ust: initial commit
authorPierre-Marc Fournier <pierre-marc.fournier@polymtl.ca>
Fri, 6 Feb 2009 21:31:05 +0000 (16:31 -0500)
committerPierre-Marc Fournier <pierre-marc.fournier@polymtl.ca>
Fri, 6 Feb 2009 21:31:05 +0000 (16:31 -0500)
hello/Makefile [new file with mode: 0644]
hello/hello.c [new file with mode: 0644]
libmarkers/marker.c [new file with mode: 0644]
libtracectl/Makefile [new file with mode: 0644]
libtracectl/tracectl.c [new file with mode: 0644]

diff --git a/hello/Makefile b/hello/Makefile
new file mode 100644 (file)
index 0000000..8a95626
--- /dev/null
@@ -0,0 +1,7 @@
+all: hello
+
+hello: hello.c
+       gcc -g -L../libtracectl -ltracectl -o hello hello.c
+
+clean:
+       rm -rf hello *.o
diff --git a/hello/hello.c b/hello/hello.c
new file mode 100644 (file)
index 0000000..990a7fa
--- /dev/null
@@ -0,0 +1,10 @@
+#include <stdio.h>
+#include <unistd.h>
+
+int main()
+{
+       printf("Hello, World!\n");
+       scanf("%*s");
+
+       return 0;
+}
diff --git a/libmarkers/marker.c b/libmarkers/marker.c
new file mode 100644 (file)
index 0000000..3820622
--- /dev/null
@@ -0,0 +1,1406 @@
+/*
+ * Copyright (C) 2007 Mathieu Desnoyers
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ */
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/types.h>
+#include <linux/jhash.h>
+#include <linux/list.h>
+#include <linux/rcupdate.h>
+#include <linux/marker.h>
+#include <linux/err.h>
+#include <linux/slab.h>
+#include <linux/immediate.h>
+#include <linux/sched.h>
+#include <linux/uaccess.h>
+#include <linux/user_marker.h>
+#include <linux/ltt-tracer.h>
+
+extern struct marker __start___markers[];
+extern struct marker __stop___markers[];
+
+/* Set to 1 to enable marker debug output */
+static const int marker_debug;
+
+/*
+ * markers_mutex nests inside module_mutex. Markers mutex protects the builtin
+ * and module markers and the hash table.
+ */
+static DEFINE_MUTEX(markers_mutex);
+
+void lock_markers(void)
+{
+       mutex_lock(&markers_mutex);
+}
+
+void unlock_markers(void)
+{
+       mutex_unlock(&markers_mutex);
+}
+
+/*
+ * Marker hash table, containing the active markers.
+ * Protected by module_mutex.
+ */
+#define MARKER_HASH_BITS 6
+#define MARKER_TABLE_SIZE (1 << MARKER_HASH_BITS)
+static struct hlist_head marker_table[MARKER_TABLE_SIZE];
+
+/*
+ * Note about RCU :
+ * It is used to make sure every handler has finished using its private data
+ * between two consecutive operation (add or remove) on a given marker.  It is
+ * also used to delay the free of multiple probes array until a quiescent state
+ * is reached.
+ * marker entries modifications are protected by the markers_mutex.
+ */
+struct marker_entry {
+       struct hlist_node hlist;
+       char *format;
+       char *name;
+                       /* Probe wrapper */
+       void (*call)(const struct marker *mdata, void *call_private, ...);
+       struct marker_probe_closure single;
+       struct marker_probe_closure *multi;
+       int refcount;   /* Number of times armed. 0 if disarmed. */
+       struct rcu_head rcu;
+       void *oldptr;
+       int rcu_pending;
+       u16 channel_id;
+       u16 event_id;
+       unsigned char ptype:1;
+       unsigned char format_allocated:1;
+       char channel[0];        /* Contains channel'\0'name'\0'format'\0' */
+};
+
+#ifdef CONFIG_MARKERS_USERSPACE
+static void marker_update_processes(void);
+#else
+static void marker_update_processes(void)
+{
+}
+#endif
+
+/**
+ * __mark_empty_function - Empty probe callback
+ * @mdata: marker data
+ * @probe_private: probe private data
+ * @call_private: call site private data
+ * @fmt: format string
+ * @...: variable argument list
+ *
+ * Empty callback provided as a probe to the markers. By providing this to a
+ * disabled marker, we make sure the  execution flow is always valid even
+ * though the function pointer change and the marker enabling are two distinct
+ * operations that modifies the execution flow of preemptible code.
+ */
+notrace void __mark_empty_function(const struct marker *mdata,
+       void *probe_private, void *call_private, const char *fmt, va_list *args)
+{
+}
+EXPORT_SYMBOL_GPL(__mark_empty_function);
+
+/*
+ * marker_probe_cb Callback that prepares the variable argument list for probes.
+ * @mdata: pointer of type struct marker
+ * @call_private: caller site private data
+ * @...:  Variable argument list.
+ *
+ * Since we do not use "typical" pointer based RCU in the 1 argument case, we
+ * need to put a full smp_rmb() in this branch. This is why we do not use
+ * rcu_dereference() for the pointer read.
+ */
+notrace void marker_probe_cb(const struct marker *mdata,
+               void *call_private, ...)
+{
+       va_list args;
+       char ptype;
+
+       /*
+        * rcu_read_lock_sched does two things : disabling preemption to make
+        * sure the teardown of the callbacks can be done correctly when they
+        * are in modules and they insure RCU read coherency.
+        */
+       rcu_read_lock_sched_notrace();
+       ptype = mdata->ptype;
+       if (likely(!ptype)) {
+               marker_probe_func *func;
+               /* Must read the ptype before ptr. They are not data dependant,
+                * so we put an explicit smp_rmb() here. */
+               smp_rmb();
+               func = mdata->single.func;
+               /* Must read the ptr before private data. They are not data
+                * dependant, so we put an explicit smp_rmb() here. */
+               smp_rmb();
+               va_start(args, call_private);
+               func(mdata, mdata->single.probe_private, call_private,
+                       mdata->format, &args);
+               va_end(args);
+       } else {
+               struct marker_probe_closure *multi;
+               int i;
+               /*
+                * Read mdata->ptype before mdata->multi.
+                */
+               smp_rmb();
+               multi = mdata->multi;
+               /*
+                * multi points to an array, therefore accessing the array
+                * depends on reading multi. However, even in this case,
+                * we must insure that the pointer is read _before_ the array
+                * data. Same as rcu_dereference, but we need a full smp_rmb()
+                * in the fast path, so put the explicit barrier here.
