fix systemtap PR14348: race condition in task_work_add() vs. buildid.exp test case

This commit is contained in:
Frank Ch. Eigler 2012-07-11 17:37:25 -04:00
parent 26f32846dd
commit 806f00ccc3
2 changed files with 418 additions and 1 deletions

412
PR14348.patch Normal file
View File

@ -0,0 +1,412 @@
commit 50bd2c22a62981ae8b7b379d48574671228f1446
Author: David Smith <dsmith@redhat.com>
Date: Mon Jul 9 16:07:22 2012 -0500
Add code to keep track of task_work callbacks to avoid a possible crash.
* runtime/stp_task_work.c: New file.
* runtime/stp_utrace.c (utrace_init): Move STAPCONF_TASK_WORK_ADD_EXPORTED
code to stp_task_work_init().
(utrace_exit): Call stp_task_work_exit().
(utrace_cleanup): Call task_work_cancel() wrapper function.
(utrace_free): Ditto.
(utrace_do_stop): Call task_work_add() wrapper function.
(utrace_control): Ditto.
(finish_report): Ditto.
(utrace_resume): Call stp_task_work_func_done().
* runtime/task_finder2.c (__stp_tf_cancel_task_work): Call
task_work_cancel() wrapper function.
(__stp_tf_quiesce_worker): Call stp_task_work_func_done().
(__stp_utrace_task_finder_target_quiesce): Call task_work_add() wrapper
function.
(__stp_tf_mmap_worker): Call stp_task_work_func_done().
(__stp_utrace_task_finder_target_syscall_exit): Call task_work_add() wrapper
function.
diff --git a/runtime/task_finder2.c b/runtime/task_finder2.c
index 19cb99e..81cb04e 100644
--- a/runtime/task_finder2.c
+++ b/runtime/task_finder2.c
@@ -9,7 +9,6 @@
#ifndef STAPCONF_TASK_UID
#include <linux/cred.h>
#endif
-#include <linux/task_work.h>
#include "syscall.h"
#include "task_finder_map.c"
@@ -170,7 +169,7 @@ static void __stp_tf_cancel_task_work(void)
list_for_each_entry(node, &__stp_tf_task_work_list, list) {
// Remove the item from the list, cancel it, then free it.
list_del(&node->list);
- task_work_cancel(node->task, node->work.func);
+ stp_task_work_cancel(node->task, node->work.func);
_stp_kfree(node);
}
spin_unlock_irqrestore(&__stp_tf_task_work_list_lock, flags);
@@ -1210,8 +1209,11 @@ __stp_tf_quiesce_worker(struct task_work *work)
might_sleep();
__stp_tf_free_task_work(work);
- if (atomic_read(&__stp_task_finder_state) != __STP_TF_RUNNING)
+ if (atomic_read(&__stp_task_finder_state) != __STP_TF_RUNNING) {
+ /* Remember that this task_work_func is finished. */
+ stp_task_work_func_done();
return;
+ }
__stp_tf_handler_start();
@@ -1227,6 +1229,9 @@ __stp_tf_quiesce_worker(struct task_work *work)
}
__stp_tf_handler_end();
+
+ /* Remember that this task_work_func is finished. */
+ stp_task_work_func_done();
return;
}
@@ -1288,12 +1293,13 @@ __stp_utrace_task_finder_target_quiesce(u32 action,
return UTRACE_RESUME;
}
init_task_work(work, &__stp_tf_quiesce_worker, tgt);
- rc = task_work_add(tsk, work, true);
- /* task_work_add() returns -ESRCH if the task has
+
+ rc = stp_task_work_add(tsk, work);
+ /* stp_task_work_add() returns -ESRCH if the task has
* already passed exit_task_work(). Just ignore this
* error. */
if (rc != 0 && rc != -ESRCH) {
- printk(KERN_ERR "%s:%d - task_work_add() returned %d\n",
+ printk(KERN_ERR "%s:%d - stp_task_work_add() returned %d\n",
__FUNCTION__, __LINE__, rc);
}
}
@@ -1397,13 +1403,19 @@ __stp_tf_mmap_worker(struct task_work *work)
might_sleep();
__stp_tf_free_task_work(work);
- if (atomic_read(&__stp_task_finder_state) != __STP_TF_RUNNING)
+ if (atomic_read(&__stp_task_finder_state) != __STP_TF_RUNNING) {
+ /* Remember that this task_work_func is finished. */
+ stp_task_work_func_done();
return;
+ }
// See if we can find saved syscall info.
