Re: [PATCH] perf/core: unaccount mmap pages from the charged owners

From: Peter Zijlstra

Date: Tue Nov 17 2026 - 10:57:29 EST


On Wed, Jul 22, 2026 at 03:03:53AM -0300, Rafael B. Dias wrote:
> perf mmap accounting splits pages between user_struct::locked_vm and
> mm_struct::pinned_vm. The first mapping that allocates a data ring stores
> the charged user and numeric pinned amount in perf_buffer, but it does not
> store the mm that received the pinned charge.
>
> A perf event fd can be mapped by more than one process. If a different mm
> owns the final VMA close, perf_mmap_unaccount() subtracts mmap_locked from
> that final mm instead of the mm charged at allocation. Repeating this
> across short-lived donor processes creates a negative pinned_vm balance
> that can fund later perf mappings beyond RLIMIT_MEMLOCK.
>
> AUX has the same last-VMA assumption. It also subtracts locked_vm from the
> data ring's mmap_user even when a different user created the AUX mapping.
>
> Retain references to the mm charged for the data ring and to the user and
> mm charged for AUX. Use those stored owners during final unaccounting and
> release the references after the counters are restored.
>
> On a two-CPU 6.12.94 KASAN guest, an unprivileged receiver with

(that's an old kernel, as in *very* old)

> RLIMIT_MEMLOCK=0 accumulated 16,385 pages of cross-mm credit and mapped
> 65,540 KiB on the vulnerable kernel. The same reproducer returned EPERM
> with this ownership fix. No KASAN report or warning occurred. The patch
> also builds kernel/events/core.o with x86_64_defconfig on the same base.
>
> Fixes: 26cb63ad11e0 ("perf: Fix perf mmap bugs")
> Fixes: 36b3db03b474 ("perf/core: Fix the mlock accounting, again")
> Cc: stable@xxxxxxxxxxxxxxx
> Assisted-by: OpenAI-Codex:gpt-5.6-sol
> Signed-off-by: Rafael B. Dias <rafaelbttdias@xxxxxxxxx>

I don't like allowing the AUX stuff to be accounted separately from the
regular stuff. But also, I think you missed perf_mmap_calc_limits().

What about something like the below?

I am also tempted to add a clause where 'vma->vm_mm != rb->mmap_mm'
would completely dis-allow the operation.

---
diff --git a/kernel/events/core.c b/kernel/events/core.c
index e870f12c3ad8..b920680aad50 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -7046,7 +7046,7 @@ static void perf_mmap_open(struct vm_area_struct *vma)
}

static void perf_pmu_output_stop(struct perf_event *event);
-static void perf_mmap_unaccount(struct vm_area_struct *vma, struct perf_buffer *rb);
+static void perf_mmap_unaccount(struct perf_buffer *rb, long user_extra, long extra);

/*
* A buffer can be mmap()ed multiple times; either directly through the same
@@ -7082,8 +7082,7 @@ static void perf_mmap_close(struct vm_area_struct *vma)
perf_pmu_output_stop(event);

/* now it's safe to free the pages */
- atomic_long_sub(rb->aux_nr_pages - rb->aux_mmap_locked, &mmap_user->perf_mlock);
- atomic64_sub(rb->aux_mmap_locked, &vma->vm_mm->pinned_vm);
+ perf_mmap_unaccount(rb, rb->aux_nr_pages, rb->aux_mmap_locked);

/* this has to be the last one */
rb_free_aux(rb);
@@ -7156,7 +7155,7 @@ static void perf_mmap_close(struct vm_area_struct *vma)
* Aside from that, this buffer is 'fully' detached and unmapped,
* undo the VM accounting.
*/
- perf_mmap_unaccount(vma, rb);
+ perf_mmap_unaccount(rb, data_page_nr(rb) + 1, rb->mmap_locked);

out_put:
ring_buffer_put(rb); /* could be last */
@@ -7256,10 +7255,10 @@ static int map_range(struct perf_buffer *rb, struct vm_area_struct *vma)
return err;
}

-static bool perf_mmap_calc_limits(struct vm_area_struct *vma, long *user_extra, long *extra)
+static bool __perf_mmap_calc_limits(struct user_struct *user, struct mm_struct *mm,
+ long *user_extra, long *extra)
{
unsigned long user_locked, user_lock_limit, locked, lock_limit;
- struct user_struct *user = current_user();

user_lock_limit = sysctl_perf_event_mlock >> (PAGE_SHIFT - 10);
/* Increase the limit linearly with more CPUs */
@@ -7281,31 +7280,36 @@ static bool perf_mmap_calc_limits(struct vm_area_struct *vma, long *user_extra,
* charge the rest from pinned_vm
*/
*extra = user_locked - user_lock_limit;
- *user_extra -= *extra;
}

lock_limit = rlimit(RLIMIT_MEMLOCK);
lock_limit >>= PAGE_SHIFT;
- locked = atomic64_read(&vma->vm_mm->pinned_vm) + *extra;
+ locked = atomic64_read(&mm->pinned_vm) + *extra;

return locked <= lock_limit || !perf_is_paranoid() || capable(CAP_IPC_LOCK);
}

-static void perf_mmap_account(struct vm_area_struct *vma, long user_extra, long extra)
+static bool perf_mmap_calc_limits(struct perf_buffer *rb, long *user_extra, long *extra)
{
- struct user_struct *user = current_user();
-
- atomic_long_add(user_extra, &user->perf_mlock);
- atomic64_add(extra, &vma->vm_mm->pinned_vm);
+ return __perf_mmap_calc_limits(rb->mmap_user, rb->mmap_mm, user_extra, extra);
}

