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",
177 static const char *str_timeout
;
178 static int got_timeout_env
;
180 extern void lttng_ring_buffer_client_overwrite_init(void);
181 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
182 extern void lttng_ring_buffer_client_discard_init(void);
183 extern void lttng_ring_buffer_client_discard_rt_init(void);
184 extern void lttng_ring_buffer_metadata_client_init(void);
185 extern void lttng_ring_buffer_client_overwrite_exit(void);
186 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
187 extern void lttng_ring_buffer_client_discard_exit(void);
188 extern void lttng_ring_buffer_client_discard_rt_exit(void);
189 extern void lttng_ring_buffer_metadata_client_exit(void);
192 * Force a read (imply TLS fixup for dlopen) of TLS variables.
195 void lttng_fixup_nest_count_tls(void)
197 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count
)));
200 int lttng_get_notify_socket(void *owner
)
202 struct sock_info
*info
= owner
;
204 return info
->notify_socket
;
208 void print_cmd(int cmd
, int handle
)
210 const char *cmd_name
= "Unknown";
212 if (cmd
>= 0 && cmd
< LTTNG_ARRAY_SIZE(cmd_name_mapping
)
213 && cmd_name_mapping
[cmd
]) {
214 cmd_name
= cmd_name_mapping
[cmd
];
216 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
218 lttng_ust_obj_get_name(handle
), handle
);
222 int setup_local_apps(void)
224 const char *home_dir
;
229 * Disallow per-user tracing for setuid binaries.
231 if (uid
!= geteuid()) {
232 assert(local_apps
.allowed
== 0);
235 home_dir
= (const char *) getenv("HOME");
237 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
238 assert(local_apps
.allowed
== 0);
241 local_apps
.allowed
= 1;
242 snprintf(local_apps
.sock_path
, PATH_MAX
, "%s/%s/%s",
244 LTTNG_DEFAULT_HOME_RUNDIR
,
245 LTTNG_UST_SOCK_FILENAME
);
246 snprintf(local_apps
.wait_shm_path
, PATH_MAX
, "/%s-%u",
247 LTTNG_UST_WAIT_FILENAME
,
253 * Get notify_sock timeout, in ms.
254 * -1: don't wait. 0: wait forever. >0: timeout, in ms.
257 long get_timeout(void)
259 long constructor_delay_ms
= LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS
;
261 if (!got_timeout_env
) {
262 str_timeout
= getenv("LTTNG_UST_REGISTER_TIMEOUT");
266 constructor_delay_ms
= strtol(str_timeout
, NULL
, 10);
267 return constructor_delay_ms
;
271 long get_notify_sock_timeout(void)
273 return get_timeout();
277 * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
280 int get_constructor_timeout(struct timespec
*constructor_timeout
)
282 long constructor_delay_ms
;
285 constructor_delay_ms
= get_timeout();
287 switch (constructor_delay_ms
) {
288 case -1:/* fall-through */
290 return constructor_delay_ms
;
296 * If we are unable to find the current time, don't wait.
298 ret
= clock_gettime(CLOCK_REALTIME
, constructor_timeout
);
302 constructor_timeout
->tv_sec
+= constructor_delay_ms
/ 1000UL;
303 constructor_timeout
->tv_nsec
+=
304 (constructor_delay_ms
% 1000UL) * 1000000UL;
305 if (constructor_timeout
->tv_nsec
>= 1000000000UL) {
306 constructor_timeout
->tv_sec
++;
307 constructor_timeout
->tv_nsec
-= 1000000000UL;
313 int register_to_sessiond(int socket
, enum ustctl_socket_type type
)
315 return ustcomm_send_reg_msg(socket
,
318 lttng_alignof(uint8_t) * CHAR_BIT
,
319 lttng_alignof(uint16_t) * CHAR_BIT
,
320 lttng_alignof(uint32_t) * CHAR_BIT
,
321 lttng_alignof(uint64_t) * CHAR_BIT
,
322 lttng_alignof(unsigned long) * CHAR_BIT
);
326 int send_reply(int sock
, struct ustcomm_ust_reply
*lur
)
330 len
= ustcomm_send_unix_sock(sock
, lur
, sizeof(*lur
