tracing/filters: distinguish between signed and unsigned fields
[safe/jmp/linux-2.6] / kernel / trace / trace_events.c
index d132997..bbbea74 100644 (file)
@@ -8,10 +8,14 @@
  *
  */
 
+#include <linux/workqueue.h>
+#include <linux/spinlock.h>
+#include <linux/kthread.h>
 #include <linux/debugfs.h>
 #include <linux/uaccess.h>
 #include <linux/module.h>
 #include <linux/ctype.h>
+#include <linux/delay.h>
 
 #include "trace_output.h"
 
 
 static DEFINE_MUTEX(event_mutex);
 
+LIST_HEAD(ftrace_events);
+
 int trace_define_field(struct ftrace_event_call *call, char *type,
-                      char *name, int offset, int size)
+                      char *name, int offset, int size, int is_signed)
 {
        struct ftrace_event_field *field;
 
@@ -38,6 +44,7 @@ int trace_define_field(struct ftrace_event_call *call, char *type,
 
        field->offset = offset;
        field->size = size;
+       field->is_signed = is_signed;
        list_add(&field->link, &call->fields);
 
        return 0;
@@ -51,19 +58,18 @@ err:
 
        return -ENOMEM;
 }
+EXPORT_SYMBOL_GPL(trace_define_field);
 
 static void ftrace_clear_events(void)
 {
-       struct ftrace_event_call *call = (void *)__start_ftrace_events;
-
+       struct ftrace_event_call *call;
 
-       while ((unsigned long)call < (unsigned long)__stop_ftrace_events) {
+       list_for_each_entry(call, &ftrace_events, list) {
 
                if (call->enabled) {
                        call->enabled = 0;
                        call->unregfunc();
                }
-               call++;
        }
 }
 
@@ -89,7 +95,7 @@ static void ftrace_event_enable_disable(struct ftrace_event_call *call,
 
 static int ftrace_set_clr_event(char *buf, int set)
 {
-       struct ftrace_event_call *call = __start_ftrace_events;
+       struct ftrace_event_call *call;
        char *event = NULL, *sub = NULL, *match;
        int ret = -EINVAL;
 
@@ -118,7 +124,7 @@ static int ftrace_set_clr_event(char *buf, int set)
        }
 
        mutex_lock(&event_mutex);
-       for_each_event(call) {
+       list_for_each_entry(call, &ftrace_events, list) {
 
                if (!call->name || !call->regfunc)
                        continue;
@@ -224,15 +230,17 @@ ftrace_event_write(struct file *file, const char __user *ubuf,
 static void *
 t_next(struct seq_file *m, void *v, loff_t *pos)
 {
-       struct ftrace_event_call *call = m->private;
-       struct ftrace_event_call *next = call;
+       struct list_head *list = m->private;
+       struct ftrace_event_call *call;
 
        (*pos)++;
 
        for (;;) {
-               if ((unsigned long)call >= (unsigned long)__stop_ftrace_events)
+               if (list == &ftrace_events)
                        return NULL;
 
+               call = list_entry(list, struct ftrace_event_call, list);
+
                /*
                 * The ftrace subsystem is for showing formats only.
                 * They can not be enabled or disabled via the event files.
@@ -240,11 +248,10 @@ t_next(struct seq_file *m, void *v, loff_t *pos)
                if (call->regfunc)
                        break;
 
-               call++;
-               next = call;
+               list = list->next;
        }
 
-       m->private = ++next;
+       m->private = list->next;
 
        return call;
 }
@@ -257,22 +264,23 @@ static void *t_start(struct seq_file *m, loff_t *pos)
 static void *
 s_next(struct seq_file *m, void *v, loff_t *pos)
 {
-       struct ftrace_event_call *call = m->private;
-       struct ftrace_event_call *next;
+       struct list_head *list = m->private;
+       struct ftrace_event_call *call;
 
        (*pos)++;
 
  retry:
-       if ((unsigned long)call >= (unsigned long)__stop_ftrace_events)
+       if (list == &ftrace_events)
                return NULL;
 