+                */
+               smp_read_barrier_depends();
+               for (i = 0; multi[i].func; i++) {
+                       va_start(args, call_private);
+                       multi[i].func(mdata, multi[i].probe_private,
+                               call_private, mdata->format, &args);
+                       va_end(args);
+               }
+       }
+       rcu_read_unlock_sched_notrace();
+}
+EXPORT_SYMBOL_GPL(marker_probe_cb);
+
+/*
+ * marker_probe_cb Callback that does not prepare the variable argument list.
+ * @mdata: pointer of type struct marker
+ * @call_private: caller site private data
+ * @...:  Variable argument list.
+ *
+ * Should be connected to markers "MARK_NOARGS".
+ */
+static notrace void marker_probe_cb_noarg(const struct marker *mdata,
+               void *call_private, ...)
+{
+       va_list args;   /* not initialized */
+       char ptype;
+
+       rcu_read_lock_sched_notrace();
+       ptype = mdata->ptype;
+       if (likely(!ptype)) {
+               marker_probe_func *func;
+               /* Must read the ptype before ptr. They are not data dependant,
+                * so we put an explicit smp_rmb() here. */
+               smp_rmb();
+               func = mdata->single.func;
+               /* Must read the ptr before private data. They are not data
+                * dependant, so we put an explicit smp_rmb() here. */
+               smp_rmb();
+               func(mdata, mdata->single.probe_private, call_private,
+                       mdata->format, &args);
+       } else {
+               struct marker_probe_closure *multi;
+               int i;
+               /*
+                * Read mdata->ptype before mdata->multi.
+                */
+               smp_rmb();
+               multi = mdata->multi;
+               /*
+                * multi points to an array, therefore accessing the array
+                * depends on reading multi. However, even in this case,
+                * we must insure that the pointer is read _before_ the array
+                * data. Same as rcu_dereference, but we need a full smp_rmb()
+                * in the fast path, so put the explicit barrier here.
+                */
+               smp_read_barrier_depends();
+               for (i = 0; multi[i].func; i++)
+                       multi[i].func(mdata, multi[i].probe_private,
+                               call_private, mdata->format, &args);
+       }
+       rcu_read_unlock_sched_notrace();
+}
+
+static void free_old_closure(struct rcu_head *head)
+{
+       struct marker_entry *entry = container_of(head,
+               struct marker_entry, rcu);
+       kfree(entry->oldptr);
+       /* Make sure we free the data before setting the pending flag to 0 */
+       smp_wmb();
+       entry->rcu_pending = 0;
+}
+
+static void debug_print_probes(struct marker_entry *entry)
+{
+       int i;
+
+       if (!marker_debug)
+               return;
+
+       if (!entry->ptype) {
+               printk(KERN_DEBUG "Single probe : %p %p\n",
+                       entry->single.func,
+                       entry->single.probe_private);
+       } else {
+               for (i = 0; entry->multi[i].func; i++)
+                       printk(KERN_DEBUG "Multi probe %d : %p %p\n", i,
+                               entry->multi[i].func,
+                               entry->multi[i].probe_private);
+       }
+}
+
+static struct marker_probe_closure *
+marker_entry_add_probe(struct marker_entry *entry,
+               marker_probe_func *probe, void *probe_private)
+{
+       int nr_probes = 0;
+       struct marker_probe_closure *old, *new;
+
+       WARN_ON(!probe);
+
+       debug_print_probes(entry);
+       old = entry->multi;
+       if (!entry->ptype) {
+               if (entry->single.func == probe &&
+                               entry->single.probe_private == probe_private)
+                       return ERR_PTR(-EBUSY);
+               if (entry->single.func == __mark_empty_function) {
+                       /* 0 -> 1 probes */
+                       entry->single.func = probe;
+                       entry->single.probe_private = probe_private;
+                       entry->refcount = 1;
+                       entry->ptype = 0;
+                       debug_print_probes(entry);
+                       return NULL;
+               } else {
+                       /* 1 -> 2 probes */
+                       nr_probes = 1;
+                       old = NULL;
+               }
+       } else {
+               /* (N -> N+1), (N != 0, 1) probes */
+               for (nr_probes = 0; old[nr_probes].func; nr_probes++)
+                       if (old[nr_probes].func == probe
+                                       && old[nr_probes].probe_private
+                                               == probe_private)
+                               return ERR_PTR(-EBUSY);
+       }
+       /* + 2 : one for new probe, one for NULL func */
+       new = kzalloc((nr_probes + 2) * sizeof(struct marker_probe_closure),
+                       GFP_KERNEL);
+       if (new == NULL)
+               return ERR_PTR(-ENOMEM);
+       if (!old)
+               new[0] = entry->single;
+       else
+               memcpy(new, old,
+                       nr_probes * sizeof(struct marker_probe_closure));
+       new[nr_probes].func = probe;
+       new[nr_probes].probe_private = probe_private;
+       entry->refcount = nr_probes + 1;
+       entry->multi = new;
+       entry->ptype = 1;
+       debug_print_probes(entry);
+       return old;
+}
+
+static struct marker_probe_closure *
+marker_entry_remove_probe(struct marker_entry *entry,
+               marker_probe_func *probe, void *probe_private)
+{
+       int nr_probes = 0, nr_del = 0, i;
+       struct marker_probe_closure *old, *new;
+
+       old = entry->multi;
+
+       debug_print_probes(entry);
+       if (!entry->ptype) {
+               /* 0 -> N is an error */
+               WARN_ON(entry->single.func == __mark_empty_function);
+               /* 1 -> 0 probes */
+               WARN_ON(probe && entry->single.func != probe);
+               WARN_ON(entry->single.probe_private != probe_private);
+               entry->single.func = __mark_empty_function;
+               entry->refcount = 0;
+               entry->ptype = 0;
+               debug_print_probes(entry);
+               return NULL;
+       } else {
+               /* (N -> M), (N > 1, M >= 0) probes */
+               for (nr_probes = 0; old[nr_probes].func; nr_probes++) {
+                       if ((!probe || old[nr_probes].func == probe)
+                                       && old[nr_probes].probe_private
+                                               == probe_private)
+                               nr_del++;
+               }
+       }
+
+       if (nr_probes - nr_del == 0) {
+               /* N -> 0, (N > 1) */
+               entry->single.func = __mark_empty_function;
+               entry->refcount = 0;
+               entry->ptype = 0;
+       } else if (nr_probes - nr_del == 1) {
+               /* N -> 1, (N > 1) */
+               for (i = 0; old[i].func; i++)
+                       if ((probe && old[i].func != probe) ||
+                                       old[i].probe_private != probe_private)
+                               entry->single = old[i];
+               entry->refcount = 1;
+               entry->ptype = 0;
+       } else {
+               int j = 0;
+               /* N -> M, (N > 1, M > 1) */
+               /* + 1 for NULL */
+               new = kzalloc((nr_probes - nr_del + 1)
+                       * sizeof(struct marker_probe_closure), GFP_KERNEL);
+               if (new == NULL)
+                       return ERR_PTR(-ENOMEM);
+               for (i = 0; old[i].func; i++)
+                       if ((probe && old[i].func != probe) ||
+                                       old[i].probe_private != probe_private)
+                               new[j++] = old[i];
+               entry->refcount = nr_probes - nr_del;
+               entry->ptype = 1;
+               entry->multi = new;
+       }
+       debug_print_probes(entry);
+       return old;
+}
+
+/*
+ * Get marker if the marker is present in the marker hash table.
