4 * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
5 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * This library is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; only
10 * version 2.1 of the License.
12 * This library is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with this library; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #include <sys/types.h>
24 #include <sys/socket.h>
27 #include <sys/types.h>
33 #include <semaphore.h>
37 #include <urcu/uatomic.h>
38 #include <urcu/futex.h>
39 #include <urcu/compiler.h>
41 #include <lttng/ust-events.h>
42 #include <lttng/ust-abi.h>
43 #include <lttng/ust.h>
44 #include <lttng/ust-error.h>
45 #include <lttng/ust-ctl.h>
46 #include <urcu/tls-compat.h>
48 #include <usterr-signal-safe.h>
50 #include "tracepoint-internal.h"
51 #include "lttng-tracer-core.h"
53 #include "../libringbuffer/tlsfixup.h"
56 * Has lttng ust comm constructor been called ?
58 static int initialized
;
61 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
62 * Held when handling a command, also held by fork() to deal with
63 * removal of threads, and by exit path.
66 /* Should the ust comm thread quit ? */
67 static int lttng_ust_comm_should_quit
;
70 * Wait for either of these before continuing to the main
72 * - the register_done message from sessiond daemon
73 * (will let the sessiond daemon enable sessions before main
75 * - sessiond daemon is not reachable.
76 * - timeout (ensuring applications are resilient to session
79 static sem_t constructor_wait
;
81 * Doing this for both the global and local sessiond.
83 static int sem_count
= { 2 };
86 * Counting nesting within lttng-ust. Used to ensure that calling fork()
87 * from liblttng-ust does not execute the pre/post fork handlers.
89 static DEFINE_URCU_TLS(int, lttng_ust_nest_count
);
92 * Info about socket and associated listener thread.
96 pthread_t ust_listener
; /* listener thread */
98 int constructor_sem_posted
;
103 char sock_path
[PATH_MAX
];
107 char wait_shm_path
[PATH_MAX
];
111 /* Socket from app (connect) to session daemon (listen) for communication */
112 struct sock_info global_apps
= {
120 .sock_path
= LTTNG_DEFAULT_RUNDIR
"/" LTTNG_UST_SOCK_FILENAME
,
124 .wait_shm_path
= "/" LTTNG_UST_WAIT_FILENAME
,
127 /* TODO: allow global_apps_sock_path override */
129 struct sock_info local_apps
= {
133 .allowed
= 0, /* Check setuid bit first */
140 static int wait_poll_fallback
;
142 static const char *cmd_name_mapping
[] = {
143 [ LTTNG_UST_RELEASE
] = "Release",
144 [ LTTNG_UST_SESSION
] = "Create Session",
145 [ LTTNG_UST_TRACER_VERSION
] = "Get Tracer Version",
147 [ LTTNG_UST_TRACEPOINT_LIST
] = "Create Tracepoint List",
148 [ LTTNG_UST_WAIT_QUIESCENT
] = "Wait for Quiescent State",
149 [ LTTNG_UST_REGISTER_DONE
] = "Registration Done",
150 [ LTTNG_UST_TRACEPOINT_FIELD_LIST
] = "Create Tracepoint Field List",
152 /* Session FD commands */
153 [ LTTNG_UST_CHANNEL
] = "Create Channel",
154 [ LTTNG_UST_SESSION_START
] = "Start Session",
155 [ LTTNG_UST_SESSION_STOP
] = "Stop Session",
157 /* Channel FD commands */
158 [ LTTNG_UST_STREAM
] = "Create Stream",
159 [ LTTNG_UST_EVENT
] = "Create Event",
161 /* Event and Channel FD commands */
162 [ LTTNG_UST_CONTEXT
] = "Create Context",
163 [ LTTNG_UST_FLUSH_BUFFER
] = "Flush Buffer",
165 /* Event, Channel and Session commands */
166 [ LTTNG_UST_ENABLE
] = "Enable",
167 [ LTTNG_UST_DISABLE
] = "Disable",
169 /* Tracepoint list commands */
170 [ LTTNG_UST_TRACEPOINT_LIST_GET
] = "List Next Tracepoint",
171 [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET
] = "List Next Tracepoint Field",
173 /* Event FD commands */
174 [ LTTNG_UST_FILTER
] = "Create Filter",
175 [ LTTNG_UST_EXCLUSION
] = "Add exclusions to event",
178 static const char *str_timeout
;
179 static int got_timeout_env
;
181 extern void lttng_ring_buffer_client_overwrite_init(void);
182 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
183 extern void lttng_ring_buffer_client_discard_init(void);
184 extern void lttng_ring_buffer_client_discard_rt_init(void);
185 extern void lttng_ring_buffer_metadata_client_init(void);
186 extern void lttng_ring_buffer_client_overwrite_exit(void);
187 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
188 extern void lttng_ring_buffer_client_discard_exit(void);
189 extern void lttng_ring_buffer_client_discard_rt_exit(void);
190 extern void lttng_ring_buffer_metadata_client_exit(void);
193 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
197 const char *get_lttng_home_dir(void)
201 val
= (const char *) getenv("LTTNG_HOME");
205 return (const char *) getenv("HOME");
209 * Force a read (imply TLS fixup for dlopen) of TLS variables.