+       call = list_entry(list, struct ftrace_event_call, list);
+
        if (!call->enabled) {
-               call++;
+               list = list->next;
                goto retry;
        }
 
-       next = call;
-       m->private = ++next;
+       m->private = list->next;
 
        return call;
 }
@@ -312,7 +320,7 @@ ftrace_event_seq_open(struct inode *inode, struct file *file)
        if (!ret) {
                struct seq_file *m = file->private_data;
 
-               m->private = __start_ftrace_events;
+               m->private = ftrace_events.next;
        }
        return ret;
 }
@@ -374,8 +382,11 @@ event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
        return cnt;
 }
 
+extern char *__bad_type_size(void);
+
 #undef FIELD
 #define FIELD(type, name)                                              \
+       sizeof(type) != sizeof(field.name) ? __bad_type_size() :        \
        #type, "common_" #name, offsetof(typeof(field), name),          \
                sizeof(field.name)
 
@@ -391,7 +402,7 @@ static int trace_write_header(struct trace_seq *s)
                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
                                "\n",
-                               FIELD(unsigned char, type),
+                               FIELD(unsigned short, type),
                                FIELD(unsigned char, flags),
                                FIELD(unsigned char, preempt_count),
                                FIELD(int, pid),
@@ -481,7 +492,7 @@ event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 
        trace_seq_init(s);
 
-       filter_print_preds(call->preds, s);
+       filter_print_preds(call, s);
        r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 
        kfree(s);
@@ -503,6 +514,7 @@ event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 
        if (copy_from_user(&buf, ubuf, cnt))
                return -EFAULT;
+       buf[cnt] = '\0';
 
        pred = kzalloc(sizeof(*pred), GFP_KERNEL);
        if (!pred)
@@ -515,16 +527,19 @@ event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
        }
 
        if (pred->clear) {
-               filter_free_preds(call);
+               filter_disable_preds(call);
                filter_free_pred(pred);
                return cnt;
        }
 
-       if (filter_add_pred(call, pred)) {
+       err = filter_add_pred(call, pred);
+       if (err < 0) {
                filter_free_pred(pred);
-               return -EINVAL;
+               return err;
        }
 
+       filter_free_pred(pred);
+
        *ppos += cnt;
 
        return cnt;
@@ -547,7 +562,7 @@ subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
 
        trace_seq_init(s);
 
-       filter_print_preds(system->preds, s);
+       filter_print_subsystem_preds(system, s);
        r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
 
        kfree(s);
@@ -569,6 +584,7 @@ subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
 
        if (copy_from_user(&buf, ubuf, cnt))
                return -EFAULT;
+       buf[cnt] = '\0';
 
        pred = kzalloc(sizeof(*pred), GFP_KERNEL);
        if (!pred)
@@ -586,10 +602,10 @@ subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
                return cnt;
        }
 
-       if (filter_add_subsystem_pred(system, pred)) {
-               filter_free_subsystem_preds(system);
+       err = filter_add_subsystem_pred(system, pred);
+       if (err < 0) {
                filter_free_pred(pred);
-               return -EINVAL;
+               return err;
        }
 
        *ppos += cnt;
@@ -597,6 +613,30 @@ subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
        return cnt;
 }
 
+static ssize_t
+show_header(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
+{
+       int (*func)(struct trace_seq *s) = filp->private_data;
+       struct trace_seq *s;
+       int r;
+
+       if (*ppos)
+               return 0;
+
+       s = kmalloc(sizeof(*s), GFP_KERNEL);
+       if (!s)
+               return -ENOMEM;
+
+       trace_seq_init(s);
+
+       func(s);
+       r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
+
+       kfree(s);
+
+       return r;
+}
+
 static const struct seq_operations show_event_seq_ops = {
        .start = t_start,
        .next = t_next,
@@ -654,6 +694,11 @@ static const struct file_operations ftrace_subsystem_filter_fops = {
        .write = subsystem_filter_write,
 };
 