));
333 DBG("message successfully sent");
336 if (len
== -ECONNRESET
) {
337 DBG("remote end closed connection");
342 DBG("incorrect message size: %zd", len
);
348 int handle_register_done(struct sock_info
*sock_info
)
352 if (sock_info
->constructor_sem_posted
)
354 sock_info
->constructor_sem_posted
= 1;
355 if (uatomic_read(&sem_count
) <= 0) {
358 ret
= uatomic_add_return(&sem_count
, -1);
360 ret
= sem_post(&constructor_wait
);
367 int handle_message(struct sock_info
*sock_info
,
368 int sock
, struct ustcomm_ust_msg
*lum
)
371 const struct lttng_ust_objd_ops
*ops
;
372 struct ustcomm_ust_reply lur
;
378 memset(&lur
, 0, sizeof(lur
));
380 if (lttng_ust_comm_should_quit
) {
381 ret
= -LTTNG_UST_ERR_EXITING
;
385 ops
= objd_ops(lum
->handle
);
392 case LTTNG_UST_REGISTER_DONE
:
393 if (lum
->handle
== LTTNG_UST_ROOT_HANDLE
)
394 ret
= handle_register_done(sock_info
);
398 case LTTNG_UST_RELEASE
:
399 if (lum
->handle
== LTTNG_UST_ROOT_HANDLE
)
402 ret
= lttng_ust_objd_unref(lum
->handle
, 1);
404 case LTTNG_UST_FILTER
:
406 /* Receive filter data */
407 struct lttng_ust_filter_bytecode_node
*bytecode
;
409 if (lum
->u
.filter
.data_size
> FILTER_BYTECODE_MAX_LEN
) {
410 ERR("Filter data size is too large: %u bytes",
411 lum
->u
.filter
.data_size
);
416 if (lum
->u
.filter
.reloc_offset
> lum
->u
.filter
.data_size
) {
417 ERR("Filter reloc offset %u is not within data",
418 lum
->u
.filter
.reloc_offset
);
423 bytecode
= zmalloc(sizeof(*bytecode
) + lum
->u
.filter
.data_size
);
428 len
= ustcomm_recv_unix_sock(sock
, bytecode
->bc
.data
,
429 lum
->u
.filter
.data_size
);
431 case 0: /* orderly shutdown */
436 if (len
== lum
->u
.filter
.data_size
) {
437 DBG("filter data received");
439 } else if (len
< 0) {
440 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
441 if (len
== -ECONNRESET
) {
442 ERR("%s remote end closed connection", sock_info
->name
);
451 DBG("incorrect filter data message size: %zd", len
);
457 bytecode
->bc
.len
= lum
->u
.filter
.data_size
;
458 bytecode
->bc
.reloc_offset
= lum
->u
.filter
.reloc_offset
;
459 bytecode
->bc
.seqnum
= lum
->u
.filter
.seqnum
;
461 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
462 (unsigned long) bytecode
,
467 /* don't free bytecode if everything went fine. */
474 case LTTNG_UST_CHANNEL
:
479 len
= ustcomm_recv_channel_from_sessiond(sock
,
480 &chan_data
, lum
->u
.channel
.len
,
483 case 0: /* orderly shutdown */
487 if (len
== lum
->u
.channel
.len
) {
488 DBG("channel data received");
490 } else if (len
< 0) {
491 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
492 if (len
== -ECONNRESET
) {
493 ERR("%s remote end closed connection", sock_info
->name
);
500 DBG("incorrect channel data message size: %zd", len
);
505 args
.channel
.chan_data
= chan_data
;
506 args
.channel
.wakeup_fd
= wakeup_fd
;
508 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
509 (unsigned long) &lum
->u
,
515 case LTTNG_UST_STREAM
:
517 /* Receive shm_fd, wakeup_fd */
518 ret
= ustcomm_recv_stream_from_sessiond(sock
,
521 &args
.stream
.wakeup_fd
);
526 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
527 (unsigned long) &lum
->u
,
535 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
536 (unsigned long) &lum
->u
,
544 lur
.handle
= lum
->handle
;
548 lur
.ret_code
= LTTNG_UST_OK
;
551 * Use -LTTNG_UST_ERR as wildcard for UST internal
552 * error that are not caused by the transport, except if
553 * we already have a more precise error message to
556 if (ret
> -LTTNG_UST_ERR
) {
557 /* Translate code to UST error. */
