rv: Add runtime reactors interface
authorDaniel Bristot de Oliveira <bristot@kernel.org>
Fri, 29 Jul 2022 09:38:41 +0000 (11:38 +0200)
committerSteven Rostedt (Google) <rostedt@goodmis.org>
Sat, 30 Jul 2022 18:01:28 +0000 (14:01 -0400)
A runtime monitor can cause a reaction to the detection of an
exception on the model's execution. By default, the monitors have
tracing reactions, printing the monitor output via tracepoints.
But other reactions can be added (on-demand) via this interface.

The user interface resembles the kernel tracing interface and
presents these files:

"available_reactors"
  - Reading shows the available reactors, one per line.

   For example:
     # cat available_reactors
     nop
     panic
     printk

 "reacting_on"
   - It is an on/off general switch for reactors, disabling
   all reactions.

 "monitors/MONITOR/reactors"
   - List available reactors, with the select reaction for the given
   MONITOR inside []. The default one is the nop (no operation)
   reactor.
   - Writing the name of a reactor enables it to the given
   MONITOR.

   For example:
     # cat monitors/wip/reactors
     [nop]
     panic
     printk
     # echo panic > monitors/wip/reactors
     # cat monitors/wip/reactors
     nop
     [panic]
     printk

Link: https://lkml.kernel.org/r/1794eb994637457bdeaa6bad0b8263d2f7eece0c.1659052063.git.bristot@kernel.org
Cc: Wim Van Sebroeck <wim@linux-watchdog.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Marco Elver <elver@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: "Paul E. McKenney" <paulmck@kernel.org>
Cc: Shuah Khan <skhan@linuxfoundation.org>
Cc: Gabriele Paoloni <gpaoloni@redhat.com>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Clark Williams <williams@redhat.com>
Cc: Tao Zhou <tao.zhou@linux.dev>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: linux-doc@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: linux-trace-devel@vger.kernel.org
Signed-off-by: Daniel Bristot de Oliveira <bristot@kernel.org>
Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
include/linux/rv.h
kernel/trace/rv/Kconfig
kernel/trace/rv/Makefile
kernel/trace/rv/rv.c
kernel/trace/rv/rv.h
kernel/trace/rv/rv_reactors.c [new file with mode: 0644]

index d8fa9e8..dcce569 100644 (file)
 union rv_task_monitor {
 };
 
+#ifdef CONFIG_RV_REACTORS
+struct rv_reactor {
+       const char              *name;
+       const char              *description;
+       void                    (*react)(char *msg);
+};
+#endif
+
 struct rv_monitor {
        const char              *name;
        const char              *description;
@@ -31,6 +39,9 @@ struct rv_monitor {
        int                     (*enable)(void);
        void                    (*disable)(void);
        void                    (*reset)(void);
+#ifdef CONFIG_RV_REACTORS
+       void                    (*react)(char *msg);
+#endif
 };
 
 bool rv_monitoring_on(void);
@@ -39,5 +50,11 @@ int rv_register_monitor(struct rv_monitor *monitor);
 int rv_get_task_monitor_slot(void);
 void rv_put_task_monitor_slot(int slot);
 
+#ifdef CONFIG_RV_REACTORS
+bool rv_reacting_on(void);
+int rv_unregister_reactor(struct rv_reactor *reactor);
+int rv_register_reactor(struct rv_reactor *reactor);
+#endif /* CONFIG_RV_REACTORS */
+
 #endif /* CONFIG_RV */
 #endif /* _LINUX_RV_H */
index 6d127cd..3eb5d48 100644 (file)
@@ -10,3 +10,14 @@ menuconfig RV
          theorem proving). RV works by analyzing the trace of the system's
          actual execution, comparing it against a formal specification of
          the system behavior.
+
+config RV_REACTORS
+       bool "Runtime verification reactors"
+       default y
+       depends on RV
+       help
+         Enables the online runtime verification reactors. A runtime
+         monitor can cause a reaction to the detection of an exception
+         on the model's execution. By default, the monitors have
+         tracing reactions, printing the monitor output via tracepoints,
+         but other reactions can be added (on-demand) via this interface.
index fd99537..8944274 100644 (file)
@@ -1,3 +1,4 @@
 # SPDX-License-Identifier: GPL-2.0
 
 obj-$(CONFIG_RV) += rv.o
+obj-$(CONFIG_RV_REACTORS) += rv_reactors.o
index 731cc96..45cf64e 100644 (file)
@@ -353,6 +353,10 @@ static int create_monitor_dir(struct rv_monitor_def *mdef)
                goto out_remove_root;
        }
 
