src/bin/lttng-sessiond/notification-thread.h

   1 /*
   2  * Copyright (C) 2017 Jérémie Galarneau <jeremie.galarneau@efficios.com>
   3  *
   4  * SPDX-License-Identifier: GPL-2.0-only
   5  *
   6  */
   7
   8 #ifndef NOTIFICATION_THREAD_H
   9 #define NOTIFICATION_THREAD_H
  10
  11 #include "action-executor.h"
  12 #include "thread.h"
  13 #include <common/compat/poll.h>
  14 #include <common/hashtable/hashtable.h>
  15 #include <common/pipe.h>
  16 #include <lttng/trigger/trigger.h>
  17 #include <lttng/domain.h>
  18 #include <pthread.h>
  19 #include <semaphore.h>
  20 #include <urcu.h>
  21 #include <urcu/list.h>
  22 #include <urcu/rculfhash.h>
  23
  24 typedef uint64_t notification_client_id;
  25
  26 /*
  27  * The notification thread holds no ownership of the tracer event source pipe
  28  * file descriptor. The tracer management logic must remove the event source
  29  * from the notification thread (see external commands) before releasing
  30  * this file descriptor.
  31  */
  32 struct notification_event_tracer_event_source_element {
  33         int fd;
  34         /*
  35          * A tracer event source can be removed from the notification thread's
  36          * poll set before the end of its lifetime (for instance, when an error
  37          * or hang-up is detected on its file descriptor). This is done to
  38          * allow the notification thread to ignore follow-up events on this
  39          * file descriptors.
  40          *
  41          * Under such circumstances, the notification thread still expects
  42          * the normal clean-up to occur through the 'REMOVE_TRACER_EVENT_SOURCE'
  43          * command.
  44          */
  45         bool is_fd_in_poll_set;
  46         enum lttng_domain_type domain;
  47         struct cds_list_head node;
  48 };
  49
  50 struct notification_trigger_tokens_ht_element {
  51         uint64_t token;
  52         /* Weak reference to the trigger. */
  53         struct lttng_trigger *trigger;
  54         struct cds_lfht_node node;
  55         /* call_rcu delayed reclaim. */
  56         struct rcu_head rcu_node;
  57 };
  58
  59 struct notification_thread_handle {
  60         /*
  61          * Queue of struct notification command.
  62          * event_pipe must be WRITE(2) to signal that a new command
  63          * has been enqueued.
  64          */
  65         struct {
  66                 struct lttng_pipe *event_pipe;
  67                 struct cds_list_head list;
  68                 pthread_mutex_t lock;
  69         } cmd_queue;
  70         /*
  71          * Read side of pipes used to receive channel status info collected
  72          * by the various consumer daemons.
  73          */
  74         struct {
  75                 int ust32_consumer;
  76                 int ust64_consumer;
  77                 int kernel_consumer;
  78         } channel_monitoring_pipes;
  79         /* Used to wait for the launch of the notification thread. */
  80         sem_t ready;
  81 };
  82
  83 /**
  84  * This thread maintains an internal state associating clients and triggers.
  85  *
  86  * In order to speed-up and simplify queries, hash tables providing the
  87  * following associations are maintained:
  88  *
  89  *   - client_socket_ht: associate a client's socket (fd) to its
  90  *             "struct notification_client".
  91  *             This hash table owns the "struct notification_client" which must
  92  *             thus be disposed-of on removal from the hash table.
  93  *
  94  *   - client_id_ht: associate a client's id to its "struct notification_client"
  95  *             This hash table holds a _weak_ reference to the
  96  *             "struct notification_client".
  97  *
  98  *   - channel_triggers_ht:
  99  *             associates a channel key to a list of
 100  *             struct lttng_trigger_list_nodes. The triggers in this list are
 101  *             those that have conditions that apply to a particular channel.
 102  *             A channel entry is only created when a channel is added; the
 103  *             list of triggers applying to such a channel is built at that
 104  *             moment.
 105  *             This hash table owns the list, but not the triggers themselves.
 106  *
 107  *   - session_triggers_ht:
 108  *             associates a session name to a list of
 109  *             struct lttng_trigger_list_nodes. The triggers in this list are
 110  *             those that have conditions that apply to a particular session.
 111  *             A session entry is only created when a session is created; the
 112  *             list of triggers applying to this new session is built at that
 113  *             moment. This happens at the time of creation of a session_info.
 114  *             Likewise, the list is destroyed at the time of the session_info's
 115  *             destruction.
 116  *
 117  *   - channel_state_ht:
 118  *             associates a pair (channel key, channel domain) to its last
 119  *             sampled state received from the consumer daemon
 120  *             (struct channel_state).
