tests: Make test_per_application_leaks more robust
[lttng-tools.git] / src / bin / lttng-sessiond / kernel.c
diff --git a/src/bin/lttng-sessiond/kernel.c b/src/bin/lttng-sessiond/kernel.c
deleted file mode 100644 (file)
index b4fc995..0000000
+++ /dev/null
@@ -1,1461 +0,0 @@
-/*
- * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License, version 2 only,
- * as published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License along
- * with this program; if not, write to the Free Software Foundation, Inc.,
- * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
- */
-
-#define _LGPL_SOURCE
-#include <fcntl.h>
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
-#include <unistd.h>
-#include <inttypes.h>
-
-#include <common/common.h>
-#include <common/kernel-ctl/kernel-ctl.h>
-#include <common/kernel-ctl/kernel-ioctl.h>
-#include <common/sessiond-comm/sessiond-comm.h>
-
-#include "consumer.h"
-#include "kernel.h"
-#include "kernel-consumer.h"
-#include "kern-modules.h"
-#include "utils.h"
-#include "rotate.h"
-
-/*
- * Key used to reference a channel between the sessiond and the consumer. This
- * is only read and updated with the session_list lock held.
- */
-static uint64_t next_kernel_channel_key;
-
-#include <lttng/userspace-probe.h>
-#include <lttng/userspace-probe-internal.h>
-/*
- * Add context on a kernel channel.
- *
- * Assumes the ownership of ctx.
- */
-int kernel_add_channel_context(struct ltt_kernel_channel *chan,
-               struct ltt_kernel_context *ctx)
-{
-       int ret;
-
-       assert(chan);
-       assert(ctx);
-
-       DBG("Adding context to channel %s", chan->channel->name);
-       ret = kernctl_add_context(chan->fd, &ctx->ctx);
-       if (ret < 0) {
-               switch (-ret) {
-               case ENOSYS:
-                       /* Exists but not available for this kernel */
-                       ret = LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE;
-                       goto error;
-               case EEXIST:
-                       /* If EEXIST, we just ignore the error */
-                       ret = 0;
-                       goto end;
-               default:
-                       PERROR("add context ioctl");
-                       ret = LTTNG_ERR_KERN_CONTEXT_FAIL;
-                       goto error;
-               }
-       }
-       ret = 0;
-
-end:
-       cds_list_add_tail(&ctx->list, &chan->ctx_list);
-       ctx->in_list = true;
-       ctx = NULL;
-error:
-       if (ctx) {
-               trace_kernel_destroy_context(ctx);
-       }
-       return ret;
-}
-
-/*
- * Create a new kernel session, register it to the kernel tracer and add it to
- * the session daemon session.
- */
-int kernel_create_session(struct ltt_session *session, int tracer_fd)
-{
-       int ret;
-       struct ltt_kernel_session *lks;
-
-       assert(session);
-
-       /* Allocate data structure */
-       lks = trace_kernel_create_session();
-       if (lks == NULL) {
-               ret = -1;
-               goto error;
-       }
-
-       /* Kernel tracer session creation */
-       ret = kernctl_create_session(tracer_fd);
-       if (ret < 0) {
-               PERROR("ioctl kernel create session");
-               goto error;
-       }
-
-       lks->fd = ret;
-       /* Prevent fd duplication after execlp() */
-       ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
-       if (ret < 0) {
-               PERROR("fcntl session fd");
-       }
-
-       lks->id = session->id;
-       lks->consumer_fds_sent = 0;
-       session->kernel_session = lks;
-
-       DBG("Kernel session created (fd: %d)", lks->fd);
-
-       return 0;
-
-error:
-       if (lks) {
-               trace_kernel_destroy_session(lks);
-       }
-       return ret;
-}
-
-/*
- * Create a kernel channel, register it to the kernel tracer and add it to the
- * kernel session.
- */
-int kernel_create_channel(struct ltt_kernel_session *session,
-               struct lttng_channel *chan)
-{
-       int ret;
-       struct ltt_kernel_channel *lkc;
-
-       assert(session);
-       assert(chan);
-
-       /* Allocate kernel channel */
-       lkc = trace_kernel_create_channel(chan);
-       if (lkc == NULL) {
-               goto error;
-       }
-
-       DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
-                       chan->name, lkc->channel->attr.overwrite,
-                       lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
-                       lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
-                       lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
-
-       /* Kernel tracer channel creation */
-       ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
-       if (ret < 0) {
-               PERROR("ioctl kernel create channel");
-               goto error;
-       }
-
-       /* Setup the channel fd */
-       lkc->fd = ret;
-       /* Prevent fd duplication after execlp() */
-       ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
-       if (ret < 0) {
-               PERROR("fcntl session fd");
-       }
-
-       /* Add channel to session */
-       cds_list_add(&lkc->list, &session->channel_list.head);
-       session->channel_count++;
-       lkc->session = session;
-       lkc->key = ++next_kernel_channel_key;
-
-       DBG("Kernel channel %s created (fd: %d, key: %" PRIu64 ")",
-                       lkc->channel->name, lkc->fd, lkc->key);
-
-       return 0;
-
-error:
-       if (lkc) {
-               free(lkc->channel);
-               free(lkc);
-       }
-       return -1;
-}
-
-/*
- * Compute the offset of the instrumentation byte in the binary based on the
- * function probe location using the ELF lookup method.