-static void perf_mmap_unaccount(struct vm_area_struct *vma, struct perf_buffer *rb)
+static void perf_mmap_account(struct perf_buffer *rb, long user_extra, long extra)
{
struct user_struct *user = rb->mmap_user;
+ struct mm_struct *mm = rb->mmap_mm;

- atomic_long_sub((perf_data_size(rb) >> PAGE_SHIFT) + 1 - rb->mmap_locked,
- &user->perf_mlock);
- atomic64_sub(rb->mmap_locked, &vma->vm_mm->pinned_vm);
+ atomic_long_add(user_extra - extra, &user->perf_mlock);
+ atomic64_add(extra, &mm->pinned_vm);
+}
+
+static void perf_mmap_unaccount(struct perf_buffer *rb, long user_extra, long extra)
+{
+ struct user_struct *user = rb->mmap_user;
+ struct mm_struct *mm = rb->mmap_mm;
+
+ atomic_long_sub(user_extra - extra, &user->perf_mlock);
+ atomic64_sub(extra, &mm->pinned_vm);
}

static int perf_mmap_rb(struct vm_area_struct *vma, struct perf_event *event,
@@ -7344,7 +7348,7 @@ static int perf_mmap_rb(struct vm_area_struct *vma, struct perf_event *event,
* Success -- managed to mmap() the same buffer
* multiple times.
*/
- perf_mmap_account(vma, user_extra, extra);
+ perf_mmap_account(event->rb, user_extra, extra);
refcount_inc(&event->mmap_count);
return 0;
}
@@ -7357,7 +7361,7 @@ static int perf_mmap_rb(struct vm_area_struct *vma, struct perf_event *event,
ring_buffer_attach(event, NULL);
}

- if (!perf_mmap_calc_limits(vma, &user_extra, &extra))
+ if (!__perf_mmap_calc_limits(current_user(), vma->vm_mm, &user_extra, &extra))
return -EPERM;

if (vma->vm_flags & VM_WRITE)
@@ -7370,6 +7374,7 @@ static int perf_mmap_rb(struct vm_area_struct *vma, struct perf_event *event,
if (!rb)
return -ENOMEM;

+ rb->mmap_mm = ({ mmgrab(vma->vm_mm); vma->vm_mm; });
rb->mmap_locked = extra;

ring_buffer_attach(event, rb);
@@ -7378,7 +7383,7 @@ static int perf_mmap_rb(struct vm_area_struct *vma, struct perf_event *event,
perf_event_init_userpage(event);
perf_event_update_userpage(event);

- perf_mmap_account(vma, user_extra, extra);
+ perf_mmap_account(rb, user_extra, extra);
refcount_set(&event->mmap_count, 1);

return 0;
@@ -7434,7 +7439,7 @@ static int perf_mmap_aux(struct vm_area_struct *vma, struct perf_event *event,
refcount_inc(&rb->aux_mmap_count);

} else {
- if (!perf_mmap_calc_limits(vma, &user_extra, &extra)) {
+ if (!perf_mmap_calc_limits(rb, &user_extra, &extra)) {
refcount_dec(&rb->mmap_count);
return -EPERM;
}
@@ -7455,7 +7460,7 @@ static int perf_mmap_aux(struct vm_area_struct *vma, struct perf_event *event,
rb->aux_mmap_locked = extra;
}

- perf_mmap_account(vma, user_extra, extra);
+ perf_mmap_account(rb, user_extra, extra);
refcount_inc(&event->mmap_count);

return 0;
@@ -7547,6 +7552,7 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma)
* attempt use this failed buffer.
*/
if (refcount_read(&event->mmap_count) == 1) {
+ struct perf_buffer *rb = event->rb;
/*
* Minimal perf_mmap_close(); there can't be AUX or
* other events on account of this being the first.
@@ -7554,7 +7560,7 @@ static int perf_mmap(struct file *file, struct vm_area_struct *vma)
mapped = get_mapped(event, event_unmapped);
if (mapped)
mapped(event, vma->vm_mm);
- perf_mmap_unaccount(vma, event->rb);
+ perf_mmap_unaccount(rb, data_page_nr(rb) + 1, rb->mmap_locked);
ring_buffer_attach(event, NULL); /* drops last rb->refcount */
refcount_set(&event->mmap_count, 0);
return ret;
diff --git a/kernel/events/internal.h b/kernel/events/internal.h
index c03c4f2eea57..8a6c620ae91e 100644
--- a/kernel/events/internal.h
+++ b/kernel/events/internal.h
@@ -5,6 +5,9 @@
#include <linux/hardirq.h>
#include <linux/uaccess.h>
#include <linux/refcount.h>
+#include <linux/sched/user.h>
+#include <linux/perf_event.h>
+#include <asm/local.h>

/* Buffer handling */

@@ -38,6 +41,7 @@ struct perf_buffer {
refcount_t mmap_count;
unsigned long mmap_locked;
struct user_struct *mmap_user;
+ struct mm_struct *mmap_mm;

/* AUX area */
struct mutex aux_mutex;
@@ -68,6 +72,7 @@ static inline void rb_free_rcu(struct rcu_head *rcu_head)

rb = container_of(rcu_head, struct perf_buffer, rcu_head);
free_uid(rb->mmap_user);
+ mmdrop(rb->mmap_mm);
rb_free(rb);
}