common: compile libfd-tracker as C++
authorSimon Marchi <simon.marchi@efficios.com>
Wed, 22 Sep 2021 01:06:30 +0000 (21:06 -0400)
committerJérémie Galarneau <jeremie.galarneau@efficios.com>
Wed, 17 Nov 2021 23:58:46 +0000 (18:58 -0500)
Change-Id: I6eed13d24dd8a306f4aa474f7d739931f0750635
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
src/common/fd-tracker/Makefile.am
src/common/fd-tracker/fd-tracker.c [deleted file]
src/common/fd-tracker/fd-tracker.cpp [new file with mode: 0644]
src/common/fd-tracker/inode.c [deleted file]
src/common/fd-tracker/inode.cpp [new file with mode: 0644]
src/common/fd-tracker/utils-poll.c [deleted file]
src/common/fd-tracker/utils-poll.cpp [new file with mode: 0644]
src/common/fd-tracker/utils.c [deleted file]
src/common/fd-tracker/utils.cpp [new file with mode: 0644]

index 05ca8e84ba93ca88844cd148fd2c398630cb90e7..68e63d42b3d997f0a319813d6f3456dc8b29e828 100644 (file)
@@ -2,7 +2,11 @@
 
 noinst_LTLIBRARIES = libfd-tracker.la
 
-libfd_tracker_la_SOURCES = fd-tracker.h fd-tracker.c \
-                           utils.h utils.c \
-                           inode.h inode.c \
-                           utils-poll.c
+libfd_tracker_la_SOURCES = \
+       fd-tracker.cpp \
+       fd-tracker.h \
+       inode.cpp \
+       inode.h \
+       utils.cpp \
+       utils.h \
+       utils-poll.cpp
diff --git a/src/common/fd-tracker/fd-tracker.c b/src/common/fd-tracker/fd-tracker.c
deleted file mode 100644 (file)
index a9921fa..0000000
+++ /dev/null
@@ -1,946 +0,0 @@
-/*
- * Copyright (C) 2018-2020 Jérémie Galarneau <jeremie.galarneau@efficios.com>
- *
- * SPDX-License-Identifier: GPL-2.0-only
- *
- */
-
-#include <urcu.h>
-#include <urcu/list.h>
-#include <urcu/rculfhash.h>
-
-#include <fcntl.h>
-#include <inttypes.h>
-#include <pthread.h>
-#include <stdbool.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-
-#include <common/defaults.h>
-#include <common/error.h>
-#include <common/fs-handle-internal.h>
-#include <common/hashtable/hashtable.h>
-#include <common/hashtable/utils.h>
-#include <common/macros.h>
-#include <common/optional.h>
-
-#include "fd-tracker.h"
-#include "inode.h"
-
-/* Tracker lock must be taken by the user. */
-#define TRACKED_COUNT(tracker)                                 \
-       (tracker->count.suspendable.active +                   \
-                       tracker->count.suspendable.suspended + \
-                       tracker->count.unsuspendable)
-
-/* Tracker lock must be taken by the user. */
-#define ACTIVE_COUNT(tracker) \
-       (tracker->count.suspendable.active + tracker->count.unsuspendable)
-
-/* Tracker lock must be taken by the user. */
-#define SUSPENDED_COUNT(tracker) (tracker->count.suspendable.suspended)
-
-/* Tracker lock must be taken by the user. */
-#define SUSPENDABLE_COUNT(tracker)           \
-       (tracker->count.suspendable.active + \
-                       tracker->count.suspendable.suspended)
-
-/* Tracker lock must be taken by the user. */
-#define UNSUSPENDABLE_COUNT(tracker) (tracker->count.unsuspendable)
-
-struct fd_tracker {
-       pthread_mutex_t lock;
-       struct {
-               struct {
-                       unsigned int active;
-                       unsigned int suspended;
-               } suspendable;
-               unsigned int unsuspendable;
-       } count;
-       unsigned int capacity;
-       struct {
-               uint64_t uses;
-               uint64_t misses;
-               /* Failures to suspend or restore fs handles. */
-               uint64_t errors;
-       } stats;
-       /*
-        * The head of the active_handles list is always the least recently
-        * used active handle. When an handle is used, it is removed from the
-        * list and added to the end. When a file has to be suspended, the
-        * first element in the list is "popped", suspended, and added to the
-        * list of suspended handles.
-        */
-       struct cds_list_head active_handles;
-       struct cds_list_head suspended_handles;
-       struct cds_lfht *unsuspendable_fds;
-       struct lttng_inode_registry *inode_registry;
-       /* Unlinked files are moved in this directory under a unique name. */
-       struct lttng_directory_handle *unlink_directory_handle;
-       struct lttng_unlinked_file_pool *unlinked_file_pool;
-};
-
-struct open_properties {
-       int flags;
-       LTTNG_OPTIONAL(mode_t) mode;
-};
-
-/*
- * A fs_handle_tracked is not ref-counted. Therefore, it is assumed that a
- * handle is never in-use while it is being reclaimed. It can be
- * shared by multiple threads, but external synchronization is required
- * to ensure it is not still being used when it is reclaimed (close method).
- * In this respect, it is not different from a regular file descriptor.
- *
- * The fs_handle lock always nests _within_ the tracker's lock.
- */
-struct fs_handle_tracked {
-       struct fs_handle parent;
-       pthread_mutex_t lock;
-       /*
-        * Weak reference to the tracker. All fs_handles are assumed to have
-        * been closed at the moment of the destruction of the fd_tracker.
-        */
-       struct fd_tracker *tracker;
-       struct open_properties properties;
-       struct lttng_inode *inode;
-       int fd;
-       /* inode number of the file at the time of the handle's creation. */
-       uint64_t ino;
-       bool in_use;
-       /* Offset to which the file should be restored. */
-       off_t offset;
-       struct cds_list_head handles_list_node;
-};
-
-struct unsuspendable_fd {
-       /*
-        * Accesses are only performed through the tracker, which is protected
-        * by its own lock.
-        */
-       int fd;
-       char *name;
-       struct cds_lfht_node tracker_node;
-       struct rcu_head rcu_head;
-};
-
-static struct {
-       pthread_mutex_t lock;
-       bool initialized;
-       unsigned long value;
-} seed = {
-       .lock = PTHREAD_MUTEX_INITIALIZER,
-};
-
-static int match_fd(struct cds_lfht_node *node, const void *key);
-static void unsuspendable_fd_destroy(struct unsuspendable_fd *entry);
-static struct unsuspendable_fd *unsuspendable_fd_create(
-               const char *name, int fd);
-static int open_from_properties(const struct lttng_directory_handle *dir_handle,
-               const char *path, struct open_properties *properties);
-
-static void fs_handle_tracked_log(struct fs_handle_tracked *handle);
-static int fs_handle_tracked_suspend(struct fs_handle_tracked *handle);
-static int fs_handle_tracked_restore(struct fs_handle_tracked *handle);
-static int fs_handle_tracked_get_fd(struct fs_handle *_handle);
-static void fs_handle_tracked_put_fd(struct fs_handle *_handle);
-static int fs_handle_tracked_unlink(struct fs_handle *_handle);
-static int fs_handle_tracked_close(struct fs_handle *_handle);
-
-static void fd_tracker_track(
-               struct fd_tracker *tracker, struct fs_handle_tracked *handle);
-static void fd_tracker_untrack(
-               struct fd_tracker *tracker, struct fs_handle_tracked *handle);
-static int fd_tracker_suspend_handles(
-               struct fd_tracker *tracker, unsigned int count);
-static int fd_tracker_restore_handle(
-               struct fd_tracker *tracker, struct fs_handle_tracked *handle);
-
-/* Match function of the tracker's unsuspendable_fds hash table. */
-static int match_fd(struct cds_lfht_node *node, const void *key)
-{
-       struct unsuspendable_fd *entry = caa_container_of(
-                       node, struct unsuspendable_fd, tracker_node);
-
-       return hash_match_key_ulong(
-                       (void *) (unsigned long) entry->fd, (void *) key);
-}
-
-static void delete_unsuspendable_fd(struct rcu_head *head)
-{
-       struct unsuspendable_fd *fd = caa_container_of(
-                       head, struct unsuspendable_fd, rcu_head);
-
-       free(fd->name);
-       free(fd);
-}
-
-static void unsuspendable_fd_destroy(struct unsuspendable_fd *entry)
-{
-       if (!entry) {
-               return;
-       }
-       call_rcu(&entry->rcu_head, delete_unsuspendable_fd);
-}
-
-static struct unsuspendable_fd *unsuspendable_fd_create(
-               const char *name, int fd)
-{
-       struct unsuspendable_fd *entry = zmalloc(sizeof(*entry));
-
-       if (!entry) {
-               goto error;
-       }
-       if (name) {
-               entry->name = strdup(name);
-               if (!entry->name) {
-                       goto error;
-               }
-       }
-       cds_lfht_node_init(&entry->tracker_node);
-       entry->fd = fd;
-       return entry;
-error:
-       unsuspendable_fd_destroy(entry);
-       return NULL;
-}
-
-static void fs_handle_tracked_log(struct fs_handle_tracked *handle)
-{
-       const char *path;
-
-       pthread_mutex_lock(&handle->lock);
-       lttng_inode_borrow_location(handle->inode, NULL, &path);
-
-       if (handle->fd >= 0) {
-               DBG_NO_LOC("    %s [active, fd %d%s]", path, handle->fd,
-                               handle->in_use ? ", in use" : "");
-       } else {
-               DBG_NO_LOC("    %s [suspended]", path);
-       }
-       pthread_mutex_unlock(&handle->lock);
-}
-
-/* Tracker lock must be held by the caller. */
-static int fs_handle_tracked_suspend(struct fs_handle_tracked *handle)
-{
-       int ret = 0;
-       struct stat fs_stat;
-       const char *path;
-       const struct lttng_directory_handle *node_directory_handle;
-
-       pthread_mutex_lock(&handle->lock);
-       lttng_inode_borrow_location(
-                       handle->inode, &node_directory_handle, &path);
-       LTTNG_ASSERT(handle->fd >= 0);
-       if (handle->in_use) {
-               /* This handle can't be suspended as it is currently in use. */
-               ret = -EAGAIN;
-               goto end;
-       }
-
-       ret = lttng_directory_handle_stat(
-                       node_directory_handle, path, &fs_stat);
-       if (ret) {
-               PERROR("Filesystem handle to %s cannot be suspended as stat() failed",
-                               path);
-               ret = -errno;
-               goto end;
-       }
-
-       if (fs_stat.st_ino != handle->ino) {
-               /* Don't suspend as the handle would not be restorable. */
-               WARN("Filesystem handle to %s cannot be suspended as its inode changed",
-                               path);
-               ret = -ENOENT;
-               goto end;
-       }
-
-       handle->offset = lseek(handle->fd, 0, SEEK_CUR);
-       if (handle->offset == -1) {
-               WARN("Filesystem handle to %s cannot be suspended as lseek() failed to sample its current position",
-                               path);
-               ret = -errno;
-               goto end;
-       }
-
-       ret = close(handle->fd);
-       if (ret) {
-               PERROR("Filesystem handle to %s cannot be suspended as close() failed",
-                               path);
-               ret = -errno;
-               goto end;
-       }
-       DBG("Suspended filesystem handle to %s (fd %i) at position %" PRId64,
-                       path, handle->fd, handle->offset);
-       handle->fd = -1;
-end:
-       if (ret) {
-               handle->tracker->stats.errors++;
-       }
-       pthread_mutex_unlock(&handle->lock);
-       return ret;
-}
-
-/* Caller must hold the tracker and handle's locks. */
-static int fs_handle_tracked_restore(struct fs_handle_tracked *handle)
-{
-       int ret, fd = -1;
-       const char *path;
-       const struct lttng_directory_handle *node_directory_handle;
-
-       lttng_inode_borrow_location(
-                       handle->inode, &node_directory_handle, &path);
-
-       LTTNG_ASSERT(handle->fd == -1);
-       LTTNG_ASSERT(path);
-       ret = open_from_properties(
-                       node_directory_handle, path, &handle->properties);
-       if (ret < 0) {
-               PERROR("Failed to restore filesystem handle to %s, open() failed",
-                               path);
-               ret = -errno;
-               goto end;
-       }
-       fd = ret;
-
-       ret = lseek(fd, handle->offset, SEEK_SET);
-       if (ret < 0) {
-               PERROR("Failed to restore filesystem handle to %s, lseek() failed",
-                               path);
-               ret = -errno;
-               goto end;
-       }
-       DBG("Restored filesystem handle to %s (fd %i) at position %" PRId64,
-                       path, fd, handle->offset);
-       ret = 0;
-       handle->fd = fd;
-       fd = -1;
-end:
-       if (fd >= 0) {
-               (void) close(fd);
-       }
-       return ret;
-}
-
-static int open_from_properties(const struct lttng_directory_handle *dir_handle,
-               const char *path, struct open_properties *properties)
-{
-       int ret;
-
-       /*
-        * open() ignores the 'flags' parameter unless the O_CREAT or O_TMPFILE
-        * flags are set. O_TMPFILE would not make sense in the context of a
-        * suspendable fs_handle as it would not be restorable (see OPEN(2)),
-        * thus it is ignored here.
-        */
-       if ((properties->flags & O_CREAT) && properties->mode.is_set) {
-               ret = lttng_directory_handle_open_file(dir_handle, path,
-                               properties->flags, properties->mode.value);
-       } else {
-               ret = lttng_directory_handle_open_file(dir_handle, path,
-                               properties->flags, 0);
-       }
-       /*
-        * Some flags should not be used beyond the initial open() of a
-        * restorable file system handle. O_CREAT and O_TRUNC must
-        * be cleared since it would be unexpected to re-use them
-        * when the handle is retored:
-        *  - O_CREAT should not be needed as the file has been created
-        *    on the initial call to open(),
-        *  - O_TRUNC would destroy the file's contents by truncating it
-        *    to length 0.
