENT lock already held by the task, resulting in the following recursive locking deadlock:
[ 60.521149] ============================================
[ 60.521164] WARNING: possible recursive locking detected
[ 60.521180] 6.12.0-55.27.1.el10_0.x86_64+debug #1 Not tainted
[ 60.521197] --------------------------------------------
[ 60.521211] mv/5099 is trying to acquire lock:
[ 60.521225] ffff888132845358 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3},
at: __kern_path_locked+0x10a/0x2f0
[ 60.521263]
but task is already holding lock:
[ 60.521280] ffff888132846b58 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3},
at: lock_two_directories+0x13f/0x2b0
[ 60.521313]
other info that might help us debug this:
[ 60.521331] Possible unsafe locking scenario: [ 60.521349] CPU0
[ 60.521357] ----
[ 60.521365] lock(&inode->i_sb->s_type->i_mutex_dir_key/1);
[ 60.521384] lock(&inode->i_sb->s_type->i_mutex_dir_key/1);
[ 60.521401]
*** DEADLOCK ***[ 60.521422] May be due to missing lock nesting notation
[ 60.521445] 6 locks held by mv/5099:
[ 60.521458] #0: ffff888112a9c440 (sb_writers#13){++++}-{0:0}, at: do_renameat2+0x34c/0xbc0
[ 60.521491] #1: ffff888112a9c790 (&type->s_vfs_rename_key#3){+.+.}-{3:3}, at: do_renameat2+0x415/0xbc0
[ 60.521525] #2: ffff888132846b58 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3},
at: lock_two_directories+0x13f/0x2b0
[ 60.521565] #3: ffff888132845358 (&inode->i_sb->s_type->i_mutex_dir_key/5){+.+.}-{3:3},
at: lock_two_directories+0x175/0x2b0
[ 60.521604] #4: ffffffffb3a1fb10 (&fsnotify_mark_srcu){.+.+}-{0:0}, at: fsnotify+0x454/0x28a0
[ 60.521636] #5: ffffffffaf886230 (audit_filter_mutex){+.+.}-{3:3}, at: audit_update_watch+0x36/0x11e0
[ 60.521670]
stack backtrace:
[ 60.521687] CPU: 0 UID: 0 PID: 5099 Comm: mv Kdump: loaded Not tainted 6.12.0-55.27.1.el10_0.x86_64+debug #1
[ 60.521718] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-2.fc40 04/01/2014
[ 60.521743] Call Trace:
[ 60.521754] <TASK>
[ 60.521764] dump_stack_lvl+0x6f/0xb0
[ 60.521781] print_deadlock_bug.cold+0xbd/0xca
[ 60.521798] validate_chain+0x83a/0xf00
[ 60.522465] __lock_acquire+0xcac/0x1d20
[ 60.524073] lock_acquire.part.0+0x11b/0x360
[ 60.528512] down_write_nested+0x9f/0x230
[ 60.530038] __kern_path_locked+0x10a/0x2f0
[ 60.532561] kern_path_locked+0x26/0x40
[ 60.533018] audit_alloc_mark+0xfb/0x4f0
[ 60.535295] audit_dupe_exe+0x6c/0xe0
[ 60.536307] audit_dupe_rule+0x6c2/0xc00
[ 60.537290] audit_update_watch+0x4cc/0x11e0
[ 60.537759] audit_watch_handle_event+0x12c/0x1b0
[ 60.538227] send_to_group+0x5d0/0x8b0
[ 60.538683] fsnotify+0x615/0x28a0
[ 60.540007] fsnotify_move+0x1d8/0x630
[ 60.540438] vfs_rename+0xdcd/0x1df0
[ 60.542176] do_renameat2+0x9d4/0xbc0
[ 60.544870] __x64_sys_renameat+0x192/0x260
[ 60.545299] do_syscall_64+0x92/0x180
[ 60.554189] entry_SYSCALL_64_after_hwframe+0x76/0x7e
[ 60.554562] RIP: 0033:0x7f0491fe8c4e
[ 60.554947] Code: 0f 1f 40 00 48 8b 15 c1 e1 16 00 f7 d8 64 89 02 b8 ff ff ff ff
c3 66 0f 1f 44 00 00 f3 0f 1e fa 49 89 ca b8 08 01 00 00 0f 05 <48> 3d 00 f0 ff ff
77 0a c3 66 0f 1f 84 00 00 00 00 00 48 8b 15 89
[ 60.555789] RSP: 002b:00007ffc7210bf38 EFLAGS: 00000246 ORIG_RAX: 0000000000000108
[ 60.556237] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0491fe8c4e
[ 60.556683] RDX: 0000000000000003 RSI: 00007ffc7210e6c8 RDI: 00000000ffffff9c
[ 60.557166] RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000001
[ 60.557618] R10: 00005575eb2dae2a R11: 0000000000000246 R12: 00005575eb2dae2a
[ 60.558080] R13: 00007ffc7210e6c8 R14: 0000000000000003 R15: 00000000ffffff9c
[ 60.558518] </TASK>The aforementioned deadlock can be consistently reproduced by running the script below:
audit-dupe-exe-deadlock.sh
--------------------------
#!/bin/bash
auditctl -D
mkdir -p /tmp/foo
touch /tmp/file
auditctl -a always,exit -F exe=/tmp/file -F path=/tmp/file -S all -k dr mv /tmp/file /tmp/foo/file
rm -Rf /tmp/foo
This patch fixes the issue by introducing struct audit_watch_ctx to pass the fsnotify event context down to audit_alloc_mark(). By utilizing the already-resolved directory inode provided by the event, we bypass the kern_path_parent() path resolution entirely, safely avoiding the recursive lock. Furthermore, it explicitly allows duplicate fsnotify marks (allow_dups = 1) during the rename update, allowing the new rule's mark to safely coexist with the old rule's mark until the old rule is freed.
ps.: this issue was identified and reproduced during a comprehensive code coverage analysis of the audit subsystem. The full report is available at the link below.
