Add --userspace-probe kernel event type
[lttng-tools.git] / src / bin / lttng-sessiond / kernel.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License, version 2 only,
6 * as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License along
14 * with this program; if not, write to the Free Software Foundation, Inc.,
15 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
16 */
17
18 #define _LGPL_SOURCE
19 #include <fcntl.h>
20 #include <stdlib.h>
21 #include <stdio.h>
22 #include <string.h>
23 #include <unistd.h>
24 #include <inttypes.h>
25
26 #include <common/common.h>
27 #include <common/kernel-ctl/kernel-ctl.h>
28 #include <common/kernel-ctl/kernel-ioctl.h>
29 #include <common/sessiond-comm/sessiond-comm.h>
30
31 #include "consumer.h"
32 #include "kernel.h"
33 #include "kernel-consumer.h"
34 #include "kern-modules.h"
35 #include "utils.h"
36 #include "rotate.h"
37
38 /*
39 * Key used to reference a channel between the sessiond and the consumer. This
40 * is only read and updated with the session_list lock held.
41 */
42 static uint64_t next_kernel_channel_key;
43
44 #include <lttng/userspace-probe.h>
45 #include <lttng/userspace-probe-internal.h>
46 /*
47 * Add context on a kernel channel.
48 *
49 * Assumes the ownership of ctx.
50 */
51 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
52 struct ltt_kernel_context *ctx)
53 {
54 int ret;
55
56 assert(chan);
57 assert(ctx);
58
59 DBG("Adding context to channel %s", chan->channel->name);
60 ret = kernctl_add_context(chan->fd, &ctx->ctx);
61 if (ret < 0) {
62 switch (-ret) {
63 case ENOSYS:
64 /* Exists but not available for this kernel */
65 ret = LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE;
66 goto error;
67 case EEXIST:
68 /* If EEXIST, we just ignore the error */
69 ret = 0;
70 goto end;
71 default:
72 PERROR("add context ioctl");
73 ret = LTTNG_ERR_KERN_CONTEXT_FAIL;
74 goto error;
75 }
76 }
77 ret = 0;
78
79 end:
80 cds_list_add_tail(&ctx->list, &chan->ctx_list);
81 ctx->in_list = true;
82 ctx = NULL;
83 error:
84 if (ctx) {
85 trace_kernel_destroy_context(ctx);
86 }
87 return ret;
88 }
89
90 /*
91 * Create a new kernel session, register it to the kernel tracer and add it to
92 * the session daemon session.
93 */
94 int kernel_create_session(struct ltt_session *session, int tracer_fd)
95 {
96 int ret;
97 struct ltt_kernel_session *lks;
98
99 assert(session);
100
101 /* Allocate data structure */
102 lks = trace_kernel_create_session();
103 if (lks == NULL) {
104 ret = -1;
105 goto error;
106 }
107
108 /* Kernel tracer session creation */
109 ret = kernctl_create_session(tracer_fd);
110 if (ret < 0) {
111 PERROR("ioctl kernel create session");
112 goto error;
113 }
114
115 lks->fd = ret;
116 /* Prevent fd duplication after execlp() */
117 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
118 if (ret < 0) {
119 PERROR("fcntl session fd");
120 }
121
122 lks->id = session->id;
123 lks->consumer_fds_sent = 0;
124 session->kernel_session = lks;
125
126 DBG("Kernel session created (fd: %d)", lks->fd);
127
128 return 0;
129
130 error:
131 if (lks) {
132 trace_kernel_destroy_session(lks);
133 }
134 return ret;
135 }
136
137 /*
138 * Create a kernel channel, register it to the kernel tracer and add it to the
139 * kernel session.
140 */
141 int kernel_create_channel(struct ltt_kernel_session *session,
142 struct lttng_channel *chan)
143 {
144 int ret;
145 struct ltt_kernel_channel *lkc;
146
147 assert(session);
148 assert(chan);
149
150 /* Allocate kernel channel */
151 lkc = trace_kernel_create_channel(chan);
152 if (lkc == NULL) {
153 goto error;
154 }
155
156 DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
157 chan->name, lkc->channel->attr.overwrite,
158 lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
159 lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
160 lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
161
162 /* Kernel tracer channel creation */
163 ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
164 if (ret < 0) {
165 PERROR("ioctl kernel create channel");
166 goto error;
167 }
168
169 /* Setup the channel fd */
170 lkc->fd = ret;
171 /* Prevent fd duplication after execlp() */
172 ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
173 if (ret < 0) {
174 PERROR("fcntl session fd");
175 }
176
177 /* Add channel to session */
178 cds_list_add(&lkc->list, &session->channel_list.head);
179 session->channel_count++;
180 lkc->session = session;
181 lkc->key = ++next_kernel_channel_key;
182
183 DBG("Kernel channel %s created (fd: %d, key: %" PRIu64 ")",
184 lkc->channel->name, lkc->fd, lkc->key);
185
186 return 0;
187
188 error:
189 if (lkc) {
190 free(lkc->channel);
191 free(lkc);
192 }
193 return -1;
194 }
195
196 /*
197 * Compute the offset of the instrumentation byte in the binary based on the
198 * function probe location using the ELF lookup method.