+ * Must be called with markers_mutex held.
+ * Returns NULL if not present.
+ */
+static struct marker_entry *get_marker(const char *channel, const char *name)
+{
+       struct hlist_head *head;
+       struct hlist_node *node;
+       struct marker_entry *e;
+       size_t channel_len = strlen(channel) + 1;
+       size_t name_len = strlen(name) + 1;
+       u32 hash;
+
+       hash = jhash(channel, channel_len-1, 0) ^ jhash(name, name_len-1, 0);
+       head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
+       hlist_for_each_entry(e, node, head, hlist) {
+               if (!strcmp(channel, e->channel) && !strcmp(name, e->name))
+                       return e;
+       }
+       return NULL;
+}
+
+/*
+ * Add the marker to the marker hash table. Must be called with markers_mutex
+ * held.
+ */
+static struct marker_entry *add_marker(const char *channel, const char *name,
+               const char *format)
+{
+       struct hlist_head *head;
+       struct hlist_node *node;
+       struct marker_entry *e;
+       size_t channel_len = strlen(channel) + 1;
+       size_t name_len = strlen(name) + 1;
+       size_t format_len = 0;
+       u32 hash;
+
+       hash = jhash(channel, channel_len-1, 0) ^ jhash(name, name_len-1, 0);
+       if (format)
+               format_len = strlen(format) + 1;
+       head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
+       hlist_for_each_entry(e, node, head, hlist) {
+               if (!strcmp(channel, e->channel) && !strcmp(name, e->name)) {
+                       printk(KERN_NOTICE
+                               "Marker %s.%s busy\n", channel, name);
+                       return ERR_PTR(-EBUSY); /* Already there */
+               }
+       }
+       /*
+        * Using kmalloc here to allocate a variable length element. Could
+        * cause some memory fragmentation if overused.
+        */
+       e = kmalloc(sizeof(struct marker_entry)
+                   + channel_len + name_len + format_len,
+                   GFP_KERNEL);
+       if (!e)
+               return ERR_PTR(-ENOMEM);
+       memcpy(e->channel, channel, channel_len);
+       e->name = &e->channel[channel_len];
+       memcpy(e->name, name, name_len);
+       if (format) {
+               e->format = &e->name[channel_len + name_len];
+               memcpy(e->format, format, format_len);
+               if (strcmp(e->format, MARK_NOARGS) == 0)
+                       e->call = marker_probe_cb_noarg;
+               else
+                       e->call = marker_probe_cb;
+               trace_mark(metadata, core_marker_format,
+                          "channel %s name %s format %s",
+                          e->channel, e->name, e->format);
+       } else {
+               e->format = NULL;
+               e->call = marker_probe_cb;
+       }
+       e->single.func = __mark_empty_function;
+       e->single.probe_private = NULL;
+       e->multi = NULL;
+       e->ptype = 0;
+       e->format_allocated = 0;
+       e->refcount = 0;
+       e->rcu_pending = 0;
+       hlist_add_head(&e->hlist, head);
+       return e;
+}
+
+/*
+ * Remove the marker from the marker hash table. Must be called with mutex_lock
+ * held.
+ */
+static int remove_marker(const char *channel, const char *name)
+{
+       struct hlist_head *head;
+       struct hlist_node *node;
+       struct marker_entry *e;
+       int found = 0;
+       size_t channel_len = strlen(channel) + 1;
+       size_t name_len = strlen(name) + 1;
+       u32 hash;
+       int ret;
+
+       hash = jhash(channel, channel_len-1, 0) ^ jhash(name, name_len-1, 0);
+       head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
+       hlist_for_each_entry(e, node, head, hlist) {
+               if (!strcmp(channel, e->channel) && !strcmp(name, e->name)) {
+                       found = 1;
+                       break;
+               }
+       }
+       if (!found)
+               return -ENOENT;
+       if (e->single.func != __mark_empty_function)
+               return -EBUSY;
+       hlist_del(&e->hlist);
+       if (e->format_allocated)
+               kfree(e->format);
+       ret = ltt_channels_unregister(e->channel);
+       WARN_ON(ret);
+       /* Make sure the call_rcu has been executed */
+       if (e->rcu_pending)
+               rcu_barrier_sched();
+       kfree(e);
+       return 0;
+}
+
+/*
+ * Set the mark_entry format to the format found in the element.
+ */
+static int marker_set_format(struct marker_entry *entry, const char *format)
+{
+       entry->format = kstrdup(format, GFP_KERNEL);
+       if (!entry->format)
+               return -ENOMEM;
+       entry->format_allocated = 1;
+
+       trace_mark(metadata, core_marker_format,
+                  "channel %s name %s format %s",
+                  entry->channel, entry->name, entry->format);
+       return 0;
+}
+
+/*
+ * Sets the probe callback corresponding to one marker.
+ */
+static int set_marker(struct marker_entry *entry, struct marker *elem,
+               int active)
+{
+       int ret = 0;
+       WARN_ON(strcmp(entry->name, elem->name) != 0);
+
+       if (entry->format) {
+               if (strcmp(entry->format, elem->format) != 0) {
+                       printk(KERN_NOTICE
+                               "Format mismatch for probe %s "
+                               "(%s), marker (%s)\n",
+                               entry->name,
+                               entry->format,
+                               elem->format);
+                       return -EPERM;
+               }
+       } else {
+               ret = marker_set_format(entry, elem->format);
+               if (ret)
+                       return ret;
+       }
+
+       /*
+        * probe_cb setup (statically known) is done here. It is
+        * asynchronous with the rest of execution, therefore we only
+        * pass from a "safe" callback (with argument) to an "unsafe"
+        * callback (does not set arguments).