--- kernel/audit.h | 13 ++++++++++--- kernel/audit_fsnotify.c | 33 ++++++++++++++++++++++----------- kernel/audit_watch.c | 25 +++++++++++++++++-------- kernel/auditfilter.c | 9 +++++---- 4 files changed, 54 insertions(+), 26 deletions(-)
kernel/audit.h +10 -3
@@ -256,8 +256,13 @@ extern int audit_del_rule(struct audit_entry *entry);extern void audit_free_rule_rcu (struct rcu_head *head);extern struct list_head audit_filter_list[]; -extern struct audit_entry *audit_dupe_rule (struct audit_krule *old); +struct audit_watch_ctx { +struct inode *dir; +struct inode *child; +}; +extern struct audit_entry *audit_dupe_rule (struct audit_krule *old, +struct audit_watch_ctx *ctx);extern void audit_log_d_path_exe (struct audit_buffer *ab,struct mm_struct *mm);
@@ -280,13 +285,15 @@ extern char *audit_watch_path(struct audit_watch *watch);extern int audit_watch_compare (struct audit_watch *watch,u64 ino,dev_t dev);extern struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule, -char *pathname,int len); +char *pathname,int len, +struct audit_watch_ctx *ctx);extern char *audit_mark_path (struct audit_fsnotify_mark *mark);extern void audit_remove_mark (struct audit_fsnotify_mark *audit_mark);extern void audit_remove_mark_rule (struct audit_krule *krule);extern int audit_mark_compare (struct audit_fsnotify_mark *mark,u64 ino,dev_t dev); -extern int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old); +extern int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old, +struct audit_watch_ctx *ctx);extern int audit_exe_compare (struct task_struct *tsk,struct audit_fsnotify_mark *mark);
kernel/audit_fsnotify.c +22 -11
@@ -71,7 +71,8 @@ static void audit_update_mark(struct audit_fsnotify_mark *audit_mark, audit_mark->ino = inode ? inode->i_ino :AUDIT_INO_UNSET ; } -struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule,char *pathname,int len) +struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule,char *pathname, +int len,struct audit_watch_ctx *ctx) {struct audit_fsnotify_mark *audit_mark;struct path path;
@@ -81,12 +82,14 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *paif (pathname[0 ] !='/' || pathname[len-1 ] =='/' )return ERR_PTR (-EINVAL ); - dentry =kern_path_parent (pathname, &path); -if (IS_ERR (dentry)) -return ERR_CAST (dentry);/* returning an error */ -if (d_really_is_negative (dentry)) { - audit_mark =ERR_PTR (-ENOENT ); - goto out; +if (!ctx) { + dentry =kern_path_parent (pathname, &path); +if (IS_ERR (dentry)) +return ERR_CAST (dentry);/* returning an error */ +if (d_really_is_negative (dentry)) { + audit_mark =ERR_PTR (-ENOENT ); + goto out; + } } audit_mark =kzalloc_obj (*audit_mark);
@@ -98,18 +101,26 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pafsnotify_init_mark (&audit_mark->mark, audit_fsnotify_group); audit_mark->mark.mask =AUDIT_FS_EVENTS ; audit_mark->path = pathname; -audit_update_mark (audit_mark, dentry->d_inode); audit_mark->rule = krule; - ret =fsnotify_add_inode_mark (&audit_mark->mark, path.dentry->d_inode,0 ); +if (ctx) { +audit_update_mark (audit_mark, ctx->child); + ret =fsnotify_add_inode_mark (&audit_mark->mark, ctx->dir,1 ); + }else { +audit_update_mark (audit_mark, dentry->d_inode); + ret =fsnotify_add_inode_mark (&audit_mark->mark, path.dentry->d_inode,0 ); + } +if (ret <0 ) { audit_mark->path =NULL ;fsnotify_put_mark (&audit_mark->mark); audit_mark =ERR_PTR (ret); } out: -dput (dentry); -path_put (&path); +if (!ctx) { +dput (dentry); +path_put (&path); + }return audit_mark; }
kernel/audit_watch.c +17 -8
@@ -244,7 +244,8 @@ static void audit_watch_log_rule_change(struct audit_krule *r, struct audit_watc/* Update inode info in audit rules based on filesystem event. */ static void audit_update_watch (struct audit_parent *parent,const struct qstr *dname,dev_t dev, -u64 ino,unsigned invalidating) +unsigned long ino,unsigned invalidating, +struct audit_watch_ctx *ctx) {struct audit_watch *owatch, *nwatch, *nextw;struct audit_krule *r, *nextr;
@@ -280,7 +281,7 @@ static void audit_update_watch(struct audit_parent *parent,list_del (&oentry->rule.rlist);list_del_rcu (&oentry->list); - nentry =audit_dupe_rule (&oentry->rule); + nentry =audit_dupe_rule (&oentry->rule, ctx);if (IS_ERR (nentry)) {list_del (&oentry->rule.list);audit_panic ("error updating watch, removing" );
@@ -479,10 +480,17 @@ static int audit_watch_handle_event(struct fsnotify_mark *inode_mark, u32 mask,if (WARN_ON_ONCE (inode_mark->group != audit_watch_group))return 0 ; -if (mask & (FS_CREATE |FS_MOVED_TO ) && inode) -audit_update_watch (parent, dname, inode->i_sb->s_dev, inode->i_ino,0 ); -else if (mask & (FS_DELETE |FS_MOVED_FROM )) -audit_update_watch (parent, dname,AUDIT_DEV_UNSET ,AUDIT_INO_UNSET ,1 ); +if (mask & (FS_CREATE |FS_MOVED_TO ) && inode) { +struct audit_watch_ctx ctx = { .dir = dir, .child = inode }; + +audit_update_watch (parent, dname, inode->i_sb->s_dev, inode->i_ino,0 , + &ctx); + }else if (mask & (FS_DELETE |FS_MOVED_FROM )) { +struct audit_watch_ctx ctx = { .dir = dir, .child =NULL }; + +audit_update_watch (parent, dname,AUDIT_DEV_UNSET ,AUDIT_INO_UNSET ,1 , + &ctx); + }else if (mask & (FS_DELETE_SELF |FS_UNMOUNT |FS_MOVE_SELF ))audit_remove_parent_watches (parent);
@@ -505,7 +513,8 @@ static int __init audit_watch_init(void) }device_initcall (audit_watch_init); -int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old) +int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old, +struct audit_watch_ctx *ctx) {struct audit_fsnotify_mark *audit_mark;char *pathname;
@@ -514,7 +523,7 @@ int audit_dupe_exe(struct audit_krule *new, struct audit_krule *old)if (!pathname)return -ENOMEM ; - audit_mark =audit_alloc_mark (new, pathname,strlen (pathname)); + audit_mark =audit_alloc_mark (new, pathname,strlen (pathname), ctx);if (IS_ERR (audit_mark)) {kfree (pathname);return PTR_ERR (audit_mark);
kernel/auditfilter.c +5 -4
@@ -589,7 +589,7 @@ static struct audit_entry *audit_data_to_entry(struct audit_rule_data *data, err =PTR_ERR (str); goto exit_free; } - audit_mark =audit_alloc_mark (&entry->rule, str, f_val); + audit_mark =audit_alloc_mark (&entry->rule, str, f_val,NULL );if (IS_ERR (audit_mark)) {kfree (str); err =PTR_ERR (audit_mark);
@@ -816,7 +816,8 @@ static inline int audit_dupe_lsm_field(struct audit_field *df,* rule with the new rule in the filterlist, then free the old rule. * The rlist element is undefined; list manipulations are handled apart from * the initial copy. */ -struct audit_entry *audit_dupe_rule (struct audit_krule *old) +struct audit_entry *audit_dupe_rule (struct audit_krule *old, +struct audit_watch_ctx *ctx) {u32 fcount = old->field_count;struct audit_entry *entry;
@@ -875,7 +876,7 @@ struct audit_entry *audit_dupe_rule(struct audit_krule *old) new->filterkey = fk;break ;case AUDIT_EXE : - err =audit_dupe_exe (new, old); + err =audit_dupe_exe (new, old, ctx);break ; }if (err) {
@@ -1414,7 +1415,7 @@ static int update_lsm_rule(struct audit_krule *r)if (!security_audit_rule_known (r))return 0 ; - nentry =audit_dupe_rule (r); + nentry =audit_dupe_rule (r,NULL );if (entry->rule.exe)audit_remove_mark (entry->rule.exe);if (IS_ERR (nentry)) {
Signature · 1 line
2.53.0
Re: [PATCH 1/2] audit: fix recursive locking deadlock in audit_dupe_exe()
Quoted 107 lines · Ricardo Robaina
A deadlock occurs in the audit subsystem when duplicating executable-related rules.