199 *
200 * Returns 0 on success and set the offset out parameter to the offset of the
201 * elf symbol
202 * Returns -1 on error
203 */
204 static
205 int extract_userspace_probe_offset_function_elf(
206 struct lttng_userspace_probe_location *probe_location,
207 struct ltt_kernel_session *session, uint64_t *offset)
208 {
209 int fd;
210 int ret = 0;
211 const char *symbol = NULL;
212 struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
213 enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
214
215
216 assert(lttng_userspace_probe_location_get_type(probe_location) ==
217 LTTNG_USERSPACE_PROBE_LOCATION_TYPE_FUNCTION);
218
219 lookup = lttng_userspace_probe_location_get_lookup_method(
220 probe_location);
221 if (!lookup) {
222 ret = -1;
223 goto end;
224 }
225
226 lookup_method_type =
227 lttng_userspace_probe_location_lookup_method_get_type(lookup);
228
229 assert(lookup_method_type ==
230 LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF);
231
232 symbol = lttng_userspace_probe_location_function_get_function_name(
233 probe_location);
234 if (!symbol) {
235 ret = -1;
236 goto end;
237 }
238
239 fd = lttng_userspace_probe_location_function_get_binary_fd(probe_location);
240 if (fd < 0) {
241 ret = -1;
242 goto end;
243 }
244
245 ret = run_as_extract_elf_symbol_offset(fd, symbol, session->uid,
246 session->gid, offset);
247 if (ret < 0) {
248 DBG("userspace probe offset calculation failed for "
249 "function %s", symbol);
250 goto end;
251 }
252
253 DBG("userspace probe elf offset for %s is 0x%jd", symbol, (intmax_t)(*offset));
254 end:
255 return ret;
256 }
257
258 /*
259 * Compute the offsets of the instrumentation bytes in the binary based on the
260 * tracepoint probe location using the SDT lookup method. This function
261 * allocates the offsets buffer, the caller must free it.
262 *
263 * Returns 0 on success and set the offset out parameter to the offsets of the
264 * SDT tracepoint.
265 * Returns -1 on error.
266 */
267 static
268 int extract_userspace_probe_offset_tracepoint_sdt(
269 struct lttng_userspace_probe_location *probe_location,
270 struct ltt_kernel_session *session, uint64_t **offsets,
271 uint32_t *offsets_count)
272 {
273 enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
274 struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
275 const char *probe_name = NULL, *provider_name = NULL;
276 int ret = 0;
277 int fd, i;
278
279 assert(lttng_userspace_probe_location_get_type(probe_location) ==
280 LTTNG_USERSPACE_PROBE_LOCATION_TYPE_TRACEPOINT);
281
282 lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
283 if (!lookup) {
284 ret = -1;
285 goto end;
286 }
287
288 lookup_method_type =
289 lttng_userspace_probe_location_lookup_method_get_type(lookup);
290
291 assert(lookup_method_type ==
292 LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT);
293
294
295 probe_name = lttng_userspace_probe_location_tracepoint_get_probe_name(
296 probe_location);
297 if (!probe_name) {
298 ret = -1;
299 goto end;
300 }
301
302 provider_name = lttng_userspace_probe_location_tracepoint_get_provider_name(
303 probe_location);
304 if (!provider_name) {
305 ret = -1;
306 goto end;
307 }
308
309 fd = lttng_userspace_probe_location_tracepoint_get_binary_fd(probe_location);
310 if (fd < 0) {
311 ret = -1;
312 goto end;
313 }
314
315 ret = run_as_extract_sdt_probe_offsets(fd, provider_name, probe_name,
316 session->uid, session->gid, offsets, offsets_count);
317 if (ret < 0) {
318 DBG("userspace probe offset calculation failed for sdt "
319 "probe %s:%s", provider_name, probe_name);
320 goto end;
321 }
322
323 if (*offsets_count == 0) {
324 DBG("no userspace probe offset found");
325 goto end;
326 }
327
328 DBG("%u userspace probe SDT offsets found for %s:%s at:",
329 *offsets_count, provider_name, probe_name);
330 for (i = 0; i < *offsets_count; i++) {
331 DBG("\t0x%jd", (intmax_t)((*offsets)[i]));
332 }
333 end:
334 return ret;
335 }
336
337 /*
338 * Extract the offsets of the instrumentation point for the different lookup
339 * methods.