212 void lttng_fixup_nest_count_tls(void)
214 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count
)));
217 int lttng_get_notify_socket(void *owner
)
219 struct sock_info
*info
= owner
;
221 return info
->notify_socket
;
225 void print_cmd(int cmd
, int handle
)
227 const char *cmd_name
= "Unknown";
229 if (cmd
>= 0 && cmd
< LTTNG_ARRAY_SIZE(cmd_name_mapping
)
230 && cmd_name_mapping
[cmd
]) {
231 cmd_name
= cmd_name_mapping
[cmd
];
233 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
235 lttng_ust_obj_get_name(handle
), handle
);
239 int setup_local_apps(void)
241 const char *home_dir
;
246 * Disallow per-user tracing for setuid binaries.
248 if (uid
!= geteuid()) {
249 assert(local_apps
.allowed
== 0);
252 home_dir
= get_lttng_home_dir();
254 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
255 assert(local_apps
.allowed
== 0);
258 local_apps
.allowed
= 1;
259 snprintf(local_apps
.sock_path
, PATH_MAX
, "%s/%s/%s",
261 LTTNG_DEFAULT_HOME_RUNDIR
,
262 LTTNG_UST_SOCK_FILENAME
);
263 snprintf(local_apps
.wait_shm_path
, PATH_MAX
, "/%s-%u",
264 LTTNG_UST_WAIT_FILENAME
,
270 * Get notify_sock timeout, in ms.
271 * -1: don't wait. 0: wait forever. >0: timeout, in ms.
274 long get_timeout(void)
276 long constructor_delay_ms
= LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS
;
278 if (!got_timeout_env
) {
279 str_timeout
= getenv("LTTNG_UST_REGISTER_TIMEOUT");
283 constructor_delay_ms
= strtol(str_timeout
, NULL
, 10);
284 return constructor_delay_ms
;
288 long get_notify_sock_timeout(void)
290 return get_timeout();
294 * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
297 int get_constructor_timeout(struct timespec
*constructor_timeout
)
299 long constructor_delay_ms
;
302 constructor_delay_ms
= get_timeout();
304 switch (constructor_delay_ms
) {
305 case -1:/* fall-through */
307 return constructor_delay_ms
;
313 * If we are unable to find the current time, don't wait.
315 ret
= clock_gettime(CLOCK_REALTIME
, constructor_timeout
);
319 constructor_timeout
->tv_sec
+= constructor_delay_ms
/ 1000UL;
320 constructor_timeout
->tv_nsec
+=
321 (constructor_delay_ms
% 1000UL) * 1000000UL;
322 if (constructor_timeout
->tv_nsec
>= 1000000000UL) {
323 constructor_timeout
->tv_sec
++;
324 constructor_timeout
->tv_nsec
-= 1000000000UL;
330 int register_to_sessiond(int socket
, enum ustctl_socket_type type
)
332 return ustcomm_send_reg_msg(socket
,
335 lttng_alignof(uint8_t) * CHAR_BIT
,
336 lttng_alignof(uint16_t) * CHAR_BIT
,
337 lttng_alignof(uint32_t) * CHAR_BIT
,
338 lttng_alignof(uint64_t) * CHAR_BIT
,
339 lttng_alignof(unsigned long) * CHAR_BIT
);
343 int send_reply(int sock
, struct ustcomm_ust_reply
*lur
)
347 len
= ustcomm_send_unix_sock(sock
, lur
, sizeof(*lur
));
350 DBG("message successfully sent");
353 if (len
== -ECONNRESET
) {
354 DBG("remote end closed connection");
359 DBG("incorrect message size: %zd", len
);
365 int handle_register_done(struct sock_info
*sock_info
)
369 if (sock_info
->constructor_sem_posted
)
371 sock_info
->constructor_sem_posted
= 1;
372 if (uatomic_read(&sem_count
) <= 0) {
375 ret
= uatomic_add_return(&sem_count
, -1);
377 ret
= sem_post(&constructor_wait
);
384 int handle_message(struct sock_info
*sock_info
,
385 int sock
, struct ustcomm_ust_msg
*lum
)
388 const struct lttng_ust_objd_ops
*ops
;
389 struct ustcomm_ust_reply lur
;
395 memset(&lur
, 0, sizeof(lur
));
397 if (lttng_ust_comm_should_quit
) {
398 ret
= -LTTNG_UST_ERR_EXITING
;
402 ops
= objd_ops(lum
->handle
);
409 case LTTNG_UST_REGISTER_DONE