When a file is moved (e.g., via
do_renameat2()), the VFS layer locks the parent directory (I_MUTEX_PARENT), which synchronously triggers anfsnotify_moveevent. If an existing executable audit rule matches the file being moved, the audit subsystem catches this event and callsaudit_dupe_exe()to duplicate the watch and update the rule. Then,audit_alloc_mark()would callkern_path_parent()to resolve the path, leading to a blind attempt to acquire the exact sameI_MUTEX_PARENTlock already held by the task, resulting in the following recursive locking deadlock:[ 60.521149] ============================================ [ 60.521164] WARNING: possible recursive locking detected [ 60.521180] 6.12.0-55.27.1.el10_0.x86_64+debug #1 Not tainted [ 60.521197] -------------------------------------------- [ 60.521211] mv/5099 is trying to acquire lock: [ 60.521225] ffff888132845358 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3}, at: __kern_path_locked+0x10a/0x2f0 [ 60.521263] but task is already holding lock: [ 60.521280] ffff888132846b58 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3}, at: lock_two_directories+0x13f/0x2b0 [ 60.521313] other info that might help us debug this: [ 60.521331] Possible unsafe locking scenario:[ 60.521349] CPU0 [ 60.521357] ---- [ 60.521365] lock(&inode->i_sb->s_type->i_mutex_dir_key/1); [ 60.521384] lock(&inode->i_sb->s_type->i_mutex_dir_key/1); [ 60.521401] *** DEADLOCK ***[ 60.521422] May be due to missing lock nesting notation[ 60.521445] 6 locks held by mv/5099: [ 60.521458] #0: ffff888112a9c440 (sb_writers#13){++++}-{0:0}, at: do_renameat2+0x34c/0xbc0 [ 60.521491] #1: ffff888112a9c790 (&type->s_vfs_rename_key#3){+.+.}-{3:3}, at: do_renameat2+0x415/0xbc0 [ 60.521525] #2: ffff888132846b58 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3}, at: lock_two_directories+0x13f/0x2b0 [ 60.521565] #3: ffff888132845358 (&inode->i_sb->s_type->i_mutex_dir_key/5){+.+.}-{3:3}, at: lock_two_directories+0x175/0x2b0 [ 60.521604] #4: ffffffffb3a1fb10 (&fsnotify_mark_srcu){.+.+}-{0:0}, at: fsnotify+0x454/0x28a0 [ 60.521636] #5: ffffffffaf886230 (audit_filter_mutex){+.+.}-{3:3}, at: audit_update_watch+0x36/0x11e0 [ 60.521670] stack backtrace: [ 60.521687] CPU: 0 UID: 0 PID: 5099 Comm: mv Kdump: loaded Not tainted 6.12.0-55.27.1.el10_0.x86_64+debug #1 [ 60.521718] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-2.fc40 04/01/2014 [ 60.521743] Call Trace: [ 60.521754] <TASK> [ 60.521764] dump_stack_lvl+0x6f/0xb0 [ 60.521781] print_deadlock_bug.cold+0xbd/0xca [ 60.521798] validate_chain+0x83a/0xf00 [ 60.522465] __lock_acquire+0xcac/0x1d20 [ 60.524073] lock_acquire.part.0+0x11b/0x360 [ 60.528512] down_write_nested+0x9f/0x230 [ 60.530038] __kern_path_locked+0x10a/0x2f0 [ 60.532561] kern_path_locked+0x26/0x40 [ 60.533018] audit_alloc_mark+0xfb/0x4f0 [ 60.535295] audit_dupe_exe+0x6c/0xe0 [ 60.536307] audit_dupe_rule+0x6c2/0xc00 [ 60.537290] audit_update_watch+0x4cc/0x11e0 [ 60.537759] audit_watch_handle_event+0x12c/0x1b0 [ 60.538227] send_to_group+0x5d0/0x8b0 [ 60.538683] fsnotify+0x615/0x28a0 [ 60.540007] fsnotify_move+0x1d8/0x630 [ 60.540438] vfs_rename+0xdcd/0x1df0 [ 60.542176] do_renameat2+0x9d4/0xbc0 [ 60.544870] __x64_sys_renameat+0x192/0x260 [ 60.545299] do_syscall_64+0x92/0x180 [ 60.554189] entry_SYSCALL_64_after_hwframe+0x76/0x7e [ 60.554562] RIP: 0033:0x7f0491fe8c4e [ 60.554947] Code: 0f 1f 40 00 48 8b 15 c1 e1 16 00 f7 d8 64 89 02 b8 ff ff ff ff c3 66 0f 1f 44 00 00 f3 0f 1e fa 49 89 ca b8 08 01 00 00 0f 05 <48> 3d 00 f0 ff ff 77 0a c3 66 0f 1f 84 00 00 00 00 00 48 8b 15 89 [ 60.555789] RSP: 002b:00007ffc7210bf38 EFLAGS: 00000246 ORIG_RAX: 0000000000000108 [ 60.556237] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0491fe8c4e [ 60.556683] RDX: 0000000000000003 RSI: 00007ffc7210e6c8 RDI: 00000000ffffff9c [ 60.557166] RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000001 [ 60.557618] R10: 00005575eb2dae2a R11: 0000000000000246 R12: 00005575eb2dae2a [ 60.558080] R13: 00007ffc7210e6c8 R14: 0000000000000003 R15: 00000000ffffff9c [ 60.558518] </TASK>The aforementioned deadlock can be consistently reproduced by running the script below:
audit-dupe-exe-deadlock.sh
--------------------------
#!/bin/bash
auditctl -D
mkdir -p/tmp/foo
touch/tmp/file
auditctl -a always,exit -F exe=/tmp/file -F path=/tmp/file -S all -k dr mv/tmp/file/tmp/foo/file
rm -Rf/tmp/fooThis patch fixes the issue by introducing struct
audit_watch_ctxto pass the fsnotify event context down toaudit_alloc_mark(). By utilizing the already-resolved directory inode provided by the event, we bypass thekern_path_parent()path resolution entirely, safely avoiding the recursive lock. Furthermore, it explicitly allows duplicate fsnotify marks (allow_dups= 1) during the rename update, allowing the new rule's mark to safely coexist with the old rule's mark until the old rule is freed.
Quoted 8 lines
ps.: this issue was identified and reproduced during a comprehensive code coverage analysis of the audit subsystem. The full report is available at the link below.
Fixes 34d99af52ad4 ("audit: implement audit by executable")Acked-by Waiman Long <[email protected]>Signed-off-by Ricardo Robaina <[email protected]>
Quoted 218 lines
--- kernel/audit.h | 13 ++++++++++--- kernel/audit_fsnotify.c | 33 ++++++++++++++++++++++----------- kernel/audit_watch.c | 25 +++++++++++++++++-------- kernel/auditfilter.c | 9 +++++---- 4 files changed, 54 insertions(+), 26 deletions(-)kernel/audit.h +10
-3@@ -256,8 +256,13 @@ extern int audit_del_rule(struct audit_entry *entry);extern void audit_free_rule_rcu (struct rcu_head *head);extern struct list_head audit_filter_list[]; -extern struct audit_entry *audit_dupe_rule (struct audit_krule *old); +struct audit_watch_ctx { +struct inode *dir; +struct inode *child; +}; +extern struct audit_entry *audit_dupe_rule (struct audit_krule *old, +struct audit_watch_ctx *ctx);extern void audit_log_d_path_exe (struct audit_buffer *ab,struct mm_struct *mm);@@ -280,13 +285,15 @@ extern char *audit_watch_path(struct audit_watch *watch);extern int audit_watch_compare (struct audit_watch *watch,u64 ino,dev_t dev);extern struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule, -char *pathname,int len); +char *pathname,int len, +struct audit_watch_ctx *ctx);extern char *audit_mark_path (struct audit_fsnotify_mark *mark);extern void audit_remove_mark (struct audit_fsnotify_mark *audit_mark);extern void audit_remove_mark_rule (struct audit_krule *krule);extern int audit_mark_compare (struct audit_fsnotify_mark *mark,u64 ino,dev_t dev); -extern int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old); +extern int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old, +struct audit_watch_ctx *ctx);extern int audit_exe_compare (struct task_struct *tsk,struct audit_fsnotify_mark *mark);kernel/audit_fsnotify.c +22