340 */
341 static
342 int userspace_probe_add_callsites(struct lttng_event *ev,
343 struct ltt_kernel_session *session, int fd)
344 {
345 struct lttng_userspace_probe_location_lookup_method *lookup_method = NULL;
346 enum lttng_userspace_probe_location_lookup_method_type type;
347 struct lttng_userspace_probe_location *location = NULL;
348 int ret;
349
350 assert(ev);
351 assert(ev->type == LTTNG_EVENT_USERSPACE_PROBE);
352
353 location = lttng_event_get_userspace_probe_location(ev);
354 if (!location) {
355 ret = -1;
356 goto end;
357 }
358 lookup_method =
359 lttng_userspace_probe_location_get_lookup_method(location);
360 if (!lookup_method) {
361 ret = -1;
362 goto end;
363 }
364
365 type = lttng_userspace_probe_location_lookup_method_get_type(lookup_method);
366 switch (type) {
367 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
368 {
369 struct lttng_kernel_event_callsite callsite;
370 uint64_t offset;
371
372 ret = extract_userspace_probe_offset_function_elf(location, session, &offset);
373 if (ret) {
374 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
375 goto end;
376 }
377
378 callsite.u.uprobe.offset = offset;
379 ret = kernctl_add_callsite(fd, &callsite);
380 if (ret) {
381 WARN("Adding callsite to userspace probe "
382 "event %s failed.", ev->name);
383 ret = LTTNG_ERR_KERN_ENABLE_FAIL;
384 goto end;
385 }
386 break;
387 }
388 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
389 {
390 int i;
391 uint64_t *offsets = NULL;
392 uint32_t offsets_count;
393 struct lttng_kernel_event_callsite callsite;
394
395 /*
396 * This call allocates the offsets buffer. This buffer must be freed
397 * by the caller
398 */
399 ret = extract_userspace_probe_offset_tracepoint_sdt(location, session,
400 &offsets, &offsets_count);
401 if (ret) {
402 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
403 goto end;
404 }
405 for (i = 0; i < offsets_count; i++) {
406 callsite.u.uprobe.offset = offsets[i];
407 ret = kernctl_add_callsite(fd, &callsite);
408 if (ret) {
409 WARN("Adding callsite to userspace probe "
410 "event %s failed.", ev->name);
411 ret = LTTNG_ERR_KERN_ENABLE_FAIL;
412 free(offsets);
413 goto end;
414 }
415 }
416 free(offsets);
417 break;
418 }
419 default:
420 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
421 goto end;
422 }
423 end:
424 return ret;
425 }
426
427 /*
428 * Create a kernel event, enable it to the kernel tracer and add it to the
429 * channel event list of the kernel session.
430 * We own filter_expression and filter.
431 */
432 int kernel_create_event(struct lttng_event *ev,
433 struct ltt_kernel_channel *channel,
434 char *filter_expression,
435 struct lttng_filter_bytecode *filter)
436 {
437 int err, fd;
438 enum lttng_error_code ret;
439 struct ltt_kernel_event *event;
440
441 assert(ev);
442 assert(channel);
443
444 /* We pass ownership of filter_expression and filter */
445 ret = trace_kernel_create_event(ev, filter_expression,
446 filter, &event);
447 if (ret != LTTNG_OK) {
448 goto error;
449 }
450
451 fd = kernctl_create_event(channel->fd, event->event);
452 if (fd < 0) {
453 switch (-fd) {
454 case EEXIST:
455 ret = LTTNG_ERR_KERN_EVENT_EXIST;
456 break;
457 case ENOSYS:
458 WARN("Event type not implemented");
459 ret = LTTNG_ERR_KERN_EVENT_ENOSYS;
460 break;
461 case ENOENT:
462 WARN("Event %s not found!", ev->name);
463 ret = LTTNG_ERR_KERN_ENABLE_FAIL;
464 break;
465 default:
466 ret = LTTNG_ERR_KERN_ENABLE_FAIL;
467 PERROR("create event ioctl");
468 }
469 goto free_event;
470 }
471
472 event->type = ev->type;
473 event->fd = fd;
474 /* Prevent fd duplication after execlp() */
475 err = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
476 if (err < 0) {
477 PERROR("fcntl session fd");
478 }
479
480 if (filter) {
481 err = kernctl_filter(event->fd, filter);
482 if (err < 0) {
483 switch (-err) {
484 case ENOMEM:
485 ret = LTTNG_ERR_FILTER_NOMEM;
486 break;
487 default:
488 ret = LTTNG_ERR_FILTER_INVAL;
489 break;
490 }
491 goto filter_error;
492 }
493 }
494
495 if (ev->type == LTTNG_EVENT_USERSPACE_PROBE) {
496 ret = userspace_probe_add_callsites(ev, channel->session, event->fd);
497 if (ret) {
498 goto add_callsite_error;
499 }
500 }
501
502 err = kernctl_enable(event->fd);
503 if (err < 0) {
504 switch (-err) {
505 case EEXIST:
506 ret = LTTNG_ERR_KERN_EVENT_EXIST;
507 break;
508 default:
509 PERROR("enable kernel event");
510 ret = LTTNG_ERR_KERN_ENABLE_FAIL;
511 break;
512 }
513 goto enable_error;
514 }
515
516 /* Add event to event list */
517 cds_list_add(&event->list, &channel->events_list.head);
518 channel->event_count++;
519
520 DBG("Event %s created (fd: %d)", ev->name, event->fd);
521
522 return 0;
523
524 add_callsite_error:
525 enable_error:
526 filter_error:
527 {
528 int closeret;
529
530 closeret = close(event->fd);
531 if (closeret) {
532 PERROR("close event fd");
533 }
534 }
535 free_event:
536 free(event);
537 error:
538 return ret;
539 }
540
541 /*
542 * Disable a kernel channel.
543 */
544 int kernel_disable_channel(struct ltt_kernel_channel *chan)
545 {
546 int ret;
547
548 assert(chan);
549
550 ret = kernctl_disable(chan->fd);
551 if (ret < 0) {
552 PERROR("disable chan ioctl");
553 goto error;
554 }
555
556 chan->enabled = 0;
557 DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64 ")",
558 chan->channel->name, chan->fd, chan->key);
559
560 return 0;
561
562 error:
563 return ret;
564 }
565
566 /*
567 * Enable a kernel channel.
