src/lib/lttng-ust-common/fd-tracker.c

   1 /*
   2  * SPDX-License-Identifier: LGPL-2.1-only
   3  *
   4  * Copyright (C) 2016 Aravind HT <aravind.ht@gmail.com>
   5  * Copyright (C) 2016 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   6  */
   7
   8 #include <limits.h>
   9 #include <stdio.h>
  10 #include <stdlib.h>
  11 #include <string.h>
  12 #include <sys/types.h>
  13 #include <unistd.h>
  14 #include <assert.h>
  15 #include <errno.h>
  16 #include <fcntl.h>
  17 #include <sys/select.h>
  18 #include <sys/resource.h>
  19 #include <sys/time.h>
  20 #include <fcntl.h>
  21 #include <pthread.h>
  22 #include <signal.h>
  23 #include <stdbool.h>
  24 #include <urcu/compiler.h>
  25 #include <urcu/tls-compat.h>
  26 #include <urcu/system.h>
  27
  28 #include "common/ust-fd.h"
  29 #include "common/macros.h"
  30 #include <lttng/ust-error.h>
  31 #include "common/logging.h"
  32
  33 #include "lib/lttng-ust-common/fd-tracker.h"
  34
  35 /* Operations on the fd set. */
  36 #define IS_FD_VALID(fd)                 ((fd) >= 0 && (fd) < lttng_ust_max_fd)
  37 #define GET_FD_SET_FOR_FD(fd, fd_sets)  (&((fd_sets)[(fd) / FD_SETSIZE]))
  38 #define CALC_INDEX_TO_SET(fd)           ((fd) % FD_SETSIZE)
  39 #define IS_FD_STD(fd)                   (IS_FD_VALID(fd) && (fd) <= STDERR_FILENO)
  40
  41 /* Check fd validity before calling these. */
  42 #define ADD_FD_TO_SET(fd, fd_sets)      \
  43                 FD_SET(CALC_INDEX_TO_SET(fd), GET_FD_SET_FOR_FD(fd, fd_sets))
  44 #define IS_FD_SET(fd, fd_sets)          \
  45                 FD_ISSET(CALC_INDEX_TO_SET(fd), GET_FD_SET_FOR_FD(fd, fd_sets))
  46 #define DEL_FD_FROM_SET(fd, fd_sets)    \
  47                 FD_CLR(CALC_INDEX_TO_SET(fd), GET_FD_SET_FOR_FD(fd, fd_sets))
  48
  49 /*
  50  * Protect the lttng_fd_set. Nests within the ust_lock, and therefore
  51  * within the libc dl lock. Therefore, we need to allocate the TLS before
  52  * nesting into this lock.
  53  *
  54  * The ust_safe_guard_fd_mutex nests within the ust_mutex. This mutex
  55  * is also held across fork.
  56  */
  57 static pthread_mutex_t ust_safe_guard_fd_mutex = PTHREAD_MUTEX_INITIALIZER;
  58
  59 /*
  60  * Cancel state when grabbing the ust_safe_guard_fd_mutex. Saved when
  61  * locking, restored on unlock. Protected by ust_safe_guard_fd_mutex.
  62  */
  63 static int ust_safe_guard_saved_cancelstate;
  64
  65 /*
  66  * Track whether we are within lttng-ust or application, for close
  67  * system call override by LD_PRELOAD library. This also tracks whether
  68  * we are invoking close() from a signal handler nested on an
  69  * application thread.
  70  */
  71 static DEFINE_URCU_TLS(int, ust_fd_mutex_nest);
  72
  73 /* fd_set used to book keep fd being used by lttng-ust. */
  74 static fd_set *lttng_fd_set;
  75 static int lttng_ust_max_fd;
  76 static int num_fd_sets;
  77 static int init_done;
  78
  79 /*
  80  * Force a read (imply TLS allocation for dlopen) of TLS variables.
  81  */
  82 void lttng_ust_fd_tracker_alloc_tls(void)
  83 {
  84         asm volatile ("" : : "m" (URCU_TLS(ust_fd_mutex_nest)));
  85 }
  86
  87 /*
  88  * Allocate the fd set array based on the hard limit set for this
  89  * process. This will be called during the constructor execution
  90  * and will also be called in the child after fork via lttng_ust_init.