-11@@ -71,7 +71,8 @@ static void audit_update_mark(struct audit_fsnotify_mark *audit_mark, audit_mark->ino = inode ? inode->i_ino :AUDIT_INO_UNSET ; } -struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule,char *pathname,int len) +struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule,char *pathname, +int len,struct audit_watch_ctx *ctx) {struct audit_fsnotify_mark *audit_mark;struct path path;@@ -81,12 +82,14 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *paif (pathname[0 ] !='/' || pathname[len-1 ] =='/' )return ERR_PTR (-EINVAL ); - dentry =kern_path_parent (pathname, &path); -if (IS_ERR (dentry)) -return ERR_CAST (dentry);/* returning an error */ -if (d_really_is_negative (dentry)) { - audit_mark =ERR_PTR (-ENOENT ); - goto out; +if (!ctx) { + dentry =kern_path_parent (pathname, &path); +if (IS_ERR (dentry)) +return ERR_CAST (dentry);/* returning an error */ +if (d_really_is_negative (dentry)) { + audit_mark =ERR_PTR (-ENOENT ); + goto out; + } } audit_mark =kzalloc_obj (*audit_mark);@@ -98,18 +101,26 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pafsnotify_init_mark (&audit_mark->mark, audit_fsnotify_group); audit_mark->mark.mask =AUDIT_FS_EVENTS ; audit_mark->path = pathname; -audit_update_mark (audit_mark, dentry->d_inode); audit_mark->rule = krule; - ret =fsnotify_add_inode_mark (&audit_mark->mark, path.dentry->d_inode,0 ); +if (ctx) { +audit_update_mark (audit_mark, ctx->child); + ret =fsnotify_add_inode_mark (&audit_mark->mark, ctx->dir,1 ); + }else { +audit_update_mark (audit_mark, dentry->d_inode); + ret =fsnotify_add_inode_mark (&audit_mark->mark, path.dentry->d_inode,0 ); + } +if (ret <0 ) { audit_mark->path =NULL ;fsnotify_put_mark (&audit_mark->mark); audit_mark =ERR_PTR (ret); } out: -dput (dentry); -path_put (&path); +if (!ctx) { +dput (dentry); +path_put (&path); + }return audit_mark; }kernel/audit_watch.c +17
-8@@ -244,7 +244,8 @@ static void audit_watch_log_rule_change(struct audit_krule *r, struct audit_watc/* Update inode info in audit rules based on filesystem event. */ static void audit_update_watch (struct audit_parent *parent,const struct qstr *dname,dev_t dev, -u64 ino,unsigned invalidating) +unsigned long ino,unsigned invalidating, +struct audit_watch_ctx *ctx) {struct audit_watch *owatch, *nwatch, *nextw;struct audit_krule *r, *nextr;@@ -280,7 +281,7 @@ static void audit_update_watch(struct audit_parent *parent,list_del (&oentry->rule.rlist);list_del_rcu (&oentry->list); - nentry =audit_dupe_rule (&oentry->rule); + nentry =audit_dupe_rule (&oentry->rule, ctx);if (IS_ERR (nentry)) {list_del (&oentry->rule.list);audit_panic ("error updating watch, removing" );@@ -479,10 +480,17 @@ static int audit_watch_handle_event(struct fsnotify_mark *inode_mark, u32 mask,if (WARN_ON_ONCE (inode_mark->group != audit_watch_group))return 0 ; -if (mask & (FS_CREATE |FS_MOVED_TO ) && inode) -audit_update_watch (parent, dname, inode->i_sb->s_dev, inode->i_ino,0 ); -else if (mask & (FS_DELETE |FS_MOVED_FROM )) -audit_update_watch (parent, dname,AUDIT_DEV_UNSET ,AUDIT_INO_UNSET ,1 ); +if (mask & (FS_CREATE |FS_MOVED_TO ) && inode) { +struct audit_watch_ctx ctx = { .dir = dir, .child = inode }; + +audit_update_watch (parent, dname, inode->i_sb->s_dev, inode->i_ino,0 , + &ctx); + }else if (mask & (FS_DELETE |FS_MOVED_FROM )) { +struct audit_watch_ctx ctx = { .dir = dir, .child =NULL }; + +audit_update_watch (parent, dname,AUDIT_DEV_UNSET ,AUDIT_INO_UNSET ,1 , + &ctx); + }else if (mask & (FS_DELETE_SELF |FS_UNMOUNT |FS_MOVE_SELF ))audit_remove_parent_watches (parent);@@ -505,7 +513,8 @@ static int __init audit_watch_init(void) }device_initcall (audit_watch_init); -int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old) +int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old, +struct audit_watch_ctx *ctx) {struct audit_fsnotify_mark *audit_mark;char *pathname;@@ -514,7 +523,7 @@ int audit_dupe_exe(struct audit_krule *new, struct audit_krule *old)if (!pathname)return -ENOMEM ; - audit_mark =audit_alloc_mark (new, pathname,strlen (pathname)); + audit_mark =audit_alloc_mark (new, pathname,strlen (pathname), ctx);if (IS_ERR (audit_mark)) {kfree (pathname);return PTR_ERR (audit_mark);kernel/auditfilter.c +5
-4@@ -589,7 +589,7 @@ static struct audit_entry *audit_data_to_entry(struct audit_rule_data *data, err =PTR_ERR (str); goto exit_free; } - audit_mark =audit_alloc_mark (&entry->rule, str, f_val); + audit_mark =audit_alloc_mark (&entry->rule, str, f_val,NULL );if (IS_ERR (audit_mark)) {kfree (str); err =PTR_ERR (audit_mark);@@ -816,7 +816,8 @@ static inline int audit_dupe_lsm_field(struct audit_field *df,* rule with the new rule in the filterlist, then free the old rule. * The rlist element is undefined; list manipulations are handled apart from * the initial copy. */ -struct audit_entry *audit_dupe_rule (struct audit_krule *old) +struct audit_entry *audit_dupe_rule (struct audit_krule *old, +struct audit_watch_ctx *ctx) {u32 fcount = old->field_count;struct audit_entry *entry;@@ -875,7 +876,7 @@ struct audit_entry *audit_dupe_rule(struct audit_krule *old) new->filterkey = fk;break ;case AUDIT_EXE : - err =audit_dupe_exe (new, old); + err =audit_dupe_exe (new, old, ctx);break ; }if (err) {@@ -1414,7 +1415,7 @@ static int update_lsm_rule(struct audit_krule *r)if (!security_audit_rule_known (r))return 0 ; - nentry =audit_dupe_rule (r); + nentry =audit_dupe_rule (r,NULL );if (entry->rule.exe)audit_remove_mark (entry->rule.exe);if (IS_ERR (nentry)) {Signature · 1 line
2.53.0
- RGB
--
Richard Guy Briggs <[email protected]>
Sr. S/W Engineer, Kernel Security, Base Operating Systems Remote, Ottawa, Red Hat Canada
Upstream IRC: SunRaycer
Voice: +1.613.860 2354 SMS: +1.613.518.6570
Paul Moore
maintainerRe: [PATCH 1/2] audit: fix recursive locking deadlock in audit_dupe_exe()
In reply to Ricardo RobainaQuoted 117 lines · On Apr 14, 2026 Ricardo Robaina <[email protected]> wrote:
A deadlock occurs in the audit subsystem when duplicating executable-related rules.
