#include <lttng/ust-events.h>
#include <lttng/ust-tracer.h>
+#include <lttng/ust-context-provider.h>
+#include <urcu-pointer.h>
+#include <usterr-signal-safe.h>
#include <helper.h>
#include <string.h>
#include <assert.h>
+/*
+ * The filter implementation requires that two consecutive "get" for the
+ * same context performed by the same thread return the same result.
+ */
+
int lttng_find_context(struct lttng_ctx *ctx, const char *name)
{
unsigned int i;
+ const char *subname;
+ if (strncmp(name, "$ctx.", strlen("$ctx.")) == 0) {
+ subname = name + strlen("$ctx.");
+ } else {
+ subname = name;
+ }
for (i = 0; i < ctx->nr_fields; i++) {
/* Skip allocated (but non-initialized) contexts */
if (!ctx->fields[i].event_field.name)
continue;
- if (!strcmp(ctx->fields[i].event_field.name, name))
+ if (!strcmp(ctx->fields[i].event_field.name, subname))
+ return 1;
+ }
+ return 0;
+}
+
+int lttng_get_context_index(struct lttng_ctx *ctx, const char *name)
+{
+ unsigned int i;
+ const char *subname;
+
+ if (!ctx)
+ return -1;
+ if (strncmp(name, "$ctx.", strlen("$ctx.")) == 0) {
+ subname = name + strlen("$ctx.");
+ } else {
+ subname = name;
+ }
+ for (i = 0; i < ctx->nr_fields; i++) {
+ /* Skip allocated (but non-initialized) contexts */
+ if (!ctx->fields[i].event_field.name)
+ continue;
+ if (!strcmp(ctx->fields[i].event_field.name, subname))
+ return i;
+ }
+ return -1;
+}
+
+static int lttng_find_context_provider(struct lttng_ctx *ctx, const char *name)
+{
+ unsigned int i;
+
+ for (i = 0; i < ctx->nr_fields; i++) {
+ /* Skip allocated (but non-initialized) contexts */
+ if (!ctx->fields[i].event_field.name)
+ continue;
+ if (!strncmp(ctx->fields[i].event_field.name, name,
+ strlen(name)))
return 1;
}
return 0;
*ctx_p = zmalloc(sizeof(struct lttng_ctx));
if (!*ctx_p)
return NULL;
+ (*ctx_p)->largest_align = 1;
}
ctx = *ctx_p;
if (ctx->nr_fields + 1 > ctx->allocated_fields) {
return field;
}
+int lttng_context_add_rcu(struct lttng_ctx **ctx_p,
+ const struct lttng_ctx_field *f)
+{
+ struct lttng_ctx *old_ctx = *ctx_p, *new_ctx = NULL;
+ struct lttng_ctx_field *new_fields = NULL;
+ struct lttng_ctx_field *nf;
+
+ if (old_ctx) {
+ new_ctx = zmalloc(sizeof(struct lttng_ctx));
+ if (!new_ctx)
+ return -ENOMEM;
+ *new_ctx = *old_ctx;
+ new_fields = zmalloc(new_ctx->allocated_fields
+ * sizeof(struct lttng_ctx_field));
+ if (!new_fields) {
+ free(new_ctx);
+ return -ENOMEM;
+ }
+ memcpy(new_fields, old_ctx->fields,
+ sizeof(*old_ctx->fields) * old_ctx->nr_fields);
+ new_ctx->fields = new_fields;
+ }
+ nf = lttng_append_context(&new_ctx);
+ if (!nf) {
+ free(new_fields);
+ free(new_ctx);
+ return -ENOMEM;
+ }
+ *nf = *f;
+ lttng_context_update(new_ctx);
+ rcu_assign_pointer(*ctx_p, new_ctx);
+ synchronize_trace();
+ if (old_ctx) {
+ free(old_ctx->fields);
+ free(old_ctx);
+ }
+ return 0;
+}
+
+/*
+ * lttng_context_update() should be called at least once between context
+ * modification and trace start.
+ */
+void lttng_context_update(struct lttng_ctx *ctx)
+{
+ int i;
+ size_t largest_align = 8; /* in bits */
+
+ for (i = 0; i < ctx->nr_fields; i++) {
+ struct lttng_type *type;
+ size_t field_align = 8;
+
+ type = &ctx->fields[i].event_field.type;
+ switch (type->atype) {
+ case atype_integer:
+ field_align = type->u.basic.integer.alignment;
+ break;
+ case atype_array:
+ {
+ struct lttng_basic_type *btype;
+
+ btype = &type->u.array.elem_type;
+ switch (btype->atype) {
+ case atype_integer:
+ field_align = btype->u.basic.integer.alignment;
+ break;
+ case atype_string:
+ break;
+
+ case atype_array:
+ case atype_sequence:
+ default:
+ WARN_ON_ONCE(1);
+ break;
+ }
+ break;
+ }
+ case atype_sequence:
+ {
+ struct lttng_basic_type *btype;
+
+ btype = &type->u.sequence.length_type;
+ switch (btype->atype) {
+ case atype_integer:
+ field_align = btype->u.basic.integer.alignment;
+ break;
+
+ case atype_string:
+ case atype_array:
+ case atype_sequence:
+ default:
+ WARN_ON_ONCE(1);
+ break;
+ }
+
+ btype = &type->u.sequence.elem_type;
+ switch (btype->atype) {
+ case atype_integer:
+ field_align = max_t(size_t,
+ field_align,
+ btype->u.basic.integer.alignment);
+ break;
+
+ case atype_string:
+ break;
+
+ case atype_array:
+ case atype_sequence:
+ default:
+ WARN_ON_ONCE(1);
+ break;
+ }
+ break;
+ }
+ case atype_string:
+ break;
+ case atype_dynamic:
+ break;
+ case atype_enum:
+ default:
+ WARN_ON_ONCE(1);
+ break;
+ }
+ largest_align = max_t(size_t, largest_align, field_align);
+ }
+ ctx->largest_align = largest_align >> 3; /* bits to bytes */
+}
+
/*
* Remove last context field.