-        */
-       properties->flags &= ~(O_CREAT | O_TRUNC);
-       if (ret < 0) {
-               ret = -errno;
-               goto end;
-       }
-end:
-       return ret;
-}
-
-struct fd_tracker *fd_tracker_create(const char *unlinked_file_path,
-               unsigned int capacity)
-{
-       struct fd_tracker *tracker = zmalloc(sizeof(struct fd_tracker));
-
-       if (!tracker) {
-               goto end;
-       }
-
-       pthread_mutex_lock(&seed.lock);
-       if (!seed.initialized) {
-               seed.value = (unsigned long) time(NULL);
-               seed.initialized = true;
-       }
-       pthread_mutex_unlock(&seed.lock);
-
-       CDS_INIT_LIST_HEAD(&tracker->active_handles);
-       CDS_INIT_LIST_HEAD(&tracker->suspended_handles);
-       tracker->capacity = capacity;
-       tracker->unsuspendable_fds = cds_lfht_new(DEFAULT_HT_SIZE, 1, 0,
-                       CDS_LFHT_AUTO_RESIZE | CDS_LFHT_ACCOUNTING, NULL);
-       if (!tracker->unsuspendable_fds) {
-               ERR("Failed to create fd-tracker's unsuspendable_fds hash table");
-               goto error;
-       }
-       tracker->inode_registry = lttng_inode_registry_create();
-       if (!tracker->inode_registry) {
-               ERR("Failed to create fd-tracker's inode registry");
-               goto error;
-       }
-
-       tracker->unlinked_file_pool =
-                       lttng_unlinked_file_pool_create(unlinked_file_path);
-       if (!tracker->unlinked_file_pool) {
-               goto error;
-       }
-       DBG("File descriptor tracker created with a limit of %u simultaneously-opened FDs",
-                       capacity);
-end:
-       return tracker;
-error:
-       fd_tracker_destroy(tracker);
-       return NULL;
-}
-
-void fd_tracker_log(struct fd_tracker *tracker)
-{
-       struct fs_handle_tracked *handle;
-       struct unsuspendable_fd *unsuspendable_fd;
-       struct cds_lfht_iter iter;
-
-       pthread_mutex_lock(&tracker->lock);
-       DBG_NO_LOC("File descriptor tracker");
-       DBG_NO_LOC("  Stats:");
-       DBG_NO_LOC("    uses:            %" PRIu64, tracker->stats.uses);
-       DBG_NO_LOC("    misses:          %" PRIu64, tracker->stats.misses);
-       DBG_NO_LOC("    errors:          %" PRIu64, tracker->stats.errors);
-       DBG_NO_LOC("  Tracked:           %u", TRACKED_COUNT(tracker));
-       DBG_NO_LOC("    active:          %u", ACTIVE_COUNT(tracker));
-       DBG_NO_LOC("      suspendable:   %u", SUSPENDABLE_COUNT(tracker));
-       DBG_NO_LOC("      unsuspendable: %u", UNSUSPENDABLE_COUNT(tracker));
-       DBG_NO_LOC("    suspended:       %u", SUSPENDED_COUNT(tracker));
-       DBG_NO_LOC("    capacity:        %u", tracker->capacity);
-
-       DBG_NO_LOC("  Tracked suspendable file descriptors");
-       cds_list_for_each_entry(
-                       handle, &tracker->active_handles, handles_list_node) {
-               fs_handle_tracked_log(handle);
-       }
-       cds_list_for_each_entry(handle, &tracker->suspended_handles,
-                       handles_list_node) {
-               fs_handle_tracked_log(handle);
-       }
-       if (!SUSPENDABLE_COUNT(tracker)) {
-               DBG_NO_LOC("    None");
-       }
-
-       DBG_NO_LOC("  Tracked unsuspendable file descriptors");
-       rcu_read_lock();
-       cds_lfht_for_each_entry(tracker->unsuspendable_fds, &iter,
-                       unsuspendable_fd, tracker_node) {
-               DBG_NO_LOC("    %s [active, fd %d]",
-                               unsuspendable_fd->name ?: "Unnamed",
-                               unsuspendable_fd->fd);
-       }
-       rcu_read_unlock();
-       if (!UNSUSPENDABLE_COUNT(tracker)) {
-               DBG_NO_LOC("    None");
-       }
-
-       pthread_mutex_unlock(&tracker->lock);
-}
-
-int fd_tracker_destroy(struct fd_tracker *tracker)
-{
-       int ret = 0;
-
-       if (!tracker) {
-               goto end;
-       }
-       /*
-        * Refuse to destroy the tracker as fs_handles may still old
-        * weak references to the tracker.
-        */
-       pthread_mutex_lock(&tracker->lock);
-       if (TRACKED_COUNT(tracker)) {
-               ERR("A file descriptor leak has been detected: %u tracked file descriptors are still being tracked",
-                               TRACKED_COUNT(tracker));
-               pthread_mutex_unlock(&tracker->lock);
-               fd_tracker_log(tracker);
-               ret = -1;
-               goto end;
-       }
-       pthread_mutex_unlock(&tracker->lock);
-
-       if (tracker->unsuspendable_fds) {
-               ret = cds_lfht_destroy(tracker->unsuspendable_fds, NULL);
-               LTTNG_ASSERT(!ret);
-       }
-
-       lttng_inode_registry_destroy(tracker->inode_registry);
-       lttng_unlinked_file_pool_destroy(tracker->unlinked_file_pool);
-       pthread_mutex_destroy(&tracker->lock);
-       free(tracker);
-end:
-       return ret;
-}
-
-struct fs_handle *fd_tracker_open_fs_handle(struct fd_tracker *tracker,
-               struct lttng_directory_handle *directory,
-               const char *path,
-               int flags,
-               mode_t *mode)
-{
-       int ret;
-       struct fs_handle_tracked *handle = NULL;
-       struct stat fd_stat;
-       struct open_properties properties = {
-               .flags = flags,
-               .mode.is_set = !!mode,
-               .mode.value = mode ? *mode : 0,
-       };
-
-       pthread_mutex_lock(&tracker->lock);
-       if (ACTIVE_COUNT(tracker) == tracker->capacity) {
-               if (tracker->count.suspendable.active > 0) {
-                       ret = fd_tracker_suspend_handles(tracker, 1);
-                       if (ret) {
-                               goto end;
-                       }
-               } else {
-                       /*
-                        * There are not enough active suspendable file
-                        * descriptors to open a new fd and still accommodate
-                        * the tracker's capacity.
-                        */
-                       WARN("Cannot open file system handle, too many unsuspendable file descriptors are opened (%u)",
-                                       tracker->count.unsuspendable);
-                       goto end;
-               }
-       }
-
-       handle = zmalloc(sizeof(*handle));
-       if (!handle) {
-               goto end;
-       }
-       handle->parent = (typeof(handle->parent)) {
-               .get_fd = fs_handle_tracked_get_fd,
-               .put_fd = fs_handle_tracked_put_fd,
-               .unlink = fs_handle_tracked_unlink,
-               .close = fs_handle_tracked_close,
-       };
-
-       handle->tracker = tracker;
-
-       ret = pthread_mutex_init(&handle->lock, NULL);
-       if (ret) {
-               PERROR("Failed to initialize handle mutex while creating fs handle");
-               goto error_mutex_init;
-       }
-
-       handle->fd = open_from_properties(directory, path, &properties);
-       if (handle->fd < 0) {
-               PERROR("Failed to open fs handle to %s, open() returned", path);
-               goto error;
-       }
-
-       handle->properties = properties;
-
-       handle->inode = lttng_inode_registry_get_inode(tracker->inode_registry,
-                       directory, path, handle->fd,
-                       tracker->unlinked_file_pool);
-       if (!handle->inode) {
-               ERR("Failed to get lttng_inode corresponding to file %s", path);
-               goto error;
-       }
-
-       if (fstat(handle->fd, &fd_stat)) {
-               PERROR("Failed to retrieve file descriptor inode while creating fs handle, fstat() returned");
-               goto error;
-       }
-       handle->ino = fd_stat.st_ino;
-
-       fd_tracker_track(tracker, handle);
-end:
-       pthread_mutex_unlock(&tracker->lock);
-       return handle ? &handle->parent : NULL;
-error:
-       if (handle->inode) {
-               lttng_inode_put(handle->inode);
-       }
-       pthread_mutex_destroy(&handle->lock);
-error_mutex_init:
-       free(handle);
-       handle = NULL;
-       goto end;
-}
-
-/* Caller must hold the tracker's lock. */
-static int fd_tracker_suspend_handles(
-               struct fd_tracker *tracker, unsigned int count)
-{
-       unsigned int left_to_close = count;
-       unsigned int attempts_left = tracker->count.suspendable.active;
-       struct fs_handle_tracked *handle, *tmp;
-
-       cds_list_for_each_entry_safe(handle, tmp, &tracker->active_handles,
-                       handles_list_node) {
-               int ret;
-
-               fd_tracker_untrack(tracker, handle);
-               ret = fs_handle_tracked_suspend(handle);
-               fd_tracker_track(tracker, handle);
-               if (!ret) {
-                       left_to_close--;
-               }
-               attempts_left--;
-
-               if (left_to_close == 0 || attempts_left == 0) {
-                       break;
-               }
-       }
-       return left_to_close ? -EMFILE : 0;
-}
-
-int fd_tracker_open_unsuspendable_fd(struct fd_tracker *tracker,
-               int *out_fds,
-               const char **names,
-               unsigned int fd_count,
-               fd_open_cb open,
-               void *user_data)
-{
-       int ret, user_ret, i, fds_to_suspend;
-       unsigned int active_fds;
-       struct unsuspendable_fd **entries;
-
-       entries = zmalloc(fd_count * sizeof(*entries));
-       if (!entries) {
-               ret = -1;
-               goto end;
-       }
-
-       pthread_mutex_lock(&tracker->lock);
-
-       active_fds = ACTIVE_COUNT(tracker);
-       fds_to_suspend = (int) active_fds + (int) fd_count -
-                       (int) tracker->capacity;
-       if (fds_to_suspend > 0) {
-               if (fds_to_suspend <= tracker->count.suspendable.active) {
-                       ret = fd_tracker_suspend_handles(
-                                       tracker, fds_to_suspend);
-                       if (ret) {
-                               goto end_unlock;
-                       }
-               } else {
-                       /*
-                        * There are not enough active suspendable file
-                        * descriptors to open a new fd and still accommodates the
-                        * tracker's capacity.
-                        */
-                       WARN("Cannot open unsuspendable fd, too many unsuspendable file descriptors are opened (%u)",
-                                       tracker->count.unsuspendable);
-                       ret = -EMFILE;
-                       goto end_unlock;
-               }
-       }
-
-       user_ret = open(user_data, out_fds);
-       if (user_ret) {
-               ret = user_ret;
-               goto end_unlock;
-       }
-
-       /*
-        * Add the fds returned by the user's callback to the hashtable
-        * of unsuspendable fds.
-        */
-       for (i = 0; i < fd_count; i++) {
-               struct unsuspendable_fd *entry = unsuspendable_fd_create(
-                               names ? names[i] : NULL, out_fds[i]);
-
-               if (!entry) {
-                       ret = -1;
-                       goto end_free_entries;
-               }
-               entries[i] = entry;
-       }
-
-       rcu_read_lock();
-       for (i = 0; i < fd_count; i++) {
-               struct cds_lfht_node *node;
-               struct unsuspendable_fd *entry = entries[i];
-
-               node = cds_lfht_add_unique(tracker->unsuspendable_fds,
-                               hash_key_ulong((void *) (unsigned long)
-                                                               out_fds[i],
-                                               seed.value),
-                               match_fd, (void *) (unsigned long) out_fds[i],
-                               &entry->tracker_node);
-
-               if (node != &entry->tracker_node) {
-                       ret = -EEXIST;
-                       rcu_read_unlock();
-                       goto end_free_entries;
-               }
-               entries[i] = NULL;
-       }
-       tracker->count.unsuspendable += fd_count;
-       rcu_read_unlock();
-       ret = user_ret;
-end_unlock:
-       pthread_mutex_unlock(&tracker->lock);
-end:
-       free(entries);
-       return ret;
-end_free_entries:
-       for (i = 0; i < fd_count; i++) {
-               unsuspendable_fd_destroy(entries[i]);
-       }
-       goto end_unlock;
-}
-
-int fd_tracker_close_unsuspendable_fd(struct fd_tracker *tracker,
-               int *fds_in,
-               unsigned int fd_count,
-               fd_close_cb close,
-               void *user_data)
-{
-       int i, ret, user_ret;
-       int *fds = NULL;
-
-       /*
-        * Maintain a local copy of fds_in as the user's callback may modify its
-        * contents (e.g. setting the fd(s) to -1 after close).
-        */
-       fds = malloc(sizeof(*fds) * fd_count);
-       if (!fds) {
-               ret = -1;
-               goto end;
-       }
-       memcpy(fds, fds_in, sizeof(*fds) * fd_count);
-
-       pthread_mutex_lock(&tracker->lock);
-       rcu_read_lock();
-
-       /* Let the user close the file descriptors. */
-       user_ret = close(user_data, fds_in);
-       if (user_ret) {
-               ret = user_ret;
-               goto end_unlock;
-       }
-
-       /* Untrack the fds that were just closed by the user's callback. */
-       for (i = 0; i < fd_count; i++) {
-               struct cds_lfht_node *node;
-               struct cds_lfht_iter iter;
-               struct unsuspendable_fd *entry;
-
-               cds_lfht_lookup(tracker->unsuspendable_fds,
-                               hash_key_ulong((void *) (unsigned long) fds[i],
-                                               seed.value),
-                               match_fd, (void *) (unsigned long) fds[i],
-                               &iter);
-               node = cds_lfht_iter_get_node(&iter);
-               if (!node) {
-                       /* Unknown file descriptor. */
-                       WARN("Untracked file descriptor %d passed to fd_tracker_close_unsuspendable_fd()",
-                                       fds[i]);
-                       ret = -EINVAL;
-                       goto end_unlock;
-               }
-               entry = caa_container_of(
-                               node, struct unsuspendable_fd, tracker_node);
-
-               cds_lfht_del(tracker->unsuspendable_fds, node);
-               unsuspendable_fd_destroy(entry);
-               fds[i] = -1;
-       }
-
-       tracker->count.unsuspendable -= fd_count;
-       ret = 0;
-end_unlock:
-       rcu_read_unlock();
-       pthread_mutex_unlock(&tracker->lock);
-       free(fds);
-end:
-       return ret;
-}
-
-/* Caller must have taken the tracker's and handle's locks. */
-static void fd_tracker_track(
-               struct fd_tracker *tracker, struct fs_handle_tracked *handle)
-{
-       if (handle->fd >= 0) {
-               tracker->count.suspendable.active++;
-               cds_list_add_tail(&handle->handles_list_node,
-                               &tracker->active_handles);
-       } else {
-               tracker->count.suspendable.suspended++;
-               cds_list_add_tail(&handle->handles_list_node,
-                               &tracker->suspended_handles);
-       }
-}
-
-/* Caller must have taken the tracker's and handle's locks. */
-static void fd_tracker_untrack(
-               struct fd_tracker *tracker, struct fs_handle_tracked *handle)
-{
-       if (handle->fd >= 0) {
-               tracker->count.suspendable.active--;
-       } else {
-               tracker->count.suspendable.suspended--;
-       }
-       cds_list_del(&handle->handles_list_node);
-}
-
-/* Caller must have taken the tracker's and handle's locks. */
-static int fd_tracker_restore_handle(
-               struct fd_tracker *tracker, struct fs_handle_tracked *handle)
-{
-       int ret;
-
-       fd_tracker_untrack(tracker, handle);
-       if (ACTIVE_COUNT(tracker) >= tracker->capacity) {
-               ret = fd_tracker_suspend_handles(tracker, 1);
-               if (ret) {
-                       goto end;
-               }
-       }
-       ret = fs_handle_tracked_restore(handle);
-end:
-       fd_tracker_track(tracker, handle);
-       return ret ? ret : handle->fd;
-}
-
-static int fs_handle_tracked_get_fd(struct fs_handle *_handle)
-{
-       int ret;
-       struct fs_handle_tracked *handle =
-                       container_of(_handle, struct fs_handle_tracked, parent);
-
-       /*
-        * TODO This should be optimized as it is a fairly hot path.