  91  */
  92 void lttng_ust_fd_tracker_init(void)
  93 {
  94         struct rlimit rlim;
  95         int i;
  96
  97         if (CMM_LOAD_SHARED(init_done))
  98                 return;
  99
 100         memset(&rlim, 0, sizeof(rlim));
 101         /* Get the current possible max number of fd for this process. */
 102         if (getrlimit(RLIMIT_NOFILE, &rlim) < 0)
 103                 abort();
 104         /*
 105          * FD set array size determined using the hard limit. Even if
 106          * the process wishes to increase its limit using setrlimit, it
 107          * can only do so with the softlimit which will be less than the
 108          * hard limit.
 109          */
 110         lttng_ust_max_fd = rlim.rlim_max;
 111         num_fd_sets = lttng_ust_max_fd / FD_SETSIZE;
 112         if (lttng_ust_max_fd % FD_SETSIZE)
 113                 ++num_fd_sets;
 114         if (lttng_fd_set != NULL) {
 115                 free(lttng_fd_set);
 116                 lttng_fd_set = NULL;
 117         }
 118         lttng_fd_set = malloc(num_fd_sets * (sizeof(fd_set)));
 119         if (!lttng_fd_set)
 120                 abort();
 121         for (i = 0; i < num_fd_sets; i++)
 122                 FD_ZERO((&lttng_fd_set[i]));
 123         CMM_STORE_SHARED(init_done, 1);
 124 }
 125
 126 void lttng_ust_lock_fd_tracker(void)
 127 {
 128         sigset_t sig_all_blocked, orig_mask;
 129         int ret, oldstate;
 130
 131         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
 132         if (ret) {
 133                 ERR("pthread_setcancelstate: %s", strerror(ret));
 134         }
 135         sigfillset(&sig_all_blocked);
 136         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
 137         if (ret) {
 138                 ERR("pthread_sigmask: %s", strerror(ret));
 139         }
 140         if (!URCU_TLS(ust_fd_mutex_nest)++) {
 141                 /*
 142                  * Ensure the compiler don't move the store after the close()
 143                  * call in case close() would be marked as leaf.
 144                  */
 145                 cmm_barrier();
 146                 pthread_mutex_lock(&ust_safe_guard_fd_mutex);
 147                 ust_safe_guard_saved_cancelstate = oldstate;
 148         }
 149         ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
 150         if (ret) {
 151                 ERR("pthread_sigmask: %s", strerror(ret));
 152         }
 153 }
 154
 155 void lttng_ust_unlock_fd_tracker(void)
 156 {
 157         sigset_t sig_all_blocked, orig_mask;
 158         int ret, newstate, oldstate;
 159         bool restore_cancel = false;
 160
 161         sigfillset(&sig_all_blocked);
 162         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
 163         if (ret) {
 164                 ERR("pthread_sigmask: %s", strerror(ret));
 165         }
 166         /*
 167          * Ensure the compiler don't move the store before the close()
 168          * call, in case close() would be marked as leaf.
 169          */
 170         cmm_barrier();
 171         if (!--URCU_TLS(ust_fd_mutex_nest)) {
 172                 newstate = ust_safe_guard_saved_cancelstate;
 173                 restore_cancel = true;
 174                 pthread_mutex_unlock(&ust_safe_guard_fd_mutex);
 175         }
 176         ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
 177         if (ret) {
 178                 ERR("pthread_sigmask: %s", strerror(ret));
 179         }
 180         if (restore_cancel) {
 181                 ret = pthread_setcancelstate(newstate, &oldstate);
 182                 if (ret) {
 183                         ERR("pthread_setcancelstate: %s", strerror(ret));
 184                 }
 185         }
 186 }
 187
 188 static int dup_std_fd(int fd)
 189 {
 190         int ret, i;
 191         int fd_to_close[STDERR_FILENO + 1];
 192         int fd_to_close_count = 0;
 193         int dup_cmd = F_DUPFD; /* Default command */
 194         int fd_valid = -1;
 195
 196         if (!(IS_FD_STD(fd))) {
 197                 /* Should not be here */
 198                 ret = -1;
 199                 goto error;
 200         }
 201
 202         /* Check for FD_CLOEXEC flag */
 203         ret = fcntl(fd, F_GETFD);
 204         if (ret < 0) {
 205                 PERROR("fcntl on f_getfd");
 206                 ret = -1;
 207                 goto error;
 208         }
 209
 210         if (ret & FD_CLOEXEC) {
 211                 dup_cmd = F_DUPFD_CLOEXEC;
 212         }
 213
 214         /* Perform dup */
 215         for (i = 0; i < STDERR_FILENO + 1; i++) {
 216                 ret = fcntl(fd, dup_cmd, 0);
 217                 if (ret < 0) {
 218                         PERROR("fcntl dup fd");
 219                         goto error;
 220                 }
 221
 222                 if (!(IS_FD_STD(ret))) {
 223                         /* fd is outside of STD range, use it. */
 224                         fd_valid = ret;
 225                         /* Close fd received as argument. */
 226                         fd_to_close[i] = fd;
 227                         fd_to_close_count++;
 228                         break;
 229                 }
 230
 231                 fd_to_close[i] = ret;
 232                 fd_to_close_count++;
 233         }
 234
 235         /* Close intermediary fds */
 236         for (i = 0; i < fd_to_close_count; i++) {
 237                 ret = close(fd_to_close[i]);
 238                 if (ret) {
 239                         PERROR("close on temporary fd: %d.", fd_to_close[i]);
 240                         /*
 241                          * Not using an abort here would yield a complicated
 242                          * error handling for the caller. If a failure occurs
 243                          * here, the system is already in a bad state.
