tracing/kprobes: Use probe_kernel_read instead of probe_mem_read
[sfrench/cifs-2.6.git] / kernel / trace / trace_kprobe.c
index adc153ab51c0e5bc26e1cbf32d8563c02f70a982..99592c27465e1156a0dce1dd2d953aef9ef3e8c8 100644 (file)
@@ -12,6 +12,7 @@
 #include <linux/rculist.h>
 #include <linux/error-injection.h>
 
+#include "trace_dynevent.h"
 #include "trace_kprobe_selftest.h"
 #include "trace_probe.h"
 #include "trace_probe_tmpl.h"
 #define KPROBE_EVENT_SYSTEM "kprobes"
 #define KRETPROBE_MAXACTIVE_MAX 4096
 
+static int trace_kprobe_create(int argc, const char **argv);
+static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
+static int trace_kprobe_release(struct dyn_event *ev);
+static bool trace_kprobe_is_busy(struct dyn_event *ev);
+static bool trace_kprobe_match(const char *system, const char *event,
+                              struct dyn_event *ev);
+
+static struct dyn_event_operations trace_kprobe_ops = {
+       .create = trace_kprobe_create,
+       .show = trace_kprobe_show,
+       .is_busy = trace_kprobe_is_busy,
+       .free = trace_kprobe_release,
+       .match = trace_kprobe_match,
+};
+
 /**
  * Kprobe event core functions
  */
 struct trace_kprobe {
-       struct list_head        list;
+       struct dyn_event        devent;
        struct kretprobe        rp;     /* Use rp.kp for kprobe use */
        unsigned long __percpu *nhit;
        const char              *symbol;        /* symbol name */
        struct trace_probe      tp;
 };
 
+static bool is_trace_kprobe(struct dyn_event *ev)
+{
+       return ev->ops == &trace_kprobe_ops;
+}
+
+static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
+{
+       return container_of(ev, struct trace_kprobe, devent);
+}
+
+/**
+ * for_each_trace_kprobe - iterate over the trace_kprobe list
+ * @pos:       the struct trace_kprobe * for each entry
+ * @dpos:      the struct dyn_event * to use as a loop cursor
+ */
+#define for_each_trace_kprobe(pos, dpos)       \
+       for_each_dyn_event(dpos)                \
+               if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
+
 #define SIZEOF_TRACE_KPROBE(n)                         \
        (offsetof(struct trace_kprobe, tp.args) +       \
        (sizeof(struct probe_arg) * (n)))
@@ -81,6 +116,22 @@ static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
        return ret;
 }
 
+static bool trace_kprobe_is_busy(struct dyn_event *ev)
+{
+       struct trace_kprobe *tk = to_trace_kprobe(ev);
+
+       return trace_probe_is_enabled(&tk->tp);
+}
+
+static bool trace_kprobe_match(const char *system, const char *event,
+                              struct dyn_event *ev)
+{
+       struct trace_kprobe *tk = to_trace_kprobe(ev);
+
+       return strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
+           (!system || strcmp(tk->tp.call.class->system, system) == 0);
+}
+
 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
 {
        unsigned long nhit = 0;
@@ -128,9 +179,6 @@ bool trace_kprobe_error_injectable(struct trace_event_call *call)
 static int register_kprobe_event(struct trace_kprobe *tk);
 static int unregister_kprobe_event(struct trace_kprobe *tk);
 
-static DEFINE_MUTEX(probe_lock);
-static LIST_HEAD(probe_list);
-
 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
                                struct pt_regs *regs);
@@ -192,7 +240,7 @@ static struct trace_kprobe *alloc_trace_kprobe(const char *group,
        if (!tk->tp.class.system)
                goto error;
 
-       INIT_LIST_HEAD(&tk->list);
+       dyn_event_init(&tk->devent, &trace_kprobe_ops);
        INIT_LIST_HEAD(&tk->tp.files);
        return tk;
 error:
@@ -207,6 +255,9 @@ static void free_trace_kprobe(struct trace_kprobe *tk)
 {
        int i;
 
+       if (!tk)
+               return;
+
        for (i = 0; i < tk->tp.nr_args; i++)
                traceprobe_free_probe_arg(&tk->tp.args[i]);
 
@@ -220,9 +271,10 @@ static void free_trace_kprobe(struct trace_kprobe *tk)
 static struct trace_kprobe *find_trace_kprobe(const char *event,
                                              const char *group)
 {
+       struct dyn_event *pos;
        struct trace_kprobe *tk;
 