-        * The fd-tracker's lock should only be taken when a fs_handle is
-        * restored (slow path). On the fast path (fs_handle is active),
-        * the only effect on the fd_tracker is marking the handle as the
-        * most recently used. Currently, it is done by a call to the
-        * track/untrack helpers, but it should be done atomically.
-        *
-        * Note that the lock's nesting order must still be respected here.
-        * The handle's lock nests inside the tracker's lock.
-        */
-       pthread_mutex_lock(&handle->tracker->lock);
-       pthread_mutex_lock(&handle->lock);
-       LTTNG_ASSERT(!handle->in_use);
-
-       handle->tracker->stats.uses++;
-       if (handle->fd >= 0) {
-               ret = handle->fd;
-               /* Mark as most recently used. */
-               fd_tracker_untrack(handle->tracker, handle);
-               fd_tracker_track(handle->tracker, handle);
-       } else {
-               handle->tracker->stats.misses++;
-               ret = fd_tracker_restore_handle(handle->tracker, handle);
-               if (ret < 0) {
-                       handle->tracker->stats.errors++;
-                       goto end;
-               }
-       }
-       handle->in_use = true;
-end:
-       pthread_mutex_unlock(&handle->lock);
-       pthread_mutex_unlock(&handle->tracker->lock);
-       return ret;
-}
-
-static void fs_handle_tracked_put_fd(struct fs_handle *_handle)
-{
-       struct fs_handle_tracked *handle =
-                       container_of(_handle, struct fs_handle_tracked, parent);
-
-       pthread_mutex_lock(&handle->lock);
-       handle->in_use = false;
-       pthread_mutex_unlock(&handle->lock);
-}
-
-static int fs_handle_tracked_unlink(struct fs_handle *_handle)
-{
-       int ret;
-       struct fs_handle_tracked *handle =
-                       container_of(_handle, struct fs_handle_tracked, parent);
-
-       pthread_mutex_lock(&handle->tracker->lock);
-       pthread_mutex_lock(&handle->lock);
-       ret = lttng_inode_unlink(handle->inode);
-       pthread_mutex_unlock(&handle->lock);
-       pthread_mutex_unlock(&handle->tracker->lock);
-       return ret;
-}
-
-static int fs_handle_tracked_close(struct fs_handle *_handle)
-{
-       int ret = 0;
-       const char *path = NULL;
-       struct fs_handle_tracked *handle =
-                       container_of(_handle, struct fs_handle_tracked, parent);
-       struct lttng_directory_handle *inode_directory_handle = NULL;
-
-       if (!handle) {
-               ret = -EINVAL;
-               goto end;
-       }
-
-       pthread_mutex_lock(&handle->tracker->lock);
-       pthread_mutex_lock(&handle->lock);
-       if (handle->inode) {
-               lttng_inode_borrow_location(handle->inode, NULL, &path);
-               /*
-                * Here a reference to the inode's directory handle is acquired
-                * to prevent the last reference to it from being released while
-                * the tracker's lock is taken.
-                *
-                * If this wasn't done, the directory handle could attempt to
-                * close its underlying directory file descriptor, which would
-                * attempt to lock the tracker's lock, resulting in a deadlock.
-                *
-                * Since a new reference to the directory handle is taken within
-                * the scope of this function, it is not possible for the last
-                * reference to the inode's location directory handle to be
-                * released during the call to lttng_inode_put().
-                *
-                * We wait until the tracker's lock is released to release the
-                * reference. Hence, the call to the tracker is delayed just
-                * enough to not attempt to recursively acquire the tracker's
-                * lock twice.
-                */
-               inode_directory_handle =
-                               lttng_inode_get_location_directory_handle(
-                                               handle->inode);
-       }
-       fd_tracker_untrack(handle->tracker, handle);
-       if (handle->fd >= 0) {
-               /*
-                * The return value of close() is not propagated as there
-                * isn't much the user can do about it.
-                */
-               if (close(handle->fd)) {
-                       PERROR("Failed to close the file descriptor (%d) of fs handle to %s, close() returned",
-                                       handle->fd, path ? path : "Unknown");
-               }
-               handle->fd = -1;
-       }
-       if (handle->inode) {
-               lttng_inode_put(handle->inode);
-       }
-       pthread_mutex_unlock(&handle->lock);
-       pthread_mutex_destroy(&handle->lock);
-       pthread_mutex_unlock(&handle->tracker->lock);
-       free(handle);
-       lttng_directory_handle_put(inode_directory_handle);
-end:
-       return ret;
-}
diff --git a/src/common/fd-tracker/fd-tracker.cpp b/src/common/fd-tracker/fd-tracker.cpp
new file mode 100644 (file)
index 0000000..f1058aa
--- /dev/null
@@ -0,0 +1,948 @@
+/*
+ * Copyright (C) 2018-2020 Jérémie Galarneau <jeremie.galarneau@efficios.com>
+ *
+ * SPDX-License-Identifier: GPL-2.0-only
+ *
+ */
+
+#include <urcu.h>
+#include <urcu/list.h>
+#include <urcu/rculfhash.h>
+
+#include <fcntl.h>
+#include <inttypes.h>
+#include <pthread.h>
+#include <stdbool.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+
+#include <common/defaults.h>
+#include <common/error.h>
+#include <common/fs-handle-internal.h>
+#include <common/hashtable/hashtable.h>
+#include <common/hashtable/utils.h>
+#include <common/macros.h>
+#include <common/optional.h>
+
+#include "fd-tracker.h"
+#include "inode.h"
+
+/* Tracker lock must be taken by the user. */
+#define TRACKED_COUNT(tracker)                                 \
+       (tracker->count.suspendable.active +                   \
+                       tracker->count.suspendable.suspended + \
+                       tracker->count.unsuspendable)
+
+/* Tracker lock must be taken by the user. */
+#define ACTIVE_COUNT(tracker) \
+       (tracker->count.suspendable.active + tracker->count.unsuspendable)
+
+/* Tracker lock must be taken by the user. */
+#define SUSPENDED_COUNT(tracker) (tracker->count.suspendable.suspended)
+
+/* Tracker lock must be taken by the user. */
+#define SUSPENDABLE_COUNT(tracker)           \
+       (tracker->count.suspendable.active + \
+                       tracker->count.suspendable.suspended)
+
+/* Tracker lock must be taken by the user. */
+#define UNSUSPENDABLE_COUNT(tracker) (tracker->count.unsuspendable)
+
+struct fd_tracker {
+       pthread_mutex_t lock;
+       struct {
+               struct {
+                       unsigned int active;
+                       unsigned int suspended;
+               } suspendable;
+               unsigned int unsuspendable;
+       } count;
+       unsigned int capacity;
+       struct {
+               uint64_t uses;
+               uint64_t misses;
+               /* Failures to suspend or restore fs handles. */
+               uint64_t errors;
+       } stats;
+       /*
+        * The head of the active_handles list is always the least recently
+        * used active handle. When an handle is used, it is removed from the
+        * list and added to the end. When a file has to be suspended, the
+        * first element in the list is "popped", suspended, and added to the
+        * list of suspended handles.
+        */
+       struct cds_list_head active_handles;
+       struct cds_list_head suspended_handles;
+       struct cds_lfht *unsuspendable_fds;
+       struct lttng_inode_registry *inode_registry;
+       /* Unlinked files are moved in this directory under a unique name. */
+       struct lttng_directory_handle *unlink_directory_handle;
+       struct lttng_unlinked_file_pool *unlinked_file_pool;
+};
+
+struct open_properties {
+       int flags;
+       LTTNG_OPTIONAL(mode_t) mode;
+};
+
+/*
+ * A fs_handle_tracked is not ref-counted. Therefore, it is assumed that a
+ * handle is never in-use while it is being reclaimed. It can be
+ * shared by multiple threads, but external synchronization is required
+ * to ensure it is not still being used when it is reclaimed (close method).
+ * In this respect, it is not different from a regular file descriptor.
+ *
+ * The fs_handle lock always nests _within_ the tracker's lock.
+ */
+struct fs_handle_tracked {
+       struct fs_handle parent;
+       pthread_mutex_t lock;
+       /*
+        * Weak reference to the tracker. All fs_handles are assumed to have
+        * been closed at the moment of the destruction of the fd_tracker.
+        */
+       struct fd_tracker *tracker;
+       struct open_properties properties;
+       struct lttng_inode *inode;
+       int fd;
+       /* inode number of the file at the time of the handle's creation. */
+       uint64_t ino;
+       bool in_use;
+       /* Offset to which the file should be restored. */
+       off_t offset;
+       struct cds_list_head handles_list_node;
+};
+
+struct unsuspendable_fd {
+       /*
+        * Accesses are only performed through the tracker, which is protected
+        * by its own lock.
+        */
+       int fd;
+       char *name;
+       struct cds_lfht_node tracker_node;
+       struct rcu_head rcu_head;
+};
+
+static struct {
+       pthread_mutex_t lock;
+       bool initialized;
+       unsigned long value;
+} seed = {
+       .lock = PTHREAD_MUTEX_INITIALIZER,
+};
+
+static int match_fd(struct cds_lfht_node *node, const void *key);
+static void unsuspendable_fd_destroy(struct unsuspendable_fd *entry);
+static struct unsuspendable_fd *unsuspendable_fd_create(
+               const char *name, int fd);
+static int open_from_properties(const struct lttng_directory_handle *dir_handle,
+               const char *path, struct open_properties *properties);
+
+static void fs_handle_tracked_log(struct fs_handle_tracked *handle);
+static int fs_handle_tracked_suspend(struct fs_handle_tracked *handle);
+static int fs_handle_tracked_restore(struct fs_handle_tracked *handle);
+static int fs_handle_tracked_get_fd(struct fs_handle *_handle);
+static void fs_handle_tracked_put_fd(struct fs_handle *_handle);
+static int fs_handle_tracked_unlink(struct fs_handle *_handle);
+static int fs_handle_tracked_close(struct fs_handle *_handle);
+
+static void fd_tracker_track(
+               struct fd_tracker *tracker, struct fs_handle_tracked *handle);
+static void fd_tracker_untrack(
+               struct fd_tracker *tracker, struct fs_handle_tracked *handle);
+static int fd_tracker_suspend_handles(
+               struct fd_tracker *tracker, unsigned int count);
+static int fd_tracker_restore_handle(
+               struct fd_tracker *tracker, struct fs_handle_tracked *handle);
+
+/* Match function of the tracker's unsuspendable_fds hash table. */
+static int match_fd(struct cds_lfht_node *node, const void *key)
+{
+       struct unsuspendable_fd *entry = caa_container_of(
+                       node, struct unsuspendable_fd, tracker_node);
+
+       return hash_match_key_ulong(
+                       (void *) (unsigned long) entry->fd, (void *) key);
+}
+
+static void delete_unsuspendable_fd(struct rcu_head *head)
+{
+       struct unsuspendable_fd *fd = caa_container_of(
+                       head, struct unsuspendable_fd, rcu_head);
+
+       free(fd->name);
+       free(fd);
+}
+
+static void unsuspendable_fd_destroy(struct unsuspendable_fd *entry)
+{
+       if (!entry) {
+               return;
+       }
+       call_rcu(&entry->rcu_head, delete_unsuspendable_fd);
+}
+
+static struct unsuspendable_fd *unsuspendable_fd_create(
+               const char *name, int fd)
+{
+       struct unsuspendable_fd *entry = (unsuspendable_fd *) zmalloc(sizeof(*entry));
+
+       if (!entry) {
+               goto error;
+       }
+       if (name) {
+               entry->name = strdup(name);
+               if (!entry->name) {
+                       goto error;
+               }
+       }
+       cds_lfht_node_init(&entry->tracker_node);
+       entry->fd = fd;
+       return entry;
+error:
+       unsuspendable_fd_destroy(entry);
+       return NULL;
+}
+
+static void fs_handle_tracked_log(struct fs_handle_tracked *handle)
+{
+       const char *path;
+
+       pthread_mutex_lock(&handle->lock);
+       lttng_inode_borrow_location(handle->inode, NULL, &path);
+
+       if (handle->fd >= 0) {
+               DBG_NO_LOC("    %s [active, fd %d%s]", path, handle->fd,
+                               handle->in_use ? ", in use" : "");
+       } else {
+               DBG_NO_LOC("    %s [suspended]", path);
+       }
+       pthread_mutex_unlock(&handle->lock);
+}
+
+/* Tracker lock must be held by the caller. */
+static int fs_handle_tracked_suspend(struct fs_handle_tracked *handle)
+{
+       int ret = 0;
+       struct stat fs_stat;
+       const char *path;
+       const struct lttng_directory_handle *node_directory_handle;
+
+       pthread_mutex_lock(&handle->lock);
+       lttng_inode_borrow_location(
+                       handle->inode, &node_directory_handle, &path);
+       LTTNG_ASSERT(handle->fd >= 0);
+       if (handle->in_use) {
+               /* This handle can't be suspended as it is currently in use. */
+               ret = -EAGAIN;
+               goto end;
+       }
+
+       ret = lttng_directory_handle_stat(
+                       node_directory_handle, path, &fs_stat);
+       if (ret) {
+               PERROR("Filesystem handle to %s cannot be suspended as stat() failed",
+                               path);
+               ret = -errno;
+               goto end;
+       }
+
+       if (fs_stat.st_ino != handle->ino) {
+               /* Don't suspend as the handle would not be restorable. */
+               WARN("Filesystem handle to %s cannot be suspended as its inode changed",
+                               path);
+               ret = -ENOENT;
+               goto end;
+       }
+
+       handle->offset = lseek(handle->fd, 0, SEEK_CUR);
+       if (handle->offset == -1) {
+               WARN("Filesystem handle to %s cannot be suspended as lseek() failed to sample its current position",
+                               path);
+               ret = -errno;
+               goto end;
+       }
+
+       ret = close(handle->fd);
+       if (ret) {
+               PERROR("Filesystem handle to %s cannot be suspended as close() failed",
+                               path);
+               ret = -errno;
+               goto end;
+       }
+       DBG("Suspended filesystem handle to %s (fd %i) at position %" PRId64,
+                       path, handle->fd, handle->offset);
+       handle->fd = -1;
+end:
+       if (ret) {
+               handle->tracker->stats.errors++;
+       }
+       pthread_mutex_unlock(&handle->lock);
+       return ret;
+}
+
+/* Caller must hold the tracker and handle's locks. */
+static int fs_handle_tracked_restore(struct fs_handle_tracked *handle)
+{
+       int ret, fd = -1;
+       const char *path;
+       const struct lttng_directory_handle *node_directory_handle;
+
+       lttng_inode_borrow_location(
+                       handle->inode, &node_directory_handle, &path);
+
+       LTTNG_ASSERT(handle->fd == -1);
+       LTTNG_ASSERT(path);
+       ret = open_from_properties(
+                       node_directory_handle, path, &handle->properties);
+       if (ret < 0) {
+               PERROR("Failed to restore filesystem handle to %s, open() failed",
+                               path);
+               ret = -errno;
+               goto end;
+       }
+       fd = ret;
+
+       ret = lseek(fd, handle->offset, SEEK_SET);
+       if (ret < 0) {
+               PERROR("Failed to restore filesystem handle to %s, lseek() failed",
+                               path);
+               ret = -errno;
+               goto end;
+       }
+       DBG("Restored filesystem handle to %s (fd %i) at position %" PRId64,
+                       path, fd, handle->offset);
+       ret = 0;
+       handle->fd = fd;
+       fd = -1;
+end:
+       if (fd >= 0) {
+               (void) close(fd);
+       }
+       return ret;
+}
+
+static int open_from_properties(const struct lttng_directory_handle *dir_handle,
+               const char *path, struct open_properties *properties)
+{
+       int ret;
+
+       /*
+        * open() ignores the 'flags' parameter unless the O_CREAT or O_TMPFILE
+        * flags are set. O_TMPFILE would not make sense in the context of a
+        * suspendable fs_handle as it would not be restorable (see OPEN(2)),
+        * thus it is ignored here.