:
410 if (lum
->handle
== LTTNG_UST_ROOT_HANDLE
)
411 ret
= handle_register_done(sock_info
);
415 case LTTNG_UST_RELEASE
:
416 if (lum
->handle
== LTTNG_UST_ROOT_HANDLE
)
419 ret
= lttng_ust_objd_unref(lum
->handle
, 1);
421 case LTTNG_UST_FILTER
:
423 /* Receive filter data */
424 struct lttng_ust_filter_bytecode_node
*bytecode
;
426 if (lum
->u
.filter
.data_size
> FILTER_BYTECODE_MAX_LEN
) {
427 ERR("Filter data size is too large: %u bytes",
428 lum
->u
.filter
.data_size
);
433 if (lum
->u
.filter
.reloc_offset
> lum
->u
.filter
.data_size
) {
434 ERR("Filter reloc offset %u is not within data",
435 lum
->u
.filter
.reloc_offset
);
440 bytecode
= zmalloc(sizeof(*bytecode
) + lum
->u
.filter
.data_size
);
445 len
= ustcomm_recv_unix_sock(sock
, bytecode
->bc
.data
,
446 lum
->u
.filter
.data_size
);
448 case 0: /* orderly shutdown */
453 if (len
== lum
->u
.filter
.data_size
) {
454 DBG("filter data received");
456 } else if (len
< 0) {
457 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
458 if (len
== -ECONNRESET
) {
459 ERR("%s remote end closed connection", sock_info
->name
);
468 DBG("incorrect filter data message size: %zd", len
);
474 bytecode
->bc
.len
= lum
->u
.filter
.data_size
;
475 bytecode
->bc
.reloc_offset
= lum
->u
.filter
.reloc_offset
;
476 bytecode
->bc
.seqnum
= lum
->u
.filter
.seqnum
;
478 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
479 (unsigned long) bytecode
,
484 /* don't free bytecode if everything went fine. */
491 case LTTNG_UST_EXCLUSION
:
493 /* Receive exclusion names */
494 struct lttng_ust_excluder_node
*node
;
497 count
= lum
->u
.exclusion
.count
;
499 /* There are no names to read */
503 node
= zmalloc(sizeof(*node
) +
504 count
* LTTNG_UST_SYM_NAME_LEN
);
509 node
->excluder
.count
= count
;
510 len
= ustcomm_recv_unix_sock(sock
, node
->excluder
.names
,
511 count
* LTTNG_UST_SYM_NAME_LEN
);
513 case 0: /* orderly shutdown */
518 if (len
== count
* LTTNG_UST_SYM_NAME_LEN
) {
519 DBG("Exclusion data received");
521 } else if (len
< 0) {
522 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
523 if (len
== -ECONNRESET
) {
524 ERR("%s remote end closed connection", sock_info
->name
);
533 DBG("Incorrect exclusion data message size: %zd", len
);
540 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
541 (unsigned long) node
,
546 /* Don't free exclusion data if everything went fine. */
553 case LTTNG_UST_CHANNEL
:
558 len
= ustcomm_recv_channel_from_sessiond(sock
,
559 &chan_data
, lum
->u
.channel
.len
,
562 case 0: /* orderly shutdown */
566 if (len
== lum
->u
.channel
.len
) {
567 DBG("channel data received");
569 } else if (len
< 0) {
570 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
571 if (len
== -ECONNRESET
) {
572 ERR("%s remote end closed connection", sock_info
->name
);
579 DBG("incorrect channel data message size: %zd", len
);
584 args
.channel
.chan_data
= chan_data
;
585 args
.channel
.wakeup_fd
= wakeup_fd
;
587 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
588 (unsigned long) &lum
->u
,
594 case LTTNG_UST_STREAM
:
596 /* Receive shm_fd, wakeup_fd */
597 ret
= ustcomm_recv_stream_from_sessiond(sock
,
600 &args
.stream
.wakeup_fd
);
605 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
606 (unsigned long) &lum
->u
,
614 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
615 (unsigned long) &lum
->u
,
623 lur
.handle
= lum
->handle
;
627 lur
.ret_code
= LTTNG_UST_OK
;
630 * Use -LTTNG_UST_ERR as wildcard for UST internal