- *
- * Returns 0 on success and set the offset out parameter to the offset of the
- * elf symbol
- * Returns -1 on error
- */
-static
-int extract_userspace_probe_offset_function_elf(
-               const struct lttng_userspace_probe_location *probe_location,
-               struct ltt_kernel_session *session, uint64_t *offset)
-{
-       int fd;
-       int ret = 0;
-       const char *symbol = NULL;
-       const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
-       enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
-
-       assert(lttng_userspace_probe_location_get_type(probe_location) ==
-                       LTTNG_USERSPACE_PROBE_LOCATION_TYPE_FUNCTION);
-
-       lookup = lttng_userspace_probe_location_get_lookup_method(
-                       probe_location);
-       if (!lookup) {
-               ret = -1;
-               goto end;
-       }
-
-       lookup_method_type =
-                       lttng_userspace_probe_location_lookup_method_get_type(lookup);
-
-       assert(lookup_method_type ==
-                       LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF);
-
-       symbol = lttng_userspace_probe_location_function_get_function_name(
-                       probe_location);
-       if (!symbol) {
-               ret = -1;
-               goto end;
-       }
-
-       fd = lttng_userspace_probe_location_function_get_binary_fd(probe_location);
-       if (fd < 0) {
-               ret = -1;
-               goto end;
-       }
-
-       ret = run_as_extract_elf_symbol_offset(fd, symbol, session->uid,
-                       session->gid, offset);
-       if (ret < 0) {
-               DBG("userspace probe offset calculation failed for "
-                               "function %s", symbol);
-               goto end;
-       }
-
-       DBG("userspace probe elf offset for %s is 0x%jd", symbol, (intmax_t)(*offset));
-end:
-       return ret;
-}
-
-/*
- * Compute the offsets of the instrumentation bytes in the binary based on the
- * tracepoint probe location using the SDT lookup method. This function
- * allocates the offsets buffer, the caller must free it.
- *
- * Returns 0 on success and set the offset out parameter to the offsets of the
- * SDT tracepoint.
- * Returns -1 on error.
- */
-static
-int extract_userspace_probe_offset_tracepoint_sdt(
-               const struct lttng_userspace_probe_location *probe_location,
-               struct ltt_kernel_session *session, uint64_t **offsets,
-               uint32_t *offsets_count)
-{
-       enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
-       const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
-       const char *probe_name = NULL, *provider_name = NULL;
-       int ret = 0;
-       int fd, i;
-
-       assert(lttng_userspace_probe_location_get_type(probe_location) ==
-                       LTTNG_USERSPACE_PROBE_LOCATION_TYPE_TRACEPOINT);
-
-       lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
-       if (!lookup) {
-               ret = -1;
-               goto end;
-       }
-
-       lookup_method_type =
-                       lttng_userspace_probe_location_lookup_method_get_type(lookup);
-
-       assert(lookup_method_type ==
-                       LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT);
-
-
-       probe_name = lttng_userspace_probe_location_tracepoint_get_probe_name(
-                       probe_location);
-       if (!probe_name) {
-               ret = -1;
-               goto end;
-       }
-
-       provider_name = lttng_userspace_probe_location_tracepoint_get_provider_name(
-                       probe_location);
-       if (!provider_name) {
-               ret = -1;
-               goto end;
-       }
-
-       fd = lttng_userspace_probe_location_tracepoint_get_binary_fd(probe_location);
-       if (fd < 0) {
-               ret = -1;
-               goto end;
-       }
-
-       ret = run_as_extract_sdt_probe_offsets(fd, provider_name, probe_name,
-                       session->uid, session->gid, offsets, offsets_count);
-       if (ret < 0) {
-               DBG("userspace probe offset calculation failed for sdt "
-                               "probe %s:%s", provider_name, probe_name);
-               goto end;
-       }
-
-       if (*offsets_count == 0) {
-               DBG("no userspace probe offset found");
-               goto end;
-       }
-
-       DBG("%u userspace probe SDT offsets found for %s:%s at:",
-                       *offsets_count, provider_name, probe_name);
-       for (i = 0; i < *offsets_count; i++) {
-               DBG("\t0x%jd", (intmax_t)((*offsets)[i]));
-       }
-end:
-       return ret;
-}
-
-/*
- * Extract the offsets of the instrumentation point for the different lookup
- * methods.