+        */
+       elem->call = entry->call;
+       elem->channel_id = entry->channel_id;
+       elem->event_id = entry->event_id;
+       /*
+        * Sanity check :
+        * We only update the single probe private data when the ptr is
+        * set to a _non_ single probe! (0 -> 1 and N -> 1, N != 1)
+        */
+       WARN_ON(elem->single.func != __mark_empty_function
+               && elem->single.probe_private != entry->single.probe_private
+               && !elem->ptype);
+       elem->single.probe_private = entry->single.probe_private;
+       /*
+        * Make sure the private data is valid when we update the
+        * single probe ptr.
+        */
+       smp_wmb();
+       elem->single.func = entry->single.func;
+       /*
+        * We also make sure that the new probe callbacks array is consistent
+        * before setting a pointer to it.
+        */
+       rcu_assign_pointer(elem->multi, entry->multi);
+       /*
+        * Update the function or multi probe array pointer before setting the
+        * ptype.
+        */
+       smp_wmb();
+       elem->ptype = entry->ptype;
+
+       if (elem->tp_name && (active ^ _imv_read(elem->state))) {
+               WARN_ON(!elem->tp_cb);
+               /*
+                * It is ok to directly call the probe registration because type
+                * checking has been done in the __trace_mark_tp() macro.
+                */
+
+               if (active) {
+                       /*
+                        * try_module_get should always succeed because we hold
+                        * markers_mutex to get the tp_cb address.
+                        */
+                       ret = try_module_get(__module_text_address(
+                               (unsigned long)elem->tp_cb));
+                       BUG_ON(!ret);
+                       ret = tracepoint_probe_register_noupdate(
+                               elem->tp_name,
+                               elem->tp_cb);
+               } else {
+                       ret = tracepoint_probe_unregister_noupdate(
+                               elem->tp_name,
+                               elem->tp_cb);
+                       /*
+                        * tracepoint_probe_update_all() must be called
+                        * before the module containing tp_cb is unloaded.
+                        */
+                       module_put(__module_text_address(
+                               (unsigned long)elem->tp_cb));
+               }
+       }
+       elem->state__imv = active;
+
+       return ret;
+}
+
+/*
+ * Disable a marker and its probe callback.
+ * Note: only waiting an RCU period after setting elem->call to the empty
+ * function insures that the original callback is not used anymore. This insured
+ * by rcu_read_lock_sched around the call site.
+ */
+static void disable_marker(struct marker *elem)
+{
+       int ret;
+
+       /* leave "call" as is. It is known statically. */
+       if (elem->tp_name && _imv_read(elem->state)) {
+               WARN_ON(!elem->tp_cb);
+               /*
+                * It is ok to directly call the probe registration because type
+                * checking has been done in the __trace_mark_tp() macro.
+                */
+               ret = tracepoint_probe_unregister_noupdate(elem->tp_name,
+                       elem->tp_cb);
+               WARN_ON(ret);
+               /*
+                * tracepoint_probe_update_all() must be called
+                * before the module containing tp_cb is unloaded.
+                */
+               module_put(__module_text_address((unsigned long)elem->tp_cb));
+       }
+       elem->state__imv = 0;
+       elem->single.func = __mark_empty_function;
+       /* Update the function before setting the ptype */
+       smp_wmb();
+       elem->ptype = 0;        /* single probe */
+       /*
+        * Leave the private data and channel_id/event_id there, because removal
+        * is racy and should be done only after an RCU period. These are never
+        * used until the next initialization anyway.
+        */
+}
+
+/**
+ * marker_update_probe_range - Update a probe range
+ * @begin: beginning of the range
+ * @end: end of the range
+ *
+ * Updates the probe callback corresponding to a range of markers.
+ */
+void marker_update_probe_range(struct marker *begin,
+       struct marker *end)
+{
+       struct marker *iter;
+       struct marker_entry *mark_entry;
+
+       mutex_lock(&markers_mutex);
+       for (iter = begin; iter < end; iter++) {
+               mark_entry = get_marker(iter->channel, iter->name);
+               if (mark_entry) {
+                       set_marker(mark_entry, iter, !!mark_entry->refcount);
+                       /*
+                        * ignore error, continue
+                        */
+               } else {
+                       disable_marker(iter);
+               }
+       }
+       mutex_unlock(&markers_mutex);
+}
+
+/*
+ * Update probes, removing the faulty probes.
+ *
+ * Internal callback only changed before the first probe is connected to it.
+ * Single probe private data can only be changed on 0 -> 1 and 2 -> 1
+ * transitions.  All other transitions will leave the old private data valid.
+ * This makes the non-atomicity of the callback/private data updates valid.
+ *
+ * "special case" updates :
+ * 0 -> 1 callback
+ * 1 -> 0 callback
+ * 1 -> 2 callbacks
+ * 2 -> 1 callbacks
+ * Other updates all behave the same, just like the 2 -> 3 or 3 -> 2 updates.
+ * Site effect : marker_set_format may delete the marker entry (creating a
+ * replacement).
+ */
+static void marker_update_probes(void)
+{
+       /* Core kernel markers */
+       marker_update_probe_range(__start___markers, __stop___markers);
+       /* Markers in modules. */
+       module_update_markers();
+       tracepoint_probe_update_all();
+       /* Update immediate values */
+       core_imv_update();
+       module_imv_update();
+       marker_update_processes();
+}
+
+/**
+ * marker_probe_register -  Connect a probe to a marker
+ * @channel: marker channel
+ * @name: marker name
+ * @format: format string
+ * @probe: probe handler
+ * @probe_private: probe private data
+ *
+ * private data must be a valid allocated memory address, or NULL.
+ * Returns 0 if ok, error value on error.
+ * The probe address must at least be aligned on the architecture pointer size.
+ */
+int marker_probe_register(const char *channel, const char *name,
+                         const char *format, marker_probe_func *probe,
+                         void *probe_private)
+{
+       struct marker_entry *entry;
+       int ret = 0, ret_err;
+       struct marker_probe_closure *old;
+       int first_probe = 0;
+
+       mutex_lock(&markers_mutex);
+       entry = get_marker(channel, name);
+       if (!entry) {
+               first_probe = 1;
+               entry = add_marker(channel, name, format);
+               if (IS_ERR(entry))
+                       ret = PTR_ERR(entry);
+               if (ret)
+                       goto end;
+               ret = ltt_channels_register(channel);
+               if (ret)
+                       goto error_remove_marker;
+               ret = ltt_channels_get_index_from_name(channel);
+               if (ret < 0)
+                       goto error_unregister_channel;
+               entry->channel_id = ret;
+               ret = ltt_channels_get_event_id(channel, name);
+               if (ret < 0)
+                       goto error_unregister_channel;
+               entry->event_id = ret;
+               ret = 0;
+               trace_mark(metadata, core_marker_id,
+                          "channel %s name %s event_id %hu "
+                          "int #1u%zu long #1u%zu pointer #1u%zu "
+                          "size_t #1u%zu alignment #1u%u",
+                          channel, name, entry->event_id,
+                          sizeof(int), sizeof(long), sizeof(void *),
+                          sizeof(size_t), ltt_get_alignment());
+       } else if (format) {
+               if (!entry->format)
+                       ret = marker_set_format(entry, format);
+               else if (strcmp(entry->format, format))
+                       ret = -EPERM;
+               if (ret)
+                       goto end;
+       }
+
+       /*
+        * If we detect that a call_rcu is pending for this marker,
+        * make sure it's executed now.