+       retval = reactor_populate_monitor(mdef);
+       if (retval)
+               goto out_remove_root;
+
        return 0;
 
 out_remove_root:
@@ -669,6 +673,7 @@ static const struct file_operations monitoring_on_fops = {
 
 static void destroy_monitor_dir(struct rv_monitor_def *mdef)
 {
+       reactor_cleanup_monitor(mdef);
        rv_remove(mdef->root_d);
 }
 
@@ -747,6 +752,7 @@ int rv_unregister_monitor(struct rv_monitor *monitor)
 int __init rv_init_interface(void)
 {
        struct dentry *tmp;
+       int retval;
 
        rv_root.root_dir = rv_create_dir("rv", NULL);
        if (!rv_root.root_dir)
@@ -770,6 +776,9 @@ int __init rv_init_interface(void)
                             &monitoring_on_fops);
        if (!tmp)
                goto out_err;
+       retval = init_rv_reactors(rv_root.root_dir);
+       if (retval)
+               goto out_err;
 
        turn_monitoring_on();
 
index 50014aa..db6cb09 100644 (file)
@@ -18,16 +18,51 @@ struct rv_interface {
 #define rv_remove                      tracefs_remove
 
 #define MAX_RV_MONITOR_NAME_SIZE       32
+#define MAX_RV_REACTOR_NAME_SIZE       32
 
 extern struct mutex rv_interface_lock;
 
+#ifdef CONFIG_RV_REACTORS
+struct rv_reactor_def {
+       struct list_head        list;
+       struct rv_reactor       *reactor;
+       /* protected by the monitor interface lock */
+       int                     counter;
+};
+#endif
+
 struct rv_monitor_def {
        struct list_head        list;
        struct rv_monitor       *monitor;
        struct dentry           *root_d;
+#ifdef CONFIG_RV_REACTORS
+       struct rv_reactor_def   *rdef;
+       bool                    reacting;
+#endif
        bool                    task_monitor;
 };
 
