Move liblttng-ust-common to 'src/lib/'
[lttng-ust.git] / src / lib / lttng-ust-common / fd-tracker.c
diff --git a/src/lib/lttng-ust-common/fd-tracker.c b/src/lib/lttng-ust-common/fd-tracker.c
new file mode 100644 (file)
index 0000000..026a932
--- /dev/null
@@ -0,0 +1,479 @@
+/*
+ * SPDX-License-Identifier: LGPL-2.1-only
+ *
+ * Copyright (C) 2016 Aravind HT <aravind.ht@gmail.com>
+ * Copyright (C) 2016 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
+ */
+
+#include <limits.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/types.h>
+#include <unistd.h>
+#include <assert.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <sys/select.h>
+#include <sys/resource.h>
+#include <sys/time.h>
+#include <fcntl.h>
+#include <pthread.h>
+#include <signal.h>
+#include <stdbool.h>
+#include <urcu/compiler.h>
+#include <urcu/tls-compat.h>
+#include <urcu/system.h>
+
+#include "common/ust-fd.h"
+#include "common/macros.h"
+#include <lttng/ust-error.h>
+#include "common/logging.h"
+
+/* Operations on the fd set. */
+#define IS_FD_VALID(fd)                        ((fd) >= 0 && (fd) < lttng_ust_max_fd)
+#define GET_FD_SET_FOR_FD(fd, fd_sets) (&((fd_sets)[(fd) / FD_SETSIZE]))
+#define CALC_INDEX_TO_SET(fd)          ((fd) % FD_SETSIZE)
+#define IS_FD_STD(fd)                  (IS_FD_VALID(fd) && (fd) <= STDERR_FILENO)
+
+/* Check fd validity before calling these. */
+#define ADD_FD_TO_SET(fd, fd_sets)     \
+               FD_SET(CALC_INDEX_TO_SET(fd), GET_FD_SET_FOR_FD(fd, fd_sets))
+#define IS_FD_SET(fd, fd_sets)         \
+               FD_ISSET(CALC_INDEX_TO_SET(fd), GET_FD_SET_FOR_FD(fd, fd_sets))
+#define DEL_FD_FROM_SET(fd, fd_sets)   \
+               FD_CLR(CALC_INDEX_TO_SET(fd), GET_FD_SET_FOR_FD(fd, fd_sets))
+
+/*
+ * Protect the lttng_fd_set. Nests within the ust_lock, and therefore
+ * within the libc dl lock. Therefore, we need to fixup the TLS before
+ * nesting into this lock.
+ *
+ * The ust_safe_guard_fd_mutex nests within the ust_mutex. This mutex
+ * is also held across fork.
+ */
+static pthread_mutex_t ust_safe_guard_fd_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+/*
+ * Cancel state when grabbing the ust_safe_guard_fd_mutex. Saved when
+ * locking, restored on unlock. Protected by ust_safe_guard_fd_mutex.
+ */
+static int ust_safe_guard_saved_cancelstate;
+
+/*
+ * Track whether we are within lttng-ust or application, for close
+ * system call override by LD_PRELOAD library. This also tracks whether
+ * we are invoking close() from a signal handler nested on an
+ * application thread.
+ */
+static DEFINE_URCU_TLS(int, ust_fd_mutex_nest);
+
+/* fd_set used to book keep fd being used by lttng-ust. */
+static fd_set *lttng_fd_set;
+static int lttng_ust_max_fd;
+static int num_fd_sets;
+static int init_done;
+
+/*
+ * Force a read (imply TLS fixup for dlopen) of TLS variables.
+ */
+void lttng_ust_fixup_fd_tracker_tls(void)
+{
+       asm volatile ("" : : "m" (URCU_TLS(ust_fd_mutex_nest)));
+}
+
+/*
+ * Allocate the fd set array based on the hard limit set for this
+ * process. This will be called during the constructor execution
+ * and will also be called in the child after fork via lttng_ust_init.