-       list_for_each_entry(tk, &probe_list, list)
+       for_each_trace_kprobe(tk, pos)
                if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
                    strcmp(tk->tp.call.class->system, group) == 0)
                        return tk;
@@ -321,7 +373,7 @@ disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
         * created with perf_event_open. We don't need to wait for these
         * trace_kprobes
         */
-       if (list_empty(&tk->list))
+       if (list_empty(&tk->devent.list))
                wait = 0;
  out:
        if (wait) {
@@ -419,7 +471,7 @@ static void __unregister_trace_kprobe(struct trace_kprobe *tk)
        }
 }
 
-/* Unregister a trace_probe and probe_event: call with locking probe_lock */
+/* Unregister a trace_probe and probe_event */
 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 {
        /* Enabled event can not be unregistered */
@@ -431,7 +483,7 @@ static int unregister_trace_kprobe(struct trace_kprobe *tk)
                return -EBUSY;
 
        __unregister_trace_kprobe(tk);
-       list_del(&tk->list);
+       dyn_event_remove(&tk->devent);
 
        return 0;
 }
@@ -442,7 +494,7 @@ static int register_trace_kprobe(struct trace_kprobe *tk)
        struct trace_kprobe *old_tk;
        int ret;
 
-       mutex_lock(&probe_lock);
+       mutex_lock(&event_mutex);
 
        /* Delete old (same name) event if exist */
        old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
@@ -471,10 +523,10 @@ static int register_trace_kprobe(struct trace_kprobe *tk)
        if (ret < 0)
                unregister_kprobe_event(tk);
        else
-               list_add_tail(&tk->list, &probe_list);
+               dyn_event_add(&tk->devent);
 
 end:
-       mutex_unlock(&probe_lock);
+       mutex_unlock(&event_mutex);
        return ret;
 }
 
@@ -483,6 +535,7 @@ static int trace_kprobe_module_callback(struct notifier_block *nb,
                                       unsigned long val, void *data)
 {
        struct module *mod = data;
+       struct dyn_event *pos;
        struct trace_kprobe *tk;
        int ret;
 
@@ -490,8 +543,8 @@ static int trace_kprobe_module_callback(struct notifier_block *nb,
                return NOTIFY_DONE;
 
        /* Update probes on coming module */
-       mutex_lock(&probe_lock);
-       list_for_each_entry(tk, &probe_list, list) {
+       mutex_lock(&event_mutex);
+       for_each_trace_kprobe(tk, pos) {
                if (trace_kprobe_within_module(tk, mod)) {
                        /* Don't need to check busy - this should have gone. */
                        __unregister_trace_kprobe(tk);
@@ -502,7 +555,7 @@ static int trace_kprobe_module_callback(struct notifier_block *nb,
                                        mod->name, ret);
                }
        }
-       mutex_unlock(&probe_lock);
+       mutex_unlock(&event_mutex);
 
        return NOTIFY_DONE;
 }
@@ -520,7 +573,7 @@ static inline void sanitize_event_name(char *name)
                        *name = '_';
 }
 
-static int create_trace_kprobe(int argc, char **argv)
+static int trace_kprobe_create(int argc, const char *argv[])
 {
        /*
         * Argument syntax:
@@ -544,37 +597,43 @@ static int create_trace_kprobe(int argc, char **argv)
         *  FETCHARG:TYPE : use TYPE instead of unsigned long.
         */
        struct trace_kprobe *tk;
-       int i, ret = 0;
-       bool is_return = false, is_delete = false;
-       char *symbol = NULL, *event = NULL, *group = NULL;
+       int i, len, ret = 0;
+       bool is_return = false;
+       char *symbol = NULL, *tmp = NULL;
+       const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
        int maxactive = 0;
-       char *arg;
        long offset = 0;
        void *addr = NULL;
        char buf[MAX_EVENT_NAME_LEN];
        unsigned int flags = TPARG_FL_KERNEL;
 