 121  *             This previous sample is kept to implement edge-triggered
 122  *             conditions as we need to detect the state transitions.
 123  *             This hash table owns the channel state.
 124  *
 125  *   - notification_trigger_clients_ht:
 126  *             associates notification-emitting triggers to clients
 127  *             (struct notification_client_list) subscribed to those
 128  *             conditions.
 129  *             The condition's hash and match functions are used directly since
 130  *             all triggers in this hash table have the "notify" action.
 131  *             This hash table holds no ownership.
 132  *
 133  *   - channels_ht:
 134  *             associates a channel_key to a struct channel_info. The hash table
 135  *             holds the ownership of the struct channel_info.
 136  *
 137  *   - sessions_ht:
 138  *             associates a session_name (hash) to a struct session_info. The
 139  *             hash table holds no ownership of the struct session_info;
 140  *             the session_info structure is owned by the session's various
 141  *             channels through their struct channel_info (ref-counting is used).
 142  *
 143  *   - triggers_ht:
 144  *             associates a trigger to a struct lttng_trigger_ht_element.
 145  *             The hash table holds the ownership of the
 146  *             lttng_trigger_ht_elements along with the triggers themselves.
 147  *   - triggers_by_name_uid_ht:
 148  *             associates a trigger (name, uid) tuple to
 149  *             a struct lttng_trigger_ht_element.
 150  *             The hash table does not hold any ownership and is used strictly
 151  *             for lookup on registration.
 152  *   - tracer_event_sources_list:
 153  *             A list of tracer event source (read side fd) of type
 154 *              struct notification_event_tracer_event_source_element.
 155 *
 156  *
 157  * The thread reacts to the following internal events:
 158  *   1) creation of a tracing channel,
 159  *   2) destruction of a tracing channel,
 160  *   3) registration of a trigger,
 161  *   4) unregistration of a trigger,
 162  *   5) reception of a channel monitor sample from the consumer daemon,
 163  *   6) Session rotation ongoing,
 164  *   7) Session rotation completed,
 165  *   8) registration of a tracer event source,
 166  *   9) unregistration of a tracer event source,
 167  *
 168  * Events specific to notification-emitting triggers:
 169  *   9) connection of a notification client,
 170  *   10) disconnection of a notification client,
 171  *   11) subscription of a client to a conditions' notifications,
 172  *   12) unsubscription of a client from a conditions' notifications,
 173  *
 174  *
 175  * 1) Creation of a tracing channel
 176  *    - notification_trigger_clients_ht is traversed to identify
 177  *      triggers which apply to this new channel,
 178  *      - triggers identified are added to the channel_triggers_ht.
 179  *    - add channel to channels_ht
 180  *    - if it is the first channel of a session, a session_info is created and
 181  *      added to the sessions_ht. A list of the triggers associated with that
 182  *      session is built, and it is added to session_triggers_ht.
 183  *
 184  * 2) Destruction of a tracing channel
 185  *    - remove entry from channel_triggers_ht, releasing the list wrapper and
 186  *      elements,
 187  *    - remove entry from the channel_state_ht.
 188  *    - remove channel from channels_ht
 189  *    - if it was the last known channel of a session, the session_info
 190  *      structure is torndown, which in return destroys the list of triggers
 191  *      applying to that session.
 192  *
 193  * 3) Registration of a trigger
 194  *    - if the trigger's action is of type "notify",
 195  *      - traverse the list of conditions of every client to build a list of
 196  *        clients which have to be notified when this trigger's condition is met,
 197  *        - add list of clients (even if it is empty) to the
 198  *          notification_trigger_clients_ht,
 199  *    - add trigger to channel_triggers_ht (if applicable),
 200  *    - add trigger to session_triggers_ht (if applicable),
 201  *    - add trigger to triggers_by_name_uid_ht
 202  *    - add trigger to triggers_ht
 203  *    - evaluate the trigger's condition right away to react if that condition
 204  *      is true from the beginning.
 205  *
 206  * 4) Unregistration of a trigger
 207  *    - if the trigger's action is of type "notify",
 208  *      - remove the trigger from the notification_trigger_clients_ht,
 209  *    - remove trigger from channel_triggers_ht (if applicable),
 210  *    - remove trigger from session_triggers_ht (if applicable),
 211  *    - remove trigger from triggers_by_name_uid_ht
 212  *    - remove trigger from triggers_ht
 213  *
 214  * 5) Reception of a channel monitor sample from the consumer daemon
 215  *    - evaluate the conditions associated with the triggers found in
 216  *      the channel_triggers_ht,
 217  *      - if a condition evaluates to "true" and the condition is of type
 218  *        "notify", query the notification_trigger_clients_ht and send
 219  *        a notification to the clients.