+        */
+       if ((properties->flags & O_CREAT) && properties->mode.is_set) {
+               ret = lttng_directory_handle_open_file(dir_handle, path,
+                               properties->flags, properties->mode.value);
+       } else {
+               ret = lttng_directory_handle_open_file(dir_handle, path,
+                               properties->flags, 0);
+       }
+       /*
+        * Some flags should not be used beyond the initial open() of a
+        * restorable file system handle. O_CREAT and O_TRUNC must
+        * be cleared since it would be unexpected to re-use them
+        * when the handle is retored:
+        *  - O_CREAT should not be needed as the file has been created
+        *    on the initial call to open(),
+        *  - O_TRUNC would destroy the file's contents by truncating it
+        *    to length 0.
+        */
+       properties->flags &= ~(O_CREAT | O_TRUNC);
+       if (ret < 0) {
+               ret = -errno;
+               goto end;
+       }
+end:
+       return ret;
+}
+
+struct fd_tracker *fd_tracker_create(const char *unlinked_file_path,
+               unsigned int capacity)
+{
+       struct fd_tracker *tracker = (fd_tracker *) zmalloc(sizeof(struct fd_tracker));
+
+       if (!tracker) {
+               goto end;
+       }
+
+       pthread_mutex_lock(&seed.lock);
+       if (!seed.initialized) {
+               seed.value = (unsigned long) time(NULL);
+               seed.initialized = true;
+       }
+       pthread_mutex_unlock(&seed.lock);
+
+       CDS_INIT_LIST_HEAD(&tracker->active_handles);
+       CDS_INIT_LIST_HEAD(&tracker->suspended_handles);
+       tracker->capacity = capacity;
+       tracker->unsuspendable_fds = cds_lfht_new(DEFAULT_HT_SIZE, 1, 0,
+                       CDS_LFHT_AUTO_RESIZE | CDS_LFHT_ACCOUNTING, NULL);
+       if (!tracker->unsuspendable_fds) {
+               ERR("Failed to create fd-tracker's unsuspendable_fds hash table");
+               goto error;
+       }
+       tracker->inode_registry = lttng_inode_registry_create();
+       if (!tracker->inode_registry) {
+               ERR("Failed to create fd-tracker's inode registry");
+               goto error;
+       }
+
+       tracker->unlinked_file_pool =
+                       lttng_unlinked_file_pool_create(unlinked_file_path);
+       if (!tracker->unlinked_file_pool) {
+               goto error;
+       }
+       DBG("File descriptor tracker created with a limit of %u simultaneously-opened FDs",
+                       capacity);
+end:
+       return tracker;
+error:
+       fd_tracker_destroy(tracker);
+       return NULL;
+}
+
+void fd_tracker_log(struct fd_tracker *tracker)
+{
+       struct fs_handle_tracked *handle;
+       struct unsuspendable_fd *unsuspendable_fd;
+       struct cds_lfht_iter iter;
+
+       pthread_mutex_lock(&tracker->lock);
+       DBG_NO_LOC("File descriptor tracker");
+       DBG_NO_LOC("  Stats:");
+       DBG_NO_LOC("    uses:            %" PRIu64, tracker->stats.uses);
+       DBG_NO_LOC("    misses:          %" PRIu64, tracker->stats.misses);
+       DBG_NO_LOC("    errors:          %" PRIu64, tracker->stats.errors);
+       DBG_NO_LOC("  Tracked:           %u", TRACKED_COUNT(tracker));
+       DBG_NO_LOC("    active:          %u", ACTIVE_COUNT(tracker));
+       DBG_NO_LOC("      suspendable:   %u", SUSPENDABLE_COUNT(tracker));
+       DBG_NO_LOC("      unsuspendable: %u", UNSUSPENDABLE_COUNT(tracker));
+       DBG_NO_LOC("    suspended:       %u", SUSPENDED_COUNT(tracker));
+       DBG_NO_LOC("    capacity:        %u", tracker->capacity);
+
+       DBG_NO_LOC("  Tracked suspendable file descriptors");
+       cds_list_for_each_entry(
+                       handle, &tracker->active_handles, handles_list_node) {
+               fs_handle_tracked_log(handle);
+       }
+       cds_list_for_each_entry(handle, &tracker->suspended_handles,
+                       handles_list_node) {
+               fs_handle_tracked_log(handle);
+       }
+       if (!SUSPENDABLE_COUNT(tracker)) {
+               DBG_NO_LOC("    None");
+       }
+
+       DBG_NO_LOC("  Tracked unsuspendable file descriptors");
+       rcu_read_lock();
+       cds_lfht_for_each_entry(tracker->unsuspendable_fds, &iter,
+                       unsuspendable_fd, tracker_node) {
+               DBG_NO_LOC("    %s [active, fd %d]",
+                               unsuspendable_fd->name ?: "Unnamed",
+                               unsuspendable_fd->fd);
+       }
+       rcu_read_unlock();
+       if (!UNSUSPENDABLE_COUNT(tracker)) {
+               DBG_NO_LOC("    None");
+       }
+
+       pthread_mutex_unlock(&tracker->lock);
+}
+
+int fd_tracker_destroy(struct fd_tracker *tracker)
+{
+       int ret = 0;
+
+       if (!tracker) {
+               goto end;
+       }
+       /*
+        * Refuse to destroy the tracker as fs_handles may still old
+        * weak references to the tracker.
+        */
+       pthread_mutex_lock(&tracker->lock);
+       if (TRACKED_COUNT(tracker)) {
+               ERR("A file descriptor leak has been detected: %u tracked file descriptors are still being tracked",
+                               TRACKED_COUNT(tracker));
+               pthread_mutex_unlock(&tracker->lock);
+               fd_tracker_log(tracker);
+               ret = -1;
+               goto end;
+       }
+       pthread_mutex_unlock(&tracker->lock);
+
+       if (tracker->unsuspendable_fds) {
+               ret = cds_lfht_destroy(tracker->unsuspendable_fds, NULL);
+               LTTNG_ASSERT(!ret);
+       }
+
+       lttng_inode_registry_destroy(tracker->inode_registry);
+       lttng_unlinked_file_pool_destroy(tracker->unlinked_file_pool);
+       pthread_mutex_destroy(&tracker->lock);
+       free(tracker);
+end:
+       return ret;
+}
+
+struct fs_handle *fd_tracker_open_fs_handle(struct fd_tracker *tracker,
+               struct lttng_directory_handle *directory,
+               const char *path,
+               int flags,
+               mode_t *mode)
+{
+       int ret;
+       struct fs_handle_tracked *handle = NULL;
+       struct stat fd_stat;
+       struct open_properties properties = {
+               .flags = flags,
+               .mode = {
+                       .is_set = !!mode,
+                       .value = mode ? *mode : 0,
+               }
+       };
+
+       pthread_mutex_lock(&tracker->lock);
+       if (ACTIVE_COUNT(tracker) == tracker->capacity) {
+               if (tracker->count.suspendable.active > 0) {
+                       ret = fd_tracker_suspend_handles(tracker, 1);
+                       if (ret) {
+                               goto end;
+                       }
+               } else {
+                       /*
+                        * There are not enough active suspendable file
+                        * descriptors to open a new fd and still accommodate
+                        * the tracker's capacity.
+                        */
+                       WARN("Cannot open file system handle, too many unsuspendable file descriptors are opened (%u)",
+                                       tracker->count.unsuspendable);
+                       goto end;
+               }
+       }
+
+       handle = (fs_handle_tracked *) zmalloc(sizeof(*handle));
+       if (!handle) {
+               goto end;
+       }
+       handle->parent = (typeof(handle->parent)) {
+               .get_fd = fs_handle_tracked_get_fd,
+               .put_fd = fs_handle_tracked_put_fd,
+               .unlink = fs_handle_tracked_unlink,
+               .close = fs_handle_tracked_close,
+       };
+
+       handle->tracker = tracker;
+
+       ret = pthread_mutex_init(&handle->lock, NULL);
+       if (ret) {
+               PERROR("Failed to initialize handle mutex while creating fs handle");
+               goto error_mutex_init;
+       }
+
+       handle->fd = open_from_properties(directory, path, &properties);
+       if (handle->fd < 0) {
+               PERROR("Failed to open fs handle to %s, open() returned", path);
+               goto error;
+       }
+
+       handle->properties = properties;
+
+       handle->inode = lttng_inode_registry_get_inode(tracker->inode_registry,
+                       directory, path, handle->fd,
+                       tracker->unlinked_file_pool);
+       if (!handle->inode) {
+               ERR("Failed to get lttng_inode corresponding to file %s", path);
+               goto error;
+       }
+
+       if (fstat(handle->fd, &fd_stat)) {
+               PERROR("Failed to retrieve file descriptor inode while creating fs handle, fstat() returned");
+               goto error;
+       }
+       handle->ino = fd_stat.st_ino;
+
+       fd_tracker_track(tracker, handle);
+end:
+       pthread_mutex_unlock(&tracker->lock);
+       return handle ? &handle->parent : NULL;
+error:
+       if (handle->inode) {
+               lttng_inode_put(handle->inode);
+       }
+       pthread_mutex_destroy(&handle->lock);
+error_mutex_init:
+       free(handle);
+       handle = NULL;
+       goto end;
+}
+
+/* Caller must hold the tracker's lock. */
+static int fd_tracker_suspend_handles(
+               struct fd_tracker *tracker, unsigned int count)
+{
+       unsigned int left_to_close = count;
+       unsigned int attempts_left = tracker->count.suspendable.active;
+       struct fs_handle_tracked *handle, *tmp;
+
+       cds_list_for_each_entry_safe(handle, tmp, &tracker->active_handles,
+                       handles_list_node) {
+               int ret;
+
+               fd_tracker_untrack(tracker, handle);
+               ret = fs_handle_tracked_suspend(handle);
+               fd_tracker_track(tracker, handle);
+               if (!ret) {
+                       left_to_close--;
+               }
+               attempts_left--;
+
+               if (left_to_close == 0 || attempts_left == 0) {
+                       break;
+               }
+       }
+       return left_to_close ? -EMFILE : 0;
+}
+
+int fd_tracker_open_unsuspendable_fd(struct fd_tracker *tracker,
+               int *out_fds,
+               const char **names,
+               unsigned int fd_count,
+               fd_open_cb open,
+               void *user_data)
+{
+       int ret, user_ret, i, fds_to_suspend;
+       unsigned int active_fds;
+       struct unsuspendable_fd **entries;
+
+       entries = (unsuspendable_fd **) zmalloc(fd_count * sizeof(*entries));
+       if (!entries) {
+               ret = -1;
+               goto end;
+       }
+
+       pthread_mutex_lock(&tracker->lock);
+
+       active_fds = ACTIVE_COUNT(tracker);
+       fds_to_suspend = (int) active_fds + (int) fd_count -
+                       (int) tracker->capacity;
+       if (fds_to_suspend > 0) {
+               if (fds_to_suspend <= tracker->count.suspendable.active) {
+                       ret = fd_tracker_suspend_handles(
+                                       tracker, fds_to_suspend);
+                       if (ret) {
+                               goto end_unlock;
+                       }
+               } else {
+                       /*
+                        * There are not enough active suspendable file
+                        * descriptors to open a new fd and still accommodates the
+                        * tracker's capacity.
+                        */
+                       WARN("Cannot open unsuspendable fd, too many unsuspendable file descriptors are opened (%u)",
+                                       tracker->count.unsuspendable);
+                       ret = -EMFILE;
+                       goto end_unlock;
+               }
+       }
+
+       user_ret = open(user_data, out_fds);
+       if (user_ret) {
+               ret = user_ret;
+               goto end_unlock;
+       }
+
+       /*
+        * Add the fds returned by the user's callback to the hashtable
+        * of unsuspendable fds.
+        */
+       for (i = 0; i < fd_count; i++) {
+               struct unsuspendable_fd *entry = unsuspendable_fd_create(
+                               names ? names[i] : NULL, out_fds[i]);
+
+               if (!entry) {
+                       ret = -1;
+                       goto end_free_entries;
+               }
+               entries[i] = entry;
+       }
+
+       rcu_read_lock();
+       for (i = 0; i < fd_count; i++) {
+               struct cds_lfht_node *node;
+               struct unsuspendable_fd *entry = entries[i];
+
+               node = cds_lfht_add_unique(tracker->unsuspendable_fds,
+                               hash_key_ulong((void *) (unsigned long)
+                                                               out_fds[i],
+                                               seed.value),
+                               match_fd, (void *) (unsigned long) out_fds[i],
+                               &entry->tracker_node);
+
+               if (node != &entry->tracker_node) {
+                       ret = -EEXIST;
+                       rcu_read_unlock();
+                       goto end_free_entries;
+               }
+               entries[i] = NULL;
+       }
+       tracker->count.unsuspendable += fd_count;
+       rcu_read_unlock();
+       ret = user_ret;
+end_unlock:
+       pthread_mutex_unlock(&tracker->lock);
+end:
+       free(entries);
+       return ret;
+end_free_entries:
+       for (i = 0; i < fd_count; i++) {
+               unsuspendable_fd_destroy(entries[i]);
+       }
+       goto end_unlock;
+}
+
+int fd_tracker_close_unsuspendable_fd(struct fd_tracker *tracker,
+               int *fds_in,
+               unsigned int fd_count,
+               fd_close_cb close,
+               void *user_data)
+{
+       int i, ret, user_ret;
+       int *fds = NULL;
+
+       /*
+        * Maintain a local copy of fds_in as the user's callback may modify its
+        * contents (e.g. setting the fd(s) to -1 after close).