560 lur
.ret_code
= -LTTNG_UST_ERR_EXIST
;
563 lur
.ret_code
= -LTTNG_UST_ERR_INVAL
;
566 lur
.ret_code
= -LTTNG_UST_ERR_NOENT
;
569 lur
.ret_code
= -LTTNG_UST_ERR_PERM
;
572 lur
.ret_code
= -LTTNG_UST_ERR_NOSYS
;
575 lur
.ret_code
= -LTTNG_UST_ERR
;
584 case LTTNG_UST_TRACER_VERSION
:
585 lur
.u
.version
= lum
->u
.version
;
587 case LTTNG_UST_TRACEPOINT_LIST_GET
:
588 memcpy(&lur
.u
.tracepoint
, &lum
->u
.tracepoint
, sizeof(lur
.u
.tracepoint
));
592 DBG("Return value: %d", lur
.ret_val
);
593 ret
= send_reply(sock
, &lur
);
595 DBG("error sending reply");
600 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
603 if (lur
.ret_code
== LTTNG_UST_OK
) {
605 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET
:
606 len
= ustcomm_send_unix_sock(sock
,
607 &args
.field_list
.entry
,
608 sizeof(args
.field_list
.entry
));
613 if (len
!= sizeof(args
.field_list
.entry
)) {
626 void cleanup_sock_info(struct sock_info
*sock_info
, int exiting
)
630 if (sock_info
->root_handle
!= -1) {
631 ret
= lttng_ust_objd_unref(sock_info
->root_handle
, 1);
633 ERR("Error unref root handle");
635 sock_info
->root_handle
= -1;
637 sock_info
->constructor_sem_posted
= 0;
640 * wait_shm_mmap, socket and notify socket are used by listener
641 * threads outside of the ust lock, so we cannot tear them down
642 * ourselves, because we cannot join on these threads. Leave
643 * responsibility of cleaning up these resources to the OS
649 if (sock_info
->socket
!= -1) {
650 ret
= ustcomm_close_unix_sock(sock_info
->socket
);
652 ERR("Error closing ust cmd socket");
654 sock_info
->socket
= -1;
656 if (sock_info
->notify_socket
!= -1) {
657 ret
= ustcomm_close_unix_sock(sock_info
->notify_socket
);
659 ERR("Error closing ust notify socket");
661 sock_info
->notify_socket
= -1;
663 if (sock_info
->wait_shm_mmap
) {
664 ret
= munmap(sock_info
->wait_shm_mmap
, sysconf(_SC_PAGE_SIZE
));
666 ERR("Error unmapping wait shm");
668 sock_info
->wait_shm_mmap
= NULL
;
673 * Using fork to set umask in the child process (not multi-thread safe).
674 * We deal with the shm_open vs ftruncate race (happening when the
675 * sessiond owns the shm and does not let everybody modify it, to ensure
676 * safety against shm_unlink) by simply letting the mmap fail and
677 * retrying after a few seconds.
678 * For global shm, everybody has rw access to it until the sessiond
682 int get_wait_shm(struct sock_info
*sock_info
, size_t mmap_size
)
684 int wait_shm_fd
, ret
;
688 * Try to open read-only.
690 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
, O_RDONLY
, 0);
691 if (wait_shm_fd
>= 0) {
694 size_t bytes_read
= 0;
697 * Try to read the fd. If unable to do so, try opening
701 len
= read(wait_shm_fd
,
702 &((char *) &tmp_read
)[bytes_read
],
703 sizeof(tmp_read
) - bytes_read
);
707 } while ((len
< 0 && errno
== EINTR
)
708 || (len
> 0 && bytes_read
< sizeof(tmp_read
)));
709 if (bytes_read
!= sizeof(tmp_read
)) {
710 ret
= close(wait_shm_fd
);
712 ERR("close wait_shm_fd");
717 } else if (wait_shm_fd
< 0 && errno
!= ENOENT
) {
719 * Real-only open did not work, and it's not because the
720 * entry was not present. It's a failure that prohibits
723 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
729 * If the open failed because the file did not exist, or because
730 * the file was not truncated yet, try creating it ourself.
732 URCU_TLS(lttng_ust_nest_count
)++;
734 URCU_TLS(lttng_ust_nest_count
)--;
739 * Parent: wait for child to return, in which case the
740 * shared memory map will have been created.