568 */
569 int kernel_enable_channel(struct ltt_kernel_channel *chan)
570 {
571 int ret;
572
573 assert(chan);
574
575 ret = kernctl_enable(chan->fd);
576 if (ret < 0 && ret != -EEXIST) {
577 PERROR("Enable kernel chan");
578 goto error;
579 }
580
581 chan->enabled = 1;
582 DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64 ")",
583 chan->channel->name, chan->fd, chan->key);
584
585 return 0;
586
587 error:
588 return ret;
589 }
590
591 /*
592 * Enable a kernel event.
593 */
594 int kernel_enable_event(struct ltt_kernel_event *event)
595 {
596 int ret;
597
598 assert(event);
599
600 ret = kernctl_enable(event->fd);
601 if (ret < 0) {
602 switch (-ret) {
603 case EEXIST:
604 ret = LTTNG_ERR_KERN_EVENT_EXIST;
605 break;
606 default:
607 PERROR("enable kernel event");
608 break;
609 }
610 goto error;
611 }
612
613 event->enabled = 1;
614 DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
615
616 return 0;
617
618 error:
619 return ret;
620 }
621
622 /*
623 * Disable a kernel event.
624 */
625 int kernel_disable_event(struct ltt_kernel_event *event)
626 {
627 int ret;
628
629 assert(event);
630
631 ret = kernctl_disable(event->fd);
632 if (ret < 0) {
633 switch (-ret) {
634 case EEXIST:
635 ret = LTTNG_ERR_KERN_EVENT_EXIST;
636 break;
637 default:
638 PERROR("disable kernel event");
639 break;
640 }
641 goto error;
642 }
643
644 event->enabled = 0;
645 DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
646
647 return 0;
648
649 error:
650 return ret;
651 }
652
653
654 int kernel_track_pid(struct ltt_kernel_session *session, int pid)
655 {
656 int ret;
657
658 DBG("Kernel track PID %d for session id %" PRIu64 ".",
659 pid, session->id);
660 ret = kernctl_track_pid(session->fd, pid);
661 if (!ret) {
662 return LTTNG_OK;
663 }
664 switch (-ret) {
665 case EINVAL:
666 return LTTNG_ERR_INVALID;
667 case ENOMEM:
668 return LTTNG_ERR_NOMEM;
669 case EEXIST:
670 return LTTNG_ERR_PID_TRACKED;
671 default:
672 return LTTNG_ERR_UNK;
673 }
674 }
675
676 int kernel_untrack_pid(struct ltt_kernel_session *session, int pid)
677 {
678 int ret;
679
680 DBG("Kernel untrack PID %d for session id %" PRIu64 ".",
681 pid, session->id);
682 ret = kernctl_untrack_pid(session->fd, pid);
683 if (!ret) {
684 return LTTNG_OK;
685 }
686 switch (-ret) {
687 case EINVAL:
688 return LTTNG_ERR_INVALID;
689 case ENOMEM:
690 return LTTNG_ERR_NOMEM;
691 case ENOENT:
692 return LTTNG_ERR_PID_NOT_TRACKED;
693 default:
694 return LTTNG_ERR_UNK;
695 }
696 }
697
698 ssize_t kernel_list_tracker_pids(struct ltt_kernel_session *session,
699 int **_pids)
700 {
701 int fd, ret;
702 int pid;
703 ssize_t nbmem, count = 0;
704 FILE *fp;
705 int *pids;
706
707 fd = kernctl_list_tracker_pids(session->fd);
708 if (fd < 0) {
709 PERROR("kernel tracker pids list");
710 goto error;
711 }
712
713 fp = fdopen(fd, "r");
714 if (fp == NULL) {
715 PERROR("kernel tracker pids list fdopen");
716 goto error_fp;
717 }
718
719 nbmem = KERNEL_TRACKER_PIDS_INIT_LIST_SIZE;
720 pids = zmalloc(sizeof(*pids) * nbmem);
721 if (pids == NULL) {
722 PERROR("alloc list pids");
723 count = -ENOMEM;
724 goto end;
725 }
726
727 while (fscanf(fp, "process { pid = %u; };\n", &pid) == 1) {
728 if (count >= nbmem) {
729 int *new_pids;
730 size_t new_nbmem;
731
732 new_nbmem = nbmem << 1;
733 DBG("Reallocating pids list from %zu to %zu entries",
734 nbmem, new_nbmem);
735 new_pids = realloc(pids, new_nbmem * sizeof(*new_pids));
736 if (new_pids == NULL) {
737 PERROR("realloc list events");
738 free(pids);
739 count = -ENOMEM;
740 goto end;
741 }
742 /* Zero the new memory */
743 memset(new_pids + nbmem, 0,
744 (new_nbmem - nbmem) * sizeof(*new_pids));
745 nbmem = new_nbmem;
746 pids = new_pids;
747 }
748 pids[count++] = pid;
749 }
750
751 *_pids = pids;
752 DBG("Kernel list tracker pids done (%zd pids)", count);
753 end:
754 ret = fclose(fp); /* closes both fp and fd */
755 if (ret) {
756 PERROR("fclose");
757 }
758 return count;
759
760 error_fp:
761 ret = close(fd);
762 if (ret) {
763 PERROR("close");
764 }
765 error:
766 return -1;
767 }
768
769 /*
770 * Create kernel metadata, open from the kernel tracer and add it to the
771 * kernel session.