-       /* argc must be >= 1 */
-       if (argv[0][0] == 'p')
-               is_return = false;
-       else if (argv[0][0] == 'r') {
+       switch (argv[0][0]) {
+       case 'r':
                is_return = true;
                flags |= TPARG_FL_RETURN;
-       } else if (argv[0][0] == '-')
-               is_delete = true;
-       else {
-               pr_info("Probe definition must be started with 'p', 'r' or"
-                       " '-'.\n");
-               return -EINVAL;
+               break;
+       case 'p':
+               break;
+       default:
+               return -ECANCELED;
        }
+       if (argc < 2)
+               return -ECANCELED;
 
        event = strchr(&argv[0][1], ':');
-       if (event) {
-               event[0] = '\0';
+       if (event)
                event++;
-       }
+
        if (is_return && isdigit(argv[0][1])) {
-               ret = kstrtouint(&argv[0][1], 0, &maxactive);
+               if (event)
+                       len = event - &argv[0][1] - 1;
+               else
+                       len = strlen(&argv[0][1]);
+               if (len > MAX_EVENT_NAME_LEN - 1)
+                       return -E2BIG;
+               memcpy(buf, &argv[0][1], len);
+               buf[len] = '\0';
+               ret = kstrtouint(buf, 0, &maxactive);
                if (ret) {
                        pr_info("Failed to parse maxactive.\n");
                        return ret;
@@ -589,74 +648,37 @@ static int create_trace_kprobe(int argc, char **argv)
                }
        }
 
-       if (event) {
-               char *slash;
-
-               slash = strchr(event, '/');
-               if (slash) {
-                       group = event;
-                       event = slash + 1;
-                       slash[0] = '\0';
-                       if (strlen(group) == 0) {
-                               pr_info("Group name is not specified\n");
-                               return -EINVAL;
-                       }
-               }
-               if (strlen(event) == 0) {
-                       pr_info("Event name is not specified\n");
-                       return -EINVAL;
-               }
-       }
-       if (!group)
-               group = KPROBE_EVENT_SYSTEM;
-
-       if (is_delete) {
-               if (!event) {
-                       pr_info("Delete command needs an event name.\n");
-                       return -EINVAL;
-               }
-               mutex_lock(&probe_lock);
-               tk = find_trace_kprobe(event, group);
-               if (!tk) {
-                       mutex_unlock(&probe_lock);
-                       pr_info("Event %s/%s doesn't exist.\n", group, event);
-                       return -ENOENT;
-               }
-               /* delete an event */
-               ret = unregister_trace_kprobe(tk);
-               if (ret == 0)
-                       free_trace_kprobe(tk);
-               mutex_unlock(&probe_lock);
-               return ret;
-       }
-
-       if (argc < 2) {
-               pr_info("Probe point is not specified.\n");
-               return -EINVAL;
-       }
-
        /* try to parse an address. if that fails, try to read the
         * input as a symbol. */
        if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
+               /* Check whether uprobe event specified */
+               if (strchr(argv[1], '/') && strchr(argv[1], ':'))
+                       return -ECANCELED;
                /* a symbol specified */
-               symbol = argv[1];
+               symbol = kstrdup(argv[1], GFP_KERNEL);
+               if (!symbol)
+                       return -ENOMEM;
                /* TODO: support .init module functions */
                ret = traceprobe_split_symbol_offset(symbol, &offset);
                if (ret || offset < 0 || offset > UINT_MAX) {
                        pr_info("Failed to parse either an address or a symbol.\n");
-                       return ret;
+                       goto out;
                }
                if (kprobe_on_func_entry(NULL, symbol, offset))
                        flags |= TPARG_FL_FENTRY;
                if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
                        pr_info("Given offset is not valid for return probe.\n");
-                       return -EINVAL;
+                       ret = -EINVAL;
+                       goto out;
                }
        }
        argc -= 2; argv += 2;
 
-       /* setup a probe */
-       if (!event) {
+       if (event) {
+               ret = traceprobe_parse_event_name(&event, &group, buf);
+               if (ret)
+                       goto out;
+       } else {
                /* Make a new event name */
                if (symbol)
                        snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
@@ -667,121 +689,67 @@ static int create_trace_kprobe(int argc, char **argv)
                sanitize_event_name(buf);
                event = buf;
        }
+
+       /* setup a probe */
        tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
                               argc, is_return);
        if (IS_ERR(tk)) {
                pr_info("Failed to allocate trace_probe.(%d)\n",
                        (int)PTR_ERR(tk));
-               return PTR_ERR(tk);
+               ret = PTR_ERR(tk);
+               goto out;
        }
 
        /* parse arguments */
-       ret = 0;
        for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
-               struct probe_arg *parg = &tk->tp.args[i];
-
-               /* Increment count for freeing args in error case */
-               tk->tp.nr_args++;
-
-               /* Parse argument name */
-               arg = strchr(argv[i], '=');
-               if (arg) {
-                       *arg++ = '\0';
-                       parg->name = kstrdup(argv[i], GFP_KERNEL);
-               } else {
-                       arg = argv[i];
-                       /* If argument name is omitted, set "argN" */
-                       snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
-                       parg->name = kstrdup(buf, GFP_KERNEL);
-               }
-
-               if (!parg->name) {
-                       pr_info("Failed to allocate argument[%d] name.\n", i);
+               tmp = kstrdup(argv[i], GFP_KERNEL);
+               if (!tmp) {
                        ret = -ENOMEM;
                        goto error;
                }
 