+static const struct file_operations ftrace_show_header_fops = {
+       .open = tracing_open_generic,
+       .read = show_header,
+};
+
 static struct dentry *event_trace_events_dir(void)
 {
        static struct dentry *d_tracer;
@@ -704,10 +749,16 @@ event_subsystem_dir(const char *name, struct dentry *d_events)
                return d_events;
        }
 
-       system->name = name;
+       system->name = kstrdup(name, GFP_KERNEL);
+       if (!system->name) {
+               debugfs_remove(system->entry);
+               kfree(system);
+               return d_events;
+       }
+
        list_add(&system->list, &event_subsystems);
 
-       system->preds = NULL;
+       system->filter = NULL;
 
        entry = debugfs_create_file("filter", 0644, system->entry, system,
                                    &ftrace_subsystem_filter_fops);
@@ -719,7 +770,11 @@ event_subsystem_dir(const char *name, struct dentry *d_events)
 }
 
 static int
-event_create_dir(struct ftrace_event_call *call, struct dentry *d_events)
+event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
+                const struct file_operations *id,
+                const struct file_operations *enable,
+                const struct file_operations *filter,
+                const struct file_operations *format)
 {
        struct dentry *entry;
        int ret;
@@ -728,7 +783,7 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events)
         * If the trace point header did not define TRACE_SYSTEM
         * then the system would be called "TRACE_SYSTEM".
         */
-       if (strcmp(call->system, "TRACE_SYSTEM") != 0)
+       if (strcmp(call->system, TRACE_SYSTEM) != 0)
                d_events = event_subsystem_dir(call->system, d_events);
 
        if (call->raw_init) {
@@ -747,21 +802,13 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events)
                return -1;
        }
 
-       if (call->regfunc) {
-               entry = debugfs_create_file("enable", 0644, call->dir, call,
-                                           &ftrace_enable_fops);
-               if (!entry)
-                       pr_warning("Could not create debugfs "
-                                  "'%s/enable' entry\n", call->name);
-       }
+       if (call->regfunc)
+               entry = trace_create_file("enable", 0644, call->dir, call,
+                                         enable);
 
-       if (call->id) {
-               entry = debugfs_create_file("id", 0444, call->dir, call,
-                               &ftrace_event_id_fops);
-               if (!entry)
-                       pr_warning("Could not create debugfs '%s/id' entry\n",
-                                       call->name);
-       }
+       if (call->id)
+               entry = trace_create_file("id", 0444, call->dir, call,
+                                         id);
 
        if (call->define_fields) {
                ret = call->define_fields();
@@ -770,33 +817,185 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events)
                                   " events/%s\n", call->name);
                        return ret;
                }
+               entry = trace_create_file("filter", 0644, call->dir, call,
+                                         filter);
        }
 
-       entry = debugfs_create_file("filter", 0644, call->dir, call,
-                                   &ftrace_event_filter_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs "
-                          "'%s/filter' entry\n", call->name);
-
        /* A trace may not want to export its format */
        if (!call->show_format)
                return 0;
 
-       entry = debugfs_create_file("format", 0444, call->dir, call,
-                                   &ftrace_event_format_fops);
-       if (!entry)
-               pr_warning("Could not create debugfs "
-                          "'%s/format' entry\n", call->name);
+       entry = trace_create_file("format", 0444, call->dir, call,
+                                 format);
 