- */
-static
-int userspace_probe_add_callsites(struct lttng_event *ev,
-                       struct ltt_kernel_session *session, int fd)
-{
-       const struct lttng_userspace_probe_location_lookup_method *lookup_method = NULL;
-       enum lttng_userspace_probe_location_lookup_method_type type;
-       const struct lttng_userspace_probe_location *location = NULL;
-       int ret;
-
-       assert(ev);
-       assert(ev->type == LTTNG_EVENT_USERSPACE_PROBE);
-
-       location = lttng_event_get_userspace_probe_location(ev);
-       if (!location) {
-               ret = -1;
-               goto end;
-       }
-       lookup_method =
-                       lttng_userspace_probe_location_get_lookup_method(location);
-       if (!lookup_method) {
-               ret = -1;
-               goto end;
-       }
-
-       type = lttng_userspace_probe_location_lookup_method_get_type(lookup_method);
-       switch (type) {
-       case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
-       {
-               struct lttng_kernel_event_callsite callsite;
-               uint64_t offset;
-
-               ret = extract_userspace_probe_offset_function_elf(location, session, &offset);
-               if (ret) {
-                       ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
-                       goto end;
-               }
-
-               callsite.u.uprobe.offset = offset;
-               ret = kernctl_add_callsite(fd, &callsite);
-               if (ret) {
-                       WARN("Adding callsite to userspace probe "
-                                       "event %s failed.", ev->name);
-                       ret = LTTNG_ERR_KERN_ENABLE_FAIL;
-                       goto end;
-               }
-               break;
-       }
-       case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
-       {
-               int i;
-               uint64_t *offsets = NULL;
-               uint32_t offsets_count;
-               struct lttng_kernel_event_callsite callsite;
-
-               /*
-                * This call allocates the offsets buffer. This buffer must be freed
-                * by the caller
-                */
-               ret = extract_userspace_probe_offset_tracepoint_sdt(location, session,
-                               &offsets, &offsets_count);
-               if (ret) {
-                       ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
-                       goto end;
-               }
-               for (i = 0; i < offsets_count; i++) {
-                       callsite.u.uprobe.offset = offsets[i];
-                       ret = kernctl_add_callsite(fd, &callsite);
-                       if (ret) {
-                               WARN("Adding callsite to userspace probe "
-                                               "event %s failed.", ev->name);
-                               ret = LTTNG_ERR_KERN_ENABLE_FAIL;
-                               free(offsets);
-                               goto end;
-                       }
-               }
-               free(offsets);
-               break;
-       }
-       default:
-               ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
-               goto end;
-       }
-end:
-       return ret;
-}
-
-/*
- * Create a kernel event, enable it to the kernel tracer and add it to the
- * channel event list of the kernel session.
- * We own filter_expression and filter.
- */
-int kernel_create_event(struct lttng_event *ev,
-               struct ltt_kernel_channel *channel,
-               char *filter_expression,
-               struct lttng_filter_bytecode *filter)
-{
-       int err, fd;
-       enum lttng_error_code ret;
-       struct ltt_kernel_event *event;
-
-       assert(ev);
-       assert(channel);
-
-       /* We pass ownership of filter_expression and filter */
-       ret = trace_kernel_create_event(ev, filter_expression,
-                       filter, &event);
-       if (ret != LTTNG_OK) {
-               goto error;
-       }
-
-       fd = kernctl_create_event(channel->fd, event->event);
-       if (fd < 0) {
-               switch (-fd) {
-               case EEXIST:
-                       ret = LTTNG_ERR_KERN_EVENT_EXIST;
-                       break;
-               case ENOSYS:
-                       WARN("Event type not implemented");
-                       ret = LTTNG_ERR_KERN_EVENT_ENOSYS;
-                       break;
-               case ENOENT:
-                       WARN("Event %s not found!", ev->name);
-                       ret = LTTNG_ERR_KERN_ENABLE_FAIL;
-                       break;
-               default:
-                       ret = LTTNG_ERR_KERN_ENABLE_FAIL;
-                       PERROR("create event ioctl");
-               }
-               goto free_event;
-       }
-
-       event->type = ev->type;
-       event->fd = fd;
-       /* Prevent fd duplication after execlp() */
-       err = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
-       if (err < 0) {
-               PERROR("fcntl session fd");
-       }
-
-       if (filter) {
-               err = kernctl_filter(event->fd, filter);
-               if (err < 0) {
-                       switch (-err) {
-                       case ENOMEM:
-                               ret = LTTNG_ERR_FILTER_NOMEM;
-                               break;
-                       default:
-                               ret = LTTNG_ERR_FILTER_INVAL;
-                               break;
-                       }
-                       goto filter_error;
-               }
-       }
-
-       if (ev->type == LTTNG_EVENT_USERSPACE_PROBE) {
-               ret = userspace_probe_add_callsites(ev, channel->session, event->fd);
-               if (ret) {
-                       goto add_callsite_error;
-               }
-       }
-
-       err = kernctl_enable(event->fd);
-       if (err < 0) {
-               switch (-err) {
-               case EEXIST:
-                       ret = LTTNG_ERR_KERN_EVENT_EXIST;
-                       break;
-               default:
-                       PERROR("enable kernel event");
-                       ret = LTTNG_ERR_KERN_ENABLE_FAIL;
-                       break;
-               }
-               goto enable_error;
-       }
-
-       /* Add event to event list */
-       cds_list_add(&event->list, &channel->events_list.head);
-       channel->event_count++;
-
-       DBG("Event %s created (fd: %d)", ev->name, event->fd);
-
-       return 0;
-
-add_callsite_error:
-enable_error:
-filter_error:
-       {
-               int closeret;
-
-               closeret = close(event->fd);
-               if (closeret) {
-                       PERROR("close event fd");
-               }
-       }
-free_event:
-       free(event);
-error:
-       return ret;
-}
-
-/*
- * Disable a kernel channel.