 struct dentry *get_monitors_root(void);
 int rv_disable_monitor(struct rv_monitor_def *mdef);
 int rv_enable_monitor(struct rv_monitor_def *mdef);
+
+#ifdef CONFIG_RV_REACTORS
+int reactor_populate_monitor(struct rv_monitor_def *mdef);
+void reactor_cleanup_monitor(struct rv_monitor_def *mdef);
+int init_rv_reactors(struct dentry *root_dir);
+#else
+static inline int reactor_populate_monitor(struct rv_monitor_def *mdef)
+{
+       return 0;
+}
+
+static inline void reactor_cleanup_monitor(struct rv_monitor_def *mdef)
+{
+       return;
+}
+
+static inline int init_rv_reactors(struct dentry *root_dir)
+{
+       return 0;
+}
+#endif
diff --git a/kernel/trace/rv/rv_reactors.c b/kernel/trace/rv/rv_reactors.c
new file mode 100644 (file)
index 0000000..a6522c1
--- /dev/null
@@ -0,0 +1,508 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2019-2022 Red Hat, Inc. Daniel Bristot de Oliveira <bristot@kernel.org>
+ *
+ * Runtime reactor interface.
+ *
+ * A runtime monitor can cause a reaction to the detection of an
+ * exception on the model's execution. By default, the monitors have
+ * tracing reactions, printing the monitor output via tracepoints.
+ * But other reactions can be added (on-demand) via this interface.
+ *
+ * == Registering reactors ==
+ *
+ * The struct rv_reactor defines a callback function to be executed
+ * in case of a model exception happens. The callback function
+ * receives a message to be (optionally) printed before executing
+ * the reaction.
+ *
+ * A RV reactor is registered via:
+ *   int rv_register_reactor(struct rv_reactor *reactor)
+ * And unregistered via:
+ *   int rv_unregister_reactor(struct rv_reactor *reactor)
+ *
+ * These functions are exported to modules, enabling reactors to be
+ * dynamically loaded.
+ *
+ * == User interface ==
+ *
+ * The user interface resembles the kernel tracing interface and
+ * presents these files:
+ *
+ *  "available_reactors"
+ *    - List the available reactors, one per line.
+ *
+ *    For example:
+ *      # cat available_reactors
+ *      nop
+ *      panic
+ *      printk
+ *
+ *  "reacting_on"
+ *    - It is an on/off general switch for reactors, disabling
+ *    all reactions.
+ *
+ *  "monitors/MONITOR/reactors"
+ *    - List available reactors, with the select reaction for the given
+ *    MONITOR inside []. The default one is the nop (no operation)
+ *    reactor.
+ *    - Writing the name of an reactor enables it to the given
+ *    MONITOR.
+ *
+ *    For example:
+ *      # cat monitors/wip/reactors
+ *      [nop]
+ *      panic
+ *      printk
+ *      # echo panic > monitors/wip/reactors
+ *      # cat monitors/wip/reactors
+ *      nop
+ *      [panic]
+ *      printk
+ */
+
+#include <linux/slab.h>
+
+#include "rv.h"
+
+/*
+ * Interface for the reactor register.
+ */
+static LIST_HEAD(rv_reactors_list);
+
+static struct rv_reactor_def *get_reactor_rdef_by_name(char *name)
+{
+       struct rv_reactor_def *r;
+
+       list_for_each_entry(r, &rv_reactors_list, list) {
+               if (strcmp(name, r->reactor->name) == 0)
+                       return r;
+       }
+       return NULL;
+}
+
+/*
+ * Available reactors seq functions.
+ */
+static int reactors_show(struct seq_file *m, void *p)
+{
+       struct rv_reactor_def *rea_def = p;
+
+       seq_printf(m, "%s\n", rea_def->reactor->name);
+       return 0;
+}
+
+static void reactors_stop(struct seq_file *m, void *p)
+{
+       mutex_unlock(&rv_interface_lock);
+}
+
+static void *reactors_start(struct seq_file *m, loff_t *pos)
+{
+       mutex_lock(&rv_interface_lock);
+       return seq_list_start(&rv_reactors_list, *pos);
+}
+
+static void *reactors_next(struct seq_file *m, void *p, loff_t *pos)
+{
+       return seq_list_next(p, &rv_reactors_list, pos);
+}
+
+/*
+ * available_reactors seq definition.
+ */
+static const struct seq_operations available_reactors_seq_ops = {
+       .start  = reactors_start,
+       .next   = reactors_next,
+       .stop   = reactors_stop,
+       .show   = reactors_show
+};
+
+/*
+ * available_reactors interface.
+ */
+static int available_reactors_open(struct inode *inode, struct file *file)
+{
+       return seq_open(file, &available_reactors_seq_ops);
+};
+
+static const struct file_operations available_reactors_ops = {
+       .open    = available_reactors_open,
+       .read    = seq_read,
+       .llseek  = seq_lseek,
+       .release = seq_release
+};
+
+/*
+ * Monitor's reactor file.
+ */
+static int monitor_reactor_show(struct seq_file *m, void *p)
+{
+       struct rv_monitor_def *mdef = m->private;
+       struct rv_reactor_def *rdef = p;
+
+       if (mdef->rdef == rdef)
+               seq_printf(m, "[%s]\n", rdef->reactor->name);
+       else
+               seq_printf(m, "%s\n", rdef->reactor->name);