entry = __stp_tf_get_map_entry(current);
- if (entry == NULL)
+ if (entry == NULL) {
+ /* Remember that this task_work_func is finished. */
+ stp_task_work_func_done();
return;
+ }
__stp_tf_handler_start();
@@ -1426,6 +1438,9 @@ __stp_tf_mmap_worker(struct task_work *work)
__stp_tf_remove_map_entry(entry);
__stp_tf_handler_end();
+
+ /* Remember that this task_work_func is finished. */
+ stp_task_work_func_done();
return;
}
@@ -1492,12 +1507,12 @@ __stp_utrace_task_finder_target_syscall_exit(u32 action,
return UTRACE_RESUME;
}
init_task_work(work, &__stp_tf_mmap_worker, tgt);
- rc = task_work_add(tsk, work, true);
- /* task_work_add() returns -ESRCH if the task has
+ rc = stp_task_work_add(tsk, work);
+ /* stp_task_work_add() returns -ESRCH if the task has
* already passed exit_task_work(). Just ignore this
* error. */
if (rc != 0 && rc != -ESRCH) {
- printk(KERN_ERR "%s:%d - task_work_add() returned %d\n",
+ printk(KERN_ERR "%s:%d - stp_task_work_add() returned %d\n",
__FUNCTION__, __LINE__, rc);
}
}
diff --git a/runtime/stp_task_work.c b/runtime/stp_task_work.c
new file mode 100644
index 0000000..b99593e
--- /dev/null
+++ b/runtime/stp_task_work.c
@@ -0,0 +1,102 @@
+#ifndef _STP_TASK_WORK_C
+#define _STP_TASK_WORK_C
+
+#include <linux/task_work.h>
+
+#if !defined(STAPCONF_TASK_WORK_ADD_EXPORTED)
+typedef int (*task_work_add_fn)(struct task_struct *task,
+ struct task_work *twork, bool notify);
+#define task_work_add (* (task_work_add_fn)kallsyms_task_work_add)
+typedef struct task_work *(*task_work_cancel_fn)(struct task_struct *,
+ task_work_func_t);
+#define task_work_cancel (* (task_work_cancel_fn)kallsyms_task_work_cancel)
+#endif
+
+/* To avoid a crash when a task_work callback gets called after the
+ * module is unloaded, keep track of the number of current callbacks. */
+static atomic_t stp_task_work_callbacks = ATOMIC_INIT(0);
+
+/*
+ * stp_task_work_init() should be called before any other
+ * stp_task_work_* functions are called to do setup.
+ */
+int
+stp_task_work_init(void)
+{
+#if !defined(STAPCONF_TASK_WORK_ADD_EXPORTED)
+ /* The task_work_add()/task_work_cancel() functions aren't
+ * exported. Look up those function addresses. */
+ kallsyms_task_work_add = (void *)kallsyms_lookup_name("task_work_add");
+ if (kallsyms_task_work_add == NULL) {
+ _stp_error("Can't resolve task_work_add!");
+ return -ENOENT;
+ }
+ kallsyms_task_work_cancel = (void *)kallsyms_lookup_name("task_work_cancel");
+ if (kallsyms_task_work_cancel == NULL) {
+ _stp_error("Can't resolve task_work_cancel!");
+ return -ENOENT;
+ }
+#endif
+ return 0;
+}
+
+/*
+ * stap_task_work_exit() should be called when no more
+ * stp_task_work_* functions will be called (before module exit).
+ *
+ * This function makes sure that all the callbacks are finished before
+ * letting the module unload. If the module unloads before a callback
+ * is called, the kernel will try to make a function call to an
+ * invalid address.
+ */
+void
+stp_task_work_exit(void)
+{
+ while (atomic_read(&stp_task_work_callbacks))
+ schedule_timeout_uninterruptible(1);
+ return;
+}
+
+/*
+ * Our task_work_add() wrapper that remembers that we've got a pending
+ * callback.
+ */
+int
+stp_task_work_add(struct task_struct *task, struct task_work *twork)
+{
+ int rc;
+
+ rc = task_work_add(task, twork, true);
+ if (rc == 0)
+ atomic_inc(&stp_task_work_callbacks);
+ return rc;
+}
+
+/*
+ * Our task_work_cancel() wrapper that remembers that a callback has
+ * been cancelled.