- */
-int kernel_disable_channel(struct ltt_kernel_channel *chan)
-{
-       int ret;
-
-       assert(chan);
-
-       ret = kernctl_disable(chan->fd);
-       if (ret < 0) {
-               PERROR("disable chan ioctl");
-               goto error;
-       }
-
-       chan->enabled = 0;
-       DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64 ")",
-                       chan->channel->name, chan->fd, chan->key);
-
-       return 0;
-
-error:
-       return ret;
-}
-
-/*
- * Enable a kernel channel.
- */
-int kernel_enable_channel(struct ltt_kernel_channel *chan)
-{
-       int ret;
-
-       assert(chan);
-
-       ret = kernctl_enable(chan->fd);
-       if (ret < 0 && ret != -EEXIST) {
-               PERROR("Enable kernel chan");
-               goto error;
-       }
-
-       chan->enabled = 1;
-       DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64 ")",
-                       chan->channel->name, chan->fd, chan->key);
-
-       return 0;
-
-error:
-       return ret;
-}
-
-/*
- * Enable a kernel event.
- */
-int kernel_enable_event(struct ltt_kernel_event *event)
-{
-       int ret;
-
-       assert(event);
-
-       ret = kernctl_enable(event->fd);
-       if (ret < 0) {
-               switch (-ret) {
-               case EEXIST:
-                       ret = LTTNG_ERR_KERN_EVENT_EXIST;
-                       break;
-               default:
-                       PERROR("enable kernel event");
-                       break;
-               }
-               goto error;
-       }
-
-       event->enabled = 1;
-       DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
-
-       return 0;
-
-error:
-       return ret;
-}
-
-/*
- * Disable a kernel event.
- */
-int kernel_disable_event(struct ltt_kernel_event *event)
-{
-       int ret;
-
-       assert(event);
-
-       ret = kernctl_disable(event->fd);
-       if (ret < 0) {
-               switch (-ret) {
-               case EEXIST:
-                       ret = LTTNG_ERR_KERN_EVENT_EXIST;
-                       break;
-               default:
-                       PERROR("disable kernel event");
-                       break;
-               }
-               goto error;
-       }
-
-       event->enabled = 0;
-       DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
-
-       return 0;
-
-error:
-       return ret;
-}
-
-
-int kernel_track_pid(struct ltt_kernel_session *session, int pid)
-{
-       int ret;
-
-       DBG("Kernel track PID %d for session id %" PRIu64 ".",
-                       pid, session->id);
-       ret = kernctl_track_pid(session->fd, pid);
-       if (!ret) {
-               return LTTNG_OK;
-       }
-       switch (-ret) {
-       case EINVAL:
-               return LTTNG_ERR_INVALID;
-       case ENOMEM:
-               return LTTNG_ERR_NOMEM;
-       case EEXIST:
-               return LTTNG_ERR_PID_TRACKED;
-       default:
-               return LTTNG_ERR_UNK;
-       }
-}
-
-int kernel_untrack_pid(struct ltt_kernel_session *session, int pid)
-{
-       int ret;
-
-       DBG("Kernel untrack PID %d for session id %" PRIu64 ".",
-                       pid, session->id);
-       ret = kernctl_untrack_pid(session->fd, pid);
-       if (!ret) {
-               return LTTNG_OK;
-       }
-       switch (-ret) {
-       case EINVAL:
-               return LTTNG_ERR_INVALID;
-       case ENOMEM:
-               return LTTNG_ERR_NOMEM;
-       case ENOENT:
-               return LTTNG_ERR_PID_NOT_TRACKED;
-       default:
-               return LTTNG_ERR_UNK;
-       }
-}
-
-ssize_t kernel_list_tracker_pids(struct ltt_kernel_session *session,
-               int **_pids)
-{
-       int fd, ret;
-       int pid;
-       ssize_t nbmem, count = 0;
-       FILE *fp;
-       int *pids;
-
-       fd = kernctl_list_tracker_pids(session->fd);
-       if (fd < 0) {
-               PERROR("kernel tracker pids list");
-               goto error;
-       }
-
-       fp = fdopen(fd, "r");
-       if (fp == NULL) {
-               PERROR("kernel tracker pids list fdopen");
-               goto error_fp;
-       }
-
-       nbmem = KERNEL_TRACKER_PIDS_INIT_LIST_SIZE;
-       pids = zmalloc(sizeof(*pids) * nbmem);
-       if (pids == NULL) {
-               PERROR("alloc list pids");
-               count = -ENOMEM;
-               goto end;
-       }
-
-       while (fscanf(fp, "process { pid = %u; };\n", &pid) == 1) {
-               if (count >= nbmem) {
-                       int *new_pids;
-                       size_t new_nbmem;
-
-                       new_nbmem = nbmem << 1;
-                       DBG("Reallocating pids list from %zu to %zu entries",
-                                       nbmem, new_nbmem);
-                       new_pids = realloc(pids, new_nbmem * sizeof(*new_pids));
-                       if (new_pids == NULL) {
-                               PERROR("realloc list events");
-                               free(pids);
-                               count = -ENOMEM;
-                               goto end;
-                       }
-                       /* Zero the new memory */
-                       memset(new_pids + nbmem, 0,
-                               (new_nbmem - nbmem) * sizeof(*new_pids));
-                       nbmem = new_nbmem;
-                       pids = new_pids;
-               }
-               pids[count++] = pid;
-       }
-
-       *_pids = pids;
-       DBG("Kernel list tracker pids done (%zd pids)", count);
-end:
-       ret = fclose(fp);       /* closes both fp and fd */
-       if (ret) {
-               PERROR("fclose");
-       }
-       return count;
-
-error_fp:
-       ret = close(fd);
-       if (ret) {
-               PERROR("close");
-       }
-error:
-       return -1;
-}
-
-/*
- * Create kernel metadata, open from the kernel tracer and add it to the
- * kernel session.