631 * error that are not caused by the transport, except if
632 * we already have a more precise error message to
635 if (ret
> -LTTNG_UST_ERR
) {
636 /* Translate code to UST error. */
639 lur
.ret_code
= -LTTNG_UST_ERR_EXIST
;
642 lur
.ret_code
= -LTTNG_UST_ERR_INVAL
;
645 lur
.ret_code
= -LTTNG_UST_ERR_NOENT
;
648 lur
.ret_code
= -LTTNG_UST_ERR_PERM
;
651 lur
.ret_code
= -LTTNG_UST_ERR_NOSYS
;
654 lur
.ret_code
= -LTTNG_UST_ERR
;
663 case LTTNG_UST_TRACER_VERSION
:
664 lur
.u
.version
= lum
->u
.version
;
666 case LTTNG_UST_TRACEPOINT_LIST_GET
:
667 memcpy(&lur
.u
.tracepoint
, &lum
->u
.tracepoint
, sizeof(lur
.u
.tracepoint
));
671 DBG("Return value: %d", lur
.ret_val
);
672 ret
= send_reply(sock
, &lur
);
674 DBG("error sending reply");
679 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
682 if (lur
.ret_code
== LTTNG_UST_OK
) {
684 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET
:
685 len
= ustcomm_send_unix_sock(sock
,
686 &args
.field_list
.entry
,
687 sizeof(args
.field_list
.entry
));
692 if (len
!= sizeof(args
.field_list
.entry
)) {
705 void cleanup_sock_info(struct sock_info
*sock_info
, int exiting
)
709 if (sock_info
->root_handle
!= -1) {
710 ret
= lttng_ust_objd_unref(sock_info
->root_handle
, 1);
712 ERR("Error unref root handle");
714 sock_info
->root_handle
= -1;
716 sock_info
->constructor_sem_posted
= 0;
719 * wait_shm_mmap, socket and notify socket are used by listener
720 * threads outside of the ust lock, so we cannot tear them down
721 * ourselves, because we cannot join on these threads. Leave
722 * responsibility of cleaning up these resources to the OS
728 if (sock_info
->socket
!= -1) {
729 ret
= ustcomm_close_unix_sock(sock_info
->socket
);
731 ERR("Error closing ust cmd socket");
733 sock_info
->socket
= -1;
735 if (sock_info
->notify_socket
!= -1) {
736 ret
= ustcomm_close_unix_sock(sock_info
->notify_socket
);
738 ERR("Error closing ust notify socket");
740 sock_info
->notify_socket
= -1;
742 if (sock_info
->wait_shm_mmap
) {
743 ret
= munmap(sock_info
->wait_shm_mmap
, sysconf(_SC_PAGE_SIZE
));
745 ERR("Error unmapping wait shm");
747 sock_info
->wait_shm_mmap
= NULL
;
752 * Using fork to set umask in the child process (not multi-thread safe).
753 * We deal with the shm_open vs ftruncate race (happening when the
754 * sessiond owns the shm and does not let everybody modify it, to ensure
755 * safety against shm_unlink) by simply letting the mmap fail and
756 * retrying after a few seconds.
757 * For global shm, everybody has rw access to it until the sessiond
761 int get_wait_shm(struct sock_info
*sock_info
, size_t mmap_size
)
763 int wait_shm_fd
, ret
;
767 * Try to open read-only.
769 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
, O_RDONLY
, 0);
770 if (wait_shm_fd
>= 0) {
773 size_t bytes_read
= 0;
776 * Try to read the fd. If unable to do so, try opening
780 len
= read(wait_shm_fd
,
781 &((char *) &tmp_read
)[bytes_read
],
782 sizeof(tmp_read
) - bytes_read
);
786 } while ((len
< 0 && errno
== EINTR
)
787 || (len
> 0 && bytes_read
< sizeof(tmp_read
)));
788 if (bytes_read
!= sizeof(tmp_read
)) {
789 ret
= close(wait_shm_fd
);
791 ERR("close wait_shm_fd");
796 } else if (wait_shm_fd
< 0 && errno
!= ENOENT
) {
798 * Real-only open did not work, and it's not because the
799 * entry was not present. It's a failure that prohibits
802 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
808 * If the open failed because the file did not exist, or because
809 * the file was not truncated yet, try creating it ourself.