 220  *
 221  * 6) Session rotation ongoing
 222  *
 223  * 7) Session rotation completed
 224  *
 225  * 8) Registration of a tracer event source
 226  *    - Add the tracer event source of the application to
 227  *      tracer_event_sources_list,
 228  *    - Add the trace event source to the pollset.
 229  *
 230  * 8) Unregistration of a tracer event source
 231  *    - Remove the tracer event source of the application from
 232  *      tracer_event_sources_list,
 233  *    - Remove the trace event source from the pollset.
 234  *
 235  * 10) Connection of a client
 236  *    - add client socket to the client_socket_ht,
 237  *    - add client socket to the client_id_ht.
 238  *
 239  * 11) Disconnection of a client
 240  *    - remove client socket from the client_id_ht,
 241  *    - remove client socket from the client_socket_ht,
 242  *    - traverse all conditions to which the client is subscribed and remove
 243  *      the client from the notification_trigger_clients_ht.
 244  *
 245  * 12) Subscription of a client to a condition's notifications
 246  *    - Add the condition to the client's list of subscribed conditions,
 247  *    - Look-up notification_trigger_clients_ht and add the client to
 248  *      list of clients.
 249  *    - Evaluate the condition for the client that subscribed if the trigger
 250  *      was already registered.
 251  *
 252  * 13) Unsubscription of a client to a condition's notifications
 253  *    - Remove the condition from the client's list of subscribed conditions,
 254  *    - Look-up notification_trigger_clients_ht and remove the client
 255  *      from the list of clients.
 256  */
 257 struct notification_thread_state {
 258         int notification_channel_socket;
 259         struct lttng_poll_event events;
 260         struct cds_lfht *client_socket_ht;
 261         struct cds_lfht *client_id_ht;
 262         struct cds_lfht *channel_triggers_ht;
 263         struct cds_lfht *session_triggers_ht;
 264         struct cds_lfht *channel_state_ht;
 265         struct cds_lfht *notification_trigger_clients_ht;
 266         struct cds_lfht *channels_ht;
 267         struct cds_lfht *sessions_ht;
 268         struct cds_lfht *triggers_ht;
 269         struct cds_lfht *triggers_by_name_uid_ht;
 270         struct cds_lfht *trigger_tokens_ht;
 271         struct {
 272                 uint64_t next_tracer_token;
 273                 uint64_t name_offset;
 274         } trigger_id;
 275         /*
 276          * Read side of the pipes used to receive tracer events. As their name
 277          * implies, tracer event source activity originate from either
 278          * registered applications (user space tracer) or from the kernel
 279          * tracer.
 280          *
 281          * The list is not protected by a lock since add and remove operations
 282          * are currently done only by the notification thread through in
 283          * response to blocking commands.
 284          */
 285         struct cds_list_head tracer_event_sources_list;
 286         notification_client_id next_notification_client_id;
 287         struct action_executor *executor;
 288
 289         /*
 290          * Indicates the thread to break for the poll event processing loop and
 291          * call _poll_wait() again.
 292          *
 293          * This is necessary because some events on one fd might trigger the
 294          * consumption of another fd.
 295          * For example, a single _poll_wait() call can return notification
 296          * thread commands and events from the tracer event source (event
 297          * notifier).
 298          * Picture a scenario where we receive two events:
 299          *  the first one is a _REMOVE_TRACER_EVENT_SOURCE command, and
 300          *  the second is an POLLIN on the tracer event source fd.
 301          *
 302          * The _REMOVE_TRACER_EVENT_SOURCE will read all the data of the
 303          * removed tracer event source.
 304          *
 305          * The second event is now invalid has we consumed all the data for
 306          * which we received the POLLIN.
 307          *
 308          * For this reason, we need to break for the event processing loop and
 309          * call _poll_wait() again to get a clean view of the activity on the
 310          * fds.
 311          */
 312         bool restart_poll;
 313 };
 314
 315 /* notification_thread_data takes ownership of the channel monitor pipes. */
 316 struct notification_thread_handle *notification_thread_handle_create(
 317                 struct lttng_pipe *ust32_channel_monitor_pipe,
 318                 struct lttng_pipe *ust64_channel_monitor_pipe,
 319                 struct lttng_pipe *kernel_channel_monitor_pipe);
 320 void notification_thread_handle_destroy(
 321                 struct notification_thread_handle *handle);
 322 struct lttng_thread *launch_notification_thread(
 323                 struct notification_thread_handle *handle);
 324
 325 #endif /* NOTIFICATION_THREAD_H */