When a file is moved (e.g., via
do_renameat2()), the VFS layer locks the parent directory (I_MUTEX_PARENT), which synchronously triggers anfsnotify_moveevent. If an existing executable audit rule matches the file being moved, the audit subsystem catches this event and callsaudit_dupe_exe()to duplicate the watch and update the rule. Then,audit_alloc_mark()would callkern_path_parent()to resolve the path, leading to a blind attempt to acquire the exact sameI_MUTEX_PARENTlock already held by the task, resulting in the following recursive locking deadlock:[ 60.521149] ============================================ [ 60.521164] WARNING: possible recursive locking detected [ 60.521180] 6.12.0-55.27.1.el10_0.x86_64+debug #1 Not tainted [ 60.521197] -------------------------------------------- [ 60.521211] mv/5099 is trying to acquire lock: [ 60.521225] ffff888132845358 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3}, at: __kern_path_locked+0x10a/0x2f0 [ 60.521263] but task is already holding lock: [ 60.521280] ffff888132846b58 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3}, at: lock_two_directories+0x13f/0x2b0 [ 60.521313] other info that might help us debug this: [ 60.521331] Possible unsafe locking scenario:[ 60.521349] CPU0 [ 60.521357] ---- [ 60.521365] lock(&inode->i_sb->s_type->i_mutex_dir_key/1); [ 60.521384] lock(&inode->i_sb->s_type->i_mutex_dir_key/1); [ 60.521401] *** DEADLOCK ***[ 60.521422] May be due to missing lock nesting notation[ 60.521445] 6 locks held by mv/5099: [ 60.521458] #0: ffff888112a9c440 (sb_writers#13){++++}-{0:0}, at: do_renameat2+0x34c/0xbc0 [ 60.521491] #1: ffff888112a9c790 (&type->s_vfs_rename_key#3){+.+.}-{3:3}, at: do_renameat2+0x415/0xbc0 [ 60.521525] #2: ffff888132846b58 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3}, at: lock_two_directories+0x13f/0x2b0 [ 60.521565] #3: ffff888132845358 (&inode->i_sb->s_type->i_mutex_dir_key/5){+.+.}-{3:3}, at: lock_two_directories+0x175/0x2b0 [ 60.521604] #4: ffffffffb3a1fb10 (&fsnotify_mark_srcu){.+.+}-{0:0}, at: fsnotify+0x454/0x28a0 [ 60.521636] #5: ffffffffaf886230 (audit_filter_mutex){+.+.}-{3:3}, at: audit_update_watch+0x36/0x11e0 [ 60.521670] stack backtrace: [ 60.521687] CPU: 0 UID: 0 PID: 5099 Comm: mv Kdump: loaded Not tainted 6.12.0-55.27.1.el10_0.x86_64+debug #1 [ 60.521718] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-2.fc40 04/01/2014 [ 60.521743] Call Trace: [ 60.521754] <TASK> [ 60.521764] dump_stack_lvl+0x6f/0xb0 [ 60.521781] print_deadlock_bug.cold+0xbd/0xca [ 60.521798] validate_chain+0x83a/0xf00 [ 60.522465] __lock_acquire+0xcac/0x1d20 [ 60.524073] lock_acquire.part.0+0x11b/0x360 [ 60.528512] down_write_nested+0x9f/0x230 [ 60.530038] __kern_path_locked+0x10a/0x2f0 [ 60.532561] kern_path_locked+0x26/0x40 [ 60.533018] audit_alloc_mark+0xfb/0x4f0 [ 60.535295] audit_dupe_exe+0x6c/0xe0 [ 60.536307] audit_dupe_rule+0x6c2/0xc00 [ 60.537290] audit_update_watch+0x4cc/0x11e0 [ 60.537759] audit_watch_handle_event+0x12c/0x1b0 [ 60.538227] send_to_group+0x5d0/0x8b0 [ 60.538683] fsnotify+0x615/0x28a0 [ 60.540007] fsnotify_move+0x1d8/0x630 [ 60.540438] vfs_rename+0xdcd/0x1df0 [ 60.542176] do_renameat2+0x9d4/0xbc0 [ 60.544870] __x64_sys_renameat+0x192/0x260 [ 60.545299] do_syscall_64+0x92/0x180 [ 60.554189] entry_SYSCALL_64_after_hwframe+0x76/0x7e [ 60.554562] RIP: 0033:0x7f0491fe8c4e [ 60.554947] Code: 0f 1f 40 00 48 8b 15 c1 e1 16 00 f7 d8 64 89 02 b8 ff ff ff ff c3 66 0f 1f 44 00 00 f3 0f 1e fa 49 89 ca b8 08 01 00 00 0f 05 <48> 3d 00 f0 ff ff 77 0a c3 66 0f 1f 84 00 00 00 00 00 48 8b 15 89 [ 60.555789] RSP: 002b:00007ffc7210bf38 EFLAGS: 00000246 ORIG_RAX: 0000000000000108 [ 60.556237] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0491fe8c4e [ 60.556683] RDX: 0000000000000003 RSI: 00007ffc7210e6c8 RDI: 00000000ffffff9c [ 60.557166] RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000001 [ 60.557618] R10: 00005575eb2dae2a R11: 0000000000000246 R12: 00005575eb2dae2a [ 60.558080] R13: 00007ffc7210e6c8 R14: 0000000000000003 R15: 00000000ffffff9c [ 60.558518] </TASK>The aforementioned deadlock can be consistently reproduced by running the script below:
audit-dupe-exe-deadlock.sh
--------------------------
#!/bin/bash
auditctl -D
mkdir -p/tmp/foo
touch/tmp/file
auditctl -a always,exit -F exe=/tmp/file -F path=/tmp/file -S all -k dr mv/tmp/file/tmp/foo/file
rm -Rf/tmp/fooThis patch fixes the issue by introducing struct
audit_watch_ctxto pass the fsnotify event context down toaudit_alloc_mark(). By utilizing the already-resolved directory inode provided by the event, we bypass thekern_path_parent()path resolution entirely, safely avoiding the recursive lock. Furthermore, it explicitly allows duplicate fsnotify marks (allow_dups= 1) during the rename update, allowing the new rule's mark to safely coexist with the old rule's mark until the old rule is freed.ps.: this issue was identified and reproduced during a comprehensive code coverage analysis of the audit subsystem. The full report is available at the link below.
Fixes 34d99af52ad4 ("audit: implement audit by executable")Acked-by Waiman Long <[email protected]>Signed-off-by Ricardo Robaina <[email protected]>
Quoted 6 lines
--- kernel/audit.h | 13 ++++++++++--- kernel/audit_fsnotify.c | 33 ++++++++++++++++++++++----------- kernel/audit_watch.c | 25 +++++++++++++++++-------- kernel/auditfilter.c | 9 +++++---- 4 files changed, 54 insertions(+), 26 deletions(-)
Thanks for looking into this, reporting the bug, and providing a fix!
I've got some relatively small comments below, but in the future please ensure that your commit description when viewed with 'git log' fits nicely in a 80 character wide terminal without line wrapping. Feel free to wrap backtrace lines and trim non-critical information to help long lines fit.