+        */
+       if (entry->rcu_pending)
+               rcu_barrier_sched();
+       old = marker_entry_add_probe(entry, probe, probe_private);
+       if (IS_ERR(old)) {
+               ret = PTR_ERR(old);
+               if (first_probe)
+                       goto error_unregister_channel;
+               else
+                       goto end;
+       }
+       mutex_unlock(&markers_mutex);
+
+       marker_update_probes();
+
+       mutex_lock(&markers_mutex);
+       entry = get_marker(channel, name);
+       if (!entry)
+               goto end;
+       if (entry->rcu_pending)
+               rcu_barrier_sched();
+       entry->oldptr = old;
+       entry->rcu_pending = 1;
+       /* write rcu_pending before calling the RCU callback */
+       smp_wmb();
+       call_rcu_sched(&entry->rcu, free_old_closure);
+       goto end;
+
+error_unregister_channel:
+       ret_err = ltt_channels_unregister(channel);
+       WARN_ON(ret_err);
+error_remove_marker:
+       ret_err = remove_marker(channel, name);
+       WARN_ON(ret_err);
+end:
+       mutex_unlock(&markers_mutex);
+       return ret;
+}
+EXPORT_SYMBOL_GPL(marker_probe_register);
+
+/**
+ * marker_probe_unregister -  Disconnect a probe from a marker
+ * @channel: marker channel
+ * @name: marker name
+ * @probe: probe function pointer
+ * @probe_private: probe private data
+ *
+ * Returns the private data given to marker_probe_register, or an ERR_PTR().
+ * We do not need to call a synchronize_sched to make sure the probes have
+ * finished running before doing a module unload, because the module unload
+ * itself uses stop_machine(), which insures that every preempt disabled section
+ * have finished.
+ */
+int marker_probe_unregister(const char *channel, const char *name,
+                           marker_probe_func *probe, void *probe_private)
+{
+       struct marker_entry *entry;
+       struct marker_probe_closure *old;
+       int ret = -ENOENT;
+
+       mutex_lock(&markers_mutex);
+       entry = get_marker(channel, name);
+       if (!entry)
+               goto end;
+       if (entry->rcu_pending)
+               rcu_barrier_sched();
+       old = marker_entry_remove_probe(entry, probe, probe_private);
+       mutex_unlock(&markers_mutex);
+
+       marker_update_probes();
+
+       mutex_lock(&markers_mutex);
+       entry = get_marker(channel, name);
+       if (!entry)
+               goto end;
+       if (entry->rcu_pending)
+               rcu_barrier_sched();
+       entry->oldptr = old;
+       entry->rcu_pending = 1;
+       /* write rcu_pending before calling the RCU callback */
+       smp_wmb();
+       call_rcu_sched(&entry->rcu, free_old_closure);
+       remove_marker(channel, name);   /* Ignore busy error message */
+       ret = 0;
+end:
+       mutex_unlock(&markers_mutex);
+       return ret;
+}
+EXPORT_SYMBOL_GPL(marker_probe_unregister);
+
+static struct marker_entry *
+get_marker_from_private_data(marker_probe_func *probe, void *probe_private)
+{
+       struct marker_entry *entry;
+       unsigned int i;
+       struct hlist_head *head;
+       struct hlist_node *node;
+
+       for (i = 0; i < MARKER_TABLE_SIZE; i++) {
+               head = &marker_table[i];
+               hlist_for_each_entry(entry, node, head, hlist) {
+                       if (!entry->ptype) {
+                               if (entry->single.func == probe
+                                               && entry->single.probe_private
+                                               == probe_private)
+                                       return entry;
+                       } else {
+                               struct marker_probe_closure *closure;
+                               closure = entry->multi;
+                               for (i = 0; closure[i].func; i++) {
+                                       if (closure[i].func == probe &&
+                                                       closure[i].probe_private
+                                                       == probe_private)
+                                               return entry;
+                               }
+                       }
+               }
+       }
+       return NULL;
+}
+
+/**
+ * marker_probe_unregister_private_data -  Disconnect a probe from a marker
+ * @probe: probe function
+ * @probe_private: probe private data
+ *
+ * Unregister a probe by providing the registered private data.
+ * Only removes the first marker found in hash table.
+ * Return 0 on success or error value.
+ * We do not need to call a synchronize_sched to make sure the probes have
+ * finished running before doing a module unload, because the module unload
+ * itself uses stop_machine(), which insures that every preempt disabled section
+ * have finished.
+ */
+int marker_probe_unregister_private_data(marker_probe_func *probe,
+               void *probe_private)
+{
+       struct marker_entry *entry;
+       int ret = 0;
+       struct marker_probe_closure *old;
+       const char *channel = NULL, *name = NULL;
+
+       mutex_lock(&markers_mutex);
+       entry = get_marker_from_private_data(probe, probe_private);
+       if (!entry) {
+               ret = -ENOENT;
+               goto end;
+       }
+       if (entry->rcu_pending)
+               rcu_barrier_sched();
+       old = marker_entry_remove_probe(entry, NULL, probe_private);
+       channel = kstrdup(entry->channel, GFP_KERNEL);
+       name = kstrdup(entry->name, GFP_KERNEL);
+       mutex_unlock(&markers_mutex);
+
+       marker_update_probes();
+
+       mutex_lock(&markers_mutex);
+       entry = get_marker(channel, name);
+       if (!entry)
+               goto end;
+       if (entry->rcu_pending)
+               rcu_barrier_sched();
+       entry->oldptr = old;
+       entry->rcu_pending = 1;
+       /* write rcu_pending before calling the RCU callback */
+       smp_wmb();
+       call_rcu_sched(&entry->rcu, free_old_closure);
+       /* Ignore busy error message */
+       remove_marker(channel, name);
+end:
+       mutex_unlock(&markers_mutex);
+       kfree(channel);
+       kfree(name);
+       return ret;
+}
+EXPORT_SYMBOL_GPL(marker_probe_unregister_private_data);
+
+/**
+ * marker_get_private_data - Get a marker's probe private data
+ * @channel: marker channel
+ * @name: marker name
+ * @probe: probe to match
+ * @num: get the nth matching probe's private data
+ *
+ * Returns the nth private data pointer (starting from 0) matching, or an
+ * ERR_PTR.