772 */
773 int kernel_open_metadata(struct ltt_kernel_session *session)
774 {
775 int ret;
776 struct ltt_kernel_metadata *lkm = NULL;
777
778 assert(session);
779
780 /* Allocate kernel metadata */
781 lkm = trace_kernel_create_metadata();
782 if (lkm == NULL) {
783 goto error;
784 }
785
786 /* Kernel tracer metadata creation */
787 ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
788 if (ret < 0) {
789 goto error_open;
790 }
791
792 lkm->fd = ret;
793 lkm->key = ++next_kernel_channel_key;
794 /* Prevent fd duplication after execlp() */
795 ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
796 if (ret < 0) {
797 PERROR("fcntl session fd");
798 }
799
800 session->metadata = lkm;
801
802 DBG("Kernel metadata opened (fd: %d)", lkm->fd);
803
804 return 0;
805
806 error_open:
807 trace_kernel_destroy_metadata(lkm);
808 error:
809 return -1;
810 }
811
812 /*
813 * Start tracing session.
814 */
815 int kernel_start_session(struct ltt_kernel_session *session)
816 {
817 int ret;
818
819 assert(session);
820
821 ret = kernctl_start_session(session->fd);
822 if (ret < 0) {
823 PERROR("ioctl start session");
824 goto error;
825 }
826
827 DBG("Kernel session started");
828
829 return 0;
830
831 error:
832 return ret;
833 }
834
835 /*
836 * Make a kernel wait to make sure in-flight probe have completed.
837 */
838 void kernel_wait_quiescent(int fd)
839 {
840 int ret;
841
842 DBG("Kernel quiescent wait on %d", fd);
843
844 ret = kernctl_wait_quiescent(fd);
845 if (ret < 0) {
846 PERROR("wait quiescent ioctl");
847 ERR("Kernel quiescent wait failed");
848 }
849 }
850
851 /*
852 * Force flush buffer of metadata.
853 */
854 int kernel_metadata_flush_buffer(int fd)
855 {
856 int ret;
857
858 DBG("Kernel flushing metadata buffer on fd %d", fd);
859
860 ret = kernctl_buffer_flush(fd);
861 if (ret < 0) {
862 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
863 }
864
865 return 0;
866 }
867
868 /*
869 * Force flush buffer for channel.
870 */
871 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
872 {
873 int ret;
874 struct ltt_kernel_stream *stream;
875
876 assert(channel);
877
878 DBG("Flush buffer for channel %s", channel->channel->name);
879
880 cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
881 DBG("Flushing channel stream %d", stream->fd);
882 ret = kernctl_buffer_flush(stream->fd);
883 if (ret < 0) {
884 PERROR("ioctl");
885 ERR("Fail to flush buffer for stream %d (ret: %d)",
886 stream->fd, ret);
887 }
888 }
889
890 return 0;
891 }
892
893 /*
894 * Stop tracing session.
895 */
896 int kernel_stop_session(struct ltt_kernel_session *session)
897 {
898 int ret;
899
900 assert(session);
901
902 ret = kernctl_stop_session(session->fd);
903 if (ret < 0) {
904 goto error;
905 }
906
907 DBG("Kernel session stopped");
908
909 return 0;
910
911 error:
912 return ret;
913 }
914
915 /*
916 * Open stream of channel, register it to the kernel tracer and add it
917 * to the stream list of the channel.
918 *
919 * Note: given that the streams may appear in random order wrt CPU
920 * number (e.g. cpu hotplug), the index value of the stream number in
921 * the stream name is not necessarily linked to the CPU number.
922 *
923 * Return the number of created stream. Else, a negative value.
924 */
925 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
926 {
927 int ret;
928 struct ltt_kernel_stream *lks;
929
930 assert(channel);
931
932 while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
933 lks = trace_kernel_create_stream(channel->channel->name,
934 channel->stream_count);
935 if (lks == NULL) {
936 ret = close(ret);
937 if (ret) {
938 PERROR("close");
939 }
940 goto error;
941 }
942
943 lks->fd = ret;
944 /* Prevent fd duplication after execlp() */
945 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
946 if (ret < 0) {
947 PERROR("fcntl session fd");
948 }
949
950 lks->tracefile_size = channel->channel->attr.tracefile_size;
951 lks->tracefile_count = channel->channel->attr.tracefile_count;
952
953 /* Add stream to channel stream list */
954 cds_list_add(&lks->list, &channel->stream_list.head);
955 channel->stream_count++;
956
957 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
958 lks->state);
959 }
960
961 return channel->stream_count;
962
963 error:
964 return -1;
965 }
966
967 /*
968 * Open the metadata stream and set it to the kernel session.