+ */
+struct task_work *
+stp_task_work_cancel(struct task_struct *task, task_work_func_t func)
+{
+ struct task_work *twork;
+
+ twork = task_work_cancel(task, func);
+ if (twork != NULL)
+ atomic_dec(&stp_task_work_callbacks);
+ return twork;
+}
+
+/*
+ * stp_task_work_func_done() should be called at the very end of a
+ * task_work callback function so that we can keep up with callback
+ * accounting.
+ */
+void
+stp_task_work_func_done(void)
+{
+ atomic_dec(&stp_task_work_callbacks);
+}
+
+
+#endif /* _STP_TASK_WORK_C */
diff --git a/runtime/stp_utrace.c b/runtime/stp_utrace.c
index d4538d9..face3f9 100644
--- a/runtime/stp_utrace.c
+++ b/runtime/stp_utrace.c
@@ -25,7 +25,7 @@
#include <linux/spinlock.h>
#include <trace/events/sched.h>
#include <trace/events/syscalls.h>
-#include <linux/task_work.h>
+#include "stp_task_work.c"
/*
* Per-thread structure private to utrace implementation.
@@ -105,38 +105,13 @@ static void utrace_report_exec(void *cb_data __attribute__ ((unused)),
#define __UTRACE_REGISTERED 1
static atomic_t utrace_state = ATOMIC_INIT(__UTRACE_UNREGISTERED);
-#if !defined(STAPCONF_TASK_WORK_ADD_EXPORTED)
-typedef int (*task_work_add_fn)(struct task_struct *task,
- struct task_work *twork, bool notify);
-#define task_work_add (* (task_work_add_fn)kallsyms_task_work_add)
-typedef struct task_work *(*task_work_cancel_fn)(struct task_struct *,
- task_work_func_t);
-#define task_work_cancel (* (task_work_cancel_fn)kallsyms_task_work_cancel)
-#endif
-
int utrace_init(void)
{
int i;
int rc = -1;
-#if !defined(STAPCONF_TASK_WORK_ADD_EXPORTED)
- /* The task_work_add()/task_work_cancel() functions aren't
- * exported. Look up those function addresses. */
- kallsyms_task_work_add = (void *)kallsyms_lookup_name ("task_work_add");
- if (kallsyms_task_work_add == NULL) {
- printk(KERN_ERR "%s can't resolve task_work_add!",
- THIS_MODULE->name);
- rc = -ENOENT;
+ if (unlikely(stp_task_work_init() != 0))
goto error;
- }
- kallsyms_task_work_cancel = (void *)kallsyms_lookup_name ("task_work_cancel");
- if (kallsyms_task_work_cancel == NULL) {
- printk(KERN_ERR "%s can't resolve task_work_cancel!",
- THIS_MODULE->name);
- rc = -ENOENT;
- goto error;
- }
-#endif
/* initialize the list heads */
for (i = 0; i < TASK_UTRACE_TABLE_SIZE; i++) {
@@ -205,6 +180,8 @@ int utrace_exit(void)
kmem_cache_destroy(utrace_cachep);
if (utrace_engine_cachep)
kmem_cache_destroy(utrace_engine_cachep);
+
+ stp_task_work_exit();
return 0;
}
@@ -239,7 +216,7 @@ static void utrace_cleanup(struct utrace *utrace)
}
if (utrace->task_work_added) {
- if (task_work_cancel(utrace->task, &utrace_resume) == NULL)
+ if (stp_task_work_cancel(utrace->task, &utrace_resume) == NULL)
printk(KERN_ERR "%s:%d - task_work_cancel() failed? task %p, %d, %s\n",
__FUNCTION__, __LINE__, utrace->task,
utrace->task->tgid,
@@ -379,7 +356,7 @@ static void utrace_free(struct utrace *utrace)
#endif
if (utrace->task_work_added) {
- if (task_work_cancel(utrace->task, &utrace_resume) == NULL)
+ if (stp_task_work_cancel(utrace->task, &utrace_resume) == NULL)
printk(KERN_ERR "%s:%d - task_work_cancel() failed? task %p, %d, %s\n",
__FUNCTION__, __LINE__, utrace->task,
utrace->task->tgid,
@@ -898,11 +875,11 @@ static bool utrace_do_stop(struct task_struct *target, struct utrace *utrace)
} else if (utrace->resume > UTRACE_REPORT) {
utrace->resume = UTRACE_REPORT;
if (! utrace->task_work_added) {
- int rc = task_work_add(target, &utrace->work, true);