+ * Returns the private data pointer, or an ERR_PTR.
+ * The private data pointer should _only_ be dereferenced if the caller is the
+ * owner of the data, or its content could vanish. This is mostly used to
+ * confirm that a caller is the owner of a registered probe.
+ */
+void *marker_get_private_data(const char *channel, const char *name,
+                             marker_probe_func *probe, int num)
+{
+       struct hlist_head *head;
+       struct hlist_node *node;
+       struct marker_entry *e;
+       size_t channel_len = strlen(channel) + 1;
+       size_t name_len = strlen(name) + 1;
+       int i;
+       u32 hash;
+
+       hash = jhash(channel, channel_len-1, 0) ^ jhash(name, name_len-1, 0);
+       head = &marker_table[hash & ((1 << MARKER_HASH_BITS)-1)];
+       hlist_for_each_entry(e, node, head, hlist) {
+               if (!strcmp(channel, e->channel) && !strcmp(name, e->name)) {
+                       if (!e->ptype) {
+                               if (num == 0 && e->single.func == probe)
+                                       return e->single.probe_private;
+                       } else {
+                               struct marker_probe_closure *closure;
+                               int match = 0;
+                               closure = e->multi;
+                               for (i = 0; closure[i].func; i++) {
+                                       if (closure[i].func != probe)
+                                               continue;
+                                       if (match++ == num)
+                                               return closure[i].probe_private;
+                               }
+                       }
+                       break;
+               }
+       }
+       return ERR_PTR(-ENOENT);
+}
+EXPORT_SYMBOL_GPL(marker_get_private_data);
+
+/**
+ * markers_compact_event_ids - Compact markers event IDs and reassign channels
+ *
+ * Called when no channel users are active by the channel infrastructure.
+ * Called with lock_markers() and channel mutex held.
+ */
+void markers_compact_event_ids(void)
+{
+       struct marker_entry *entry;
+       unsigned int i;
+       struct hlist_head *head;
+       struct hlist_node *node;
+       int ret;
+
+       for (i = 0; i < MARKER_TABLE_SIZE; i++) {
+               head = &marker_table[i];
+               hlist_for_each_entry(entry, node, head, hlist) {
+                       ret = ltt_channels_get_index_from_name(entry->channel);
+                       WARN_ON(ret < 0);
+                       entry->channel_id = ret;
+                       ret = _ltt_channels_get_event_id(entry->channel,
+                                                        entry->name);
+                       WARN_ON(ret < 0);
+                       entry->event_id = ret;
+               }
+       }
+}
+
+#ifdef CONFIG_MODULES
+
+/**
+ * marker_get_iter_range - Get a next marker iterator given a range.
+ * @marker: current markers (in), next marker (out)
+ * @begin: beginning of the range
+ * @end: end of the range
+ *
+ * Returns whether a next marker has been found (1) or not (0).
+ * Will return the first marker in the range if the input marker is NULL.
+ */
+int marker_get_iter_range(struct marker **marker, struct marker *begin,
+       struct marker *end)
+{
+       if (!*marker && begin != end) {
+               *marker = begin;
+               return 1;
+       }
+       if (*marker >= begin && *marker < end)
+               return 1;
+       return 0;
+}
+EXPORT_SYMBOL_GPL(marker_get_iter_range);
+
+static void marker_get_iter(struct marker_iter *iter)
+{
+       int found = 0;
+
+       /* Core kernel markers */
+       if (!iter->module) {
+               found = marker_get_iter_range(&iter->marker,
+                               __start___markers, __stop___markers);
+               if (found)
+                       goto end;
+       }
+       /* Markers in modules. */
+       found = module_get_iter_markers(iter);
+end:
+       if (!found)
+               marker_iter_reset(iter);
+}
+
+void marker_iter_start(struct marker_iter *iter)
+{
+       marker_get_iter(iter);
+}
+EXPORT_SYMBOL_GPL(marker_iter_start);
+
+void marker_iter_next(struct marker_iter *iter)
+{
+       iter->marker++;
+       /*
+        * iter->marker may be invalid because we blindly incremented it.
+        * Make sure it is valid by marshalling on the markers, getting the
+        * markers from following modules if necessary.