+        */
+       fds = (int *) malloc(sizeof(*fds) * fd_count);
+       if (!fds) {
+               ret = -1;
+               goto end;
+       }
+       memcpy(fds, fds_in, sizeof(*fds) * fd_count);
+
+       pthread_mutex_lock(&tracker->lock);
+       rcu_read_lock();
+
+       /* Let the user close the file descriptors. */
+       user_ret = close(user_data, fds_in);
+       if (user_ret) {
+               ret = user_ret;
+               goto end_unlock;
+       }
+
+       /* Untrack the fds that were just closed by the user's callback. */
+       for (i = 0; i < fd_count; i++) {
+               struct cds_lfht_node *node;
+               struct cds_lfht_iter iter;
+               struct unsuspendable_fd *entry;
+
+               cds_lfht_lookup(tracker->unsuspendable_fds,
+                               hash_key_ulong((void *) (unsigned long) fds[i],
+                                               seed.value),
+                               match_fd, (void *) (unsigned long) fds[i],
+                               &iter);
+               node = cds_lfht_iter_get_node(&iter);
+               if (!node) {
+                       /* Unknown file descriptor. */
+                       WARN("Untracked file descriptor %d passed to fd_tracker_close_unsuspendable_fd()",
+                                       fds[i]);
+                       ret = -EINVAL;
+                       goto end_unlock;
+               }
+               entry = caa_container_of(
+                               node, struct unsuspendable_fd, tracker_node);
+
+               cds_lfht_del(tracker->unsuspendable_fds, node);
+               unsuspendable_fd_destroy(entry);
+               fds[i] = -1;
+       }
+
+       tracker->count.unsuspendable -= fd_count;
+       ret = 0;
+end_unlock:
+       rcu_read_unlock();
+       pthread_mutex_unlock(&tracker->lock);
+       free(fds);
+end:
+       return ret;
+}
+
+/* Caller must have taken the tracker's and handle's locks. */
+static void fd_tracker_track(
+               struct fd_tracker *tracker, struct fs_handle_tracked *handle)
+{
+       if (handle->fd >= 0) {
+               tracker->count.suspendable.active++;
+               cds_list_add_tail(&handle->handles_list_node,
+                               &tracker->active_handles);
+       } else {
+               tracker->count.suspendable.suspended++;
+               cds_list_add_tail(&handle->handles_list_node,
+                               &tracker->suspended_handles);
+       }
+}
+
+/* Caller must have taken the tracker's and handle's locks. */
+static void fd_tracker_untrack(
+               struct fd_tracker *tracker, struct fs_handle_tracked *handle)
+{
+       if (handle->fd >= 0) {
+               tracker->count.suspendable.active--;
+       } else {
+               tracker->count.suspendable.suspended--;
+       }
+       cds_list_del(&handle->handles_list_node);
+}
+
+/* Caller must have taken the tracker's and handle's locks. */
+static int fd_tracker_restore_handle(
+               struct fd_tracker *tracker, struct fs_handle_tracked *handle)
+{
+       int ret;
+
+       fd_tracker_untrack(tracker, handle);
+       if (ACTIVE_COUNT(tracker) >= tracker->capacity) {
+               ret = fd_tracker_suspend_handles(tracker, 1);
+               if (ret) {
+                       goto end;
+               }
+       }
+       ret = fs_handle_tracked_restore(handle);
+end:
+       fd_tracker_track(tracker, handle);
+       return ret ? ret : handle->fd;
+}
+
+static int fs_handle_tracked_get_fd(struct fs_handle *_handle)
+{
+       int ret;
+       struct fs_handle_tracked *handle =
+                       container_of(_handle, struct fs_handle_tracked, parent);
+
+       /*
+        * TODO This should be optimized as it is a fairly hot path.
+        * The fd-tracker's lock should only be taken when a fs_handle is
+        * restored (slow path). On the fast path (fs_handle is active),
+        * the only effect on the fd_tracker is marking the handle as the
+        * most recently used. Currently, it is done by a call to the
+        * track/untrack helpers, but it should be done atomically.
+        *
+        * Note that the lock's nesting order must still be respected here.
+        * The handle's lock nests inside the tracker's lock.
+        */
+       pthread_mutex_lock(&handle->tracker->lock);
+       pthread_mutex_lock(&handle->lock);
+       LTTNG_ASSERT(!handle->in_use);
+
+       handle->tracker->stats.uses++;
+       if (handle->fd >= 0) {
+               ret = handle->fd;
+               /* Mark as most recently used. */
+               fd_tracker_untrack(handle->tracker, handle);
+               fd_tracker_track(handle->tracker, handle);
+       } else {
+               handle->tracker->stats.misses++;
+               ret = fd_tracker_restore_handle(handle->tracker, handle);
+               if (ret < 0) {
+                       handle->tracker->stats.errors++;
+                       goto end;
+               }
+       }
+       handle->in_use = true;
+end:
+       pthread_mutex_unlock(&handle->lock);
+       pthread_mutex_unlock(&handle->tracker->lock);
+       return ret;
+}
+
+static void fs_handle_tracked_put_fd(struct fs_handle *_handle)
+{
+       struct fs_handle_tracked *handle =
+                       container_of(_handle, struct fs_handle_tracked, parent);
+
+       pthread_mutex_lock(&handle->lock);
+       handle->in_use = false;
+       pthread_mutex_unlock(&handle->lock);
+}
+
+static int fs_handle_tracked_unlink(struct fs_handle *_handle)
+{
+       int ret;
+       struct fs_handle_tracked *handle =
+                       container_of(_handle, struct fs_handle_tracked, parent);
+
+       pthread_mutex_lock(&handle->tracker->lock);
+       pthread_mutex_lock(&handle->lock);
+       ret = lttng_inode_unlink(handle->inode);
+       pthread_mutex_unlock(&handle->lock);
+       pthread_mutex_unlock(&handle->tracker->lock);
+       return ret;
+}
+
+static int fs_handle_tracked_close(struct fs_handle *_handle)
+{
+       int ret = 0;
+       const char *path = NULL;
+       struct fs_handle_tracked *handle =
+                       container_of(_handle, struct fs_handle_tracked, parent);
+       struct lttng_directory_handle *inode_directory_handle = NULL;
+
+       if (!handle) {
+               ret = -EINVAL;
+               goto end;
+       }
+
+       pthread_mutex_lock(&handle->tracker->lock);
+       pthread_mutex_lock(&handle->lock);
+       if (handle->inode) {
+               lttng_inode_borrow_location(handle->inode, NULL, &path);
+               /*
+                * Here a reference to the inode's directory handle is acquired
+                * to prevent the last reference to it from being released while
+                * the tracker's lock is taken.
+                *
+                * If this wasn't done, the directory handle could attempt to
+                * close its underlying directory file descriptor, which would
+                * attempt to lock the tracker's lock, resulting in a deadlock.
+                *
+                * Since a new reference to the directory handle is taken within
+                * the scope of this function, it is not possible for the last
+                * reference to the inode's location directory handle to be
+                * released during the call to lttng_inode_put().
+                *
+                * We wait until the tracker's lock is released to release the
+                * reference. Hence, the call to the tracker is delayed just
+                * enough to not attempt to recursively acquire the tracker's
+                * lock twice.
+                */
+               inode_directory_handle =
+                               lttng_inode_get_location_directory_handle(
+                                               handle->inode);
+       }
+       fd_tracker_untrack(handle->tracker, handle);
+       if (handle->fd >= 0) {
+               /*
+                * The return value of close() is not propagated as there
+                * isn't much the user can do about it.
+                */
+               if (close(handle->fd)) {
+                       PERROR("Failed to close the file descriptor (%d) of fs handle to %s, close() returned",
+                                       handle->fd, path ? path : "Unknown");
+               }
+               handle->fd = -1;
+       }
+       if (handle->inode) {
+               lttng_inode_put(handle->inode);
+       }
+       pthread_mutex_unlock(&handle->lock);
+       pthread_mutex_destroy(&handle->lock);
+       pthread_mutex_unlock(&handle->tracker->lock);
+       free(handle);
+       lttng_directory_handle_put(inode_directory_handle);
+end:
+       return ret;
+}
diff --git a/src/common/fd-tracker/inode.c b/src/common/fd-tracker/inode.c
deleted file mode 100644 (file)
index 88e392b..0000000
+++ /dev/null
@@ -1,543 +0,0 @@
-/*
- * Copyright (C) 2020 Jérémie Galarneau <jeremie.galarneau@efficios.com>
- *
- * SPDX-License-Identifier: GPL-2.0-only
- *
- */
-
-#include <common/defaults.h>
-#include <common/error.h>
-#include <common/hashtable/utils.h>
-#include <common/macros.h>
-#include <common/optional.h>
-#include <common/string-utils/format.h>
-#include <common/utils.h>
-#include <inttypes.h>
-#include <lttng/constant.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <unistd.h>
-#include <urcu.h>
-#include <urcu/rculfhash.h>
-#include <urcu/ref.h>
-
-#include "inode.h"
-
-struct inode_id {
-       dev_t device;
-       ino_t inode;
-};
-
-struct lttng_inode_registry {
-       /* Hashtable of inode_id to lttng_inode. */
-       struct cds_lfht *inodes;
-};
-
-struct lttng_inode {
-       struct inode_id id;
-       /* Node in the lttng_inode_registry's ht. */
-       struct cds_lfht_node registry_node;
-       /* Weak reference to ht containing the node. */
-       struct cds_lfht *registry_ht;
-       struct urcu_ref ref;
-       struct rcu_head rcu_head;
-       /* Location from which this file can be opened. */
-       struct {
-               struct lttng_directory_handle *directory_handle;
-               char *path;
-       } location;
-       /* Unlink the underlying file at the release of the inode. */
-       bool unlink_pending;
-       LTTNG_OPTIONAL(unsigned int) unlinked_id;
-       /* Weak reference. */
-       struct lttng_unlinked_file_pool *unlinked_file_pool;
-};
-
-struct lttng_unlinked_file_pool {
-       struct lttng_directory_handle *unlink_directory_handle;
-       char *unlink_directory_path;
-       unsigned int file_count;
-       unsigned int next_id;
-};
-
-static struct {
-       pthread_mutex_t lock;
-       bool initialized;
-       unsigned long value;
-} seed = {
-               .lock = PTHREAD_MUTEX_INITIALIZER,
-};
-
-static unsigned long lttng_inode_id_hash(const struct inode_id *id)
-{
-       uint64_t device = id->device, inode_no = id->inode;
-
-       return hash_key_u64(&device, seed.value) ^
-              hash_key_u64(&inode_no, seed.value);
-}
-
-static int lttng_inode_match(struct cds_lfht_node *node, const void *key)
-{
-       const struct inode_id *id = key;
-       const struct lttng_inode *inode = caa_container_of(
-                       node, struct lttng_inode, registry_node);
-
-       return inode->id.device == id->device && inode->id.inode == id->inode;
-}
-
-static void lttng_inode_free(struct rcu_head *head)
-{
-       struct lttng_inode *inode =
-                       caa_container_of(head, struct lttng_inode, rcu_head);
-
-       free(inode);
-}
-
-static int lttng_unlinked_file_pool_add_inode(
-               struct lttng_unlinked_file_pool *pool,
-               struct lttng_inode *inode)
-{
-       int ret;
-       const unsigned int unlinked_id = pool->next_id++;
-       char *inode_unlinked_name;
-       bool reference_acquired;
-
-       DBG("Adding inode of %s to unlinked file pool as id %u",
-                       inode->location.path, unlinked_id);
-       ret = asprintf(&inode_unlinked_name, "%u", unlinked_id);
-       if (ret < 0) {
-               ERR("Failed to format unlinked inode name");
-               ret = -1;
-               goto end;
-       }
-
-       if (pool->file_count == 0) {
-               DBG("Creating unlinked files directory at %s",
-                               pool->unlink_directory_path);
-               LTTNG_ASSERT(!pool->unlink_directory_handle);
-               ret = utils_mkdir(pool->unlink_directory_path,
-                               S_IRWXU | S_IRWXG, -1, -1);
-               if (ret) {
-                       if (errno == EEXIST) {
-                               /*
-                                * Unexpected (previous crash?), but not an
-                                * error.
-                                */
-                               DBG("Unlinked file directory \"%s\" already exists",
-                                               pool->unlink_directory_path);
-                       } else {
-                               PERROR("Failed to create unlinked files directory at %s",
-                                               pool->unlink_directory_path);
-                               goto end;
-                       }
-               }
-               pool->unlink_directory_handle = lttng_directory_handle_create(
-                               pool->unlink_directory_path);
-               if (!pool->unlink_directory_handle) {
-                       ERR("Failed to create directory handle to unlinked file pool at %s",
-                                       pool->unlink_directory_path);
-                       ret = -1;
-                       goto end;
-               }
-       }
-
-       ret = lttng_directory_handle_rename(inode->location.directory_handle,
-                       inode->location.path, pool->unlink_directory_handle,
-                       inode_unlinked_name);
-       if (ret) {
-               goto end;
-       }
-
-       lttng_directory_handle_put(inode->location.directory_handle);
-       inode->location.directory_handle = NULL;
-       reference_acquired = lttng_directory_handle_get(
-                       pool->unlink_directory_handle);
-       LTTNG_ASSERT(reference_acquired);
-       inode->location.directory_handle = pool->unlink_directory_handle;
-
-       free(inode->location.path);
-       inode->location.path = inode_unlinked_name;
-       inode_unlinked_name = NULL;
-       LTTNG_OPTIONAL_SET(&inode->unlinked_id, unlinked_id);
-       pool->file_count++;
-end:
-       free(inode_unlinked_name);
-       return ret;
-}
-
-static int lttng_unlinked_file_pool_remove_inode(
-               struct lttng_unlinked_file_pool *pool,
-               struct lttng_inode *inode)
-{
-       int ret;
-
-       DBG("Removing inode with unlinked id %u from unlinked file pool",
-                       LTTNG_OPTIONAL_GET(inode->unlinked_id));
-
-       ret = lttng_directory_handle_unlink_file(
-                       inode->location.directory_handle, inode->location.path);
-       if (ret) {
-               PERROR("Failed to unlink file %s from unlinked file directory",
-                               inode->location.path);
-               goto end;
-       }
-       free(inode->location.path);
-       inode->location.path = NULL;
-       lttng_directory_handle_put(inode->location.directory_handle);
-       inode->location.directory_handle = NULL;
-
-       pool->file_count--;
-       if (pool->file_count == 0) {
-               ret = utils_recursive_rmdir(pool->unlink_directory_path);
-               if (ret) {
-                       /*
-                        * There is nothing the caller can do, don't report an
-                        * error except through logging.