743 if (pid
< 0 || !WIFEXITED(status
) || WEXITSTATUS(status
) != 0) {
748 * Try to open read-only again after creation.
750 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
, O_RDONLY
, 0);
751 if (wait_shm_fd
< 0) {
753 * Real-only open did not work. It's a failure
754 * that prohibits using shm.
756 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
760 } else if (pid
== 0) {
764 create_mode
= S_IRUSR
| S_IWUSR
| S_IRGRP
;
765 if (sock_info
->global
)
766 create_mode
|= S_IROTH
| S_IWGRP
| S_IWOTH
;
768 * We're alone in a child process, so we can modify the
769 * process-wide umask.
773 * Try creating shm (or get rw access).
774 * We don't do an exclusive open, because we allow other
775 * processes to create+ftruncate it concurrently.
777 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
,
778 O_RDWR
| O_CREAT
, create_mode
);
779 if (wait_shm_fd
>= 0) {
780 ret
= ftruncate(wait_shm_fd
, mmap_size
);
788 * For local shm, we need to have rw access to accept
789 * opening it: this means the local sessiond will be
790 * able to wake us up. For global shm, we open it even
791 * if rw access is not granted, because the root.root
792 * sessiond will be able to override all rights and wake
795 if (!sock_info
->global
&& errno
!= EACCES
) {
796 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
800 * The shm exists, but we cannot open it RW. Report
808 if (wait_shm_fd
>= 0 && !sock_info
->global
) {
812 * Ensure that our user is the owner of the shm file for
813 * local shm. If we do not own the file, it means our
814 * sessiond will not have access to wake us up (there is
815 * probably a rogue process trying to fake our
816 * sessiond). Fallback to polling method in this case.
818 ret
= fstat(wait_shm_fd
, &statbuf
);
823 if (statbuf
.st_uid
!= getuid())
829 ret
= close(wait_shm_fd
);
831 PERROR("Error closing fd");
837 char *get_map_shm(struct sock_info
*sock_info
)
839 size_t mmap_size
= sysconf(_SC_PAGE_SIZE
);
840 int wait_shm_fd
, ret
;
843 wait_shm_fd
= get_wait_shm(sock_info
, mmap_size
);
844 if (wait_shm_fd
< 0) {
847 wait_shm_mmap
= mmap(NULL
, mmap_size
, PROT_READ
,
848 MAP_SHARED
, wait_shm_fd
, 0);
849 /* close shm fd immediately after taking the mmap reference */
850 ret
= close(wait_shm_fd
);
852 PERROR("Error closing fd");
854 if (wait_shm_mmap
== MAP_FAILED
) {
855 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
858 return wait_shm_mmap
;
865 void wait_for_sessiond(struct sock_info
*sock_info
)
870 if (lttng_ust_comm_should_quit
) {
873 if (wait_poll_fallback
) {
876 if (!sock_info
->wait_shm_mmap
) {
877 sock_info
->wait_shm_mmap
= get_map_shm(sock_info
);
878 if (!sock_info
->wait_shm_mmap
)
883 DBG("Waiting for %s apps sessiond", sock_info
->name
);
884 /* Wait for futex wakeup */
885 if (uatomic_read((int32_t *) sock_info
->wait_shm_mmap
) == 0) {
886 ret
= futex_async((int32_t *) sock_info
->wait_shm_mmap
,
887 FUTEX_WAIT
, 0, NULL
, NULL
, 0);
889 if (errno
== EFAULT
) {
890 wait_poll_fallback
= 1;
892 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
893 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
894 "Please upgrade your kernel "
895 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
896 "mainline). LTTng-UST will use polling mode fallback.");
914 * This thread does not allocate any resource, except within
915 * handle_message, within mutex protection. This mutex protects against
917 * The other moment it allocates resources is at socket connection, which
918 * is also protected by the mutex.
921 void *ust_listener_thread(void *arg
)
923 struct sock_info
*sock_info
= arg
;
924 int sock
, ret
, prev_connect_failed
= 0, has_waited
= 0;
927 /* Restart trying to connect to the session daemon */
929 if (prev_connect_failed
) {
930 /* Wait for sessiond availability with pipe */
931 wait_for_sessiond(sock_info
);
935 * Sleep for 5 seconds before retrying after a
936 * sequence of failure / wait / failure. This
937 * deals with a killed or broken session daemon.