969 */
970 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
971 {
972 int ret;
973
974 assert(session);
975
976 ret = kernctl_create_stream(session->metadata->fd);
977 if (ret < 0) {
978 PERROR("kernel create metadata stream");
979 goto error;
980 }
981
982 DBG("Kernel metadata stream created (fd: %d)", ret);
983 session->metadata_stream_fd = ret;
984 /* Prevent fd duplication after execlp() */
985 ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
986 if (ret < 0) {
987 PERROR("fcntl session fd");
988 }
989
990 return 0;
991
992 error:
993 return -1;
994 }
995
996 /*
997 * Get the event list from the kernel tracer and return the number of elements.
998 */
999 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
1000 {
1001 int fd, ret;
1002 char *event;
1003 size_t nbmem, count = 0;
1004 FILE *fp;
1005 struct lttng_event *elist;
1006
1007 assert(events);
1008
1009 fd = kernctl_tracepoint_list(tracer_fd);
1010 if (fd < 0) {
1011 PERROR("kernel tracepoint list");
1012 goto error;
1013 }
1014
1015 fp = fdopen(fd, "r");
1016 if (fp == NULL) {
1017 PERROR("kernel tracepoint list fdopen");
1018 goto error_fp;
1019 }
1020
1021 /*
1022 * Init memory size counter
1023 * See kernel-ctl.h for explanation of this value
1024 */
1025 nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
1026 elist = zmalloc(sizeof(struct lttng_event) * nbmem);
1027 if (elist == NULL) {
1028 PERROR("alloc list events");
1029 count = -ENOMEM;
1030 goto end;
1031 }
1032
1033 while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
1034 if (count >= nbmem) {
1035 struct lttng_event *new_elist;
1036 size_t new_nbmem;
1037
1038 new_nbmem = nbmem << 1;
1039 DBG("Reallocating event list from %zu to %zu bytes",
1040 nbmem, new_nbmem);
1041 new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
1042 if (new_elist == NULL) {
1043 PERROR("realloc list events");
1044 free(event);
1045 free(elist);
1046 count = -ENOMEM;
1047 goto end;
1048 }
1049 /* Zero the new memory */
1050 memset(new_elist + nbmem, 0,
1051 (new_nbmem - nbmem) * sizeof(struct lttng_event));
1052 nbmem = new_nbmem;
1053 elist = new_elist;
1054 }
1055 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
1056 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
1057 elist[count].enabled = -1;
1058 count++;
1059 free(event);
1060 }
1061
1062 *events = elist;
1063 DBG("Kernel list events done (%zu events)", count);
1064 end:
1065 ret = fclose(fp); /* closes both fp and fd */
1066 if (ret) {
1067 PERROR("fclose");
1068 }
1069 return count;
1070
1071 error_fp:
1072 ret = close(fd);
1073 if (ret) {
1074 PERROR("close");
1075 }
1076 error:
1077 return -1;
1078 }
1079
1080 /*
1081 * Get kernel version and validate it.
1082 */
1083 int kernel_validate_version(int tracer_fd,
1084 struct lttng_kernel_tracer_version *version,
1085 struct lttng_kernel_tracer_abi_version *abi_version)
1086 {
1087 int ret;
1088
1089 ret = kernctl_tracer_version(tracer_fd, version);
1090 if (ret < 0) {
1091 ERR("Failed to retrieve the lttng-modules version");
1092 goto error;
1093 }
1094
1095 /* Validate version */
1096 if (version->major != VERSION_MAJOR) {
1097 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
1098 version->major, VERSION_MAJOR);
1099 goto error_version;
1100 }
1101 ret = kernctl_tracer_abi_version(tracer_fd, abi_version);
1102 if (ret < 0) {
1103 ERR("Failed to retrieve lttng-modules ABI version");
1104 goto error;
1105 }
1106 if (abi_version->major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
1107 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
1108 abi_version->major, abi_version->minor,
1109 LTTNG_MODULES_ABI_MAJOR_VERSION);
1110 goto error;
1111 }
1112 DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
1113 version->major, version->minor,
1114 abi_version->major, abi_version->minor);
1115 return 0;
1116
1117 error_version:
1118 ret = -1;
1119
1120 error:
1121 ERR("Kernel tracer version check failed; kernel tracing will not be available");
1122 return ret;
1123 }
1124
1125 /*
1126 * Kernel work-arounds called at the start of sessiond main().
1127 */
1128 int init_kernel_workarounds(void)
1129 {
1130 int ret;
1131 FILE *fp;
1132
1133 /*
1134 * boot_id needs to be read once before being used concurrently
1135 * to deal with a Linux kernel race. A fix is proposed for
1136 * upstream, but the work-around is needed for older kernels.
1137 */
1138 fp = fopen("/proc/sys/kernel/random/boot_id", "r");
1139 if (!fp) {
1140 goto end_boot_id;
1141 }
1142 while (!feof(fp)) {
1143 char buf[37] = "";
1144
1145 ret = fread(buf, 1, sizeof(buf), fp);
1146 if (ret < 0) {
1147 /* Ignore error, we don't really care */
1148 }
1149 }
1150 ret = fclose(fp);
1151 if (ret) {
1152 PERROR("fclose");
1153 }
1154 end_boot_id:
1155 return 0;
1156 }
1157
1158 /*
1159 * Complete teardown of a kernel session.