Quoted 77 lines
kernel/audit.h +10
-3@@ -256,8 +256,13 @@ extern int audit_del_rule(struct audit_entry *entry);extern void audit_free_rule_rcu (struct rcu_head *head);extern struct list_head audit_filter_list[]; -extern struct audit_entry *audit_dupe_rule (struct audit_krule *old); +struct audit_watch_ctx { +struct inode *dir; +struct inode *child; +}; +extern struct audit_entry *audit_dupe_rule (struct audit_krule *old, +struct audit_watch_ctx *ctx);extern void audit_log_d_path_exe (struct audit_buffer *ab,struct mm_struct *mm);@@ -280,13 +285,15 @@ extern char *audit_watch_path(struct audit_watch *watch);extern int audit_watch_compare (struct audit_watch *watch,u64 ino,dev_t dev);extern struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule, -char *pathname,int len); +char *pathname,int len, +struct audit_watch_ctx *ctx);extern char *audit_mark_path (struct audit_fsnotify_mark *mark);extern void audit_remove_mark (struct audit_fsnotify_mark *audit_mark);extern void audit_remove_mark_rule (struct audit_krule *krule);extern int audit_mark_compare (struct audit_fsnotify_mark *mark,u64 ino,dev_t dev); -extern int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old); +extern int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old, +struct audit_watch_ctx *ctx);extern int audit_exe_compare (struct task_struct *tsk,struct audit_fsnotify_mark *mark);kernel/audit_fsnotify.c +10
-8@@ -71,7 +71,8 @@ static void audit_update_mark(struct audit_fsnotify_mark *audit_mark, audit_mark->ino = inode ? inode->i_ino :AUDIT_INO_UNSET ; } -struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule,char *pathname,int len) +struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule,char *pathname, +int len,struct audit_watch_ctx *ctx) {struct audit_fsnotify_mark *audit_mark;struct path path;@@ -81,12 +82,14 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *paif (pathname[0 ] !='/' || pathname[len-1 ] =='/' )return ERR_PTR (-EINVAL ); - dentry =kern_path_parent (pathname, &path); -if (IS_ERR (dentry)) -return ERR_CAST (dentry);/* returning an error */ -if (d_really_is_negative (dentry)) { - audit_mark =ERR_PTR (-ENOENT ); - goto out; +if (!ctx) { + dentry =kern_path_parent (pathname, &path); +if (IS_ERR (dentry)) +return ERR_CAST (dentry);/* returning an error */ +if (d_really_is_negative (dentry)) { + audit_mark =ERR_PTR (-ENOENT ); + goto out; + } } audit_mark =kzalloc_obj (*audit_mark);@@ -98,18 +101,26 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pafsnotify_init_mark (&audit_mark->mark, audit_fsnotify_group); audit_mark->mark.mask =AUDIT_FS_EVENTS ; audit_mark->path = pathname; -audit_update_mark (audit_mark, dentry->d_inode);
kernel/audit_fsnotify.c struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pa · line 112
Quoted 10 lines
kernel/audit_fsnotify.c +7
-1audit_mark->rule = krule; - ret =fsnotify_add_inode_mark (&audit_mark->mark, path.dentry->d_inode,0 ); +if (ctx) { +audit_update_mark (audit_mark, ctx->child); + ret =fsnotify_add_inode_mark (&audit_mark->mark, ctx->dir,1 ); + }else { +audit_update_mark (audit_mark, dentry->d_inode); + ret =fsnotify_add_inode_mark (&audit_mark->mark, path.dentry->d_inode,0 ); + }
Generally speaking, I dislike when critical code paths are duplicated and conditionalized like this when the only differences are a small number of variables/parameters. It makes it all to easy to introduce small bugs in the future. I would suggest a small change list this to unify this split:
kernel/audit_fsnotify.c +8 -7
@@ -77,6 +77,7 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pastruct audit_fsnotify_mark *audit_mark;struct path path;struct dentry *dentry; +struct inode *dir, *child;int ret;if (pathname[0 ] !='/' || pathname[len-1 ] =='/' )
@@ -90,6 +91,11 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pa audit_mark =ERR_PTR (-ENOENT ); goto out; } + dir =d_inode (path.dentry); + child =d_inode (dentry); + }else { + dir = ctx->dir; + child = ctx->child; } audit_mark =kzalloc_obj (*audit_mark);
@@ -103,13 +109,8 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pa audit_mark->path = pathname; audit_mark->rule = krule; -if (ctx) { -audit_update_mark (audit_mark, ctx->child); - ret =fsnotify_add_inode_mark (&audit_mark->mark, ctx->dir,1 ); - }else { -audit_update_mark (audit_mark, dentry->d_inode); - ret =fsnotify_add_inode_mark (&audit_mark->mark, path.dentry->d_inode,0 ); - } +audit_update_mark (audit_mark, child); + ret =fsnotify_add_inode_mark (&audit_mark->mark, dir,1 );if (ret <0 ) { audit_mark->path =NULL ;
Quoted 4 lines
kernel/audit_fsnotify.c +0
-0if (ret <0 ) { audit_mark->path =NULL ;fsnotify_put_mark (&audit_mark->mark); audit_mark =ERR_PTR (ret);
kernel/audit_fsnotify.c struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pa · line 124
Quoted 10 lines
kernel/audit_fsnotify.c +4
-2} out: -dput (dentry); -path_put (&path); +if (!ctx) { +dput (dentry); +path_put (&path); + }return audit_mark; }
--
paul-moore.com
Re: [PATCH 1/2] audit: fix recursive locking deadlock in audit_dupe_exe()
Quoted 126 lines · Paul Moore
Quoted 117 lines · On Apr 14, 2026 Ricardo Robaina <[email protected]> wrote:
A deadlock occurs in the audit subsystem when duplicating executable-related rules.
When a file is moved (e.g., via
do_renameat2()), the VFS layer locks the parent directory (I_MUTEX_PARENT), which synchronously triggers anfsnotify_moveevent. If an existing executable audit rule matches the file being moved, the audit subsystem catches this event and callsaudit_dupe_exe()to duplicate the watch and update the rule. Then,audit_alloc_mark()would callkern_path_parent()to resolve the path, leading to a blind attempt to acquire the exact sameI_MUTEX_PARENTlock already held by the task, resulting in the following recursive locking deadlock:[ 60.521149] ============================================ [ 60.521164] WARNING: possible recursive locking detected [ 60.521180] 6.12.0-55.27.1.el10_0.x86_64+debug #1 Not tainted [ 60.521197] -------------------------------------------- [ 60.521211] mv/5099 is trying to acquire lock: [ 60.521225] ffff888132845358 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3}, at: __kern_path_locked+0x10a/0x2f0 [ 60.521263] but task is already holding lock: [ 60.521280] ffff888132846b58 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3}, at: lock_two_directories+0x13f/0x2b0 [ 60.521313] other info that might help us debug this: [ 60.521331] Possible unsafe locking scenario:[ 60.521349] CPU0 [ 60.521357] ---- [ 60.521365] lock(&inode->i_sb->s_type->i_mutex_dir_key/1); [ 60.521384] lock(&inode->i_sb->s_type->i_mutex_dir_key/1); [ 60.521401] *** DEADLOCK ***[ 60.521422] May be due to missing lock nesting notation[ 60.521445] 6 locks held by mv/5099: [ 60.521458] #0: ffff888112a9c440 (sb_writers#13){++++}-{0:0}, at: do_renameat2+0x34c/0xbc0 [ 60.521491] #1: ffff888112a9c790 (&type->s_vfs_rename_key#3){+.+.}-{3:3}, at: do_renameat2+0x415/0xbc0 [ 60.521525] #2: ffff888132846b58 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3}, at: lock_two_directories+0x13f/0x2b0 [ 60.521565] #3: ffff888132845358 (&inode->i_sb->s_type->i_mutex_dir_key/5){+.+.}-{3:3}, at: lock_two_directories+0x175/0x2b0 [ 60.521604] #4: ffffffffb3a1fb10 (&fsnotify_mark_srcu){.+.+}-{0:0}, at: fsnotify+0x454/0x28a0 [ 60.521636] #5: ffffffffaf886230 (audit_filter_mutex){+.+.}-{3:3}, at: audit_update_watch+0x36/0x11e0 [ 60.521670] stack backtrace: [ 60.521687] CPU: 0 UID: 0 PID: 5099 Comm: mv Kdump: loaded Not tainted 6.12.0-55.27.1.el10_0.x86_64+debug #1 [ 60.521718] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-2.fc40 04/01/2014 [ 60.521743] Call Trace: [ 60.521754] <TASK> [ 60.521764] dump_stack_lvl+0x6f/0xb0 [ 60.521781] print_deadlock_bug.cold+0xbd/0xca [ 60.521798] validate_chain+0x83a/0xf00 [ 60.522465] __lock_acquire+0xcac/0x1d20 [ 60.524073] lock_acquire.part.0+0x11b/0x360 [ 60.528512] down_write_nested+0x9f/0x230 [ 60.530038] __kern_path_locked+0x10a/0x2f0 [ 60.532561] kern_path_locked+0x26/0x40 [ 60.533018] audit_alloc_mark+0xfb/0x4f0 [ 60.535295] audit_dupe_exe+0x6c/0xe0 [ 60.536307] audit_dupe_rule+0x6c2/0xc00 [ 60.537290] audit_update_watch+0x4cc/0x11e0 [ 60.537759] audit_watch_handle_event+0x12c/0x1b0 [ 60.538227] send_to_group+0x5d0/0x8b0 [ 60.538683] fsnotify+0x615/0x28a0 [ 60.540007] fsnotify_move+0x1d8/0x630 [ 60.540438] vfs_rename+0xdcd/0x1df0 [ 60.542176] do_renameat2+0x9d4/0xbc0 [ 60.544870] __x64_sys_renameat+0x192/0x260 [ 60.545299] do_syscall_64+0x92/0x180 [ 60.554189] entry_SYSCALL_64_after_hwframe+0x76/0x7e [ 60.554562] RIP: 0033:0x7f0491fe8c4e [ 60.554947] Code: 0f 1f 40 00 48 8b 15 c1 e1 16 00 f7 d8 64 89 02 b8 ff ff ff ff c3 66 0f 1f 44 00 00 f3 0f 1e fa 49 89 ca b8 08 01 00 00 0f 05 <48> 3d 00 f0 ff ff 77 0a c3 66 0f 1f 84 00 00 00 00 00 48 8b 15 89 [ 60.555789] RSP: 002b:00007ffc7210bf38 EFLAGS: 00000246 ORIG_RAX: 0000000000000108 [ 60.556237] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0491fe8c4e [ 60.556683] RDX: 0000000000000003 RSI: 00007ffc7210e6c8 RDI: 00000000ffffff9c [ 60.557166] RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000001 [ 60.557618] R10: 00005575eb2dae2a R11: 0000000000000246 R12: 00005575eb2dae2a [ 60.558080] R13: 00007ffc7210e6c8 R14: 0000000000000003 R15: 00000000ffffff9c [ 60.558518] </TASK>The aforementioned deadlock can be consistently reproduced by running the script below:
audit-dupe-exe-deadlock.sh
--------------------------
#!/bin/bash
auditctl -D
mkdir -p/tmp/foo
touch/tmp/file
auditctl -a always,exit -F exe=/tmp/file -F path=/tmp/file -S all -k dr mv/tmp/file/tmp/foo/file
rm -Rf/tmp/fooThis patch fixes the issue by introducing struct
audit_watch_ctxto pass the fsnotify event context down toaudit_alloc_mark(). By utilizing the already-resolved directory inode provided by the event, we bypass thekern_path_parent()path resolution entirely, safely avoiding the recursive lock. Furthermore, it explicitly allows duplicate fsnotify marks (allow_dups= 1) during the rename update, allowing the new rule's mark to safely coexist with the old rule's mark until the old rule is freed.ps.: this issue was identified and reproduced during a comprehensive code coverage analysis of the audit subsystem. The full report is available at the link below.