*/
ctx = *ctx_p;
ctx->nr_fields--;
assert(&ctx->fields[ctx->nr_fields] == field);
+ assert(field->field_name == NULL);
memset(&ctx->fields[ctx->nr_fields], 0, sizeof(struct lttng_ctx_field));
}
for (i = 0; i < ctx->nr_fields; i++) {
if (ctx->fields[i].destroy)
ctx->fields[i].destroy(&ctx->fields[i]);
+ free(ctx->fields[i].field_name);
}
free(ctx->fields);
free(ctx);
}
+
+/*
+ * Can be safely performed concurrently with tracing using the struct
+ * lttng_ctx. Using RCU update. Needs to match RCU read-side handling of
+ * contexts.
+ *
+ * This does not allow adding, removing, or changing typing of the
+ * contexts, since this needs to stay invariant for metadata. However,
+ * it allows updating the handlers associated with all contexts matching
+ * a provider (by name) while tracing is using it, in a way that ensures
+ * a single RCU read-side critical section see either all old, or all
+ * new handlers.
+ */
+int lttng_ust_context_set_provider_rcu(struct lttng_ctx **_ctx,
+ const char *name,
+ size_t (*get_size)(struct lttng_ctx_field *field, size_t offset),
+ void (*record)(struct lttng_ctx_field *field,
+ struct lttng_ust_lib_ring_buffer_ctx *ctx,
+ struct lttng_channel *chan),
+ void (*get_value)(struct lttng_ctx_field *field,
+ struct lttng_ctx_value *value))
+{
+ int i, ret;
+ struct lttng_ctx *ctx = *_ctx, *new_ctx;
+ struct lttng_ctx_field *new_fields;
+
+ if (!ctx || !lttng_find_context_provider(ctx, name))
+ return 0;
+ /*
+ * We have at least one instance of context for the provider.
+ */
+ new_ctx = zmalloc(sizeof(*new_ctx));
+ if (!new_ctx)
+ return -ENOMEM;
+ *new_ctx = *ctx;
+ new_fields = zmalloc(sizeof(*new_fields) * ctx->allocated_fields);
+ if (!new_fields) {
+ ret = -ENOMEM;
+ goto field_error;
+ }
+ memcpy(new_fields, ctx->fields,
+ sizeof(*new_fields) * ctx->allocated_fields);
+ for (i = 0; i < ctx->nr_fields; i++) {
+ if (strncmp(new_fields[i].event_field.name,
+ name, strlen(name)) != 0)
+ continue;
+ new_fields[i].get_size = get_size;
+ new_fields[i].record = record;
+ new_fields[i].get_value = get_value;
+ }
+ new_ctx->fields = new_fields;
+ rcu_assign_pointer(*_ctx, new_ctx);
+ synchronize_trace();
+ free(ctx->fields);
+ free(ctx);
+ return 0;
+
+field_error:
+ free(new_ctx);
+ return ret;
+}
+
+int lttng_session_context_init(struct lttng_ctx **ctx)
+{
+ int ret;
+
+ ret = lttng_add_pthread_id_to_ctx(ctx);
+ if (ret) {
+ WARN("Cannot add context lttng_add_pthread_id_to_ctx");
+ goto error;
+ }
+ ret = lttng_add_vtid_to_ctx(ctx);
+ if (ret) {
+ WARN("Cannot add context lttng_add_vtid_to_ctx");
+ goto error;
+ }
+ ret = lttng_add_vpid_to_ctx(ctx);
+ if (ret) {
+ WARN("Cannot add context lttng_add_vpid_to_ctx");
+ goto error;
+ }
+ ret = lttng_add_procname_to_ctx(ctx);
+ if (ret) {
+ WARN("Cannot add context lttng_add_procname_to_ctx");
+ goto error;
+ }
+ ret = lttng_add_cpu_id_to_ctx(ctx);
+ if (ret) {
+ WARN("Cannot add context lttng_add_cpu_id_to_ctx");
+ goto error;
+ }
+ lttng_context_update(*ctx);
+ return 0;
+
+error:
+ lttng_destroy_context(*ctx);
+ return ret;
+}
+
+/* For backward compatibility. Leave those exported symbols in place. */
+struct lttng_ctx *lttng_static_ctx;
+
+void lttng_context_init(void)
+{
+}
+
+void lttng_context_exit(void)
+{
+}