1160 */
1161 void kernel_destroy_session(struct ltt_kernel_session *ksess)
1162 {
1163 if (ksess == NULL) {
1164 DBG3("No kernel session when tearing down session");
1165 return;
1166 }
1167
1168 DBG("Tearing down kernel session");
1169
1170 /*
1171 * Destroy channels on the consumer if at least one FD has been sent and we
1172 * are in no output mode because the streams are in *no* monitor mode so we
1173 * have to send a command to clean them up or else they leaked.
1174 */
1175 if (!ksess->output_traces && ksess->consumer_fds_sent) {
1176 int ret;
1177 struct consumer_socket *socket;
1178 struct lttng_ht_iter iter;
1179
1180 /* For each consumer socket. */
1181 rcu_read_lock();
1182 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1183 socket, node.node) {
1184 struct ltt_kernel_channel *chan;
1185
1186 /* For each channel, ask the consumer to destroy it. */
1187 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1188 ret = kernel_consumer_destroy_channel(socket, chan);
1189 if (ret < 0) {
1190 /* Consumer is probably dead. Use next socket. */
1191 continue;
1192 }
1193 }
1194 }
1195 rcu_read_unlock();
1196 }
1197
1198 /* Close any relayd session */
1199 consumer_output_send_destroy_relayd(ksess->consumer);
1200
1201 trace_kernel_destroy_session(ksess);
1202 }
1203
1204 /*
1205 * Destroy a kernel channel object. It does not do anything on the tracer side.
1206 */
1207 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
1208 {
1209 struct ltt_kernel_session *ksess = NULL;
1210
1211 assert(kchan);
1212 assert(kchan->channel);
1213
1214 DBG3("Kernel destroy channel %s", kchan->channel->name);
1215
1216 /* Update channel count of associated session. */
1217 if (kchan->session) {
1218 /* Keep pointer reference so we can update it after the destroy. */
1219 ksess = kchan->session;
1220 }
1221
1222 trace_kernel_destroy_channel(kchan);
1223
1224 /*
1225 * At this point the kernel channel is not visible anymore. This is safe
1226 * since in order to work on a visible kernel session, the tracing session
1227 * lock (ltt_session.lock) MUST be acquired.
1228 */
1229 if (ksess) {
1230 ksess->channel_count--;
1231 }
1232 }
1233
1234 /*
1235 * Take a snapshot for a given kernel session.
1236 *
1237 * Return 0 on success or else return a LTTNG_ERR code.
1238 */
1239 int kernel_snapshot_record(struct ltt_kernel_session *ksess,
1240 struct snapshot_output *output, int wait,
1241 uint64_t nb_packets_per_stream)
1242 {
1243 int err, ret, saved_metadata_fd;
1244 struct consumer_socket *socket;
1245 struct lttng_ht_iter iter;
1246 struct ltt_kernel_metadata *saved_metadata;
1247 struct ltt_session *session;
1248 uint64_t trace_archive_id;
1249
1250 assert(ksess);
1251 assert(ksess->consumer);
1252 assert(output);
1253
1254 DBG("Kernel snapshot record started");
1255
1256 session = session_find_by_id(ksess->id);
1257 assert(session);
1258 assert(pthread_mutex_trylock(&session->lock));
1259 assert(session_trylock_list());
1260 trace_archive_id = session->current_archive_id;
1261
1262 /* Save current metadata since the following calls will change it. */
1263 saved_metadata = ksess->metadata;
1264 saved_metadata_fd = ksess->metadata_stream_fd;
1265
1266 rcu_read_lock();
1267
1268 ret = kernel_open_metadata(ksess);
1269 if (ret < 0) {
1270 ret = LTTNG_ERR_KERN_META_FAIL;
1271 goto error;
1272 }
1273
1274 ret = kernel_open_metadata_stream(ksess);
1275 if (ret < 0) {
1276 ret = LTTNG_ERR_KERN_META_FAIL;
1277 goto error_open_stream;
1278 }
1279
1280 /* Send metadata to consumer and snapshot everything. */
1281 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1282 socket, node.node) {
1283 struct consumer_output *saved_output;
1284 struct ltt_kernel_channel *chan;
1285
1286 /*
1287 * Temporarly switch consumer output for our snapshot output. As long
1288 * as the session lock is taken, this is safe.
1289 */
1290 saved_output = ksess->consumer;
1291 ksess->consumer = output->consumer;
1292
1293 pthread_mutex_lock(socket->lock);
1294 /* This stream must not be monitored by the consumer. */
1295 ret = kernel_consumer_add_metadata(socket, ksess, 0);
1296 pthread_mutex_unlock(socket->lock);
1297 /* Put back the saved consumer output into the session. */
1298 ksess->consumer = saved_output;
1299 if (ret < 0) {
1300 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1301 goto error_consumer;
1302 }
1303
1304 /* For each channel, ask the consumer to snapshot it. */
1305 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1306 ret = consumer_snapshot_channel(socket, chan->key, output, 0,
1307 ksess->uid, ksess->gid,
1308 DEFAULT_KERNEL_TRACE_DIR, wait,
1309 nb_packets_per_stream,
1310 trace_archive_id);
1311 if (ret < 0) {
1312 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1313 (void) kernel_consumer_destroy_metadata(socket,
1314 ksess->metadata);
1315 goto error_consumer;
1316 }
1317 }
1318
1319 /* Snapshot metadata, */
1320 ret = consumer_snapshot_channel(socket, ksess->metadata->key, output,
1321 1, ksess->uid, ksess->gid,
1322 DEFAULT_KERNEL_TRACE_DIR, wait, 0,
1323 trace_archive_id);
1324 if (ret < 0) {
1325 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1326 goto error_consumer;
1327 }
1328
1329 /*
1330 * The metadata snapshot is done, ask the consumer to destroy it since
1331 * it's not monitored on the consumer side.