811 URCU_TLS(lttng_ust_nest_count
)++;
813 URCU_TLS(lttng_ust_nest_count
)--;
818 * Parent: wait for child to return, in which case the
819 * shared memory map will have been created.
822 if (pid
< 0 || !WIFEXITED(status
) || WEXITSTATUS(status
) != 0) {
827 * Try to open read-only again after creation.
829 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
, O_RDONLY
, 0);
830 if (wait_shm_fd
< 0) {
832 * Real-only open did not work. It's a failure
833 * that prohibits using shm.
835 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
839 } else if (pid
== 0) {
843 create_mode
= S_IRUSR
| S_IWUSR
| S_IRGRP
;
844 if (sock_info
->global
)
845 create_mode
|= S_IROTH
| S_IWGRP
| S_IWOTH
;
847 * We're alone in a child process, so we can modify the
848 * process-wide umask.
852 * Try creating shm (or get rw access).
853 * We don't do an exclusive open, because we allow other
854 * processes to create+ftruncate it concurrently.
856 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
,
857 O_RDWR
| O_CREAT
, create_mode
);
858 if (wait_shm_fd
>= 0) {
859 ret
= ftruncate(wait_shm_fd
, mmap_size
);
867 * For local shm, we need to have rw access to accept
868 * opening it: this means the local sessiond will be
869 * able to wake us up. For global shm, we open it even
870 * if rw access is not granted, because the root.root
871 * sessiond will be able to override all rights and wake
874 if (!sock_info
->global
&& errno
!= EACCES
) {
875 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
879 * The shm exists, but we cannot open it RW. Report
887 if (wait_shm_fd
>= 0 && !sock_info
->global
) {
891 * Ensure that our user is the owner of the shm file for
892 * local shm. If we do not own the file, it means our
893 * sessiond will not have access to wake us up (there is
894 * probably a rogue process trying to fake our
895 * sessiond). Fallback to polling method in this case.
897 ret
= fstat(wait_shm_fd
, &statbuf
);
902 if (statbuf
.st_uid
!= getuid())
908 ret
= close(wait_shm_fd
);
910 PERROR("Error closing fd");
916 char *get_map_shm(struct sock_info
*sock_info
)
918 size_t mmap_size
= sysconf(_SC_PAGE_SIZE
);
919 int wait_shm_fd
, ret
;
922 wait_shm_fd
= get_wait_shm(sock_info
, mmap_size
);
923 if (wait_shm_fd
< 0) {
926 wait_shm_mmap
= mmap(NULL
, mmap_size
, PROT_READ
,
927 MAP_SHARED
, wait_shm_fd
, 0);
928 /* close shm fd immediately after taking the mmap reference */
929 ret
= close(wait_shm_fd
);
931 PERROR("Error closing fd");
933 if (wait_shm_mmap
== MAP_FAILED
) {
934 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
937 return wait_shm_mmap
;
944 void wait_for_sessiond(struct sock_info
*sock_info
)
949 if (lttng_ust_comm_should_quit
) {
952 if (wait_poll_fallback
) {
955 if (!sock_info
->wait_shm_mmap
) {
956 sock_info
->wait_shm_mmap
= get_map_shm(sock_info
);
957 if (!sock_info
->wait_shm_mmap
)
962 DBG("Waiting for %s apps sessiond", sock_info
->name
);
963 /* Wait for futex wakeup */
964 if (uatomic_read((int32_t *) sock_info
->wait_shm_mmap
) == 0) {
965 ret
= futex_async((int32_t *) sock_info
->wait_shm_mmap
,
966 FUTEX_WAIT
, 0, NULL
, NULL
, 0);
968 if (errno
== EFAULT
) {
969 wait_poll_fallback
= 1;
971 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
972 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
973 "Please upgrade your kernel "
974 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
975 "mainline). LTTng-UST will use polling mode fallback.");
993 * This thread does not allocate any resource, except within
994 * handle_message, within mutex protection. This mutex protects against
996 * The other moment it allocates resources is at socket connection, which
997 * is also protected by the mutex.