 244                          */
 245                         abort();
 246                 }
 247         }
 248
 249         ret = fd_valid;
 250 error:
 251         return ret;
 252 }
 253
 254 /*
 255  * Needs to be called with ust_safe_guard_fd_mutex held when opening the fd.
 256  * Has strict checking of fd validity.
 257  *
 258  * If fd <= 2, dup the fd until fd > 2. This enables us to bypass
 259  * problems that can be encountered if UST uses stdin, stdout, stderr
 260  * fds for internal use (daemon etc.). This can happen if the
 261  * application closes either of those file descriptors. Intermediary fds
 262  * are closed as needed.
 263  *
 264  * Return -1 on error.
 265  *
 266  */
 267 int lttng_ust_add_fd_to_tracker(int fd)
 268 {
 269         int ret;
 270         /*
 271          * Ensure the tracker is initialized when called from
 272          * constructors.
 273          */
 274         lttng_ust_fd_tracker_init();
 275         assert(URCU_TLS(ust_fd_mutex_nest));
 276
 277         if (IS_FD_STD(fd)) {
 278                 ret = dup_std_fd(fd);
 279                 if (ret < 0) {
 280                         goto error;
 281                 }
 282                 fd = ret;
 283         }
 284
 285         /* Trying to add an fd which we can not accommodate. */
 286         assert(IS_FD_VALID(fd));
 287         /* Setting an fd that's already set. */
 288         assert(!IS_FD_SET(fd, lttng_fd_set));
 289
 290         ADD_FD_TO_SET(fd, lttng_fd_set);
 291         return fd;
 292 error:
 293         return ret;
 294 }
 295
 296 /*
 297  * Needs to be called with ust_safe_guard_fd_mutex held when opening the fd.
 298  * Has strict checking for fd validity.
 299  */
 300 void lttng_ust_delete_fd_from_tracker(int fd)
 301 {
 302         /*
 303          * Ensure the tracker is initialized when called from
 304          * constructors.
 305          */
 306         lttng_ust_fd_tracker_init();
 307
 308         assert(URCU_TLS(ust_fd_mutex_nest));
 309         /* Not a valid fd. */
 310         assert(IS_FD_VALID(fd));
 311         /* Deleting an fd which was not set. */
 312         assert(IS_FD_SET(fd, lttng_fd_set));
 313
 314         DEL_FD_FROM_SET(fd, lttng_fd_set);
 315 }
 316
 317 /*
 318  * Interface allowing applications to close arbitrary file descriptors.
 319  * We check if it is owned by lttng-ust, and return -1, errno=EBADF
 320  * instead of closing it if it is the case.
 321  */
 322 int lttng_ust_safe_close_fd(int fd, int (*close_cb)(int fd))
 323 {
 324         int ret = 0;
 325
 326         lttng_ust_fd_tracker_alloc_tls();
 327
 328         /*
 329          * Ensure the tracker is initialized when called from
 330          * constructors.
 331          */
 332         lttng_ust_fd_tracker_init();
 333
 334         /*
 335          * If called from lttng-ust, we directly call close without
 336          * validating whether the FD is part of the tracked set.