942 prev_connect_failed
= 0;
945 if (sock_info
->socket
!= -1) {
946 ret
= ustcomm_close_unix_sock(sock_info
->socket
);
948 ERR("Error closing %s ust cmd socket",
951 sock_info
->socket
= -1;
953 if (sock_info
->notify_socket
!= -1) {
954 ret
= ustcomm_close_unix_sock(sock_info
->notify_socket
);
956 ERR("Error closing %s ust notify socket",
959 sock_info
->notify_socket
= -1;
963 * Register. We need to perform both connect and sending
964 * registration message before doing the next connect otherwise
965 * we may reach unix socket connect queue max limits and block
966 * on the 2nd connect while the session daemon is awaiting the
967 * first connect registration message.
969 /* Connect cmd socket */
970 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
);
972 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
973 prev_connect_failed
= 1;
977 if (lttng_ust_comm_should_quit
) {
982 * If we cannot find the sessiond daemon, don't delay
983 * constructor execution.
985 ret
= handle_register_done(sock_info
);
990 sock_info
->socket
= ret
;
994 if (lttng_ust_comm_should_quit
) {
999 * Create only one root handle per listener thread for the whole
1000 * process lifetime, so we ensure we get ID which is statically
1001 * assigned to the root handle.
1003 if (sock_info
->root_handle
== -1) {
1004 ret
= lttng_abi_create_root_handle();
1006 ERR("Error creating root handle");
1009 sock_info
->root_handle
= ret
;
1012 ret
= register_to_sessiond(sock_info
->socket
, USTCTL_SOCKET_CMD
);
1014 ERR("Error registering to %s ust cmd socket",
1016 prev_connect_failed
= 1;
1018 * If we cannot register to the sessiond daemon, don't
1019 * delay constructor execution.
1021 ret
= handle_register_done(sock_info
);
1029 /* Connect notify socket */
1030 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
);
1032 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
1033 prev_connect_failed
= 1;
1037 if (lttng_ust_comm_should_quit
) {
1042 * If we cannot find the sessiond daemon, don't delay
1043 * constructor execution.
1045 ret
= handle_register_done(sock_info
);
1050 sock_info
->notify_socket
= ret
;
1052 timeout
= get_notify_sock_timeout();
1055 * Give at least 10ms to sessiond to reply to
1060 ret
= ustcomm_setsockopt_rcv_timeout(sock_info
->notify_socket
,
1063 WARN("Error setting socket receive timeout");
1065 ret
= ustcomm_setsockopt_snd_timeout(sock_info
->notify_socket
,
1068 WARN("Error setting socket send timeout");
1070 } else if (timeout
< -1) {
1071 WARN("Unsupported timeout value %ld", timeout
);
1076 if (lttng_ust_comm_should_quit
) {
1080 ret
= register_to_sessiond(sock_info
->notify_socket
,
1081 USTCTL_SOCKET_NOTIFY
);
1083 ERR("Error registering to %s ust notify socket",
1085 prev_connect_failed
= 1;
1087 * If we cannot register to the sessiond daemon, don't
1088 * delay constructor execution.
1090 ret
= handle_register_done(sock_info
);
1095 sock
= sock_info
->socket
;
1101 struct ustcomm_ust_msg lum
;
1103 len
= ustcomm_recv_unix_sock(sock
, &lum
, sizeof(lum
));
1105 case 0: /* orderly shutdown */
1106 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info
->name
);
1108 if (lttng_ust_comm_should_quit
) {
1112 * Either sessiond has shutdown or refused us by closing the socket.
1113 * In either case, we don't want to delay construction execution,
1114 * and we need to wait before retry.
1116 prev_connect_failed
= 1;
1118 * If we cannot register to the sessiond daemon, don't
1119 * delay constructor execution.
1121 ret
= handle_register_done(sock_info
);
1126 print_cmd(lum
.cmd
, lum
.handle
);
1127 ret
= handle_message(sock_info
, sock
, &lum
);
1129 ERR("Error handling message for %s socket", sock_info
->name
);
1134 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1136 DBG("incorrect message size (%s socket): %zd", sock_info
->name
, len
);
1138 if (len
== -ECONNRESET
) {
1139 DBG("%s remote end closed connection", sock_info
->name
);
1148 if (lttng_ust_comm_should_quit
) {
1151 /* Cleanup socket handles before trying to reconnect */
1152 lttng_ust_objd_table_owner_cleanup(sock_info
);
1154 goto restart
; /* try to reconnect */
1157 sock_info
->thread_active
= 0;
1163 * sessiond monitoring thread: monitor presence of global and per-user
1164 * sessiond by polling the application common named pipe.