1000 void *ust_listener_thread(void *arg
)
1002 struct sock_info
*sock_info
= arg
;
1003 int sock
, ret
, prev_connect_failed
= 0, has_waited
= 0;
1006 /* Restart trying to connect to the session daemon */
1008 if (prev_connect_failed
) {
1009 /* Wait for sessiond availability with pipe */
1010 wait_for_sessiond(sock_info
);
1014 * Sleep for 5 seconds before retrying after a
1015 * sequence of failure / wait / failure. This
1016 * deals with a killed or broken session daemon.
1021 prev_connect_failed
= 0;
1024 if (sock_info
->socket
!= -1) {
1025 ret
= ustcomm_close_unix_sock(sock_info
->socket
);
1027 ERR("Error closing %s ust cmd socket",
1030 sock_info
->socket
= -1;
1032 if (sock_info
->notify_socket
!= -1) {
1033 ret
= ustcomm_close_unix_sock(sock_info
->notify_socket
);
1035 ERR("Error closing %s ust notify socket",
1038 sock_info
->notify_socket
= -1;
1042 * Register. We need to perform both connect and sending
1043 * registration message before doing the next connect otherwise
1044 * we may reach unix socket connect queue max limits and block
1045 * on the 2nd connect while the session daemon is awaiting the
1046 * first connect registration message.
1048 /* Connect cmd socket */
1049 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
);
1051 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
1052 prev_connect_failed
= 1;
1056 if (lttng_ust_comm_should_quit
) {
1061 * If we cannot find the sessiond daemon, don't delay
1062 * constructor execution.
1064 ret
= handle_register_done(sock_info
);
1069 sock_info
->socket
= ret
;
1073 if (lttng_ust_comm_should_quit
) {
1078 * Create only one root handle per listener thread for the whole
1079 * process lifetime, so we ensure we get ID which is statically
1080 * assigned to the root handle.
1082 if (sock_info
->root_handle
== -1) {
1083 ret
= lttng_abi_create_root_handle();
1085 ERR("Error creating root handle");
1088 sock_info
->root_handle
= ret
;
1091 ret
= register_to_sessiond(sock_info
->socket
, USTCTL_SOCKET_CMD
);
1093 ERR("Error registering to %s ust cmd socket",
1095 prev_connect_failed
= 1;
1097 * If we cannot register to the sessiond daemon, don't
1098 * delay constructor execution.
1100 ret
= handle_register_done(sock_info
);
1108 /* Connect notify socket */
1109 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
);
1111 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
1112 prev_connect_failed
= 1;
1116 if (lttng_ust_comm_should_quit
) {
1121 * If we cannot find the sessiond daemon, don't delay
1122 * constructor execution.
1124 ret
= handle_register_done(sock_info
);
1129 sock_info
->notify_socket
= ret
;
1131 timeout
= get_notify_sock_timeout();
1134 * Give at least 10ms to sessiond to reply to
1139 ret
= ustcomm_setsockopt_rcv_timeout(sock_info
->notify_socket
,
1142 WARN("Error setting socket receive timeout");
1144 ret
= ustcomm_setsockopt_snd_timeout(sock_info
->notify_socket
,
1147 WARN("Error setting socket send timeout");
1149 } else if (timeout
< -1) {
1150 WARN("Unsupported timeout value %ld", timeout
);
1155 if (lttng_ust_comm_should_quit
) {
1159 ret
= register_to_sessiond(sock_info
->notify_socket
,
1160 USTCTL_SOCKET_NOTIFY
);
1162 ERR("Error registering to %s ust notify socket",
1164 prev_connect_failed
= 1;
1166 * If we cannot register to the sessiond daemon, don't
1167 * delay constructor execution.
1169 ret
= handle_register_done(sock_info
);
1174 sock
= sock_info
->socket
;
1180 struct ustcomm_ust_msg lum
;
1182 len
= ustcomm_recv_unix_sock(sock
, &lum
, sizeof(lum
));
1184 case 0: /* orderly shutdown */
1185 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info
->name
);
1187 if (lttng_ust_comm_should_quit
) {
1191 * Either sessiond has shutdown or refused us by closing the socket.
1192 * In either case, we don't want to delay construction execution,
1193 * and we need to wait before retry.
1195 prev_connect_failed
= 1;
1197 * If we cannot register to the sessiond daemon, don't
1198 * delay constructor execution.