-                        */
-                       PERROR("Failed to remove unlinked files directory at %s",
-                                       pool->unlink_directory_path);
-               }
-               lttng_directory_handle_put(pool->unlink_directory_handle);
-               pool->unlink_directory_handle = NULL;
-       }
-end:
-       return ret;
-}
-
-static void lttng_inode_destroy(struct lttng_inode *inode)
-{
-       if (!inode) {
-               return;
-       }
-
-       rcu_read_lock();
-       cds_lfht_del(inode->registry_ht, &inode->registry_node);
-       rcu_read_unlock();
-
-       if (inode->unlink_pending) {
-               int ret;
-
-               LTTNG_ASSERT(inode->location.directory_handle);
-               LTTNG_ASSERT(inode->location.path);
-               DBG("Removing %s from unlinked file pool",
-                               inode->location.path);
-               ret = lttng_unlinked_file_pool_remove_inode(inode->unlinked_file_pool, inode);
-               if (ret) {
-                       PERROR("Failed to unlink %s", inode->location.path);
-               }
-       }
-
-       lttng_directory_handle_put(
-                       inode->location.directory_handle);
-       inode->location.directory_handle = NULL;
-       free(inode->location.path);
-       inode->location.path = NULL;
-       call_rcu(&inode->rcu_head, lttng_inode_free);
-}
-
-static void lttng_inode_release(struct urcu_ref *ref)
-{
-       lttng_inode_destroy(caa_container_of(ref, struct lttng_inode, ref));
-}
-
-static void lttng_inode_get(struct lttng_inode *inode)
-{
-       urcu_ref_get(&inode->ref);
-}
-
-struct lttng_unlinked_file_pool *lttng_unlinked_file_pool_create(
-               const char *path)
-{
-       struct lttng_unlinked_file_pool *pool = zmalloc(sizeof(*pool));
-
-       if (!pool) {
-               goto error;
-       }
-
-       if (!path || *path != '/') {
-               ERR("Unlinked file pool must be created with an absolute path, path = \"%s\"",
-                               path ? path : "NULL");
-               goto error;
-       }
-
-       pool->unlink_directory_path = strdup(path);
-       if (!pool->unlink_directory_path) {
-               PERROR("Failed to allocation unlinked file pool path");
-               goto error;
-       }
-       DBG("Unlinked file pool created at: %s", path);
-       return pool;
-error:
-       lttng_unlinked_file_pool_destroy(pool);
-       return NULL;
-}
-
-void lttng_unlinked_file_pool_destroy(
-               struct lttng_unlinked_file_pool *pool)
-{
-       if (!pool) {
-               return;
-       }
-
-       LTTNG_ASSERT(pool->file_count == 0);
-       lttng_directory_handle_put(pool->unlink_directory_handle);
-       free(pool->unlink_directory_path);
-       free(pool);
-}
-
-void lttng_inode_put(struct lttng_inode *inode)
-{
-       urcu_ref_put(&inode->ref, lttng_inode_release);
-}
-
-struct lttng_directory_handle *
-lttng_inode_get_location_directory_handle(
-               struct lttng_inode *inode)
-{
-       if (inode->location.directory_handle) {
-               const bool reference_acquired = lttng_directory_handle_get(
-                               inode->location.directory_handle);
-
-               LTTNG_ASSERT(reference_acquired);
-       }
-       return inode->location.directory_handle;
-}
-
-void lttng_inode_borrow_location(struct lttng_inode *inode,
-               const struct lttng_directory_handle **out_directory_handle,
-               const char **out_path)
-{
-       if (out_directory_handle) {
-               *out_directory_handle = inode->location.directory_handle;
-       }
-       if (out_path) {
-               *out_path = inode->location.path;
-       }
-}
-
-int lttng_inode_rename(
-               struct lttng_inode *inode,
-               struct lttng_directory_handle *old_directory_handle,
-               const char *old_path,
-               struct lttng_directory_handle *new_directory_handle,
-               const char *new_path,
-               bool overwrite)
-{
-       int ret = 0;
-       char *new_path_copy = strdup(new_path);
-       bool reference_acquired;
-
-       DBG("Performing rename of inode from %s to %s with %s directory handles",
-                       old_path, new_path,
-                       lttng_directory_handle_equals(old_directory_handle,
-                                       new_directory_handle) ?
-                                       "identical" :
-                                       "different");
-
-       if (!new_path_copy) {
-               ret = -ENOMEM;
-               goto end;
-       }
-
-       if (inode->unlink_pending) {
-               WARN("An attempt to rename an unlinked file from %s to %s has been performed",
-                               old_path, new_path);
-               ret = -ENOENT;
-               goto end;
-       }
-
-       if (!overwrite) {
-               /* Verify that file doesn't exist. */
-               struct stat statbuf;
-
-               ret = lttng_directory_handle_stat(
-                               new_directory_handle, new_path, &statbuf);
-               if (ret == 0) {
-                       ERR("Refusing to rename %s as the destination already exists",
-                                       old_path);
-                       ret = -EEXIST;
-                       goto end;
-               } else if (ret < 0 && errno != ENOENT) {
-                       PERROR("Failed to stat() %s", new_path);
-                       ret = -errno;
-                       goto end;
-               }
-       }
-
-       ret = lttng_directory_handle_rename(old_directory_handle, old_path,
-                       new_directory_handle, new_path);
-       if (ret) {
-               PERROR("Failed to rename file %s to %s", old_path, new_path);
-               ret = -errno;
-               goto end;
-       }
-
-       reference_acquired = lttng_directory_handle_get(new_directory_handle);
-       LTTNG_ASSERT(reference_acquired);
-       lttng_directory_handle_put(inode->location.directory_handle);
-       free(inode->location.path);
-       inode->location.directory_handle = new_directory_handle;
-       /* Ownership transferred. */
-       inode->location.path = new_path_copy;
-       new_path_copy = NULL;
-end:
-       free(new_path_copy);
-       return ret;
-}
-
-int lttng_inode_unlink(struct lttng_inode *inode)
-{
-       int ret = 0;
-
-       DBG("Attempting unlink of inode %s", inode->location.path);
-
-       if (inode->unlink_pending) {
-               WARN("An attempt to re-unlink %s has been performed, ignoring.",
-                               inode->location.path);
-               ret = -ENOENT;
-               goto end;
-       }
-
-       /*
-        * Move to the temporary "deleted" directory until all
-        * references are released.
-        */
-       ret = lttng_unlinked_file_pool_add_inode(
-                       inode->unlinked_file_pool, inode);
-       if (ret) {
-               PERROR("Failed to add inode \"%s\" to the unlinked file pool",
-                               inode->location.path);
-               goto end;
-       }
-       inode->unlink_pending = true;
-end:
-       return ret;
-}
-
-static struct lttng_inode *lttng_inode_create(const struct inode_id *id,
-               struct cds_lfht *ht,
-               struct lttng_unlinked_file_pool *unlinked_file_pool,
-               struct lttng_directory_handle *directory_handle,
-               const char *path)
-{
-       struct lttng_inode *inode = NULL;
-       char *path_copy;
-       bool reference_acquired;
-
-       path_copy = strdup(path);
-       if (!path_copy) {
-               goto end;
-       }
-
-       reference_acquired = lttng_directory_handle_get(directory_handle);
-       LTTNG_ASSERT(reference_acquired);
-
-       inode = zmalloc(sizeof(*inode));
-       if (!inode) {
-               goto end;
-       }
-
-       urcu_ref_init(&inode->ref);
-       cds_lfht_node_init(&inode->registry_node);
-       inode->id = *id;
-       inode->registry_ht = ht;
-       inode->unlinked_file_pool = unlinked_file_pool;
-       /* Ownership of path copy is transferred to inode. */
-       inode->location.path = path_copy;
-       path_copy = NULL;
-       inode->location.directory_handle = directory_handle;
-end:
-       free(path_copy);
-       return inode;
-}
-
-struct lttng_inode_registry *lttng_inode_registry_create(void)
-{
-       struct lttng_inode_registry *registry = zmalloc(sizeof(*registry));
-
-       if (!registry) {
-               goto end;
-       }
-
-       pthread_mutex_lock(&seed.lock);
-       if (!seed.initialized) {
-               seed.value = (unsigned long) time(NULL);
-               seed.initialized = true;
-       }
-       pthread_mutex_unlock(&seed.lock);
-
-       registry->inodes = cds_lfht_new(DEFAULT_HT_SIZE, 1, 0,
-                       CDS_LFHT_AUTO_RESIZE | CDS_LFHT_ACCOUNTING, NULL);
-       if (!registry->inodes) {
-               goto error;
-       }
-end:
-       return registry;
-error:
-       lttng_inode_registry_destroy(registry);
-       return NULL;
-}
-
-void lttng_inode_registry_destroy(
-               struct lttng_inode_registry *registry)
-{
-       if (!registry) {
-               return;
-       }
-       if (registry->inodes) {
-               int ret = cds_lfht_destroy(registry->inodes, NULL);
-
-               LTTNG_ASSERT(!ret);
-       }
-       free(registry);
-}
-
-struct lttng_inode *lttng_inode_registry_get_inode(
-               struct lttng_inode_registry *registry,
-               struct lttng_directory_handle *handle,
-               const char *path,
-               int fd,
-               struct lttng_unlinked_file_pool *unlinked_file_pool)
-{
-       int ret;
-       struct stat statbuf;
-       struct inode_id id;
-       struct cds_lfht_iter iter;
-       struct cds_lfht_node *node;
-       struct lttng_inode *inode = NULL;
-
-       ret = fstat(fd, &statbuf);
-       if (ret < 0) {
-               PERROR("stat() failed on fd %i", fd);
-               goto end;
-       }
-
-       id.device = statbuf.st_dev;
-       id.inode = statbuf.st_ino;
-
-       rcu_read_lock();
-       cds_lfht_lookup(registry->inodes, lttng_inode_id_hash(&id),
-                       lttng_inode_match, &id, &iter);
-       node = cds_lfht_iter_get_node(&iter);
-       if (node) {
-               inode = caa_container_of(
-                               node, struct lttng_inode, registry_node);
-               lttng_inode_get(inode);
-               goto end_unlock;
-       }
-
-       inode = lttng_inode_create(&id, registry->inodes, unlinked_file_pool,
-                       handle, path);
-       if (!inode) {
-               goto end_unlock;
-       }
-
-       node = cds_lfht_add_unique(registry->inodes,
-                       lttng_inode_id_hash(&inode->id), lttng_inode_match,
-                       &inode->id, &inode->registry_node);
-       LTTNG_ASSERT(node == &inode->registry_node);
-end_unlock:
-       rcu_read_unlock();
-end:
-       return inode;
-}
diff --git a/src/common/fd-tracker/inode.cpp b/src/common/fd-tracker/inode.cpp
new file mode 100644 (file)
index 0000000..404ba36
--- /dev/null
@@ -0,0 +1,543 @@
+/*
+ * Copyright (C) 2020 Jérémie Galarneau <jeremie.galarneau@efficios.com>
+ *
+ * SPDX-License-Identifier: GPL-2.0-only
+ *
+ */
+
+#include <common/defaults.h>
+#include <common/error.h>
+#include <common/hashtable/utils.h>
+#include <common/macros.h>
+#include <common/optional.h>
+#include <common/string-utils/format.h>
+#include <common/utils.h>
+#include <inttypes.h>
+#include <lttng/constant.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <unistd.h>
+#include <urcu.h>
+#include <urcu/rculfhash.h>
+#include <urcu/ref.h>
+
+#include "inode.h"
+
+struct inode_id {
+       dev_t device;
+       ino_t inode;
+};
+
+struct lttng_inode_registry {
+       /* Hashtable of inode_id to lttng_inode. */
+       struct cds_lfht *inodes;
+};
+
+struct lttng_inode {
+       struct inode_id id;
+       /* Node in the lttng_inode_registry's ht. */
+       struct cds_lfht_node registry_node;
+       /* Weak reference to ht containing the node. */
+       struct cds_lfht *registry_ht;
+       struct urcu_ref ref;
+       struct rcu_head rcu_head;
+       /* Location from which this file can be opened. */
+       struct {
+               struct lttng_directory_handle *directory_handle;
+               char *path;
+       } location;
+       /* Unlink the underlying file at the release of the inode. */
+       bool unlink_pending;
+       LTTNG_OPTIONAL(unsigned int) unlinked_id;
+       /* Weak reference. */
+       struct lttng_unlinked_file_pool *unlinked_file_pool;
+};
+
+struct lttng_unlinked_file_pool {
+       struct lttng_directory_handle *unlink_directory_handle;
+       char *unlink_directory_path;
+       unsigned int file_count;
+       unsigned int next_id;
+};
+
+static struct {
+       pthread_mutex_t lock;
+       bool initialized;
+       unsigned long value;
+} seed = {
+               .lock = PTHREAD_MUTEX_INITIALIZER,
+};
+
+static unsigned long lttng_inode_id_hash(const struct inode_id *id)
+{
+       uint64_t device = id->device, inode_no = id->inode;
+
+       return hash_key_u64(&device, seed.value) ^
+              hash_key_u64(&inode_no, seed.value);
+}
+
+static int lttng_inode_match(struct cds_lfht_node *node, const void *key)
+{
+       const struct inode_id *id = (inode_id *) key;
+       const struct lttng_inode *inode = caa_container_of(
+                       node, struct lttng_inode, registry_node);
+
+       return inode->id.device == id->device && inode->id.inode == id->inode;
+}
+
+static void lttng_inode_free(struct rcu_head *head)
+{
+       struct lttng_inode *inode =
+                       caa_container_of(head, struct lttng_inode, rcu_head);
+
+       free(inode);
+}
+
+static int lttng_unlinked_file_pool_add_inode(
+               struct lttng_unlinked_file_pool *pool,
+               struct lttng_inode *inode)
+{
+       int ret;
+       const unsigned int unlinked_id = pool->next_id++;
+       char *inode_unlinked_name;
+       bool reference_acquired;
+
+       DBG("Adding inode of %s to unlinked file pool as id %u",
+                       inode->location.path, unlinked_id);
+       ret = asprintf(&inode_unlinked_name, "%u", unlinked_id);
+       if (ret < 0) {
+               ERR("Failed to format unlinked inode name");
+               ret = -1;
+               goto end;
+       }
+
+       if (pool->file_count == 0) {
+               DBG("Creating unlinked files directory at %s",
+                               pool->unlink_directory_path);
+               LTTNG_ASSERT(!pool->unlink_directory_handle);
+               ret = utils_mkdir(pool->unlink_directory_path,
+                               S_IRWXU | S_IRWXG, -1, -1);
+               if (ret) {
+                       if (errno == EEXIST) {
+                               /*
+                                * Unexpected (previous crash?), but not an
+                                * error.