        return 0;
 }
 
+#define for_each_event(event, start, end)                      \
+       for (event = start;                                     \
+            (unsigned long)event < (unsigned long)end;         \
+            event++)
+
+#ifdef CONFIG_MODULES
+
+static LIST_HEAD(ftrace_module_file_list);
+
+/*
+ * Modules must own their file_operations to keep up with
+ * reference counting.
+ */
+struct ftrace_module_file_ops {
+       struct list_head                list;
+       struct module                   *mod;
+       struct file_operations          id;
+       struct file_operations          enable;
+       struct file_operations          format;
+       struct file_operations          filter;
+};
+
+static struct ftrace_module_file_ops *
+trace_create_file_ops(struct module *mod)
+{
+       struct ftrace_module_file_ops *file_ops;
+
+       /*
+        * This is a bit of a PITA. To allow for correct reference
+        * counting, modules must "own" their file_operations.
+        * To do this, we allocate the file operations that will be
+        * used in the event directory.
+        */
+
+       file_ops = kmalloc(sizeof(*file_ops), GFP_KERNEL);
+       if (!file_ops)
+               return NULL;
+
+       file_ops->mod = mod;
+
+       file_ops->id = ftrace_event_id_fops;
+       file_ops->id.owner = mod;
+
+       file_ops->enable = ftrace_enable_fops;
+       file_ops->enable.owner = mod;
+
+       file_ops->filter = ftrace_event_filter_fops;
+       file_ops->filter.owner = mod;
+
+       file_ops->format = ftrace_event_format_fops;
+       file_ops->format.owner = mod;
+
+       list_add(&file_ops->list, &ftrace_module_file_list);
+
+       return file_ops;
+}
+
+static void trace_module_add_events(struct module *mod)
+{
+       struct ftrace_module_file_ops *file_ops = NULL;
+       struct ftrace_event_call *call, *start, *end;
+       struct dentry *d_events;
+
+       start = mod->trace_events;
+       end = mod->trace_events + mod->num_trace_events;
+
+       if (start == end)
+               return;
+
+       d_events = event_trace_events_dir();
+       if (!d_events)
+               return;
+
+       for_each_event(call, start, end) {
+               /* The linker may leave blanks */
+               if (!call->name)
+                       continue;
+
+               /*
+                * This module has events, create file ops for this module
+                * if not already done.
+                */
+               if (!file_ops) {
+                       file_ops = trace_create_file_ops(mod);
+                       if (!file_ops)
+                               return;
+               }
+               call->mod = mod;
+               list_add(&call->list, &ftrace_events);
+               event_create_dir(call, d_events,
+                                &file_ops->id, &file_ops->enable,
+                                &file_ops->filter, &file_ops->format);
+       }
+}
+
+static void trace_module_remove_events(struct module *mod)
+{
+       struct ftrace_module_file_ops *file_ops;
+       struct ftrace_event_call *call, *p;
+
+       list_for_each_entry_safe(call, p, &ftrace_events, list) {
+               if (call->mod == mod) {
+                       if (call->enabled) {
+                               call->enabled = 0;
+                               call->unregfunc();
+                       }
+                       if (call->event)
+                               unregister_ftrace_event(call->event);
+                       debugfs_remove_recursive(call->dir);
+                       list_del(&call->list);
+               }
+       }
+
+       /* Now free the file_operations */
+       list_for_each_entry(file_ops, &ftrace_module_file_list, list) {
+               if (file_ops->mod == mod)
+                       break;
+       }
+       if (&file_ops->list != &ftrace_module_file_list) {
+               list_del(&file_ops->list);
+               kfree(file_ops);
+       }
+}
+
+static int trace_module_notify(struct notifier_block *self,
+                              unsigned long val, void *data)
+{
+       struct module *mod = data;
+
+       mutex_lock(&event_mutex);
+       switch (val) {
+       case MODULE_STATE_COMING:
+               trace_module_add_events(mod);
+               break;
+       case MODULE_STATE_GOING:
+               trace_module_remove_events(mod);
+               break;
+       }
+       mutex_unlock(&event_mutex);
+
+       return 0;
+}
+#else
+static int trace_module_notify(struct notifier_block *self,
+                              unsigned long val, void *data)
+{
+       return 0;
+}
+#endif /* CONFIG_MODULES */
+
+struct notifier_block trace_module_nb = {
+       .notifier_call = trace_module_notify,
+       .priority = 0,
+};
+
+extern struct ftrace_event_call __start_ftrace_events[];
+extern struct ftrace_event_call __stop_ftrace_events[];
+
 static __init int event_trace_init(void)
 {
-       struct ftrace_event_call *call = __start_ftrace_events;
+       struct ftrace_event_call *call;
        struct dentry *d_tracer;
        struct dentry *entry;
        struct dentry *d_events;
+       int ret;
 
        d_tracer = tracing_init_dentry();
        if (!d_tracer)
@@ -820,13 +1019,269 @@ static __init int event_trace_init(void)
        if (!d_events)
                return 0;
 
