src/lxc/monitor.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2008
   5  *
   6  * Authors:
   7  * Daniel Lezcano <daniel.lezcano at free.fr>
   8  * Dwight Engen <dwight.engen@oracle.com>
   9  *
  10  * This library is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2.1 of the License, or (at your option) any later version.
  14  *
  15  * This library is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with this library; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  */
  24
  25 #define _GNU_SOURCE
  26 #include <errno.h>
  27 #include <fcntl.h>
  28 #include <inttypes.h>
  29 #include <poll.h>
  30 #include <stddef.h>
  31 #include <stdint.h>
  32 #include <stdio.h>
  33 #include <stdlib.h>
  34 #include <string.h>
  35 #include <unistd.h>
  36 #include <net/if.h>
  37 #include <netinet/in.h>
  38 #include <sys/param.h>
  39 #include <sys/socket.h>
  40 #include <sys/stat.h>
  41 #include <sys/types.h>
  42 #include <sys/wait.h>
  43
  44 #include "config.h"
  45 #include "af_unix.h"
  46 #include "error.h"
  47 #include "log.h"
  48 #include "lxclock.h"
  49 #include "macro.h"
  50 #include "monitor.h"
  51 #include "state.h"
  52 #include "utils.h"
  53
  54 #ifndef HAVE_STRLCPY
  55 #include "include/strlcpy.h"
  56 #endif
  57
  58 lxc_log_define(monitor, lxc);
  59
  60 /* routines used by monitor publishers (containers) */
  61 int lxc_monitor_fifo_name(const char *lxcpath, char *fifo_path, size_t fifo_path_sz,
  62                           int do_mkdirp)
  63 {
  64         int ret;
  65         char *rundir;
  66
  67         rundir = get_rundir();
  68         if (!rundir)
  69                 return -1;
  70
  71         if (do_mkdirp) {
  72                 ret = snprintf(fifo_path, fifo_path_sz, "%s/lxc/%s", rundir, lxcpath);
  73                 if (ret < 0 || (size_t)ret >= fifo_path_sz) {
  74                         ERROR("rundir/lxcpath (%s/%s) too long for monitor fifo.", rundir, lxcpath);
  75                         free(rundir);
  76                         return -1;
  77                 }
  78                 ret = mkdir_p(fifo_path, 0755);
  79                 if (ret < 0) {
  80                         ERROR("Unable to create monitor fifo directory %s.", fifo_path);
  81                         free(rundir);
  82                         return ret;
  83                 }
  84         }
  85         ret = snprintf(fifo_path, fifo_path_sz, "%s/lxc/%s/monitor-fifo", rundir, lxcpath);
  86         if (ret < 0 || (size_t)ret >= fifo_path_sz) {
  87                 ERROR("rundir/lxcpath (%s/%s) too long for monitor fifo.", rundir, lxcpath);
  88                 free(rundir);
  89                 return -1;
  90         }
  91         free(rundir);
  92         return 0;
  93 }
  94
  95 static void lxc_monitor_fifo_send(struct lxc_msg *msg, const char *lxcpath)
  96 {
  97         int fd,ret;
  98         char fifo_path[PATH_MAX];
  99
 100         BUILD_BUG_ON(sizeof(*msg) > PIPE_BUF); /* write not guaranteed atomic */
 101
 102         ret = lxc_monitor_fifo_name(lxcpath, fifo_path, sizeof(fifo_path), 0);
 103         if (ret < 0)
 104                 return;
 105
 106         /* Open the fifo nonblock in case the monitor is dead, we don't want the
 107          * open to wait for a reader since it may never come.
 108          */
 109         fd = open(fifo_path, O_WRONLY | O_NONBLOCK);
 110         if (fd < 0) {
 111                 /* It is normal for this open() to fail with ENXIO when there is
 112                  * no monitor running, so we don't log it.
 113                  */
 114                 if (errno == ENXIO || errno == ENOENT)
 115                         return;
 116
 117                 SYSWARN("Failed to open fifo to send message");
 118                 return;
 119         }
 120
 121         if (fcntl(fd, F_SETFL, O_WRONLY) < 0) {
 122                 close(fd);
 123                 return;
 124         }
 125
 126         ret = lxc_write_nointr(fd, msg, sizeof(*msg));
 127         if (ret != sizeof(*msg)) {
 128                 close(fd);
 129                 SYSERROR("Failed to write to monitor fifo \"%s\".", fifo_path);
 130                 return;
 131         }
 132
 133         close(fd);
 134 }
 135
 136 void lxc_monitor_send_state(const char *name, lxc_state_t state,
 137                             const char *lxcpath)
 138 {
 139         struct lxc_msg msg = {.type = lxc_msg_state, .value = state};
 140
 141         (void)strlcpy(msg.name, name, sizeof(msg.name));
 142         lxc_monitor_fifo_send(&msg, lxcpath);
 143 }
 144
 145 void lxc_monitor_send_exit_code(const char *name, int exit_code,
 146                                 const char *lxcpath)
 147 {
 148         struct lxc_msg msg = {.type = lxc_msg_exit_code, .value = exit_code};
 149
 150         (void)strlcpy(msg.name, name, sizeof(msg.name));
 151         lxc_monitor_fifo_send(&msg, lxcpath);
 152 }
 153
 154 /* routines used by monitor subscribers (lxc-monitor) */
 155 int lxc_monitor_close(int fd)
 156 {
 157         return close(fd);
 158 }
 159
 160 /* Enforces \0-termination for the abstract unix socket. This is not required
 161  * but allows us to print it out.