Fixes 34d99af52ad4 ("audit: implement audit by executable")Acked-by Waiman Long <[email protected]>Signed-off-by Ricardo Robaina <[email protected]>Quoted 6 lines
--- kernel/audit.h | 13 ++++++++++--- kernel/audit_fsnotify.c | 33 ++++++++++++++++++++++----------- kernel/audit_watch.c | 25 +++++++++++++++++-------- kernel/auditfilter.c | 9 +++++---- 4 files changed, 54 insertions(+), 26 deletions(-)
Quoted 2 lines
Thanks for looking into this, reporting the bug, and providing a fix!
Thanks for reviewing this patch series, Paul. Sure thing, I'm happy to help.
Quoted 6 lines
I've got some relatively small comments below, but in the future please ensure that your commit description when viewed with 'git log' fits nicely in a 80 character wide terminal without line wrapping. Feel free to wrap backtrace lines and trim non-critical information to help long lines fit.
Ack.
Quoted 125 lines
Quoted 77 lines
kernel/audit.h +10
-3@@ -256,8 +256,13 @@ extern int audit_del_rule(struct audit_entry *entry);extern void audit_free_rule_rcu (struct rcu_head *head);extern struct list_head audit_filter_list[]; -extern struct audit_entry *audit_dupe_rule (struct audit_krule *old); +struct audit_watch_ctx { +struct inode *dir; +struct inode *child; +}; +extern struct audit_entry *audit_dupe_rule (struct audit_krule *old, +struct audit_watch_ctx *ctx);extern void audit_log_d_path_exe (struct audit_buffer *ab,struct mm_struct *mm);@@ -280,13 +285,15 @@ extern char *audit_watch_path(struct audit_watch *watch);extern int audit_watch_compare (struct audit_watch *watch,u64 ino,dev_t dev);extern struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule, -char *pathname,int len); +char *pathname,int len, +struct audit_watch_ctx *ctx);extern char *audit_mark_path (struct audit_fsnotify_mark *mark);extern void audit_remove_mark (struct audit_fsnotify_mark *audit_mark);extern void audit_remove_mark_rule (struct audit_krule *krule);extern int audit_mark_compare (struct audit_fsnotify_mark *mark,u64 ino,dev_t dev); -extern int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old); +extern int audit_dupe_exe (struct audit_krule *new,struct audit_krule *old, +struct audit_watch_ctx *ctx);extern int audit_exe_compare (struct task_struct *tsk,struct audit_fsnotify_mark *mark);kernel/audit_fsnotify.c +10
-8@@ -71,7 +71,8 @@ static void audit_update_mark(struct audit_fsnotify_mark *audit_mark, audit_mark->ino = inode ? inode->i_ino :AUDIT_INO_UNSET ; } -struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule,char *pathname,int len) +struct audit_fsnotify_mark *audit_alloc_mark (struct audit_krule *krule,char *pathname, +int len,struct audit_watch_ctx *ctx) {struct audit_fsnotify_mark *audit_mark;struct path path;@@ -81,12 +82,14 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *paif (pathname[0 ] !='/' || pathname[len-1 ] =='/' )return ERR_PTR (-EINVAL ); - dentry =kern_path_parent (pathname, &path); -if (IS_ERR (dentry)) -return ERR_CAST (dentry);/* returning an error */ -if (d_really_is_negative (dentry)) { - audit_mark =ERR_PTR (-ENOENT ); - goto out; +if (!ctx) { + dentry =kern_path_parent (pathname, &path); +if (IS_ERR (dentry)) +return ERR_CAST (dentry);/* returning an error */ +if (d_really_is_negative (dentry)) { + audit_mark =ERR_PTR (-ENOENT ); + goto out; + } } audit_mark =kzalloc_obj (*audit_mark);@@ -98,18 +101,26 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pafsnotify_init_mark (&audit_mark->mark, audit_fsnotify_group); audit_mark->mark.mask =AUDIT_FS_EVENTS ; audit_mark->path = pathname; -audit_update_mark (audit_mark, dentry->d_inode);kernel/audit_fsnotify.c struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pa · line 112
Quoted 10 lines
kernel/audit_fsnotify.c +7
-1audit_mark->rule = krule; - ret =fsnotify_add_inode_mark (&audit_mark->mark, path.dentry->d_inode,0 ); +if (ctx) { +audit_update_mark (audit_mark, ctx->child); + ret =fsnotify_add_inode_mark (&audit_mark->mark, ctx->dir,1 ); + }else { +audit_update_mark (audit_mark, dentry->d_inode); + ret =fsnotify_add_inode_mark (&audit_mark->mark, path.dentry->d_inode,0 ); + }Generally speaking, I dislike when critical code paths are duplicated and conditionalized like this when the only differences are a small number of variables/parameters. It makes it all to easy to introduce small bugs in the future. I would suggest a small change list this to unify this split:
kernel/audit_fsnotify.c +6
-2@@ -77,6 +77,7 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pastruct audit_fsnotify_mark *audit_mark;struct path path;struct dentry *dentry; +struct inode *dir, *child;int ret;if (pathname[0 ] !='/' || pathname[len-1 ] =='/' )@@ -90,6 +91,11 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pa audit_mark =ERR_PTR (-ENOENT ); goto out; } + dir =d_inode (path.dentry); + child =d_inode (dentry); + }else { + dir = ctx->dir; + child = ctx->child; } audit_mark =kzalloc_obj (*audit_mark);@@ -103,13 +109,8 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pa audit_mark->path = pathname; audit_mark->rule = krule; -if (ctx) { -audit_update_mark (audit_mark, ctx->child);
kernel/audit_fsnotify.c struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pa · line 116
Quoted 10 lines
kernel/audit_fsnotify.c +2
-5- ret =fsnotify_add_inode_mark (&audit_mark->mark, ctx->dir,1 ); - }else { -audit_update_mark (audit_mark, dentry->d_inode); - ret =fsnotify_add_inode_mark (&audit_mark->mark, path.dentry->d_inode,0 ); - } +audit_update_mark (audit_mark, child); + ret =fsnotify_add_inode_mark (&audit_mark->mark, dir,1 );if (ret <0 ) { audit_mark->path =NULL ;
Nice, the refactored code looks much cleaner. One quick catch, though:
hardcoding 1 in fsnotify_add_inode_mark() drops the -EEXIST check for standard rule additions, which would allow duplicate rules.