+       return 0;
+}
+
+/*
+ * available_reactors seq definition.
+ */
+static const struct seq_operations monitor_reactors_seq_ops = {
+       .start  = reactors_start,
+       .next   = reactors_next,
+       .stop   = reactors_stop,
+       .show   = monitor_reactor_show
+};
+
+static void monitor_swap_reactors(struct rv_monitor_def *mdef, struct rv_reactor_def *rdef,
+                                   bool reacting)
+{
+       bool monitor_enabled;
+
+       /* nothing to do */
+       if (mdef->rdef == rdef)
+               return;
+
+       monitor_enabled = mdef->monitor->enabled;
+       if (monitor_enabled)
+               rv_disable_monitor(mdef);
+
+       /* swap reactor's usage */
+       mdef->rdef->counter--;
+       rdef->counter++;
+
+       mdef->rdef = rdef;
+       mdef->reacting = reacting;
+       mdef->monitor->react = rdef->reactor->react;
+
+       if (monitor_enabled)
+               rv_enable_monitor(mdef);
+}
+
+static ssize_t
+monitor_reactors_write(struct file *file, const char __user *user_buf,
+                     size_t count, loff_t *ppos)
+{
+       char buff[MAX_RV_REACTOR_NAME_SIZE + 2];
+       struct rv_monitor_def *mdef;
+       struct rv_reactor_def *rdef;
+       struct seq_file *seq_f;
+       int retval = -EINVAL;
+       bool enable;
+       char *ptr;
+       int len;
+
+       if (count < 1 || count > MAX_RV_REACTOR_NAME_SIZE + 1)
+               return -EINVAL;
+
+       memset(buff, 0, sizeof(buff));
+
+       retval = simple_write_to_buffer(buff, sizeof(buff) - 1, ppos, user_buf, count);
+       if (retval < 0)
+               return -EFAULT;
+
+       ptr = strim(buff);
+
+       len = strlen(ptr);
+       if (!len)
+               return count;
+
+       /*
+        * See monitor_reactors_open()
+        */
+       seq_f = file->private_data;
+       mdef = seq_f->private;
+
+       mutex_lock(&rv_interface_lock);
+
+       retval = -EINVAL;
+
+       list_for_each_entry(rdef, &rv_reactors_list, list) {
+               if (strcmp(ptr, rdef->reactor->name) != 0)
+                       continue;
+
+               if (rdef == get_reactor_rdef_by_name("nop"))
+                       enable = false;
+               else
+                       enable = true;
+
+               monitor_swap_reactors(mdef, rdef, enable);
+
+               retval = count;
+               break;
+       }
+
+       mutex_unlock(&rv_interface_lock);
+
+       return retval;
+}
+
+/*
+ * available_reactors interface.
+ */
+static int monitor_reactors_open(struct inode *inode, struct file *file)
+{
+       struct rv_monitor_def *mdef = inode->i_private;
+       struct seq_file *seq_f;
+       int ret;
+
+       ret = seq_open(file, &monitor_reactors_seq_ops);
+       if (ret < 0)
+               return ret;
+
+       /*
+        * seq_open stores the seq_file on the file->private data.
+        */
+       seq_f = file->private_data;
+
+       /*
+        * Copy the create file "private" data to the seq_file private data.
+        */
+       seq_f->private = mdef;
+
+       return 0;
+};
+
+static const struct file_operations monitor_reactors_ops = {
+       .open    = monitor_reactors_open,
+       .read    = seq_read,
+       .llseek  = seq_lseek,
+       .release = seq_release,
+       .write = monitor_reactors_write
+};
+
+static int __rv_register_reactor(struct rv_reactor *reactor)
+{
+       struct rv_reactor_def *r;
+
+       list_for_each_entry(r, &rv_reactors_list, list) {
+               if (strcmp(reactor->name, r->reactor->name) == 0) {
+                       pr_info("Reactor %s is already registered\n", reactor->name);
+                       return -EINVAL;
+               }
+       }
+
+       r = kzalloc(sizeof(struct rv_reactor_def), GFP_KERNEL);
+       if (!r)
+               return -ENOMEM;
+
+       r->reactor = reactor;
+       r->counter = 0;
+
+       list_add_tail(&r->list, &rv_reactors_list);
+
+       return 0;
+}
+
+/**
+ * rv_register_reactor - register a rv reactor.
+ * @reactor:   The rv_reactor to be registered.
+ *
+ * Returns 0 if successful, error otherwise.
+ */
+int rv_register_reactor(struct rv_reactor *reactor)
+{
+       int retval = 0;
+
+       if (strlen(reactor->name) >= MAX_RV_REACTOR_NAME_SIZE) {
+               pr_info("Reactor %s has a name longer than %d\n",
+                       reactor->name, MAX_RV_MONITOR_NAME_SIZE);
+               return -EINVAL;
+       }
+
+       mutex_lock(&rv_interface_lock);
+       retval = __rv_register_reactor(reactor);
+       mutex_unlock(&rv_interface_lock);
+       return retval;
+}
+
+/**
+ * rv_unregister_reactor - unregister a rv reactor.
+ * @reactor:   The rv_reactor to be unregistered.
+ *
+ * Returns 0 if successful, error otherwise.
+ */
+int rv_unregister_reactor(struct rv_reactor *reactor)