+                                */
+                               DBG("Unlinked file directory \"%s\" already exists",
+                                               pool->unlink_directory_path);
+                       } else {
+                               PERROR("Failed to create unlinked files directory at %s",
+                                               pool->unlink_directory_path);
+                               goto end;
+                       }
+               }
+               pool->unlink_directory_handle = lttng_directory_handle_create(
+                               pool->unlink_directory_path);
+               if (!pool->unlink_directory_handle) {
+                       ERR("Failed to create directory handle to unlinked file pool at %s",
+                                       pool->unlink_directory_path);
+                       ret = -1;
+                       goto end;
+               }
+       }
+
+       ret = lttng_directory_handle_rename(inode->location.directory_handle,
+                       inode->location.path, pool->unlink_directory_handle,
+                       inode_unlinked_name);
+       if (ret) {
+               goto end;
+       }
+
+       lttng_directory_handle_put(inode->location.directory_handle);
+       inode->location.directory_handle = NULL;
+       reference_acquired = lttng_directory_handle_get(
+                       pool->unlink_directory_handle);
+       LTTNG_ASSERT(reference_acquired);
+       inode->location.directory_handle = pool->unlink_directory_handle;
+
+       free(inode->location.path);
+       inode->location.path = inode_unlinked_name;
+       inode_unlinked_name = NULL;
+       LTTNG_OPTIONAL_SET(&inode->unlinked_id, unlinked_id);
+       pool->file_count++;
+end:
+       free(inode_unlinked_name);
+       return ret;
+}
+
+static int lttng_unlinked_file_pool_remove_inode(
+               struct lttng_unlinked_file_pool *pool,
+               struct lttng_inode *inode)
+{
+       int ret;
+
+       DBG("Removing inode with unlinked id %u from unlinked file pool",
+                       LTTNG_OPTIONAL_GET(inode->unlinked_id));
+
+       ret = lttng_directory_handle_unlink_file(
+                       inode->location.directory_handle, inode->location.path);
+       if (ret) {
+               PERROR("Failed to unlink file %s from unlinked file directory",
+                               inode->location.path);
+               goto end;
+       }
+       free(inode->location.path);
+       inode->location.path = NULL;
+       lttng_directory_handle_put(inode->location.directory_handle);
+       inode->location.directory_handle = NULL;
+
+       pool->file_count--;
+       if (pool->file_count == 0) {
+               ret = utils_recursive_rmdir(pool->unlink_directory_path);
+               if (ret) {
+                       /*
+                        * There is nothing the caller can do, don't report an
+                        * error except through logging.
+                        */
+                       PERROR("Failed to remove unlinked files directory at %s",
+                                       pool->unlink_directory_path);
+               }
+               lttng_directory_handle_put(pool->unlink_directory_handle);
+               pool->unlink_directory_handle = NULL;
+       }
+end:
+       return ret;
+}
+
+static void lttng_inode_destroy(struct lttng_inode *inode)
+{
+       if (!inode) {
+               return;
+       }
+
+       rcu_read_lock();
+       cds_lfht_del(inode->registry_ht, &inode->registry_node);
+       rcu_read_unlock();
+
+       if (inode->unlink_pending) {
+               int ret;
+
+               LTTNG_ASSERT(inode->location.directory_handle);
+               LTTNG_ASSERT(inode->location.path);
+               DBG("Removing %s from unlinked file pool",
+                               inode->location.path);
+               ret = lttng_unlinked_file_pool_remove_inode(inode->unlinked_file_pool, inode);
+               if (ret) {
+                       PERROR("Failed to unlink %s", inode->location.path);
+               }
+       }
+
+       lttng_directory_handle_put(
+                       inode->location.directory_handle);
+       inode->location.directory_handle = NULL;
+       free(inode->location.path);
+       inode->location.path = NULL;
+       call_rcu(&inode->rcu_head, lttng_inode_free);
+}
+
+static void lttng_inode_release(struct urcu_ref *ref)
+{
+       lttng_inode_destroy(caa_container_of(ref, struct lttng_inode, ref));
+}
+
+static void lttng_inode_get(struct lttng_inode *inode)
+{
+       urcu_ref_get(&inode->ref);
+}
+
+struct lttng_unlinked_file_pool *lttng_unlinked_file_pool_create(
+               const char *path)
+{
+       struct lttng_unlinked_file_pool *pool = (lttng_unlinked_file_pool *) zmalloc(sizeof(*pool));
+
+       if (!pool) {
+               goto error;
+       }
+
+       if (!path || *path != '/') {
+               ERR("Unlinked file pool must be created with an absolute path, path = \"%s\"",
+                               path ? path : "NULL");
+               goto error;
+       }
+
+       pool->unlink_directory_path = strdup(path);
+       if (!pool->unlink_directory_path) {
+               PERROR("Failed to allocation unlinked file pool path");
+               goto error;
+       }
+       DBG("Unlinked file pool created at: %s", path);
+       return pool;
+error:
+       lttng_unlinked_file_pool_destroy(pool);
+       return NULL;
+}
+
+void lttng_unlinked_file_pool_destroy(
+               struct lttng_unlinked_file_pool *pool)
+{
+       if (!pool) {
+               return;
+       }
+
+       LTTNG_ASSERT(pool->file_count == 0);
+       lttng_directory_handle_put(pool->unlink_directory_handle);
+       free(pool->unlink_directory_path);
+       free(pool);
+}
+
+void lttng_inode_put(struct lttng_inode *inode)
+{
+       urcu_ref_put(&inode->ref, lttng_inode_release);
+}
+
+struct lttng_directory_handle *
+lttng_inode_get_location_directory_handle(
+               struct lttng_inode *inode)
+{
+       if (inode->location.directory_handle) {
+               const bool reference_acquired = lttng_directory_handle_get(
+                               inode->location.directory_handle);
+
+               LTTNG_ASSERT(reference_acquired);
+       }
+       return inode->location.directory_handle;
+}
+
+void lttng_inode_borrow_location(struct lttng_inode *inode,
+               const struct lttng_directory_handle **out_directory_handle,
+               const char **out_path)
+{
+       if (out_directory_handle) {
+               *out_directory_handle = inode->location.directory_handle;
+       }
+       if (out_path) {
+               *out_path = inode->location.path;
+       }
+}
+
+int lttng_inode_rename(
+               struct lttng_inode *inode,
+               struct lttng_directory_handle *old_directory_handle,
+               const char *old_path,
+               struct lttng_directory_handle *new_directory_handle,
+               const char *new_path,
+               bool overwrite)
+{
+       int ret = 0;
+       char *new_path_copy = strdup(new_path);
+       bool reference_acquired;
+
+       DBG("Performing rename of inode from %s to %s with %s directory handles",
+                       old_path, new_path,
+                       lttng_directory_handle_equals(old_directory_handle,
+                                       new_directory_handle) ?
+                                       "identical" :
+                                       "different");
+
+       if (!new_path_copy) {
+               ret = -ENOMEM;
+               goto end;
+       }
+
+       if (inode->unlink_pending) {
+               WARN("An attempt to rename an unlinked file from %s to %s has been performed",
+                               old_path, new_path);
+               ret = -ENOENT;
+               goto end;
+       }
+
+       if (!overwrite) {
+               /* Verify that file doesn't exist. */
+               struct stat statbuf;
+
+               ret = lttng_directory_handle_stat(
+                               new_directory_handle, new_path, &statbuf);
+               if (ret == 0) {
+                       ERR("Refusing to rename %s as the destination already exists",
+                                       old_path);
+                       ret = -EEXIST;
+                       goto end;
+               } else if (ret < 0 && errno != ENOENT) {
+                       PERROR("Failed to stat() %s", new_path);
+                       ret = -errno;
+                       goto end;
+               }
+       }
+
+       ret = lttng_directory_handle_rename(old_directory_handle, old_path,
+                       new_directory_handle, new_path);
+       if (ret) {
+               PERROR("Failed to rename file %s to %s", old_path, new_path);
+               ret = -errno;
+               goto end;
+       }
+
+       reference_acquired = lttng_directory_handle_get(new_directory_handle);
+       LTTNG_ASSERT(reference_acquired);
+       lttng_directory_handle_put(inode->location.directory_handle);
+       free(inode->location.path);
+       inode->location.directory_handle = new_directory_handle;
+       /* Ownership transferred. */
+       inode->location.path = new_path_copy;
+       new_path_copy = NULL;
+end:
+       free(new_path_copy);
+       return ret;
+}
+
+int lttng_inode_unlink(struct lttng_inode *inode)
+{
+       int ret = 0;
+
+       DBG("Attempting unlink of inode %s", inode->location.path);
+
+       if (inode->unlink_pending) {
+               WARN("An attempt to re-unlink %s has been performed, ignoring.",
+                               inode->location.path);
+               ret = -ENOENT;
+               goto end;
+       }
+
+       /*
+        * Move to the temporary "deleted" directory until all
+        * references are released.
+        */
+       ret = lttng_unlinked_file_pool_add_inode(
+                       inode->unlinked_file_pool, inode);
+       if (ret) {
+               PERROR("Failed to add inode \"%s\" to the unlinked file pool",
+                               inode->location.path);
+               goto end;
+       }
+       inode->unlink_pending = true;
+end:
+       return ret;
+}
+
+static struct lttng_inode *lttng_inode_create(const struct inode_id *id,
+               struct cds_lfht *ht,
+               struct lttng_unlinked_file_pool *unlinked_file_pool,
+               struct lttng_directory_handle *directory_handle,
+               const char *path)
+{
+       struct lttng_inode *inode = NULL;
+       char *path_copy;
+       bool reference_acquired;
+
+       path_copy = strdup(path);
+       if (!path_copy) {
+               goto end;
+       }
+
+       reference_acquired = lttng_directory_handle_get(directory_handle);
+       LTTNG_ASSERT(reference_acquired);
+
+       inode = (lttng_inode *) zmalloc(sizeof(*inode));
+       if (!inode) {
+               goto end;
+       }
+
+       urcu_ref_init(&inode->ref);
+       cds_lfht_node_init(&inode->registry_node);
+       inode->id = *id;
+       inode->registry_ht = ht;
+       inode->unlinked_file_pool = unlinked_file_pool;
+       /* Ownership of path copy is transferred to inode. */
+       inode->location.path = path_copy;
+       path_copy = NULL;
+       inode->location.directory_handle = directory_handle;
+end:
+       free(path_copy);
+       return inode;
+}
+
+struct lttng_inode_registry *lttng_inode_registry_create(void)
+{
+       struct lttng_inode_registry *registry = (lttng_inode_registry *) zmalloc(sizeof(*registry));
+
+       if (!registry) {
+               goto end;
+       }
+
+       pthread_mutex_lock(&seed.lock);
+       if (!seed.initialized) {
+               seed.value = (unsigned long) time(NULL);
+               seed.initialized = true;
+       }
+       pthread_mutex_unlock(&seed.lock);
+
+       registry->inodes = cds_lfht_new(DEFAULT_HT_SIZE, 1, 0,
+                       CDS_LFHT_AUTO_RESIZE | CDS_LFHT_ACCOUNTING, NULL);
+       if (!registry->inodes) {
+               goto error;
+       }
+end:
+       return registry;
+error:
+       lttng_inode_registry_destroy(registry);
+       return NULL;
+}
+
+void lttng_inode_registry_destroy(
+               struct lttng_inode_registry *registry)
+{
+       if (!registry) {
+               return;
+       }
+       if (registry->inodes) {
+               int ret = cds_lfht_destroy(registry->inodes, NULL);
+
+               LTTNG_ASSERT(!ret);
+       }
+       free(registry);
+}
+
+struct lttng_inode *lttng_inode_registry_get_inode(
+               struct lttng_inode_registry *registry,
+               struct lttng_directory_handle *handle,
+               const char *path,
+               int fd,
+               struct lttng_unlinked_file_pool *unlinked_file_pool)
+{
+       int ret;
+       struct stat statbuf;
+       struct inode_id id;
+       struct cds_lfht_iter iter;
+       struct cds_lfht_node *node;
+       struct lttng_inode *inode = NULL;
+
+       ret = fstat(fd, &statbuf);
+       if (ret < 0) {
+               PERROR("stat() failed on fd %i", fd);
+               goto end;
+       }
+
+       id.device = statbuf.st_dev;
+       id.inode = statbuf.st_ino;
+
+       rcu_read_lock();
+       cds_lfht_lookup(registry->inodes, lttng_inode_id_hash(&id),
+                       lttng_inode_match, &id, &iter);
+       node = cds_lfht_iter_get_node(&iter);
+       if (node) {
+               inode = caa_container_of(
+                               node, struct lttng_inode, registry_node);
+               lttng_inode_get(inode);
+               goto end_unlock;
+       }
+
+       inode = lttng_inode_create(&id, registry->inodes, unlinked_file_pool,
+                       handle, path);
+       if (!inode) {
+               goto end_unlock;
+       }
+
+       node = cds_lfht_add_unique(registry->inodes,
+                       lttng_inode_id_hash(&inode->id), lttng_inode_match,
+                       &inode->id, &inode->registry_node);
+       LTTNG_ASSERT(node == &inode->registry_node);
+end_unlock:
+       rcu_read_unlock();
+end:
+       return inode;
+}
diff --git a/src/common/fd-tracker/utils-poll.c b/src/common/fd-tracker/utils-poll.c
deleted file mode 100644 (file)
index 17349a4..0000000
+++ /dev/null
@@ -1,92 +0,0 @@
-/*
- * Copyright (C) 2018 Jérémie Galarneau <jeremie.galarneau@efficios.com>
- *
- * SPDX-License-Identifier: GPL-2.0-only
- *
- */
-
-#include <common/compat/poll.h>
-
-#include "utils.h"
-
-#if HAVE_EPOLL
-
-struct create_args {
-       struct lttng_poll_event *events;
-       int size;
-       int flags;
-};
-
-static int open_epoll(void *data, int *out_fd)
-{
-       int ret;
-       struct create_args *args = data;
-
-       ret = lttng_poll_create(args->events, args->size, args->flags);
-       if (ret < 0) {
-               goto end;
-       }
-
-       *out_fd = args->events->epfd;
-end:
-       return ret;
-}
-
-static int close_epoll(void *data, int *in_fd)
-{
-       /* Will close the epfd. */
-       lttng_poll_clean((struct lttng_poll_event *) data);
-       return 0;
-}
-
-/*
- * The epoll variant of the poll compat layer creates an unsuspendable fd which
- * must be tracked.
- */
-int fd_tracker_util_poll_create(struct fd_tracker *tracker,
-               const char *name,
-               struct lttng_poll_event *events,
-               int size,
-               int flags)
-{
-       int out_fd;
-       struct create_args create_args = {
-               .events = events,
-               .size = size,
-               .flags = flags,
-       };
-
-       return fd_tracker_open_unsuspendable_fd(
-                       tracker, &out_fd, &name, 1, open_epoll, &create_args);
-}
-
-int fd_tracker_util_poll_clean(
-               struct fd_tracker *tracker, struct lttng_poll_event *events)
-{
-       return fd_tracker_close_unsuspendable_fd(
-                       tracker, &events->epfd, 1, close_epoll, events);
-}
-
-#else /* HAVE_EPOLL */
-
-/*
- * The epoll variant of the poll compat layer creates an unsuspendable fd which
- * must be tracked.