+ int rc = stp_task_work_add(target, &utrace->work);
if (rc == 0) {
utrace->task_work_added = 1;
}
- /* task_work_add() returns -ESRCH if the task
+ /* stp_task_work_add() returns -ESRCH if the task
* has already passed exit_task_work(). Just
* ignore this error. */
else if (rc != -ESRCH) {
@@ -1041,16 +1018,16 @@ static void utrace_stop(struct task_struct *task, struct utrace *utrace,
*/
utrace->resume = action;
if (! utrace->task_work_added) {
- int rc = task_work_add(task, &utrace->work, true);
+ int rc = stp_task_work_add(task, &utrace->work);
if (rc == 0) {
utrace->task_work_added = 1;
}
- /* task_work_add() returns -ESRCH if the task
+ /* stp_task_work_add() returns -ESRCH if the task
* has already passed exit_task_work(). Just
* ignore this error. */
else if (rc != -ESRCH) {
printk(KERN_ERR
- "%s:%d - task_work_add() returned %d\n",
+ "%s:%d - stp_task_work_add() returned %d\n",
__FUNCTION__, __LINE__, rc);
}
}
@@ -1397,18 +1374,17 @@ int utrace_control(struct task_struct *target,
if (action < utrace->resume) {
utrace->resume = action;
if (! utrace->task_work_added) {
- ret = task_work_add(target, &utrace->work,
- true);
+ ret = stp_task_work_add(target, &utrace->work);
if (ret == 0) {
utrace->task_work_added = 1;
}
- /* task_work_add() returns -ESRCH if
+ /* stp_task_work_add() returns -ESRCH if
* the task has already passed
* exit_task_work(). Just ignore this
* error. */
else if (ret != -ESRCH) {
printk(KERN_ERR
- "%s:%d - task_work_add() returned %d\n",
+ "%s:%d - stp_task_work_add() returned %d\n",
__FUNCTION__, __LINE__, ret);
}
}
@@ -1560,11 +1536,11 @@ static void finish_report(struct task_struct *task, struct utrace *utrace,
spin_lock(&utrace->lock);
utrace->resume = resume;
if (! utrace->task_work_added) {
- int rc = task_work_add(task, &utrace->work, true);
+ int rc = stp_task_work_add(task, &utrace->work);
if (rc == 0) {
utrace->task_work_added = 1;
}
- /* task_work_add() returns -ESRCH if the task
+ /* stp_task_work_add() returns -ESRCH if the task
* has already passed exit_task_work(). Just
* ignore this error. */
else if (rc != -ESRCH) {
@@ -2132,6 +2108,9 @@ void utrace_resume(struct task_work *work)
* call. (The arch might use TIF_NOTIFY_RESUME for other
* purposes as well as calling us.)
*/
+
+ /* Remember that this task_work_func is finished. */
+ stp_task_work_func_done();
return;
case UTRACE_REPORT:
if (unlikely(!(utrace->utrace_flags & UTRACE_EVENT(QUIESCE))))
@@ -2160,6 +2139,9 @@ void utrace_resume(struct task_work *work)
* effect now (i.e. step or interrupt).
*/
finish_resume_report(task, utrace, &report);
+
+ /* Remember that this task_work_func is finished. */
+ stp_task_work_func_done();
}
#endif /* _STP_UTRACE_C */

View File

@ -20,7 +20,7 @@
Name: systemtap
Version: 1.8
Release: 1%{?dist}
Release: 3%{?dist}
# for version, see also configure.ac
@ -91,6 +91,7 @@ BuildRequires: /usr/share/publican/Common_Content/%{publican_brand}/defaults.cfg
%endif
Patch2: bz837641-staprun-no-linux-types.patch
Patch3: PR14348.patch
# Install requirements
Requires: systemtap-client = %{version}-%{release}
@ -253,6 +254,7 @@ cd ..
# bz837641-staprun-no-linux-types.patch
%patch2 -p1
%patch3 -p1
%build
@ -584,6 +586,9 @@ exit 0
# ------------------------------------------------------------------------
%changelog
* Wed Jul 11 2012 Frank Ch. Eigler <fche@redhat.com> - 1.8-3
- PR14348 task_work_add race condition fix
* Mon Jul 09 2012 Josh Stone <jistone@redhat.com>
- bz837641 build fix