+ */
+void lttng_ust_init_fd_tracker(void)
+{
+       struct rlimit rlim;
+       int i;
+
+       if (CMM_LOAD_SHARED(init_done))
+               return;
+
+       memset(&rlim, 0, sizeof(rlim));
+       /* Get the current possible max number of fd for this process. */
+       if (getrlimit(RLIMIT_NOFILE, &rlim) < 0)
+               abort();
+       /*
+        * FD set array size determined using the hard limit. Even if
+        * the process wishes to increase its limit using setrlimit, it
+        * can only do so with the softlimit which will be less than the
+        * hard limit.
+        */
+       lttng_ust_max_fd = rlim.rlim_max;
+       num_fd_sets = lttng_ust_max_fd / FD_SETSIZE;
+       if (lttng_ust_max_fd % FD_SETSIZE)
+               ++num_fd_sets;
+       if (lttng_fd_set != NULL) {
+               free(lttng_fd_set);
+               lttng_fd_set = NULL;
+       }
+       lttng_fd_set = malloc(num_fd_sets * (sizeof(fd_set)));
+       if (!lttng_fd_set)
+               abort();
+       for (i = 0; i < num_fd_sets; i++)
+               FD_ZERO((&lttng_fd_set[i]));
+       CMM_STORE_SHARED(init_done, 1);
+}
+
+void lttng_ust_lock_fd_tracker(void)
+{
+       sigset_t sig_all_blocked, orig_mask;
+       int ret, oldstate;
+
+       ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
+       if (ret) {
+               ERR("pthread_setcancelstate: %s", strerror(ret));
+       }
+       sigfillset(&sig_all_blocked);
+       ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
+       if (ret) {
+               ERR("pthread_sigmask: %s", strerror(ret));
+       }
+       if (!URCU_TLS(ust_fd_mutex_nest)++) {
+               /*
+                * Ensure the compiler don't move the store after the close()
+                * call in case close() would be marked as leaf.
+                */
+               cmm_barrier();
+               pthread_mutex_lock(&ust_safe_guard_fd_mutex);
+               ust_safe_guard_saved_cancelstate = oldstate;
+       }
+       ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
+       if (ret) {
+               ERR("pthread_sigmask: %s", strerror(ret));
+       }
+}
+
+void lttng_ust_unlock_fd_tracker(void)
+{
+       sigset_t sig_all_blocked, orig_mask;
+       int ret, newstate, oldstate;
+       bool restore_cancel = false;
+
+       sigfillset(&sig_all_blocked);
+       ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
+       if (ret) {
+               ERR("pthread_sigmask: %s", strerror(ret));
+       }
+       /*
+        * Ensure the compiler don't move the store before the close()
+        * call, in case close() would be marked as leaf.
+        */
+       cmm_barrier();
+       if (!--URCU_TLS(ust_fd_mutex_nest)) {
+               newstate = ust_safe_guard_saved_cancelstate;
+               restore_cancel = true;
+               pthread_mutex_unlock(&ust_safe_guard_fd_mutex);
+       }
+       ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
+       if (ret) {
+               ERR("pthread_sigmask: %s", strerror(ret));
+       }
+       if (restore_cancel) {
+               ret = pthread_setcancelstate(newstate, &oldstate);
+               if (ret) {
+                       ERR("pthread_setcancelstate: %s", strerror(ret));
+               }
+       }
+}
+
+static int dup_std_fd(int fd)
+{
+       int ret, i;
+       int fd_to_close[STDERR_FILENO + 1];
+       int fd_to_close_count = 0;
+       int dup_cmd = F_DUPFD; /* Default command */
+       int fd_valid = -1;
+
+       if (!(IS_FD_STD(fd))) {
+               /* Should not be here */
+               ret = -1;
+               goto error;
+       }
+
+       /* Check for FD_CLOEXEC flag */
+       ret = fcntl(fd, F_GETFD);
+       if (ret < 0) {
+               PERROR("fcntl on f_getfd");
+               ret = -1;
+               goto error;
+       }
+
+       if (ret & FD_CLOEXEC) {
+               dup_cmd = F_DUPFD_CLOEXEC;
+       }
+
+       /* Perform dup */
+       for (i = 0; i < STDERR_FILENO + 1; i++) {
+               ret = fcntl(fd, dup_cmd, 0);
+               if (ret < 0) {
+                       PERROR("fcntl dup fd");
+                       goto error;
+               }
+
+               if (!(IS_FD_STD(ret))) {
+                       /* fd is outside of STD range, use it. */
+                       fd_valid = ret;
+                       /* Close fd received as argument. */
+                       fd_to_close[i] = fd;
+                       fd_to_close_count++;
+                       break;
+               }
+
+               fd_to_close[i] = ret;
+               fd_to_close_count++;
+       }
+
+       /* Close intermediary fds */
+       for (i = 0; i < fd_to_close_count; i++) {
+               ret = close(fd_to_close[i]);
+               if (ret) {
+                       PERROR("close on temporary fd: %d.", fd_to_close[i]);
+                       /*
+                        * Not using an abort here would yield a complicated
+                        * error handling for the caller. If a failure occurs
+                        * here, the system is already in a bad state.