-       for_each_event(call) {
+       /* ring buffer internal formats */
+       trace_create_file("header_page", 0444, d_events,
+                         ring_buffer_print_page_header,
+                         &ftrace_show_header_fops);
+
+       trace_create_file("header_event", 0444, d_events,
+                         ring_buffer_print_entry_header,
+                         &ftrace_show_header_fops);
+
+       for_each_event(call, __start_ftrace_events, __stop_ftrace_events) {
                /* The linker may leave blanks */
                if (!call->name)
                        continue;
-               event_create_dir(call, d_events);
+               list_add(&call->list, &ftrace_events);
+               event_create_dir(call, d_events, &ftrace_event_id_fops,
+                                &ftrace_enable_fops, &ftrace_event_filter_fops,
+                                &ftrace_event_format_fops);
        }
 
+       ret = register_module_notifier(&trace_module_nb);
+       if (!ret)
+               pr_warning("Failed to register trace events module notifier\n");
+
        return 0;
 }
 fs_initcall(event_trace_init);
+
+#ifdef CONFIG_FTRACE_STARTUP_TEST
+
+static DEFINE_SPINLOCK(test_spinlock);
+static DEFINE_SPINLOCK(test_spinlock_irq);
+static DEFINE_MUTEX(test_mutex);
+
+static __init void test_work(struct work_struct *dummy)
+{
+       spin_lock(&test_spinlock);
+       spin_lock_irq(&test_spinlock_irq);
+       udelay(1);
+       spin_unlock_irq(&test_spinlock_irq);
+       spin_unlock(&test_spinlock);
+
+       mutex_lock(&test_mutex);
+       msleep(1);
+       mutex_unlock(&test_mutex);
+}
+
+static __init int event_test_thread(void *unused)
+{
+       void *test_malloc;
+
+       test_malloc = kmalloc(1234, GFP_KERNEL);
+       if (!test_malloc)
+               pr_info("failed to kmalloc\n");
+
+       schedule_on_each_cpu(test_work);
+
+       kfree(test_malloc);
+
+       set_current_state(TASK_INTERRUPTIBLE);
+       while (!kthread_should_stop())
+               schedule();
+
+       return 0;
+}
+
+/*
+ * Do various things that may trigger events.
+ */
+static __init void event_test_stuff(void)
+{
+       struct task_struct *test_thread;
+
+       test_thread = kthread_run(event_test_thread, NULL, "test-events");
+       msleep(1);
+       kthread_stop(test_thread);
+}
+
+/*
+ * For every trace event defined, we will test each trace point separately,
+ * and then by groups, and finally all trace points.
+ */
+static __init void event_trace_self_tests(void)
+{
+       struct ftrace_event_call *call;
+       struct event_subsystem *system;
+       char *sysname;
+       int ret;
+
+       pr_info("Running tests on trace events:\n");
+
+       list_for_each_entry(call, &ftrace_events, list) {
+
+               /* Only test those that have a regfunc */
+               if (!call->regfunc)
+                       continue;
+
+               pr_info("Testing event %s: ", call->name);
+
+               /*
+                * If an event is already enabled, someone is using
+                * it and the self test should not be on.
+                */
+               if (call->enabled) {
+                       pr_warning("Enabled event during self test!\n");
+                       WARN_ON_ONCE(1);
+                       continue;
+               }
+
+               call->enabled = 1;
+               call->regfunc();
+
+               event_test_stuff();
+
+               call->unregfunc();
+               call->enabled = 0;
+
+               pr_cont("OK\n");
+       }
+
+       /* Now test at the sub system level */
+
+       pr_info("Running tests on trace event systems:\n");
+
+       list_for_each_entry(system, &event_subsystems, list) {
+
+               /* the ftrace system is special, skip it */
+               if (strcmp(system->name, "ftrace") == 0)
+                       continue;
+