- */
-int fd_tracker_util_poll_create(struct fd_tracker *tracker,
-               const char *name,
-               struct lttng_poll_event *events,
-               int size,
-               int flags)
-{
-       return lttng_poll_create(events, size, flags);
-}
-
-int fd_tracker_util_poll_clean(
-               struct fd_tracker *tracker, struct lttng_poll_event *events)
-{
-       lttng_poll_clean(events);
-       return 0;
-}
-
-#endif /* !HAVE_EPOLL */
diff --git a/src/common/fd-tracker/utils-poll.cpp b/src/common/fd-tracker/utils-poll.cpp
new file mode 100644 (file)
index 0000000..4949d81
--- /dev/null
@@ -0,0 +1,92 @@
+/*
+ * Copyright (C) 2018 Jérémie Galarneau <jeremie.galarneau@efficios.com>
+ *
+ * SPDX-License-Identifier: GPL-2.0-only
+ *
+ */
+
+#include <common/compat/poll.h>
+
+#include "utils.h"
+
+#if HAVE_EPOLL
+
+struct create_args {
+       struct lttng_poll_event *events;
+       int size;
+       int flags;
+};
+
+static int open_epoll(void *data, int *out_fd)
+{
+       int ret;
+       struct create_args *args = (create_args *) data;
+
+       ret = lttng_poll_create(args->events, args->size, args->flags);
+       if (ret < 0) {
+               goto end;
+       }
+
+       *out_fd = args->events->epfd;
+end:
+       return ret;
+}
+
+static int close_epoll(void *data, int *in_fd)
+{
+       /* Will close the epfd. */
+       lttng_poll_clean((struct lttng_poll_event *) data);
+       return 0;
+}
+
+/*
+ * The epoll variant of the poll compat layer creates an unsuspendable fd which
+ * must be tracked.
+ */
+int fd_tracker_util_poll_create(struct fd_tracker *tracker,
+               const char *name,
+               struct lttng_poll_event *events,
+               int size,
+               int flags)
+{
+       int out_fd;
+       struct create_args create_args = {
+               .events = events,
+               .size = size,
+               .flags = flags,
+       };
+
+       return fd_tracker_open_unsuspendable_fd(
+                       tracker, &out_fd, &name, 1, open_epoll, &create_args);
+}
+
+int fd_tracker_util_poll_clean(
+               struct fd_tracker *tracker, struct lttng_poll_event *events)
+{
+       return fd_tracker_close_unsuspendable_fd(
+                       tracker, &events->epfd, 1, close_epoll, events);
+}
+
+#else /* HAVE_EPOLL */
+
+/*
+ * The epoll variant of the poll compat layer creates an unsuspendable fd which
+ * must be tracked.
+ */
+int fd_tracker_util_poll_create(struct fd_tracker *tracker,
+               const char *name,
+               struct lttng_poll_event *events,
+               int size,
+               int flags)
+{
+       return lttng_poll_create(events, size, flags);
+}
+
+int fd_tracker_util_poll_clean(
+               struct fd_tracker *tracker, struct lttng_poll_event *events)
+{
+       lttng_poll_clean(events);
+       return 0;
+}
+
+#endif /* !HAVE_EPOLL */
diff --git a/src/common/fd-tracker/utils.c b/src/common/fd-tracker/utils.c
deleted file mode 100644 (file)
index fc307fe..0000000
+++ /dev/null
@@ -1,187 +0,0 @@
-/*
- * Copyright (C) 2018 Jérémie Galarneau <jeremie.galarneau@efficios.com>
- *
- * SPDX-License-Identifier: GPL-2.0-only
- *
- */
-
-#include <common/error.h>
-#include <common/fd-tracker/utils.h>
-#include <common/utils.h>
-#include <lttng/constant.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-
-static
-int open_pipe_cloexec(void *data, int *fds)
-{
-       return utils_create_pipe_cloexec(fds);
-}
-
-static
-int close_pipe(void *data, int *pipe)
-{
-       utils_close_pipe(pipe);
-       pipe[0] = pipe[1] = -1;
-       return 0;
-}
-
-int fd_tracker_util_close_fd(void *unused, int *fd)
-{
-       return close(*fd);
-}
-
-int fd_tracker_util_pipe_open_cloexec(
-               struct fd_tracker *tracker, const char *name, int *pipe)
-{
-       int ret;
-       const char *name_prefix;
-       char *names[2];
-
-       name_prefix = name ? name : "Unknown pipe";
-       ret = asprintf(&names[0], "%s (read end)", name_prefix);
-       if (ret < 0) {
-               goto end;
-       }
-       ret = asprintf(&names[1], "%s (write end)", name_prefix);
-       if (ret < 0) {
-               goto end;
-       }
-
-       ret = fd_tracker_open_unsuspendable_fd(tracker, pipe,
-                       (const char **) names, 2, open_pipe_cloexec, NULL);
-       free(names[0]);
-       free(names[1]);
-end:
-       return ret;
-}
-
-int fd_tracker_util_pipe_close(struct fd_tracker *tracker, int *pipe)
-{
-       return fd_tracker_close_unsuspendable_fd(
-                       tracker, pipe, 2, close_pipe, NULL);
-}
-
-struct open_directory_handle_args {
-       const struct lttng_directory_handle *in_handle;
-       struct lttng_directory_handle *ret_handle;
-       const char *path;
-};
-
-static
-int open_directory_handle(void *_args, int *out_fds)
-{
-       int ret = 0;
-       struct open_directory_handle_args *args = _args;
-       struct lttng_directory_handle *new_handle = NULL;
-
-       new_handle = args->in_handle ?
-                       lttng_directory_handle_create_from_handle(
-                               args->path, args->in_handle) :
-                       lttng_directory_handle_create(args->path);
-       if (!new_handle) {
-               ret = -errno;
-               goto end;
-       }
-
-       args->ret_handle = new_handle;
-
-       /*
-        * Reserved to indicate that the handle does not use a handle; there is
-        * nothing to track. We want to indicate an error to the fd-tracker so
-        * that it doesn't attempt to track the file descriptor, but also want
-        * the caller to retrieve the newly-created handle.
-        *
-        * Calling this a hack is a fair assessment.
-        */
-       if (!lttng_directory_handle_uses_fd(new_handle)) {
-               ret = ENOTSUP;
-       } else {
-#ifdef HAVE_DIRFD
-               *out_fds = new_handle->dirfd;
-#else
-               abort();
-#endif
-
-       }
-end:
-       return ret;
-}
-
-#ifdef HAVE_DIRFD
-static
-int fd_close(void *unused, int *in_fds)
-{
-       const int ret = close(in_fds[0]);
-
-       in_fds[0] = -1;
-       return ret;
-}
-
-static
-void directory_handle_destroy(
-               struct lttng_directory_handle *handle, void *data)
-{
-       struct fd_tracker *tracker = data;
-       const int ret = fd_tracker_close_unsuspendable_fd(
-                       tracker, &handle->dirfd, 1, fd_close, NULL);
-
-       if (ret) {
-               ERR("Failed to untrack directory handle file descriptor");
-       }
-}
-#endif
-
-struct lttng_directory_handle *fd_tracker_create_directory_handle(
-               struct fd_tracker *tracker, const char *path)
-{
-       return fd_tracker_create_directory_handle_from_handle(
-                       tracker, NULL, path);
-}
-
-struct lttng_directory_handle *fd_tracker_create_directory_handle_from_handle(
-               struct fd_tracker *tracker,
-               struct lttng_directory_handle *in_handle,
-               const char *path)
-{
-       int ret;
-       int dirfd = -1;
-       char *handle_name = NULL;
-       char cwd_path[LTTNG_PATH_MAX] = "working directory";
-       struct lttng_directory_handle *new_handle = NULL;
-       struct open_directory_handle_args open_args = {
-               .in_handle = in_handle,
-               .path = path,
-       };
-
-       if (!path) {
-               if (!getcwd(cwd_path, sizeof(cwd_path))) {
-                       PERROR("Failed to get current working directory to name directory handle");
-                       goto end;
-               }
-       }
-
-       ret = asprintf(&handle_name, "Directory handle to %s",
-                       path ? path : cwd_path);
-       if (ret < 0) {
-               PERROR("Failed to format directory handle name");
-               goto end;
-       }
-
-       ret = fd_tracker_open_unsuspendable_fd(tracker, &dirfd,
-                       (const char **) &handle_name, 1, open_directory_handle,
-                       &open_args);
-       if (ret && ret != ENOTSUP) {
-               ERR("Failed to open directory handle to %s through the fd tracker", path ? path : cwd_path);
-       }
-       new_handle = open_args.ret_handle;
-
-#ifdef HAVE_DIRFD
-       new_handle->destroy_cb = directory_handle_destroy;
-       new_handle->destroy_cb_data = tracker;
-#endif
-end:
-       free(handle_name);
-       return new_handle;
-}
diff --git a/src/common/fd-tracker/utils.cpp b/src/common/fd-tracker/utils.cpp
new file mode 100644 (file)
index 0000000..be6c65d
--- /dev/null
@@ -0,0 +1,187 @@
+/*
+ * Copyright (C) 2018 Jérémie Galarneau <jeremie.galarneau@efficios.com>
+ *
+ * SPDX-License-Identifier: GPL-2.0-only
+ *
+ */
+
+#include <common/error.h>
+#include <common/fd-tracker/utils.h>
+#include <common/utils.h>
+#include <lttng/constant.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+static
+int open_pipe_cloexec(void *data, int *fds)
+{
+       return utils_create_pipe_cloexec(fds);
+}
+
+static
+int close_pipe(void *data, int *pipe)
+{
+       utils_close_pipe(pipe);
+       pipe[0] = pipe[1] = -1;
+       return 0;
+}
+
+int fd_tracker_util_close_fd(void *unused, int *fd)
+{
+       return close(*fd);
+}
+
+int fd_tracker_util_pipe_open_cloexec(
+               struct fd_tracker *tracker, const char *name, int *pipe)
+{
+       int ret;
+       const char *name_prefix;
+       char *names[2];
+
+       name_prefix = name ? name : "Unknown pipe";
+       ret = asprintf(&names[0], "%s (read end)", name_prefix);
+       if (ret < 0) {
+               goto end;
+       }
+       ret = asprintf(&names[1], "%s (write end)", name_prefix);
+       if (ret < 0) {
+               goto end;
+       }
+
+       ret = fd_tracker_open_unsuspendable_fd(tracker, pipe,
+                       (const char **) names, 2, open_pipe_cloexec, NULL);
+       free(names[0]);
+       free(names[1]);
+end:
+       return ret;
+}
+
+int fd_tracker_util_pipe_close(struct fd_tracker *tracker, int *pipe)
+{
+       return fd_tracker_close_unsuspendable_fd(
+                       tracker, pipe, 2, close_pipe, NULL);
+}
+
+struct open_directory_handle_args {
+       const struct lttng_directory_handle *in_handle;
+       struct lttng_directory_handle *ret_handle;
+       const char *path;
+};
+
+static
+int open_directory_handle(void *_args, int *out_fds)
+{
+       int ret = 0;
+       struct open_directory_handle_args *args = (open_directory_handle_args *) _args;
+       struct lttng_directory_handle *new_handle = NULL;
+
+       new_handle = args->in_handle ?
+                       lttng_directory_handle_create_from_handle(
+                               args->path, args->in_handle) :
+                       lttng_directory_handle_create(args->path);
+       if (!new_handle) {
+               ret = -errno;
+               goto end;
+       }
+
+       args->ret_handle = new_handle;
+
+       /*
+        * Reserved to indicate that the handle does not use a handle; there is
+        * nothing to track. We want to indicate an error to the fd-tracker so
+        * that it doesn't attempt to track the file descriptor, but also want
+        * the caller to retrieve the newly-created handle.
+        *
+        * Calling this a hack is a fair assessment.
+        */
+       if (!lttng_directory_handle_uses_fd(new_handle)) {
+               ret = ENOTSUP;
+       } else {
+#ifdef HAVE_DIRFD
+               *out_fds = new_handle->dirfd;
+#else
+               abort();
+#endif
+
+       }
+end:
+       return ret;
+}
+
+#ifdef HAVE_DIRFD
+static
+int fd_close(void *unused, int *in_fds)
+{
+       const int ret = close(in_fds[0]);
+
+       in_fds[0] = -1;
+       return ret;
+}
+
+static
+void directory_handle_destroy(
+               struct lttng_directory_handle *handle, void *data)
+{
+       struct fd_tracker *tracker = (fd_tracker *) data;
+       const int ret = fd_tracker_close_unsuspendable_fd(
+                       tracker, &handle->dirfd, 1, fd_close, NULL);
+
+       if (ret) {
+               ERR("Failed to untrack directory handle file descriptor");
+       }
+}
+#endif
+
+struct lttng_directory_handle *fd_tracker_create_directory_handle(
+               struct fd_tracker *tracker, const char *path)
+{
+       return fd_tracker_create_directory_handle_from_handle(
+                       tracker, NULL, path);
+}
+
+struct lttng_directory_handle *fd_tracker_create_directory_handle_from_handle(
+               struct fd_tracker *tracker,
+               struct lttng_directory_handle *in_handle,
+               const char *path)
+{
+       int ret;
+       int dirfd = -1;
+       char *handle_name = NULL;
+       char cwd_path[LTTNG_PATH_MAX] = "working directory";
+       struct lttng_directory_handle *new_handle = NULL;
+       open_directory_handle_args open_args {};
+
+       open_args.in_handle = in_handle;
+       open_args.path = path;
+
+       if (!path) {
+               if (!getcwd(cwd_path, sizeof(cwd_path))) {
+                       PERROR("Failed to get current working directory to name directory handle");
+                       goto end;
+               }
+       }
+
+       ret = asprintf(&handle_name, "Directory handle to %s",
+                       path ? path : cwd_path);
+       if (ret < 0) {
+               PERROR("Failed to format directory handle name");
+               goto end;
+       }
+
+       ret = fd_tracker_open_unsuspendable_fd(tracker, &dirfd,
+                       (const char **) &handle_name, 1, open_directory_handle,
+                       &open_args);
+       if (ret && ret != ENOTSUP) {
+               ERR("Failed to open directory handle to %s through the fd tracker", path ? path : cwd_path);
+       }
+       new_handle = open_args.ret_handle;
+
+#ifdef HAVE_DIRFD
+       new_handle->destroy_cb = directory_handle_destroy;
+       new_handle->destroy_cb_data = tracker;
+#endif
+end:
+       free(handle_name);
+       return new_handle;
+}
This page took 0.079103 seconds and 4 git commands to generate.