+        */
+       marker_get_iter(iter);
+}
+EXPORT_SYMBOL_GPL(marker_iter_next);
+
+void marker_iter_stop(struct marker_iter *iter)
+{
+}
+EXPORT_SYMBOL_GPL(marker_iter_stop);
+
+void marker_iter_reset(struct marker_iter *iter)
+{
+       iter->module = NULL;
+       iter->marker = NULL;
+}
+EXPORT_SYMBOL_GPL(marker_iter_reset);
+
+#ifdef CONFIG_MARKERS_USERSPACE
+/*
+ * must be called with current->user_markers_mutex held
+ */
+static void free_user_marker(char __user *state, struct hlist_head *head)
+{
+       struct user_marker *umark;
+       struct hlist_node *pos, *n;
+
+       hlist_for_each_entry_safe(umark, pos, n, head, hlist) {
+               if (umark->state == state) {
+                       hlist_del(&umark->hlist);
+                       kfree(umark);
+               }
+       }
+}
+
+asmlinkage long sys_marker(char __user *name, char __user *format,
+               char __user *state, int reg)
+{
+       struct user_marker *umark;
+       long len;
+       struct marker_entry *entry;
+       int ret = 0;
+
+       printk(KERN_DEBUG "Program %s %s marker [%p, %p]\n",
+               current->comm, reg ? "registers" : "unregisters",
+               name, state);
+       if (reg) {
+               umark = kmalloc(sizeof(struct user_marker), GFP_KERNEL);
+               umark->name[MAX_USER_MARKER_NAME_LEN - 1] = '\0';
+               umark->format[MAX_USER_MARKER_FORMAT_LEN - 1] = '\0';
+               umark->state = state;
+               len = strncpy_from_user(umark->name, name,
+                       MAX_USER_MARKER_NAME_LEN - 1);
+               if (len < 0) {
+                       ret = -EFAULT;
+                       goto error;
+               }
+               len = strncpy_from_user(umark->format, format,
+                       MAX_USER_MARKER_FORMAT_LEN - 1);
+               if (len < 0) {
+                       ret = -EFAULT;
+                       goto error;
+               }
+               printk(KERN_DEBUG "Marker name : %s, format : %s", umark->name,
+                       umark->format);
+               mutex_lock(&markers_mutex);
+               entry = get_marker("userspace", umark->name);
+               if (entry) {
+                       if (entry->format &&
+                               strcmp(entry->format, umark->format) != 0) {
+                               printk(" error, wrong format in process %s",
+                                       current->comm);
+                               ret = -EPERM;
+                               goto error_unlock;
+                       }
+                       printk(" %s", !!entry->refcount
+                                       ? "enabled" : "disabled");
+                       if (put_user(!!entry->refcount, state)) {
+                               ret = -EFAULT;
+                               goto error_unlock;
+                       }
+                       printk("\n");
+               } else {
+                       printk(" disabled\n");
+                       if (put_user(0, umark->state)) {
+                               printk(KERN_WARNING
+                                       "Marker in %s caused a fault\n",
+                                       current->comm);
+                               goto error_unlock;
+                       }
+               }
+               mutex_lock(&current->group_leader->user_markers_mutex);
+               hlist_add_head(&umark->hlist,
+                       &current->group_leader->user_markers);
+               current->group_leader->user_markers_sequence++;
+               mutex_unlock(&current->group_leader->user_markers_mutex);
+               mutex_unlock(&markers_mutex);
+       } else {
+               mutex_lock(&current->group_leader->user_markers_mutex);
+               free_user_marker(state,
+                       &current->group_leader->user_markers);
+               current->group_leader->user_markers_sequence++;
+               mutex_unlock(&current->group_leader->user_markers_mutex);
+       }
+       goto end;
+error_unlock:
+       mutex_unlock(&markers_mutex);
+error:
+       kfree(umark);
+end:
+       return ret;
+}
+
+/*
+ * Types :
+ * string : 0
+ */
+asmlinkage long sys_trace(int type, uint16_t id,
+               char __user *ubuf)
+{
+       long ret = -EPERM;
+       char *page;
+       int len;
+
+       switch (type) {
+       case 0: /* String */
+               ret = -ENOMEM;
+               page = (char *)__get_free_page(GFP_TEMPORARY);
+               if (!page)
+                       goto string_out;
+               len = strncpy_from_user(page, ubuf, PAGE_SIZE);
+               if (len < 0) {
+                       ret = -EFAULT;
+                       goto string_err;
+               }
+               trace_mark(userspace, string, "string %s", page);
+string_err:
+               free_page((unsigned long) page);
+string_out:
+               break;
+       default:
+               break;
+       }
+       return ret;
+}
+
+static void marker_update_processes(void)
+{
+       struct task_struct *g, *t;
+
+       /*
+        * markers_mutex is taken to protect the p->user_markers read.
+        */
+       mutex_lock(&markers_mutex);
+       read_lock(&tasklist_lock);
+       for_each_process(g) {
+               WARN_ON(!thread_group_leader(g));
+               if (hlist_empty(&g->user_markers))
+                       continue;
+               if (strcmp(g->comm, "testprog") == 0)
+                       printk(KERN_DEBUG "set update pending for testprog\n");
+               t = g;
+               do {
+                       /* TODO : implement this thread flag in each arch. */
+                       set_tsk_thread_flag(t, TIF_MARKER_PENDING);
+               } while ((t = next_thread(t)) != g);
+       }
+       read_unlock(&tasklist_lock);
+       mutex_unlock(&markers_mutex);
+}
+
+/*
+ * Update current process.
+ * Note that we have to wait a whole scheduler period before we are sure that
+ * every running userspace threads have their markers updated.
+ * (synchronize_sched() can be used to insure this).
+ */
+void marker_update_process(void)
+{
+       struct user_marker *umark;
+       struct hlist_node *pos;
+       struct marker_entry *entry;
+
+       mutex_lock(&markers_mutex);
+       mutex_lock(&current->group_leader->user_markers_mutex);
+       if (strcmp(current->comm, "testprog") == 0)
+               printk(KERN_DEBUG "do update pending for testprog\n");
+       hlist_for_each_entry(umark, pos,
+                       &current->group_leader->user_markers, hlist) {
+               printk(KERN_DEBUG "Updating marker %s in %s\n",
+                       umark->name, current->comm);
+               entry = get_marker("userspace", umark->name);
+               if (entry) {
+                       if (entry->format &&
+                               strcmp(entry->format, umark->format) != 0) {
+                               printk(KERN_WARNING
+                                       " error, wrong format in process %s\n",
+                                       current->comm);
+                               break;
+                       }
+                       if (put_user(!!entry->refcount, umark->state)) {
+                               printk(KERN_WARNING
+                                       "Marker in %s caused a fault\n",
+                                       current->comm);
+                               break;
+                       }
+               } else {
+                       if (put_user(0, umark->state)) {
+                               printk(KERN_WARNING
+                                       "Marker in %s caused a fault\n",
+                                       current->comm);
+                               break;
+                       }
+               }
+       }
+       clear_thread_flag(TIF_MARKER_PENDING);
+       mutex_unlock(&current->group_leader->user_markers_mutex);
+       mutex_unlock(&markers_mutex);
+}
+
+/*
+ * Called at process exit and upon do_execve().
+ * We assume that when the leader exits, no more references can be done to the
+ * leader structure by the other threads.