- */
-int kernel_open_metadata(struct ltt_kernel_session *session)
-{
-       int ret;
-       struct ltt_kernel_metadata *lkm = NULL;
-
-       assert(session);
-
-       /* Allocate kernel metadata */
-       lkm = trace_kernel_create_metadata();
-       if (lkm == NULL) {
-               goto error;
-       }
-
-       /* Kernel tracer metadata creation */
-       ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
-       if (ret < 0) {
-               goto error_open;
-       }
-
-       lkm->fd = ret;
-       lkm->key = ++next_kernel_channel_key;
-       /* Prevent fd duplication after execlp() */
-       ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
-       if (ret < 0) {
-               PERROR("fcntl session fd");
-       }
-
-       session->metadata = lkm;
-
-       DBG("Kernel metadata opened (fd: %d)", lkm->fd);
-
-       return 0;
-
-error_open:
-       trace_kernel_destroy_metadata(lkm);
-error:
-       return -1;
-}
-
-/*
- * Start tracing session.
- */
-int kernel_start_session(struct ltt_kernel_session *session)
-{
-       int ret;
-
-       assert(session);
-
-       ret = kernctl_start_session(session->fd);
-       if (ret < 0) {
-               PERROR("ioctl start session");
-               goto error;
-       }
-
-       DBG("Kernel session started");
-
-       return 0;
-
-error:
-       return ret;
-}
-
-/*
- * Make a kernel wait to make sure in-flight probe have completed.
- */
-void kernel_wait_quiescent(int fd)
-{
-       int ret;
-
-       DBG("Kernel quiescent wait on %d", fd);
-
-       ret = kernctl_wait_quiescent(fd);
-       if (ret < 0) {
-               PERROR("wait quiescent ioctl");
-               ERR("Kernel quiescent wait failed");
-       }
-}
-
-/*
- *  Force flush buffer of metadata.
- */
-int kernel_metadata_flush_buffer(int fd)
-{
-       int ret;
-
-       DBG("Kernel flushing metadata buffer on fd %d", fd);
-
-       ret = kernctl_buffer_flush(fd);
-       if (ret < 0) {
-               ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
-       }
-
-       return 0;
-}
-
-/*
- * Force flush buffer for channel.
- */
-int kernel_flush_buffer(struct ltt_kernel_channel *channel)
-{
-       int ret;
-       struct ltt_kernel_stream *stream;
-
-       assert(channel);
-
-       DBG("Flush buffer for channel %s", channel->channel->name);
-
-       cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
-               DBG("Flushing channel stream %d", stream->fd);
-               ret = kernctl_buffer_flush(stream->fd);
-               if (ret < 0) {
-                       PERROR("ioctl");
-                       ERR("Fail to flush buffer for stream %d (ret: %d)",
-                                       stream->fd, ret);
-               }
-       }
-
-       return 0;
-}
-
-/*
- * Stop tracing session.
- */
-int kernel_stop_session(struct ltt_kernel_session *session)
-{
-       int ret;
-
-       assert(session);
-
-       ret = kernctl_stop_session(session->fd);
-       if (ret < 0) {
-               goto error;
-       }
-
-       DBG("Kernel session stopped");
-
-       return 0;
-
-error:
-       return ret;
-}
-
-/*
- * Open stream of channel, register it to the kernel tracer and add it
- * to the stream list of the channel.
- *
- * Note: given that the streams may appear in random order wrt CPU
- * number (e.g. cpu hotplug), the index value of the stream number in
- * the stream name is not necessarily linked to the CPU number.
- *
- * Return the number of created stream. Else, a negative value.
