From: Lucas De Marchi <lucas.demarchi@intel.com>
To: intel-gfx@lists.freedesktop.org, linux-perf-users@vger.kernel.org
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>,
dri-devel@lists.freedesktop.org,
Peter Zijlstra <peterz@infradead.org>,
Ingo Molnar <mingo@redhat.com>,
Arnaldo Carvalho de Melo <acme@kernel.org>,
linux-kernel@vger.kernel.org,
Lucas De Marchi <lucas.demarchi@intel.com>
Subject: [PATCH 1/7] perf/core: Add pmu get/put
Date: Mon, 22 Jul 2024 14:06:42 -0700 [thread overview]
Message-ID: <20240722210648.80892-2-lucas.demarchi@intel.com> (raw)
In-Reply-To: <20240722210648.80892-1-lucas.demarchi@intel.com>
If a pmu is unregistered while there's an active event, perf will still
access the pmu via event->pmu, even after the event is destroyed. This
makes it difficult for drivers like i915 that take a reference on the
device when the event is created and put it when it's destroyed.
Currently the following use-after-free happens just after destroying the
event:
BUG: KASAN: use-after-free in exclusive_event_destroy+0xd8/0xf0
Read of size 4 at addr ffff88816e2bb63c by task perf/7748
Whenever and event is created, get a pmu reference to use in event->pmu
and just before calling module_put(), drop the reference..
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
---
include/linux/perf_event.h | 3 +++
kernel/events/core.c | 32 ++++++++++++++++++++++++++++----
2 files changed, 31 insertions(+), 4 deletions(-)
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index a5304ae8c654..7048a505e93c 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -540,6 +540,9 @@ struct pmu {
* Check period value for PERF_EVENT_IOC_PERIOD ioctl.
*/
int (*check_period) (struct perf_event *event, u64 value); /* optional */
+
+ struct pmu *(*get) (struct pmu *pmu); /* optional: get a reference */
+ void (*put) (struct pmu *pmu); /* optional: put a reference */
};
enum perf_addr_filter_action_t {
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 1b6f5dc7ed32..cc7541b644b0 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -5208,6 +5208,8 @@ static void perf_addr_filters_splice(struct perf_event *event,
static void _free_event(struct perf_event *event)
{
+ struct module *module;
+
irq_work_sync(&event->pending_irq);
unaccount_event(event);
@@ -5259,7 +5261,13 @@ static void _free_event(struct perf_event *event)
put_ctx(event->ctx);
exclusive_event_destroy(event);
- module_put(event->pmu->module);
+
+ module = event->pmu->module;
+ event->pmu->put(event->pmu);
+ /* can't touch pmu anymore */
+ event->pmu = NULL;
+
+ module_put(module);
call_rcu(&event->rcu_head, free_event_rcu);
}
@@ -11331,6 +11339,11 @@ static int perf_pmu_nop_int(struct pmu *pmu)
return 0;
}
+static struct pmu *perf_pmu_nop_pmu(struct pmu *pmu)
+{
+ return pmu;
+}
+
static int perf_event_nop_int(struct perf_event *event, u64 value)
{
return 0;
@@ -11617,6 +11630,12 @@ int perf_pmu_register(struct pmu *pmu, const char *name, int type)
if (!pmu->event_idx)
pmu->event_idx = perf_event_idx_default;
+ if (!pmu->get)
+ pmu->get = perf_pmu_nop_pmu;
+
+ if (!pmu->put)
+ pmu->put = perf_pmu_nop_void;
+
list_add_rcu(&pmu->entry, &pmus);
atomic_set(&pmu->exclusive_cnt, 0);
ret = 0;
@@ -11695,7 +11714,8 @@ static int perf_try_init_event(struct pmu *pmu, struct perf_event *event)
BUG_ON(!ctx);
}
- event->pmu = pmu;
+ event->pmu = pmu->get(pmu);
+
ret = pmu->event_init(event);
if (ctx)
@@ -11714,8 +11734,12 @@ static int perf_try_init_event(struct pmu *pmu, struct perf_event *event)
event->destroy(event);
}
- if (ret)
- module_put(pmu->module);
+ if (ret) {
+ struct module *module = pmu->module;
+
+ pmu->put(pmu);
+ module_put(module);
+ }
return ret;
}
--
2.43.0
next prev parent reply other threads:[~2024-07-22 21:07 UTC|newest]
Thread overview: 20+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-07-22 21:06 [PATCH 0/7] Fix i915 pmu on bind/unbind Lucas De Marchi
2024-07-22 21:06 ` Lucas De Marchi [this message]
2024-07-23 23:07 ` [PATCH 1/7] perf/core: Add pmu get/put Ian Rogers
2024-07-22 21:06 ` [PATCH 2/7] drm/i915/pmu: Fix crash due to use-after-free Lucas De Marchi
2024-07-22 21:06 ` [PATCH 3/7] drm/i915/pmu: Use event_to_pmu() Lucas De Marchi
2024-07-23 4:35 ` Dixit, Ashutosh
2024-07-22 21:06 ` [PATCH 4/7] drm/i915/pmu: Drop is_igp() Lucas De Marchi
2024-07-22 23:25 ` Dixit, Ashutosh
2024-07-23 7:52 ` Tvrtko Ursulin
2024-07-22 21:06 ` [PATCH 5/7] drm/i915/pmu: Let resource survive unbind Lucas De Marchi
2024-07-23 7:58 ` Tvrtko Ursulin
2024-07-22 21:06 ` [PATCH 6/7] drm/i915/pmu: Lazy unregister Lucas De Marchi
2024-07-23 8:03 ` Tvrtko Ursulin
2024-07-23 15:30 ` Lucas De Marchi
2024-07-24 7:48 ` Tvrtko Ursulin
2024-07-24 12:41 ` Peter Zijlstra
2024-07-24 15:39 ` Lucas De Marchi
2024-09-09 21:03 ` Lucas De Marchi
2024-07-22 21:06 ` [PATCH 7/7] drm/i915/pmu: Do not set event_init to NULL Lucas De Marchi
2024-08-05 6:55 ` kernel test robot
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20240722210648.80892-2-lucas.demarchi@intel.com \
--to=lucas.demarchi@intel.com \
--cc=acme@kernel.org \
--cc=dri-devel@lists.freedesktop.org \
--cc=intel-gfx@lists.freedesktop.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-perf-users@vger.kernel.org \
--cc=mingo@redhat.com \
--cc=peterz@infradead.org \
--cc=tvrtko.ursulin@linux.intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).