1166 void __attribute__((constructor
)) lttng_ust_init(void)
1168 struct timespec constructor_timeout
;
1169 sigset_t sig_all_blocked
, orig_parent_mask
;
1170 pthread_attr_t thread_attr
;
1174 if (uatomic_xchg(&initialized
, 1) == 1)
1178 * Fixup interdependency between TLS fixup mutex (which happens
1179 * to be the dynamic linker mutex) and ust_lock, taken within
1182 lttng_fixup_ringbuffer_tls();
1183 lttng_fixup_vtid_tls();
1184 lttng_fixup_nest_count_tls();
1185 lttng_fixup_procname_tls();
1188 * We want precise control over the order in which we construct
1189 * our sub-libraries vs starting to receive commands from
1190 * sessiond (otherwise leading to errors when trying to create
1191 * sessiond before the init functions are completed).
1195 lttng_ring_buffer_metadata_client_init();
1196 lttng_ring_buffer_client_overwrite_init();
1197 lttng_ring_buffer_client_overwrite_rt_init();
1198 lttng_ring_buffer_client_discard_init();
1199 lttng_ring_buffer_client_discard_rt_init();
1200 lttng_context_init();
1202 timeout_mode
= get_constructor_timeout(&constructor_timeout
);
1204 ret
= sem_init(&constructor_wait
, 0, 0);
1207 ret
= setup_local_apps();
1209 DBG("local apps setup returned %d", ret
);
1212 /* A new thread created by pthread_create inherits the signal mask
1213 * from the parent. To avoid any signal being received by the
1214 * listener thread, we block all signals temporarily in the parent,
1215 * while we create the listener thread.
1217 sigfillset(&sig_all_blocked
);
1218 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_parent_mask
);
1220 ERR("pthread_sigmask: %s", strerror(ret
));
1223 ret
= pthread_attr_init(&thread_attr
);
1225 ERR("pthread_attr_init: %s", strerror(ret
));
1227 ret
= pthread_attr_setdetachstate(&thread_attr
, PTHREAD_CREATE_DETACHED
);
1229 ERR("pthread_attr_setdetachstate: %s", strerror(ret
));
1233 ret
= pthread_create(&global_apps
.ust_listener
, &thread_attr
,
1234 ust_listener_thread
, &global_apps
);
1236 ERR("pthread_create global: %s", strerror(ret
));
1238 global_apps
.thread_active
= 1;
1241 if (local_apps
.allowed
) {
1243 ret
= pthread_create(&local_apps
.ust_listener
, &thread_attr
,
1244 ust_listener_thread
, &local_apps
);
1246 ERR("pthread_create local: %s", strerror(ret
));
1248 local_apps
.thread_active
= 1;
1251 handle_register_done(&local_apps
);
1253 ret
= pthread_attr_destroy(&thread_attr
);
1255 ERR("pthread_attr_destroy: %s", strerror(ret
));
1258 /* Restore original signal mask in parent */
1259 ret
= pthread_sigmask(SIG_SETMASK
, &orig_parent_mask
, NULL
);
1261 ERR("pthread_sigmask: %s", strerror(ret
));
1264 switch (timeout_mode
) {
1265 case 1: /* timeout wait */
1267 ret
= sem_timedwait(&constructor_wait
,
1268 &constructor_timeout
);
1269 } while (ret
< 0 && errno
== EINTR
);
1270 if (ret
< 0 && errno
== ETIMEDOUT
) {
1271 ERR("Timed out waiting for lttng-sessiond");
1276 case -1:/* wait forever */
1278 ret
= sem_wait(&constructor_wait
);
1279 } while (ret
< 0 && errno
== EINTR
);
1282 case 0: /* no timeout */
1288 void lttng_ust_cleanup(int exiting
)
1290 cleanup_sock_info(&global_apps
, exiting
);
1291 if (local_apps
.allowed
) {
1292 cleanup_sock_info(&local_apps
, exiting
);
1295 * The teardown in this function all affect data structures
1296 * accessed under the UST lock by the listener thread. This
1297 * lock, along with the lttng_ust_comm_should_quit flag, ensure
1298 * that none of these threads are accessing this data at this
1301 lttng_ust_abi_exit();
1302 lttng_ust_events_exit();
1303 lttng_context_exit();
1304 lttng_ring_buffer_client_discard_rt_exit();
1305 lttng_ring_buffer_client_discard_exit();
1306 lttng_ring_buffer_client_overwrite_rt_exit();
1307 lttng_ring_buffer_client_overwrite_exit();
1308 lttng_ring_buffer_metadata_client_exit();
1311 /* Reinitialize values for fork */
1313 lttng_ust_comm_should_quit
= 0;
1318 void __attribute__((destructor
)) lttng_ust_exit(void)
1323 * Using pthread_cancel here because:
1324 * A) we don't want to hang application teardown.