- */
-int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
-{
-       int ret;
-       struct ltt_kernel_stream *lks;
-
-       assert(channel);
-
-       while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
-               lks = trace_kernel_create_stream(channel->channel->name,
-                               channel->stream_count);
-               if (lks == NULL) {
-                       ret = close(ret);
-                       if (ret) {
-                               PERROR("close");
-                       }
-                       goto error;
-               }
-
-               lks->fd = ret;
-               /* Prevent fd duplication after execlp() */
-               ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
-               if (ret < 0) {
-                       PERROR("fcntl session fd");
-               }
-
-               lks->tracefile_size = channel->channel->attr.tracefile_size;
-               lks->tracefile_count = channel->channel->attr.tracefile_count;
-
-               /* Add stream to channel stream list */
-               cds_list_add(&lks->list, &channel->stream_list.head);
-               channel->stream_count++;
-
-               DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
-                               lks->state);
-       }
-
-       return channel->stream_count;
-
-error:
-       return -1;
-}
-
-/*
- * Open the metadata stream and set it to the kernel session.
- */
-int kernel_open_metadata_stream(struct ltt_kernel_session *session)
-{
-       int ret;
-
-       assert(session);
-
-       ret = kernctl_create_stream(session->metadata->fd);
-       if (ret < 0) {
-               PERROR("kernel create metadata stream");
-               goto error;
-       }
-
-       DBG("Kernel metadata stream created (fd: %d)", ret);
-       session->metadata_stream_fd = ret;
-       /* Prevent fd duplication after execlp() */
-       ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
-       if (ret < 0) {
-               PERROR("fcntl session fd");
-       }
-
-       return 0;
-
-error:
-       return -1;
-}
-
-/*
- * Get the event list from the kernel tracer and return the number of elements.
- */
-ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
-{
-       int fd, ret;
-       char *event;
-       size_t nbmem, count = 0;
-       FILE *fp;
-       struct lttng_event *elist;
-
-       assert(events);
-
-       fd = kernctl_tracepoint_list(tracer_fd);
-       if (fd < 0) {
-               PERROR("kernel tracepoint list");
-               goto error;
-       }
-
-       fp = fdopen(fd, "r");
-       if (fp == NULL) {
-               PERROR("kernel tracepoint list fdopen");
-               goto error_fp;
-       }
-
-       /*
-        * Init memory size counter
-        * See kernel-ctl.h for explanation of this value
-        */
-       nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
-       elist = zmalloc(sizeof(struct lttng_event) * nbmem);
-       if (elist == NULL) {
-               PERROR("alloc list events");
-               count = -ENOMEM;
-               goto end;
-       }
-
-       while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
-               if (count >= nbmem) {
-                       struct lttng_event *new_elist;
-                       size_t new_nbmem;
-
-                       new_nbmem = nbmem << 1;
-                       DBG("Reallocating event list from %zu to %zu bytes",
-                                       nbmem, new_nbmem);
-                       new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
-                       if (new_elist == NULL) {
-                               PERROR("realloc list events");
-                               free(event);
-                               free(elist);
-                               count = -ENOMEM;
-                               goto end;
-                       }
-                       /* Zero the new memory */
-                       memset(new_elist + nbmem, 0,
-                               (new_nbmem - nbmem) * sizeof(struct lttng_event));
-                       nbmem = new_nbmem;
-                       elist = new_elist;
-               }
-               strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
-               elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
-               elist[count].enabled = -1;
-               count++;
-               free(event);
-       }
-
-       *events = elist;
-       DBG("Kernel list events done (%zu events)", count);
-end:
-       ret = fclose(fp);       /* closes both fp and fd */
-       if (ret) {
-               PERROR("fclose");
-       }
-       return count;
-
-error_fp:
-       ret = close(fd);
-       if (ret) {
-               PERROR("close");
-       }
-error:
-       return -1;
-}
-
-/*
- * Get kernel version and validate it.
- */
-int kernel_validate_version(int tracer_fd,
-               struct lttng_kernel_tracer_version *version,
-               struct lttng_kernel_tracer_abi_version *abi_version)
-{
-       int ret;
-
-       ret = kernctl_tracer_version(tracer_fd, version);
-       if (ret < 0) {
-               ERR("Failed to retrieve the lttng-modules version");
-               goto error;
-       }
-
-       /* Validate version */
-       if (version->major != VERSION_MAJOR) {
-               ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
-                       version->major, VERSION_MAJOR);
-               goto error_version;
-       }
-       ret = kernctl_tracer_abi_version(tracer_fd, abi_version);
-       if (ret < 0) {
-               ERR("Failed to retrieve lttng-modules ABI version");
-               goto error;
-       }
-       if (abi_version->major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
-               ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
-                       abi_version->major, abi_version->minor,
-                       LTTNG_MODULES_ABI_MAJOR_VERSION);
-               goto error;
-       }
-       DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
-                       version->major, version->minor,
-                       abi_version->major, abi_version->minor);
-       return 0;
-
-error_version:
-       ret = -1;
-
-error:
-       ERR("Kernel tracer version check failed; kernel tracing will not be available");
-       return ret;
-}
-
-/*
- * Kernel work-arounds called at the start of sessiond main().
- */
-int init_kernel_workarounds(void)
-{
-       int ret;
-       FILE *fp;
-
-       /*
-        * boot_id needs to be read once before being used concurrently
-        * to deal with a Linux kernel race. A fix is proposed for
-        * upstream, but the work-around is needed for older kernels.