 337          */
 338         if (URCU_TLS(ust_fd_mutex_nest))
 339                 return close_cb(fd);
 340
 341         lttng_ust_lock_fd_tracker();
 342         if (IS_FD_VALID(fd) && IS_FD_SET(fd, lttng_fd_set)) {
 343                 ret = -1;
 344                 errno = EBADF;
 345         } else {
 346                 ret = close_cb(fd);
 347         }
 348         lttng_ust_unlock_fd_tracker();
 349
 350         return ret;
 351 }
 352
 353 /*
 354  * Interface allowing applications to close arbitrary streams.
 355  * We check if it is owned by lttng-ust, and return -1, errno=EBADF
 356  * instead of closing it if it is the case.
 357  */
 358 int lttng_ust_safe_fclose_stream(FILE *stream, int (*fclose_cb)(FILE *stream))
 359 {
 360         int ret = 0, fd;
 361
 362         lttng_ust_fd_tracker_alloc_tls();
 363
 364         /*
 365          * Ensure the tracker is initialized when called from
 366          * constructors.
 367          */
 368         lttng_ust_fd_tracker_init();
 369
 370         /*
 371          * If called from lttng-ust, we directly call fclose without
 372          * validating whether the FD is part of the tracked set.
 373          */
 374         if (URCU_TLS(ust_fd_mutex_nest))
 375                 return fclose_cb(stream);
 376
 377         fd = fileno(stream);
 378
 379         lttng_ust_lock_fd_tracker();
 380         if (IS_FD_VALID(fd) && IS_FD_SET(fd, lttng_fd_set)) {
 381                 ret = -1;
 382                 errno = EBADF;
 383         } else {
 384                 ret = fclose_cb(stream);
 385         }
 386         lttng_ust_unlock_fd_tracker();
 387
 388         return ret;
 389 }
 390
 391 #ifdef __OpenBSD__
 392 static void set_close_success(int *p)
 393 {
 394         *p = 1;
 395 }
 396 static int test_close_success(const int *p)
 397 {
 398         return *p;
 399 }
 400 #else
 401 static void set_close_success(int *p  __attribute__((unused)))
 402 {
 403 }
 404 static int test_close_success(const int *p __attribute__((unused)))
 405 {
 406         return 1;
 407 }
 408 #endif
 409
 410 /*
 411  * Implement helper for closefrom() override.
 412  */
 413 int lttng_ust_safe_closefrom_fd(int lowfd, int (*close_cb)(int fd))
 414 {
 415         int ret = 0, close_success = 0, i;
 416
 417         lttng_ust_fd_tracker_alloc_tls();
 418
 419         /*
 420          * Ensure the tracker is initialized when called from
 421          * constructors.
 422          */
 423         lttng_ust_fd_tracker_init();
 424
 425         if (lowfd < 0) {
 426                 /*
 427                  * NetBSD return EBADF if fd is invalid.
 428                  */
 429                 errno = EBADF;
 430                 ret = -1;
 431                 goto end;
 432         }
 433         /*
 434          * If called from lttng-ust, we directly call close without
 435          * validating whether the FD is part of the tracked set.
 436          */
 437         if (URCU_TLS(ust_fd_mutex_nest)) {
 438                 for (i = lowfd; i < lttng_ust_max_fd; i++) {
 439                         if (close_cb(i) < 0) {
 440                                 switch (errno) {
 441                                 case EBADF:
 442                                         continue;
 443                                 case EINTR:
 444                                 default:
 445                                         ret = -1;
 446                                         goto end;
 447                                 }
 448                         }
 449                         set_close_success(&close_success);
 450                 }
 451         } else {
 452                 lttng_ust_lock_fd_tracker();
 453                 for (i = lowfd; i < lttng_ust_max_fd; i++) {
 454                         if (IS_FD_VALID(i) && IS_FD_SET(i, lttng_fd_set))
 455                                 continue;
 456                         if (close_cb(i) < 0) {
 457                                 switch (errno) {
 458                                 case EBADF:
 459                                         continue;
 460                                 case EINTR:
 461                                 default:
 462                                         ret = -1;
 463                                         lttng_ust_unlock_fd_tracker();
 464                                         goto end;
 465                                 }
 466                         }
 467                         set_close_success(&close_success);
 468                 }
 469                 lttng_ust_unlock_fd_tracker();
 470         }
 471         if (!test_close_success(&close_success)) {
 472                 /*
 473                  * OpenBSD return EBADF if fd is greater than all open
 474                  * file descriptors.
 475                  */
 476                 ret = -1;
 477                 errno = EBADF;
 478         }
 479 end:
 480         return ret;
 481 }