+                        */
+                       abort();
+               }
+       }
+
+       ret = fd_valid;
+error:
+       return ret;
+}
+
+/*
+ * Needs to be called with ust_safe_guard_fd_mutex held when opening the fd.
+ * Has strict checking of fd validity.
+ *
+ * If fd <= 2, dup the fd until fd > 2. This enables us to bypass
+ * problems that can be encountered if UST uses stdin, stdout, stderr
+ * fds for internal use (daemon etc.). This can happen if the
+ * application closes either of those file descriptors. Intermediary fds
+ * are closed as needed.
+ *
+ * Return -1 on error.
+ *
+ */
+int lttng_ust_add_fd_to_tracker(int fd)
+{
+       int ret;
+       /*
+        * Ensure the tracker is initialized when called from
+        * constructors.
+        */
+       lttng_ust_init_fd_tracker();
+       assert(URCU_TLS(ust_fd_mutex_nest));
+
+       if (IS_FD_STD(fd)) {
+               ret = dup_std_fd(fd);
+               if (ret < 0) {
+                       goto error;
+               }
+               fd = ret;
+       }
+
+       /* Trying to add an fd which we can not accommodate. */
+       assert(IS_FD_VALID(fd));
+       /* Setting an fd thats already set. */
+       assert(!IS_FD_SET(fd, lttng_fd_set));
+
+       ADD_FD_TO_SET(fd, lttng_fd_set);
+       return fd;
+error:
+       return ret;
+}
+
+/*
+ * Needs to be called with ust_safe_guard_fd_mutex held when opening the fd.
+ * Has strict checking for fd validity.
+ */
+void lttng_ust_delete_fd_from_tracker(int fd)
+{
+       /*
+        * Ensure the tracker is initialized when called from
+        * constructors.
+        */
+       lttng_ust_init_fd_tracker();
+
+       assert(URCU_TLS(ust_fd_mutex_nest));
+       /* Not a valid fd. */
+       assert(IS_FD_VALID(fd));
+       /* Deleting an fd which was not set. */
+       assert(IS_FD_SET(fd, lttng_fd_set));
+
+       DEL_FD_FROM_SET(fd, lttng_fd_set);
+}
+
+/*
+ * Interface allowing applications to close arbitrary file descriptors.
+ * We check if it is owned by lttng-ust, and return -1, errno=EBADF
+ * instead of closing it if it is the case.
+ */
+int lttng_ust_safe_close_fd(int fd, int (*close_cb)(int fd))
+{
+       int ret = 0;
+
+       lttng_ust_fixup_fd_tracker_tls();
+
+       /*
+        * Ensure the tracker is initialized when called from
+        * constructors.
+        */
+       lttng_ust_init_fd_tracker();
+
+       /*
+        * If called from lttng-ust, we directly call close without
+        * validating whether the FD is part of the tracked set.