+               pr_info("Testing event system %s: ", system->name);
+
+               /* ftrace_set_clr_event can modify the name passed in. */
+               sysname = kstrdup(system->name, GFP_KERNEL);
+               if (WARN_ON(!sysname)) {
+                       pr_warning("Can't allocate memory, giving up!\n");
+                       return;
+               }
+               ret = ftrace_set_clr_event(sysname, 1);
+               kfree(sysname);
+               if (WARN_ON_ONCE(ret)) {
+                       pr_warning("error enabling system %s\n",
+                                  system->name);
+                       continue;
+               }
+
+               event_test_stuff();
+
+               sysname = kstrdup(system->name, GFP_KERNEL);
+               if (WARN_ON(!sysname)) {
+                       pr_warning("Can't allocate memory, giving up!\n");
+                       return;
+               }
+               ret = ftrace_set_clr_event(sysname, 0);
+               kfree(sysname);
+
+               if (WARN_ON_ONCE(ret))
+                       pr_warning("error disabling system %s\n",
+                                  system->name);
+
+               pr_cont("OK\n");
+       }
+
+       /* Test with all events enabled */
+
+       pr_info("Running tests on all trace events:\n");
+       pr_info("Testing all events: ");
+
+       sysname = kmalloc(4, GFP_KERNEL);
+       if (WARN_ON(!sysname)) {
+               pr_warning("Can't allocate memory, giving up!\n");
+               return;
+       }
+       memcpy(sysname, "*:*", 4);
+       ret = ftrace_set_clr_event(sysname, 1);
+       if (WARN_ON_ONCE(ret)) {
+               kfree(sysname);
+               pr_warning("error enabling all events\n");
+               return;
+       }
+
+       event_test_stuff();
+
+       /* reset sysname */
+       memcpy(sysname, "*:*", 4);
+       ret = ftrace_set_clr_event(sysname, 0);
+       kfree(sysname);
+
+       if (WARN_ON_ONCE(ret)) {
+               pr_warning("error disabling all events\n");
+               return;
+       }
+
+       pr_cont("OK\n");
+}
+
+#ifdef CONFIG_FUNCTION_TRACER
+
+static DEFINE_PER_CPU(atomic_t, test_event_disable);
+
+static void
+function_test_events_call(unsigned long ip, unsigned long parent_ip)
+{
+       struct ring_buffer_event *event;
+       struct ftrace_entry *entry;
+       unsigned long flags;
+       long disabled;
+       int resched;
+       int cpu;
+       int pc;
+
+       pc = preempt_count();
+       resched = ftrace_preempt_disable();
+       cpu = raw_smp_processor_id();
+       disabled = atomic_inc_return(&per_cpu(test_event_disable, cpu));
+
+       if (disabled != 1)
+               goto out;
+
+       local_save_flags(flags);
+
+       event = trace_current_buffer_lock_reserve(TRACE_FN, sizeof(*entry),
+                                                 flags, pc);
+       if (!event)
+               goto out;
+       entry   = ring_buffer_event_data(event);
+       entry->ip                       = ip;
+       entry->parent_ip                = parent_ip;
+
+       trace_nowake_buffer_unlock_commit(event, flags, pc);
+
+ out:
+       atomic_dec(&per_cpu(test_event_disable, cpu));
+       ftrace_preempt_enable(resched);
+}
+
+static struct ftrace_ops trace_ops __initdata  =
+{
+       .func = function_test_events_call,
+};
+
+static __init void event_trace_self_test_with_function(void)
+{
+       register_ftrace_function(&trace_ops);
+       pr_info("Running tests again, along with the function tracer\n");
+       event_trace_self_tests();
+       unregister_ftrace_function(&trace_ops);
+}
+#else
+static __init void event_trace_self_test_with_function(void)
+{
+}
+#endif
+
+static __init int event_trace_self_tests_init(void)
+{
+
+       event_trace_self_tests();
+
+       event_trace_self_test_with_function();
+
+       return 0;
+}
+
+late_initcall(event_trace_self_tests_init);
+
+#endif