1332 */
1333 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
1334 }
1335
1336 ret = LTTNG_OK;
1337
1338 error_consumer:
1339 /* Close newly opened metadata stream. It's now on the consumer side. */
1340 err = close(ksess->metadata_stream_fd);
1341 if (err < 0) {
1342 PERROR("close snapshot kernel");
1343 }
1344
1345 error_open_stream:
1346 trace_kernel_destroy_metadata(ksess->metadata);
1347 error:
1348 /* Restore metadata state.*/
1349 ksess->metadata = saved_metadata;
1350 ksess->metadata_stream_fd = saved_metadata_fd;
1351
1352 rcu_read_unlock();
1353 return ret;
1354 }
1355
1356 /*
1357 * Get the syscall mask array from the kernel tracer.
1358 *
1359 * Return 0 on success else a negative value. In both case, syscall_mask should
1360 * be freed.
1361 */
1362 int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
1363 {
1364 assert(syscall_mask);
1365 assert(nr_bits);
1366
1367 return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
1368 }
1369
1370 /*
1371 * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1372 * version number.
1373 *
1374 * Return 1 on success, 0 when feature is not supported, negative value in case
1375 * of errors.
1376 */
1377 int kernel_supports_ring_buffer_snapshot_sample_positions(int tracer_fd)
1378 {
1379 int ret = 0; // Not supported by default
1380 struct lttng_kernel_tracer_abi_version abi;
1381
1382 ret = kernctl_tracer_abi_version(tracer_fd, &abi);
1383 if (ret < 0) {
1384 ERR("Failed to retrieve lttng-modules ABI version");
1385 goto error;
1386 }
1387
1388 /*
1389 * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1390 */
1391 if (abi.major >= 2 && abi.minor >= 3) {
1392 /* Supported */
1393 ret = 1;
1394 } else {
1395 /* Not supported */
1396 ret = 0;
1397 }
1398 error:
1399 return ret;
1400 }
1401
1402 /*
1403 * Rotate a kernel session.
1404 *
1405 * Return 0 on success or else return a LTTNG_ERR code.
1406 */
1407 int kernel_rotate_session(struct ltt_session *session)
1408 {
1409 int ret;
1410 struct consumer_socket *socket;
1411 struct lttng_ht_iter iter;
1412 struct ltt_kernel_session *ksess = session->kernel_session;
1413
1414 assert(ksess);
1415 assert(ksess->consumer);
1416
1417 DBG("Rotate kernel session %s started (session %" PRIu64 ")",
1418 session->name, session->id);
1419
1420 rcu_read_lock();
1421
1422 /*
1423 * Note that this loop will end after one iteration given that there is
1424 * only one kernel consumer.
1425 */
1426 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1427 socket, node.node) {
1428 struct ltt_kernel_channel *chan;
1429
1430 /*
1431 * Account the metadata channel first to make sure the
1432 * number of channels waiting for a rotation cannot
1433 * reach 0 before we complete the iteration over all
1434 * the channels.
1435 */
1436 ret = rotate_add_channel_pending(ksess->metadata->key,
1437 LTTNG_DOMAIN_KERNEL, session);
1438 if (ret < 0) {
1439 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1440 goto error;
1441 }
1442
1443 /* For each channel, ask the consumer to rotate it. */
1444 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1445 ret = rotate_add_channel_pending(chan->key,
1446 LTTNG_DOMAIN_KERNEL, session);
1447 if (ret < 0) {
1448 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1449 goto error;
1450 }
1451
1452 DBG("Rotate channel %" PRIu64 ", session %s", chan->key, session->name);
1453 ret = consumer_rotate_channel(socket, chan->key,
1454 ksess->uid, ksess->gid, ksess->consumer,
1455 ksess->consumer->subdir,
1456 /* is_metadata_channel */ false,
1457 session->current_archive_id,
1458 &session->rotate_pending_relay);
1459 if (ret < 0) {
1460 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1461 goto error;
1462 }
1463 }
1464
1465 /*
1466 * Rotate the metadata channel.
1467 */
1468 ret = consumer_rotate_channel(socket, ksess->metadata->key,
1469 ksess->uid, ksess->gid, ksess->consumer,
1470 ksess->consumer->subdir,
1471 /* is_metadata_channel */ true,
1472 session->current_archive_id,
1473 &session->rotate_pending_relay);
1474 if (ret < 0) {
1475 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1476 goto error;
1477 }
1478 }
1479
1480 ret = LTTNG_OK;
1481
1482 error:
1483 rcu_read_unlock();
1484 return ret;
1485 }
This page took 0.095927 seconds and 6 git commands to generate.