+        */
+       if (URCU_TLS(ust_fd_mutex_nest))
+               return close_cb(fd);
+
+       lttng_ust_lock_fd_tracker();
+       if (IS_FD_VALID(fd) && IS_FD_SET(fd, lttng_fd_set)) {
+               ret = -1;
+               errno = EBADF;
+       } else {
+               ret = close_cb(fd);
+       }
+       lttng_ust_unlock_fd_tracker();
+
+       return ret;
+}
+
+/*
+ * Interface allowing applications to close arbitrary streams.
+ * We check if it is owned by lttng-ust, and return -1, errno=EBADF
+ * instead of closing it if it is the case.
+ */
+int lttng_ust_safe_fclose_stream(FILE *stream, int (*fclose_cb)(FILE *stream))
+{
+       int ret = 0, fd;
+
+       lttng_ust_fixup_fd_tracker_tls();
+
+       /*
+        * Ensure the tracker is initialized when called from
+        * constructors.
+        */
+       lttng_ust_init_fd_tracker();
+
+       /*
+        * If called from lttng-ust, we directly call fclose without
+        * validating whether the FD is part of the tracked set.
+        */
+       if (URCU_TLS(ust_fd_mutex_nest))
+               return fclose_cb(stream);
+
+       fd = fileno(stream);
+
+       lttng_ust_lock_fd_tracker();
+       if (IS_FD_VALID(fd) && IS_FD_SET(fd, lttng_fd_set)) {
+               ret = -1;
+               errno = EBADF;
+       } else {
+               ret = fclose_cb(stream);
+       }
+       lttng_ust_unlock_fd_tracker();
+
+       return ret;
+}
+
+#ifdef __OpenBSD__
+static void set_close_success(int *p)
+{
+       *p = 1;
+}
+static int test_close_success(const int *p)
+{
+       return *p;
+}
+#else
+static void set_close_success(int *p  __attribute__((unused)))
+{
+}
+static int test_close_success(const int *p __attribute__((unused)))
+{
+       return 1;
+}
+#endif
+
+/*
+ * Implement helper for closefrom() override.
+ */
+int lttng_ust_safe_closefrom_fd(int lowfd, int (*close_cb)(int fd))
+{
+       int ret = 0, close_success = 0, i;
+
+       lttng_ust_fixup_fd_tracker_tls();
+
+       /*
+        * Ensure the tracker is initialized when called from
+        * constructors.
+        */
+       lttng_ust_init_fd_tracker();
+
+       if (lowfd < 0) {
+               /*
+                * NetBSD return EBADF if fd is invalid.
+                */
+               errno = EBADF;
+               ret = -1;
+               goto end;
+       }
+       /*
+        * If called from lttng-ust, we directly call close without
+        * validating whether the FD is part of the tracked set.
+        */
+       if (URCU_TLS(ust_fd_mutex_nest)) {
+               for (i = lowfd; i < lttng_ust_max_fd; i++) {
+                       if (close_cb(i) < 0) {
+                               switch (errno) {
+                               case EBADF:
+                                       continue;
+                               case EINTR:
+                               default:
+                                       ret = -1;
+                                       goto end;
+                               }
+                       }
+                       set_close_success(&close_success);
+               }
+       } else {
+               lttng_ust_lock_fd_tracker();
+               for (i = lowfd; i < lttng_ust_max_fd; i++) {
+                       if (IS_FD_VALID(i) && IS_FD_SET(i, lttng_fd_set))
+                               continue;
+                       if (close_cb(i) < 0) {
+                               switch (errno) {
+                               case EBADF:
+                                       continue;
+                               case EINTR:
+                               default:
+                                       ret = -1;
+                                       lttng_ust_unlock_fd_tracker();
+                                       goto end;
+                               }
+                       }
+                       set_close_success(&close_success);
+               }
+               lttng_ust_unlock_fd_tracker();
+       }
+       if (!test_close_success(&close_success)) {
+               /*
+                * OpenBSD return EBADF if fd is greater than all open
+                * file descriptors.
+                */
+               ret = -1;
+               errno = EBADF;
+       }
+end:
+       return ret;
+}
This page took 0.028376 seconds and 4 git commands to generate.