1200 ret
= handle_register_done(sock_info
);
1205 print_cmd(lum
.cmd
, lum
.handle
);
1206 ret
= handle_message(sock_info
, sock
, &lum
);
1208 ERR("Error handling message for %s socket", sock_info
->name
);
1213 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1215 DBG("incorrect message size (%s socket): %zd", sock_info
->name
, len
);
1217 if (len
== -ECONNRESET
) {
1218 DBG("%s remote end closed connection", sock_info
->name
);
1227 if (lttng_ust_comm_should_quit
) {
1230 /* Cleanup socket handles before trying to reconnect */
1231 lttng_ust_objd_table_owner_cleanup(sock_info
);
1233 goto restart
; /* try to reconnect */
1236 sock_info
->thread_active
= 0;
1242 * sessiond monitoring thread: monitor presence of global and per-user
1243 * sessiond by polling the application common named pipe.
1245 void __attribute__((constructor
)) lttng_ust_init(void)
1247 struct timespec constructor_timeout
;
1248 sigset_t sig_all_blocked
, orig_parent_mask
;
1249 pthread_attr_t thread_attr
;
1253 if (uatomic_xchg(&initialized
, 1) == 1)
1257 * Fixup interdependency between TLS fixup mutex (which happens
1258 * to be the dynamic linker mutex) and ust_lock, taken within
1261 lttng_fixup_ringbuffer_tls();
1262 lttng_fixup_vtid_tls();
1263 lttng_fixup_nest_count_tls();
1264 lttng_fixup_procname_tls();
1267 * We want precise control over the order in which we construct
1268 * our sub-libraries vs starting to receive commands from
1269 * sessiond (otherwise leading to errors when trying to create
1270 * sessiond before the init functions are completed).
1274 lttng_ring_buffer_metadata_client_init();
1275 lttng_ring_buffer_client_overwrite_init();
1276 lttng_ring_buffer_client_overwrite_rt_init();
1277 lttng_ring_buffer_client_discard_init();
1278 lttng_ring_buffer_client_discard_rt_init();
1279 lttng_context_init();
1281 timeout_mode
= get_constructor_timeout(&constructor_timeout
);
1283 ret
= sem_init(&constructor_wait
, 0, 0);
1286 ret
= setup_local_apps();
1288 DBG("local apps setup returned %d", ret
);
1291 /* A new thread created by pthread_create inherits the signal mask
1292 * from the parent. To avoid any signal being received by the
1293 * listener thread, we block all signals temporarily in the parent,
1294 * while we create the listener thread.
1296 sigfillset(&sig_all_blocked
);
1297 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_parent_mask
);
1299 ERR("pthread_sigmask: %s", strerror(ret
));
1302 ret
= pthread_attr_init(&thread_attr
);
1304 ERR("pthread_attr_init: %s", strerror(ret
));
1306 ret
= pthread_attr_setdetachstate(&thread_attr
, PTHREAD_CREATE_DETACHED
);
1308 ERR("pthread_attr_setdetachstate: %s", strerror(ret
));
1312 ret
= pthread_create(&global_apps
.ust_listener
, &thread_attr
,
1313 ust_listener_thread
, &global_apps
);
1315 ERR("pthread_create global: %s", strerror(ret
));
1317 global_apps
.thread_active
= 1;
1320 if (local_apps
.allowed
) {
1322 ret
= pthread_create(&local_apps
.ust_listener
, &thread_attr
,
1323 ust_listener_thread
, &local_apps
);
1325 ERR("pthread_create local: %s", strerror(ret
));
1327 local_apps
.thread_active
= 1;
1330 handle_register_done(&local_apps
);
1332 ret
= pthread_attr_destroy(&thread_attr
);
1334 ERR("pthread_attr_destroy: %s", strerror(ret
));
1337 /* Restore original signal mask in parent */
1338 ret
= pthread_sigmask(SIG_SETMASK
, &orig_parent_mask
, NULL
);
1340 ERR("pthread_sigmask: %s", strerror(ret
));
1343 switch (timeout_mode
) {
1344 case 1: /* timeout wait */
1346 ret
= sem_timedwait(&constructor_wait
,
1347 &constructor_timeout
);
1348 } while (ret
< 0 && errno
== EINTR
);
1349 if (ret
< 0 && errno
== ETIMEDOUT
) {
1350 ERR("Timed out waiting for lttng-sessiond");
1355 case -1:/* wait forever */
1357 ret
= sem_wait(&constructor_wait
);
1358 } while (ret
< 0 && errno
== EINTR
);
1361 case 0: /* no timeout */
1367 void lttng_ust_cleanup(int exiting
)
1369 cleanup_sock_info(&global_apps
, exiting
);
1370 if (local_apps
.allowed
) {
1371 cleanup_sock_info(&local_apps
, exiting
);
1374 * The teardown in this function all affect data structures
1375 * accessed under the UST lock by the listener thread. This
1376 * lock, along with the lttng_ust_comm_should_quit flag, ensure
1377 * that none of these threads are accessing this data at this
1380 lttng_ust_abi_exit();
1381 lttng_ust_events_exit();
1382 lttng_context_exit();
1383 lttng_ring_buffer_client_discard_rt_exit();
1384 lttng_ring_buffer_client_discard_exit();
1385 lttng_ring_buffer_client_overwrite_rt_exit();
1386 lttng_ring_buffer_client_overwrite_exit();
1387 lttng_ring_buffer_metadata_client_exit();
1390 /* Reinitialize values for fork */
1392 lttng_ust_comm_should_quit
= 0;
1397 void __attribute__((destructor
)) lttng_ust_exit(void)
1402 * Using pthread_cancel here because:
1403 * A) we don't want to hang application teardown.