1325 * B) the thread is not allocating any resource.
1329 * Require the communication thread to quit. Synchronize with
1330 * mutexes to ensure it is not in a mutex critical section when
1331 * pthread_cancel is later called.
1334 lttng_ust_comm_should_quit
= 1;
1336 /* cancel threads */
1337 if (global_apps
.thread_active
) {
1338 ret
= pthread_cancel(global_apps
.ust_listener
);
1340 ERR("Error cancelling global ust listener thread: %s",
1343 global_apps
.thread_active
= 0;
1346 if (local_apps
.thread_active
) {
1347 ret
= pthread_cancel(local_apps
.ust_listener
);
1349 ERR("Error cancelling local ust listener thread: %s",
1352 local_apps
.thread_active
= 0;
1358 * Do NOT join threads: use of sys_futex makes it impossible to
1359 * join the threads without using async-cancel, but async-cancel
1360 * is delivered by a signal, which could hit the target thread
1361 * anywhere in its code path, including while the ust_lock() is
1362 * held, causing a deadlock for the other thread. Let the OS
1363 * cleanup the threads if there are stalled in a syscall.
1365 lttng_ust_cleanup(1);
1369 * We exclude the worker threads across fork and clone (except
1370 * CLONE_VM), because these system calls only keep the forking thread
1371 * running in the child. Therefore, we don't want to call fork or clone
1372 * in the middle of an tracepoint or ust tracing state modification.
1373 * Holding this mutex protects these structures across fork and clone.
1375 void ust_before_fork(sigset_t
*save_sigset
)
1378 * Disable signals. This is to avoid that the child intervenes
1379 * before it is properly setup for tracing. It is safer to
1380 * disable all signals, because then we know we are not breaking
1381 * anything by restoring the original mask.
1386 if (URCU_TLS(lttng_ust_nest_count
))
1388 /* Disable signals */
1389 sigfillset(&all_sigs
);
1390 ret
= sigprocmask(SIG_BLOCK
, &all_sigs
, save_sigset
);
1392 PERROR("sigprocmask");
1395 rcu_bp_before_fork();
1398 static void ust_after_fork_common(sigset_t
*restore_sigset
)
1402 DBG("process %d", getpid());
1404 /* Restore signals */
1405 ret
= sigprocmask(SIG_SETMASK
, restore_sigset
, NULL
);
1407 PERROR("sigprocmask");
1411 void ust_after_fork_parent(sigset_t
*restore_sigset
)
1413 if (URCU_TLS(lttng_ust_nest_count
))
1415 DBG("process %d", getpid());
1416 rcu_bp_after_fork_parent();
1417 /* Release mutexes and reenable signals */
1418 ust_after_fork_common(restore_sigset
);
1422 * After fork, in the child, we need to cleanup all the leftover state,
1423 * except the worker thread which already magically disappeared thanks
1424 * to the weird Linux fork semantics. After tyding up, we call
1425 * lttng_ust_init() again to start over as a new PID.
1427 * This is meant for forks() that have tracing in the child between the
1428 * fork and following exec call (if there is any).
1430 void ust_after_fork_child(sigset_t
*restore_sigset
)
1432 if (URCU_TLS(lttng_ust_nest_count
))
1434 DBG("process %d", getpid());
1435 /* Release urcu mutexes */
1436 rcu_bp_after_fork_child();
1437 lttng_ust_cleanup(0);
1438 lttng_context_vtid_reset();
1439 /* Release mutexes and reenable signals */
1440 ust_after_fork_common(restore_sigset
);