-        */
-       fp = fopen("/proc/sys/kernel/random/boot_id", "r");
-       if (!fp) {
-               goto end_boot_id;
-       }
-       while (!feof(fp)) {
-               char buf[37] = "";
-
-               ret = fread(buf, 1, sizeof(buf), fp);
-               if (ret < 0) {
-                       /* Ignore error, we don't really care */
-               }
-       }
-       ret = fclose(fp);
-       if (ret) {
-               PERROR("fclose");
-       }
-end_boot_id:
-       return 0;
-}
-
-/*
- * Complete teardown of a kernel session.
- */
-void kernel_destroy_session(struct ltt_kernel_session *ksess)
-{
-       if (ksess == NULL) {
-               DBG3("No kernel session when tearing down session");
-               return;
-       }
-
-       DBG("Tearing down kernel session");
-
-       /*
-        * Destroy channels on the consumer if at least one FD has been sent and we
-        * are in no output mode because the streams are in *no* monitor mode so we
-        * have to send a command to clean them up or else they leaked.
-        */
-       if (!ksess->output_traces && ksess->consumer_fds_sent) {
-               int ret;
-               struct consumer_socket *socket;
-               struct lttng_ht_iter iter;
-
-               /* For each consumer socket. */
-               rcu_read_lock();
-               cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
-                               socket, node.node) {
-                       struct ltt_kernel_channel *chan;
-
-                       /* For each channel, ask the consumer to destroy it. */
-                       cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
-                               ret = kernel_consumer_destroy_channel(socket, chan);
-                               if (ret < 0) {
-                                       /* Consumer is probably dead. Use next socket. */
-                                       continue;
-                               }
-                       }
-               }
-               rcu_read_unlock();
-       }
-
-       /* Close any relayd session */
-       consumer_output_send_destroy_relayd(ksess->consumer);
-
-       trace_kernel_destroy_session(ksess);
-}
-
-/*
- * Destroy a kernel channel object. It does not do anything on the tracer side.
- */
-void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
-{
-       struct ltt_kernel_session *ksess = NULL;
-
-       assert(kchan);
-       assert(kchan->channel);
-
-       DBG3("Kernel destroy channel %s", kchan->channel->name);
-
-       /* Update channel count of associated session. */
-       if (kchan->session) {
-               /* Keep pointer reference so we can update it after the destroy. */
-               ksess = kchan->session;
-       }
-
-       trace_kernel_destroy_channel(kchan);
-
-       /*
-        * At this point the kernel channel is not visible anymore. This is safe
-        * since in order to work on a visible kernel session, the tracing session
-        * lock (ltt_session.lock) MUST be acquired.
-        */
-       if (ksess) {
-               ksess->channel_count--;
-       }
-}
-
-/*
- * Take a snapshot for a given kernel session.
- *
- * Return LTTNG_OK on success or else return a LTTNG_ERR code.
- */
-enum lttng_error_code kernel_snapshot_record(struct ltt_kernel_session *ksess,
-               struct snapshot_output *output, int wait,
-               uint64_t nb_packets_per_stream)
-{
-       int err, ret, saved_metadata_fd;
-       enum lttng_error_code status = LTTNG_OK;
-       struct consumer_socket *socket;
-       struct lttng_ht_iter iter;
-       struct ltt_kernel_metadata *saved_metadata;
-       struct ltt_session *session = NULL;
-       uint64_t trace_archive_id;
-
-       assert(ksess);
-       assert(ksess->consumer);
-       assert(output);
-
-       DBG("Kernel snapshot record started");
-
-       session = session_find_by_id(ksess->id);
-       assert(session);
-       assert(pthread_mutex_trylock(&session->lock));
-       assert(session_trylock_list());
-       trace_archive_id = session->current_archive_id;
-
-       /* Save current metadata since the following calls will change it. */
-       saved_metadata = ksess->metadata;
-       saved_metadata_fd = ksess->metadata_stream_fd;
-
-       rcu_read_lock();
-
-       ret = kernel_open_metadata(ksess);
-       if (ret < 0) {
-               status = LTTNG_ERR_KERN_META_FAIL;
-               goto error;
-       }
-
-       ret = kernel_open_metadata_stream(ksess);
-       if (ret < 0) {
-               status = LTTNG_ERR_KERN_META_FAIL;
-               goto error_open_stream;
-       }
-
-       /* Send metadata to consumer and snapshot everything. */
-       cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
-                       socket, node.node) {
-               struct consumer_output *saved_output;
-               struct ltt_kernel_channel *chan;
-
-               /*
-                * Temporarly switch consumer output for our snapshot output. As long
-                * as the session lock is taken, this is safe.