1404 * B) the thread is not allocating any resource.
1408 * Require the communication thread to quit. Synchronize with
1409 * mutexes to ensure it is not in a mutex critical section when
1410 * pthread_cancel is later called.
1413 lttng_ust_comm_should_quit
= 1;
1415 /* cancel threads */
1416 if (global_apps
.thread_active
) {
1417 ret
= pthread_cancel(global_apps
.ust_listener
);
1419 ERR("Error cancelling global ust listener thread: %s",
1422 global_apps
.thread_active
= 0;
1425 if (local_apps
.thread_active
) {
1426 ret
= pthread_cancel(local_apps
.ust_listener
);
1428 ERR("Error cancelling local ust listener thread: %s",
1431 local_apps
.thread_active
= 0;
1437 * Do NOT join threads: use of sys_futex makes it impossible to
1438 * join the threads without using async-cancel, but async-cancel
1439 * is delivered by a signal, which could hit the target thread
1440 * anywhere in its code path, including while the ust_lock() is
1441 * held, causing a deadlock for the other thread. Let the OS
1442 * cleanup the threads if there are stalled in a syscall.
1444 lttng_ust_cleanup(1);
1448 * We exclude the worker threads across fork and clone (except
1449 * CLONE_VM), because these system calls only keep the forking thread
1450 * running in the child. Therefore, we don't want to call fork or clone
1451 * in the middle of an tracepoint or ust tracing state modification.
1452 * Holding this mutex protects these structures across fork and clone.
1454 void ust_before_fork(sigset_t
*save_sigset
)
1457 * Disable signals. This is to avoid that the child intervenes
1458 * before it is properly setup for tracing. It is safer to
1459 * disable all signals, because then we know we are not breaking
1460 * anything by restoring the original mask.
1465 if (URCU_TLS(lttng_ust_nest_count
))
1467 /* Disable signals */
1468 sigfillset(&all_sigs
);
1469 ret
= sigprocmask(SIG_BLOCK
, &all_sigs
, save_sigset
);
1471 PERROR("sigprocmask");
1474 rcu_bp_before_fork();
1477 static void ust_after_fork_common(sigset_t
*restore_sigset
)
1481 DBG("process %d", getpid());
1483 /* Restore signals */
1484 ret
= sigprocmask(SIG_SETMASK
, restore_sigset
, NULL
);
1486 PERROR("sigprocmask");
1490 void ust_after_fork_parent(sigset_t
*restore_sigset
)
1492 if (URCU_TLS(lttng_ust_nest_count
))
1494 DBG("process %d", getpid());
1495 rcu_bp_after_fork_parent();
1496 /* Release mutexes and reenable signals */
1497 ust_after_fork_common(restore_sigset
);
1501 * After fork, in the child, we need to cleanup all the leftover state,
1502 * except the worker thread which already magically disappeared thanks
1503 * to the weird Linux fork semantics. After tyding up, we call
1504 * lttng_ust_init() again to start over as a new PID.
1506 * This is meant for forks() that have tracing in the child between the
1507 * fork and following exec call (if there is any).
1509 void ust_after_fork_child(sigset_t
*restore_sigset
)
1511 if (URCU_TLS(lttng_ust_nest_count
))
1513 DBG("process %d", getpid());
1514 /* Release urcu mutexes */
1515 rcu_bp_after_fork_child();
1516 lttng_ust_cleanup(0);
1517 lttng_context_vtid_reset();
1518 /* Release mutexes and reenable signals */
1519 ust_after_fork_common(restore_sigset
);