-               if (!is_good_name(parg->name)) {
-                       pr_info("Invalid argument[%d] name: %s\n",
-                               i, parg->name);
-                       ret = -EINVAL;
-                       goto error;
-               }
-
-               if (traceprobe_conflict_field_name(parg->name,
-                                                       tk->tp.args, i)) {
-                       pr_info("Argument[%d] name '%s' conflicts with "
-                               "another field.\n", i, argv[i]);
-                       ret = -EINVAL;
-                       goto error;
-               }
-
-               /* Parse fetch argument */
-               ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
-                                                flags);
-               if (ret) {
-                       pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
+               ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
+               kfree(tmp);
+               if (ret)
                        goto error;
-               }
        }
 
        ret = register_trace_kprobe(tk);
        if (ret)
                goto error;
-       return 0;
+out:
+       kfree(symbol);
+       return ret;
 
 error:
        free_trace_kprobe(tk);
-       return ret;
+       goto out;
 }
 
-static int release_all_trace_kprobes(void)
+static int create_or_delete_trace_kprobe(int argc, char **argv)
 {
-       struct trace_kprobe *tk;
-       int ret = 0;
-
-       mutex_lock(&probe_lock);
-       /* Ensure no probe is in use. */
-       list_for_each_entry(tk, &probe_list, list)
-               if (trace_probe_is_enabled(&tk->tp)) {
-                       ret = -EBUSY;
-                       goto end;
-               }
-       /* TODO: Use batch unregistration */
-       while (!list_empty(&probe_list)) {
-               tk = list_entry(probe_list.next, struct trace_kprobe, list);
-               ret = unregister_trace_kprobe(tk);
-               if (ret)
-                       goto end;
-               free_trace_kprobe(tk);
-       }
-
-end:
-       mutex_unlock(&probe_lock);
+       int ret;
 
-       return ret;
-}
+       if (argv[0][0] == '-')
+               return dyn_event_release(argc, argv, &trace_kprobe_ops);
 
-/* Probes listing interfaces */
-static void *probes_seq_start(struct seq_file *m, loff_t *pos)
-{
-       mutex_lock(&probe_lock);
-       return seq_list_start(&probe_list, *pos);
+       ret = trace_kprobe_create(argc, (const char **)argv);
+       return ret == -ECANCELED ? -EINVAL : ret;
 }
 
-static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
+static int trace_kprobe_release(struct dyn_event *ev)
 {
-       return seq_list_next(v, &probe_list, pos);
-}
+       struct trace_kprobe *tk = to_trace_kprobe(ev);
+       int ret = unregister_trace_kprobe(tk);
 
-static void probes_seq_stop(struct seq_file *m, void *v)
-{
-       mutex_unlock(&probe_lock);
+       if (!ret)
+               free_trace_kprobe(tk);
+       return ret;
 }
 
-static int probes_seq_show(struct seq_file *m, void *v)
+static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
 {
-       struct trace_kprobe *tk = v;
+       struct trace_kprobe *tk = to_trace_kprobe(ev);
        int i;
 
        seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
@@ -803,10 +771,20 @@ static int probes_seq_show(struct seq_file *m, void *v)
        return 0;
 }
 
+static int probes_seq_show(struct seq_file *m, void *v)
+{
+       struct dyn_event *ev = v;
+
+       if (!is_trace_kprobe(ev))
+               return 0;
+
+       return trace_kprobe_show(m, ev);
+}
+
 static const struct seq_operations probes_seq_op = {
-       .start  = probes_seq_start,
-       .next   = probes_seq_next,
-       .stop   = probes_seq_stop,
+       .start  = dyn_event_seq_start,
+       .next   = dyn_event_seq_next,
+       .stop   = dyn_event_seq_stop,
        .show   = probes_seq_show
 };
 