In the v2 I'm about to send, I used your exact structure but added an allow_dups flag to preserve that protection.
Quoted 18 lines
Quoted 4 lines
kernel/audit_fsnotify.c +0
-0if (ret <0 ) { audit_mark->path =NULL ;fsnotify_put_mark (&audit_mark->mark); audit_mark =ERR_PTR (ret);kernel/audit_fsnotify.c struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pa · line 124
Quoted 10 lines
kernel/audit_fsnotify.c +4
-2} out: -dput (dentry); -path_put (&path); +if (!ctx) { +dput (dentry); +path_put (&path); + }return audit_mark; }--
paul-moore.com
PATCH 2/2 audit: fix removal of dangling executable rules
In reply to Ricardo RobainaWhen an audited executable is deleted from the disk, its dentry becomes negative. Any later attempt to delete the associated audit rule will lead to audit_alloc_mark() encountering this negative dentry and immediately aborting, returning -ENOENT.
This early abort prevents the subsystem from allocating the temporary fsnotify mark needed to construct the search key, meaning the kernel cannot find the existing rule in its own lists to delete it. This leaves a dangling rule in memory, resulting in the following error while attempting to delete the rule:
# ./audit-dupe-exe-deadlock.sh No rules Error deleting rule (No such file or directory) There was an error while processing parameters
# auditctl -l -a always,exit -S all -F exe=/tmp/file -F path=/tmp/file -F key=dr
# auditctl -D Error deleting rule (No such file or directory) There was an error while processing parameters
This patch fixes this issue by removing the d_really_is_negative() check. By doing so, a dummy mark can be successfully generated for the deleted path, which allows the audit subsystem to properly match and flush the dangling rule.
--- kernel/audit_fsnotify.c | 4 ---- 1 file changed, 4 deletions(-)
kernel/audit_fsnotify.c +0 -4
@@ -86,10 +86,6 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pa dentry =kern_path_parent (pathname, &path);if (IS_ERR (dentry))return ERR_CAST (dentry);/* returning an error */ -if (d_really_is_negative (dentry)) { - audit_mark =ERR_PTR (-ENOENT ); - goto out; - } } audit_mark =kzalloc_obj (*audit_mark);
Signature · 1 line
2.53.0
Re: [PATCH 2/2] audit: fix removal of dangling executable rules
Quoted 23 lines · Ricardo Robaina
When an audited executable is deleted from the disk, its dentry becomes negative. Any later attempt to delete the associated audit rule will lead to
audit_alloc_mark()encountering this negative dentry and immediately aborting, returning-ENOENT.This early abort prevents the subsystem from allocating the temporary fsnotify mark needed to construct the search key, meaning the kernel cannot find the existing rule in its own lists to delete it. This leaves a dangling rule in memory, resulting in the following error while attempting to delete the rule:
# ./audit-dupe-exe-deadlock.sh No rules Error deleting rule (No such file or directory) There was an error while processing parameters# auditctl -l -a always,exit -S all -F exe=/tmp/file -F path=/tmp/file -F key=dr# auditctl -D Error deleting rule (No such file or directory) There was an error while processing parameters
Quoted 8 lines
This patch fixes this issue by removing the
d_really_is_negative()check. By doing so, a dummy mark can be successfully generated for the deleted path, which allows the audit subsystem to properly match and flush the dangling rule.Fixes 76a53de6f7ff ("VFS/audit: introduce kern_path_parent() for audit")Acked-by Waiman Long <[email protected]>Signed-off-by Ricardo Robaina <[email protected]>
Quoted 22 lines
--- kernel/audit_fsnotify.c | 4 ---- 1 file changed, 4 deletions(-)kernel/audit_fsnotify.c +0
-4@@ -86,10 +86,6 @@ struct audit_fsnotify_mark *audit_alloc_mark(struct audit_krule *krule, char *pa dentry =kern_path_parent (pathname, &path);if (IS_ERR (dentry))return ERR_CAST (dentry);/* returning an error */ -if (d_really_is_negative (dentry)) { - audit_mark =ERR_PTR (-ENOENT ); - goto out; - } } audit_mark =kzalloc_obj (*audit_mark);Signature · 1 line
2.53.0
- RGB
--
Richard Guy Briggs <[email protected]>
Sr. S/W Engineer, Kernel Security, Base Operating Systems Remote, Ottawa, Red Hat Canada
Upstream IRC: SunRaycer
Voice: +1.613.860 2354 SMS: +1.613.518.6570
Paul Moore
maintainerRe: [PATCH 2/2] audit: fix removal of dangling executable rules
In reply to Ricardo RobainaQuoted 29 lines · On Apr 14, 2026 Ricardo Robaina <[email protected]> wrote:
When an audited executable is deleted from the disk, its dentry becomes negative. Any later attempt to delete the associated audit rule will lead to
audit_alloc_mark()encountering this negative dentry and immediately aborting, returning-ENOENT.This early abort prevents the subsystem from allocating the temporary fsnotify mark needed to construct the search key, meaning the kernel cannot find the existing rule in its own lists to delete it. This leaves a dangling rule in memory, resulting in the following error while attempting to delete the rule:
# ./audit-dupe-exe-deadlock.sh No rules Error deleting rule (No such file or directory) There was an error while processing parameters# auditctl -l -a always,exit -S all -F exe=/tmp/file -F path=/tmp/file -F key=dr# auditctl -D Error deleting rule (No such file or directory) There was an error while processing parametersThis patch fixes this issue by removing the
d_really_is_negative()check. By doing so, a dummy mark can be successfully generated for the deleted path, which allows the audit subsystem to properly match and flush the dangling rule.
Quoted 7 lines
Fixes 76a53de6f7ff ("VFS/audit: introduce kern_path_parent() for audit")Acked-by Waiman Long <[email protected]>Signed-off-by Ricardo Robaina <[email protected]>--- kernel/audit_fsnotify.c | 4 ---- 1 file changed, 4 deletions(-)
Good catch. I might reorder the patchset so this patch comes first in the patchset, but this isn't a big deal either way.
--
paul-moore.com
Re: [PATCH 2/2] audit: fix removal of dangling executable rules
Quoted 39 lines · Paul Moore
Quoted 29 lines · On Apr 14, 2026 Ricardo Robaina <[email protected]> wrote:
When an audited executable is deleted from the disk, its dentry becomes negative. Any later attempt to delete the associated audit rule will lead to
audit_alloc_mark()encountering this negative dentry and immediately aborting, returning-ENOENT.This early abort prevents the subsystem from allocating the temporary fsnotify mark needed to construct the search key, meaning the kernel cannot find the existing rule in its own lists to delete it. This leaves a dangling rule in memory, resulting in the following error while attempting to delete the rule:
# ./audit-dupe-exe-deadlock.sh No rules Error deleting rule (No such file or directory) There was an error while processing parameters# auditctl -l -a always,exit -S all -F exe=/tmp/file -F path=/tmp/file -F key=dr# auditctl -D Error deleting rule (No such file or directory) There was an error while processing parametersThis patch fixes this issue by removing the
d_really_is_negative()check. By doing so, a dummy mark can be successfully generated for the deleted path, which allows the audit subsystem to properly match and flush the dangling rule.Quoted 7 lines
Fixes 76a53de6f7ff ("VFS/audit: introduce kern_path_parent() for audit")Acked-by Waiman Long <[email protected]>Signed-off-by Ricardo Robaina <[email protected]>--- kernel/audit_fsnotify.c | 4 ---- 1 file changed, 4 deletions(-)
Quoted 6 lines
Good catch. I might reorder the patchset so this patch comes first in the patchset, but this isn't a big deal either way.
--
paul-moore.com
I'm sending the v2 reordered. Thanks for reviewing it as well!