+{
+       struct rv_reactor_def *ptr, *next;
+
+       mutex_lock(&rv_interface_lock);
+
+       list_for_each_entry_safe(ptr, next, &rv_reactors_list, list) {
+               if (strcmp(reactor->name, ptr->reactor->name) == 0) {
+
+                       if (!ptr->counter) {
+                               list_del(&ptr->list);
+                       } else {
+                               printk(KERN_WARNING
+                                      "rv: the rv_reactor %s is in use by %d monitor(s)\n",
+                                      ptr->reactor->name, ptr->counter);
+                               printk(KERN_WARNING "rv: the rv_reactor %s cannot be removed\n",
+                                      ptr->reactor->name);
+                               return -EBUSY;
+                       }
+               }
+       }
+
+       mutex_unlock(&rv_interface_lock);
+       return 0;
+}
+
+/*
+ * reacting_on interface.
+ */
+static bool __read_mostly reacting_on;
+
+/**
+ * rv_reacting_on - checks if reacting is on
+ *
+ * Returns 1 if on, 0 otherwise.
+ */
+bool rv_reacting_on(void)
+{
+       /* Ensures that concurrent monitors read consistent reacting_on */
+       smp_rmb();
+       return READ_ONCE(reacting_on);
+}
+
+static ssize_t reacting_on_read_data(struct file *filp,
+                                    char __user *user_buf,
+                                    size_t count, loff_t *ppos)
+{
+       char *buff;
+
+       buff = rv_reacting_on() ? "1\n" : "0\n";
+
+       return simple_read_from_buffer(user_buf, count, ppos, buff, strlen(buff)+1);
+}
+
+static void turn_reacting_off(void)
+{
+       WRITE_ONCE(reacting_on, false);
+       /* Ensures that concurrent monitors read consistent reacting_on */
+       smp_wmb();
+}
+
+static void turn_reacting_on(void)
+{
+       WRITE_ONCE(reacting_on, true);
+       /* Ensures that concurrent monitors read consistent reacting_on */
+       smp_wmb();
+}
+
+static ssize_t reacting_on_write_data(struct file *filp, const char __user *user_buf,
+                                     size_t count, loff_t *ppos)
+{
+       int retval;
+       bool val;
+
+       retval = kstrtobool_from_user(user_buf, count, &val);
+       if (retval)
+               return retval;
+
+       mutex_lock(&rv_interface_lock);
+
+       if (val)
+               turn_reacting_on();
+       else
+               turn_reacting_off();
+
+       /*
+        * Wait for the execution of all events to finish
+        * before returning to user-space.
+        */
+       tracepoint_synchronize_unregister();
+
+       mutex_unlock(&rv_interface_lock);
+
+       return count;
+}
+
+static const struct file_operations reacting_on_fops = {
+       .open   = simple_open,
+       .llseek = no_llseek,
+       .write  = reacting_on_write_data,
+       .read   = reacting_on_read_data,
+};
+
+/**
+ * reactor_populate_monitor - creates per monitor reactors file
+ * @mdef:      monitor's definition.
+ *
+ * Returns 0 if successful, error otherwise.
+ */
+int reactor_populate_monitor(struct rv_monitor_def *mdef)
+{
+       struct dentry *tmp;
+
+       tmp = rv_create_file("reactors", RV_MODE_WRITE, mdef->root_d, mdef, &monitor_reactors_ops);
+       if (!tmp)
+               return -ENOMEM;
+
+       /*
+        * Configure as the rv_nop reactor.
+        */
+       mdef->rdef = get_reactor_rdef_by_name("nop");
+       mdef->rdef->counter++;
+       mdef->reacting = false;
+
+       return 0;
+}
+
+/**
+ * reactor_cleanup_monitor - cleanup a monitor reference
+ * @mdef:       monitor's definition.
+ */
+void reactor_cleanup_monitor(struct rv_monitor_def *mdef)
+{
+       lockdep_assert_held(&rv_interface_lock);
+       mdef->rdef->counter--;
+       WARN_ON_ONCE(mdef->rdef->counter < 0);
+}
+
+/*
+ * Nop reactor register
+ */
+static void rv_nop_reaction(char *msg)
+{
+}
+
+static struct rv_reactor rv_nop = {
+       .name = "nop",
+       .description = "no-operation reactor: do nothing.",
+       .react = rv_nop_reaction
+};
+
+int init_rv_reactors(struct dentry *root_dir)
+{
+       struct dentry *available, *reacting;
+       int retval;
+
+       available = rv_create_file("available_reactors", RV_MODE_READ, root_dir, NULL,
+                                  &available_reactors_ops);
+       if (!available)
+               goto out_err;
+
+       reacting = rv_create_file("reacting_on", RV_MODE_WRITE, root_dir, NULL, &reacting_on_fops);
+       if (!reacting)
+               goto rm_available;
+
+       retval = __rv_register_reactor(&rv_nop);
+       if (retval)
+               goto rm_reacting;
+
+       turn_reacting_on();
+
+       return 0;
+
+rm_reacting:
+       rv_remove(reacting);
+rm_available:
+       rv_remove(available);
+out_err:
+       return -ENOMEM;
+}