+ */
+void exit_user_markers(struct task_struct *p)
+{
+       struct user_marker *umark;
+       struct hlist_node *pos, *n;
+
+       if (thread_group_leader(p)) {
+               mutex_lock(&markers_mutex);
+               mutex_lock(&p->user_markers_mutex);
+               hlist_for_each_entry_safe(umark, pos, n, &p->user_markers,
+                       hlist)
+                   kfree(umark);
+               INIT_HLIST_HEAD(&p->user_markers);
+               p->user_markers_sequence++;
+               mutex_unlock(&p->user_markers_mutex);
+               mutex_unlock(&markers_mutex);
+       }
+}
+
+int is_marker_enabled(const char *channel, const char *name)
+{
+       struct marker_entry *entry;
+
+       mutex_lock(&markers_mutex);
+       entry = get_marker(channel, name);
+       mutex_unlock(&markers_mutex);
+
+       return entry && !!entry->refcount;
+}
+#endif
+
+int marker_module_notify(struct notifier_block *self,
+                        unsigned long val, void *data)
+{
+       struct module *mod = data;
+
+       switch (val) {
+       case MODULE_STATE_COMING:
+               marker_update_probe_range(mod->markers,
+                       mod->markers + mod->num_markers);
+               break;
+       case MODULE_STATE_GOING:
+               marker_update_probe_range(mod->markers,
+                       mod->markers + mod->num_markers);
+               break;
+       }
+       return 0;
+}
+
+struct notifier_block marker_module_nb = {
+       .notifier_call = marker_module_notify,
+       .priority = 0,
+};
+
+static int init_markers(void)
+{
+       return register_module_notifier(&marker_module_nb);
+}
+__initcall(init_markers);
+
+#endif /* CONFIG_MODULES */
+
+void ltt_dump_marker_state(struct ltt_trace_struct *trace)
+{
+       struct marker_iter iter;
+       struct ltt_probe_private_data call_data;
+       const char *channel;
+
+       call_data.trace = trace;
+       call_data.serializer = NULL;
+
+       marker_iter_reset(&iter);
+       marker_iter_start(&iter);
+       for (; iter.marker != NULL; marker_iter_next(&iter)) {
+               if (!_imv_read(iter.marker->state))
+                       continue;
+               channel = ltt_channels_get_name_from_index(
+                               iter.marker->channel_id);
+               __trace_mark(0, metadata, core_marker_id,
+                       &call_data,
+                       "channel %s name %s event_id %hu "
+                       "int #1u%zu long #1u%zu pointer #1u%zu "
+                       "size_t #1u%zu alignment #1u%u",
+                       channel,
+                       iter.marker->name,
+                       iter.marker->event_id,
+                       sizeof(int), sizeof(long),
+                       sizeof(void *), sizeof(size_t),
+                       ltt_get_alignment());
+               if (iter.marker->format)
+                       __trace_mark(0, metadata,
+                               core_marker_format,
+                               &call_data,
+                               "channel %s name %s format %s",
+                               channel,
+                               iter.marker->name,
+                               iter.marker->format);
+       }
+       marker_iter_stop(&iter);
+}
+EXPORT_SYMBOL_GPL(ltt_dump_marker_state);
diff --git a/libtracectl/Makefile b/libtracectl/Makefile
new file mode 100644 (file)
index 0000000..754e8ff
--- /dev/null
@@ -0,0 +1,7 @@
+all: libtracectl.so
+
+libtracectl.so: tracectl.c
+       gcc -g -shared -fPIC -o libtracectl.so tracectl.c
+
+clean:
+       rm -rf *.so *.o
diff --git a/libtracectl/tracectl.c b/libtracectl/tracectl.c
new file mode 100644 (file)
index 0000000..aca3abc
--- /dev/null
@@ -0,0 +1,155 @@
+#include <stdio.h>
+#include <stdint.h>
+#include <signal.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/un.h>
+#define UNIX_PATH_MAX 108
+
+//#define SOCKETDIR "/var/run/ust/socks"
+#define SOCKETDIR "/tmp/socks"
+#define SOCKETDIRLEN sizeof(SOCKETDIR)
+#define USTSIGNAL SIGIO
+
+#define DBG(fmt, args...) fprintf(stderr, fmt "\n", ## args)
+#define WARN(fmt, args...) fprintf(stderr, "usertrace: WARNING: " fmt "\n", ## args)
+#define ERR(fmt, args...) fprintf(stderr, "usertrace: ERROR: " fmt "\n", ## args)
+#define PERROR(call) perror("usertrace: ERROR: " call)
+
+struct tracecmd { /* no padding */
+       uint32_t size;
+       uint16_t command;
+};
+
+
+pid_t mypid;
+char mysocketfile[UNIX_PATH_MAX] = "";
+
+void do_command(struct tracecmd *cmd)
+{
+}
+
+void receive_commands()
+{
+}
+
+/* The signal handler itself. */
+
+void sighandler(int sig)
+{
+       DBG("sighandler");
+       receive_commands();
+}
+
+/* Called by the app signal handler to chain it to us. */
+
+void chain_signal()
+{
+       sighandler(USTSIGNAL);
+}
+
+int init_socket()
+{
+       int result;
+       int fd;
+       char pidstr[6];
+       int pidlen;
+
+       struct sockaddr_un addr;
+       
+       result = fd = socket(PF_UNIX, SOCK_SEQPACKET, 0);
+       if(result == -1) {
+               PERROR("socket");
+               return -1;
+       }
+
+       addr.sun_family = AF_UNIX;
+
+       result = snprintf(addr.sun_path, UNIX_PATH_MAX, "%s/%d", SOCKETDIR, mypid);
+       if(result >= UNIX_PATH_MAX) {
+               ERR("string overflow allocating socket name");
+               goto close_sock;
+       }
+       //DBG("opening socket at %s", addr.sun_path);
+
+       result = bind(fd, (struct sockaddr *)&addr, sizeof(addr));
+       if(result == -1) {
+               PERROR("bind");
+               goto close_sock;
+       }
+
+       strcpy(mysocketfile, addr.sun_path);
+
+       close_sock:
+       close(fd);
+
+       return -1;
+}
+
+void destroy_socket()
+{
+       int result;
+
+       if(mysocketfile[0] == '\0')
+               return;
+
+       result = unlink(mysocketfile);
+       if(result == -1) {
+               PERROR("unlink");
+       }
+}
+
+int init_signal_handler(void)
+{
+       /* Attempt to handler SIGIO. If the main program wants to
+        * handle it, fine, it'll override us. They it'll have to
+        * use the chaining function.
+        */
+
+       int result;
+       struct sigaction act;
+
+       result = sigemptyset(&act.sa_mask);
+       if(result == -1) {
+               PERROR("sigemptyset");
+               return -1;
+       }
+
+       act.sa_handler = sighandler;
+       act.sa_flags = SA_RESTART;
+
+       /* Only defer ourselves. Also, try to restart interrupted
+        * syscalls to disturb the traced program as little as possible.
+        */
+       result = sigaction(SIGIO, &act, NULL);
+       if(result == -1) {
+               PERROR("sigaction");
+               return -1;
+       }
+
+       return 0;
+}
+
+void __attribute__((constructor)) init()
+{
+       int result;
+
+       mypid = getpid();
+
+       /* if using instead seperate thread, then should create thread */
+       result = init_signal_handler();
+       result = init_socket();
+
+       return;
+
+       /* should decrementally destroy stuff if error */
+
+}
+
+/* This is only called if we terminate normally, not with an unhandled signal,
+ * so we cannot rely on it. */
+
+void __attribute__((destructor)) fini()
+{
+       destroy_socket();
+}
This page took 0.039589 seconds and 4 git commands to generate.