-                */
-               saved_output = ksess->consumer;
-               ksess->consumer = output->consumer;
-
-               pthread_mutex_lock(socket->lock);
-               /* This stream must not be monitored by the consumer. */
-               ret = kernel_consumer_add_metadata(socket, ksess, 0);
-               pthread_mutex_unlock(socket->lock);
-               /* Put back the saved consumer output into the session. */
-               ksess->consumer = saved_output;
-               if (ret < 0) {
-                       status = LTTNG_ERR_KERN_META_FAIL;
-                       goto error_consumer;
-               }
-
-               /* For each channel, ask the consumer to snapshot it. */
-               cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
-                       status = consumer_snapshot_channel(socket, chan->key, output, 0,
-                                       ksess->uid, ksess->gid,
-                                       DEFAULT_KERNEL_TRACE_DIR, wait,
-                                       nb_packets_per_stream,
-                                       trace_archive_id);
-                       if (status != LTTNG_OK) {
-                               (void) kernel_consumer_destroy_metadata(socket,
-                                               ksess->metadata);
-                               goto error_consumer;
-                       }
-               }
-
-               /* Snapshot metadata, */
-               status = consumer_snapshot_channel(socket, ksess->metadata->key, output,
-                               1, ksess->uid, ksess->gid,
-                               DEFAULT_KERNEL_TRACE_DIR, wait, 0,
-                               trace_archive_id);
-               if (status != LTTNG_OK) {
-                       goto error_consumer;
-               }
-
-               /*
-                * The metadata snapshot is done, ask the consumer to destroy it since
-                * it's not monitored on the consumer side.
-                */
-               (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
-       }
-
-error_consumer:
-       /* Close newly opened metadata stream. It's now on the consumer side. */
-       err = close(ksess->metadata_stream_fd);
-       if (err < 0) {
-               PERROR("close snapshot kernel");
-       }
-
-error_open_stream:
-       trace_kernel_destroy_metadata(ksess->metadata);
-error:
-       /* Restore metadata state.*/
-       ksess->metadata = saved_metadata;
-       ksess->metadata_stream_fd = saved_metadata_fd;
-       if (session) {
-               session_put(session);
-       }
-       rcu_read_unlock();
-       return status;
-}
-
-/*
- * Get the syscall mask array from the kernel tracer.
- *
- * Return 0 on success else a negative value. In both case, syscall_mask should
- * be freed.
- */
-int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
-{
-       assert(syscall_mask);
-       assert(nr_bits);
-
-       return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
-}
-
-/*
- * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
- * version number.
- *
- * Return 1 on success, 0 when feature is not supported, negative value in case
- * of errors.
- */
-int kernel_supports_ring_buffer_snapshot_sample_positions(int tracer_fd)
-{
-       int ret = 0; // Not supported by default
-       struct lttng_kernel_tracer_abi_version abi;
-
-       ret = kernctl_tracer_abi_version(tracer_fd, &abi);
-       if (ret < 0) {
-               ERR("Failed to retrieve lttng-modules ABI version");
-               goto error;
-       }
-
-       /*
-        * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
-        */
-       if (abi.major >= 2 && abi.minor >= 3) {
-               /* Supported */
-               ret = 1;
-       } else {
-               /* Not supported */
-               ret = 0;
-       }
-error:
-       return ret;
-}
-
-/*
- * Rotate a kernel session.
- *
- * Return LTTNG_OK on success or else an LTTng error code.
- */
-enum lttng_error_code kernel_rotate_session(struct ltt_session *session)
-{
-       int ret;
-       enum lttng_error_code status = LTTNG_OK;
-       struct consumer_socket *socket;
-       struct lttng_ht_iter iter;
-       struct ltt_kernel_session *ksess = session->kernel_session;
-
-       assert(ksess);
-       assert(ksess->consumer);
-
-       DBG("Rotate kernel session %s started (session %" PRIu64 ")",
-                       session->name, session->id);
-
-       rcu_read_lock();
-
-       /*
-        * Note that this loop will end after one iteration given that there is
-        * only one kernel consumer.
-        */
-       cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
-                       socket, node.node) {
-               struct ltt_kernel_channel *chan;
-
-               /* For each channel, ask the consumer to rotate it. */
-               cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
-                       DBG("Rotate kernel channel %" PRIu64 ", session %s",
-                                       chan->key, session->name);
-                       ret = consumer_rotate_channel(socket, chan->key,
-                                       ksess->uid, ksess->gid, ksess->consumer,
-                                       ksess->consumer->domain_subdir,
-                                       /* is_metadata_channel */ false,
-                                       session->current_archive_id);
-                       if (ret < 0) {
-                               status = LTTNG_ERR_KERN_CONSUMER_FAIL;
-                               goto error;
-                       }
-               }
-
-               /*
-                * Rotate the metadata channel.
-                */
-               ret = consumer_rotate_channel(socket, ksess->metadata->key,
-                               ksess->uid, ksess->gid, ksess->consumer,
-                               ksess->consumer->domain_subdir,
-                               /* is_metadata_channel */ true,
-                               session->current_archive_id);
-               if (ret < 0) {
-                       status = LTTNG_ERR_KERN_CONSUMER_FAIL;
-                       goto error;
-               }
-       }
-
-error:
-       rcu_read_unlock();
-       return status;
-}
This page took 0.038035 seconds and 4 git commands to generate.