@@ -815,7 +793,7 @@ static int probes_open(struct inode *inode, struct file *file)
        int ret;
 
        if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
-               ret = release_all_trace_kprobes();
+               ret = dyn_events_release_all(&trace_kprobe_ops);
                if (ret < 0)
                        return ret;
        }
@@ -827,7 +805,7 @@ static ssize_t probes_write(struct file *file, const char __user *buffer,
                            size_t count, loff_t *ppos)
 {
        return trace_parse_run_command(file, buffer, count, ppos,
-                                      create_trace_kprobe);
+                                      create_or_delete_trace_kprobe);
 }
 
 static const struct file_operations kprobe_events_ops = {
@@ -842,8 +820,13 @@ static const struct file_operations kprobe_events_ops = {
 /* Probes profiling interfaces */
 static int probes_profile_seq_show(struct seq_file *m, void *v)
 {
-       struct trace_kprobe *tk = v;
+       struct dyn_event *ev = v;
+       struct trace_kprobe *tk;
 
+       if (!is_trace_kprobe(ev))
+               return 0;
+
+       tk = to_trace_kprobe(ev);
        seq_printf(m, "  %-44s %15lu %15lu\n",
                   trace_event_name(&tk->tp.call),
                   trace_kprobe_nhit(tk),
@@ -853,9 +836,9 @@ static int probes_profile_seq_show(struct seq_file *m, void *v)
 }
 
 static const struct seq_operations profile_seq_op = {
-       .start  = probes_seq_start,
-       .next   = probes_seq_next,
-       .stop   = probes_seq_stop,
+       .start  = dyn_event_seq_start,
+       .next   = dyn_event_seq_next,
+       .stop   = dyn_event_seq_stop,
        .show   = probes_profile_seq_show
 };
 
@@ -878,22 +861,14 @@ static const struct file_operations kprobe_profile_ops = {
 static nokprobe_inline int
 fetch_store_strlen(unsigned long addr)
 {
-       mm_segment_t old_fs;
        int ret, len = 0;
        u8 c;
 
-       old_fs = get_fs();
-       set_fs(KERNEL_DS);
-       pagefault_disable();
-
        do {
-               ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
+               ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
                len++;
        } while (c && ret == 0 && len < MAX_STRING_SIZE);
 
-       pagefault_enable();
-       set_fs(old_fs);
-
        return (ret < 0) ? ret : len;
 }
 
@@ -1408,7 +1383,7 @@ create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
        char *event;
 
        /*
-        * local trace_kprobes are not added to probe_list, so they are never
+        * local trace_kprobes are not added to dyn_event, so they are never
         * searched in find_trace_kprobe(). Therefore, there is no concern of
         * duplicated name here.
         */
@@ -1466,6 +1441,11 @@ static __init int init_kprobe_trace(void)
 {
        struct dentry *d_tracer;
        struct dentry *entry;
+       int ret;
+
+       ret = dyn_event_register(&trace_kprobe_ops);
+       if (ret)
+               return ret;
 
        if (register_module_notifier(&trace_kprobe_module_nb))
                return -EINVAL;
@@ -1523,9 +1503,8 @@ static __init int kprobe_trace_self_tests_init(void)
 
        pr_info("Testing kprobe tracing: ");
 
-       ret = trace_run_command("p:testprobe kprobe_trace_selftest_target "
-                               "$stack $stack0 +0($stack)",
-                               create_trace_kprobe);
+       ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
+                               create_or_delete_trace_kprobe);
        if (WARN_ON_ONCE(ret)) {
                pr_warn("error on probing function entry.\n");
                warn++;
@@ -1545,8 +1524,8 @@ static __init int kprobe_trace_self_tests_init(void)
                }
        }
 
-       ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target "
-                               "$retval", create_trace_kprobe);
+       ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
+                               create_or_delete_trace_kprobe);
        if (WARN_ON_ONCE(ret)) {
                pr_warn("error on probing function return.\n");
                warn++;
@@ -1616,20 +1595,24 @@ static __init int kprobe_trace_self_tests_init(void)
                        disable_trace_kprobe(tk, file);
        }
 
-       ret = trace_run_command("-:testprobe", create_trace_kprobe);
+       ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
        if (WARN_ON_ONCE(ret)) {
                pr_warn("error on deleting a probe.\n");
                warn++;
        }
 
-       ret = trace_run_command("-:testprobe2", create_trace_kprobe);
+       ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
        if (WARN_ON_ONCE(ret)) {
                pr_warn("error on deleting a probe.\n");
                warn++;
        }
 
 end:
-       release_all_trace_kprobes();
+       ret = dyn_events_release_all(&trace_kprobe_ops);
+       if (WARN_ON_ONCE(ret)) {
+               pr_warn("error on cleaning up probes.\n");
+               warn++;
+       }
        /*
         * Wait for the optimizer work to finish. Otherwise it might fiddle
         * with probes in already freed __init text.