 162  *
 163  * Older version of liblxc only allowed for 105 bytes to be used for the
 164  * abstract unix domain socket name because the code for our abstract unix
 165  * socket handling performed invalid checks. Since we \0-terminate we could now
 166  * have a maximum of 106 chars. But to not break backwards compatibility we keep
 167  * the limit at 105.
 168  */
 169 int lxc_monitor_sock_name(const char *lxcpath, struct sockaddr_un *addr) {
 170         size_t len;
 171         int ret;
 172         char *path;
 173         uint64_t hash;
 174
 175         /* addr.sun_path is only 108 bytes, so we hash the full name and
 176          * then append as much of the name as we can fit.
 177          */
 178         memset(addr, 0, sizeof(*addr));
 179         addr->sun_family = AF_UNIX;
 180
 181         /* strlen("lxc/") + strlen("/monitor-sock") + 1 = 18 */
 182         len = strlen(lxcpath) + 18;
 183         path = alloca(len);
 184         ret = snprintf(path, len, "lxc/%s/monitor-sock", lxcpath);
 185         if (ret < 0 || (size_t)ret >= len) {
 186                 ERROR("failed to create name for monitor socket");
 187                 return -1;
 188         }
 189
 190         /* Note: snprintf() will \0-terminate addr->sun_path on the 106th byte
 191          * and so the abstract socket name has 105 "meaningful" characters. This
 192          * is absolutely intentional. For further info read the comment for this
 193          * function above!
 194          */
 195         len = sizeof(addr->sun_path) - 1;
 196         hash = fnv_64a_buf(path, ret, FNV1A_64_INIT);
 197         ret = snprintf(addr->sun_path, len, "@lxc/%016" PRIx64 "/%s", hash, lxcpath);
 198         if (ret < 0) {
 199                 ERROR("failed to create hashed name for monitor socket");
 200                 return -1;
 201         }
 202
 203         /* replace @ with \0 */
 204         addr->sun_path[0] = '\0';
 205         INFO("using monitor socket name \"%s\" (length of socket name %zu must be <= %zu)", &addr->sun_path[1], strlen(&addr->sun_path[1]), sizeof(addr->sun_path) - 3);
 206
 207         return 0;
 208 }
 209
 210 int lxc_monitor_open(const char *lxcpath)
 211 {
 212         struct sockaddr_un addr;
 213         int fd;
 214         size_t retry;
 215         size_t len;
 216         int backoff_ms[] = {10, 50, 100};
 217
 218         if (lxc_monitor_sock_name(lxcpath, &addr) < 0)
 219                 return -1;
 220
 221         len = strlen(&addr.sun_path[1]);
 222         DEBUG("opening monitor socket %s with len %zu", &addr.sun_path[1], len);
 223         if (len >= sizeof(addr.sun_path) - 1) {
 224                 errno = ENAMETOOLONG;
 225                 SYSERROR("The name of monitor socket too long (%zu bytes)", len);
 226                 return -1;
 227         }
 228
 229         for (retry = 0; retry < sizeof(backoff_ms) / sizeof(backoff_ms[0]); retry++) {
 230                 fd = lxc_abstract_unix_connect(addr.sun_path);
 231                 if (fd != -1 || errno != ECONNREFUSED)
 232                         break;
 233
 234                 SYSERROR("Failed to connect to monitor socket. Retrying in %d ms", backoff_ms[retry]);
 235                 usleep(backoff_ms[retry] * 1000);
 236         }
 237
 238         if (fd < 0) {
 239                 SYSERROR("Failed to connect to monitor socket");
 240                 return -1;
 241         }
 242
 243         return fd;
 244 }
 245
 246 int lxc_monitor_read_fdset(struct pollfd *fds, nfds_t nfds, struct lxc_msg *msg,
 247                            int timeout)
 248 {
 249         long i;
 250         int ret;
 251
 252         ret = poll(fds, nfds, timeout * 1000);
 253         if (ret == -1)
 254                 return -1;
 255         else if (ret == 0)
 256                 return -2;  /* timed out */
 257
 258         /* Only read from the first ready fd, the others will remain ready for
 259          * when this routine is called again.
 260          */
 261         for (i = 0; i < nfds; i++) {
 262                 if (fds[i].revents != 0) {
 263                         fds[i].revents = 0;
 264                         ret = recv(fds[i].fd, msg, sizeof(*msg), 0);
 265                         if (ret <= 0) {
 266                                 SYSERROR("Failed to receive message. Did monitord die?");
 267                                 return -1;
 268                         }
 269                         return ret;
 270                 }
 271         }
 272
 273         SYSERROR("No ready fd found.");
 274
 275         return -1;
 276 }
 277
 278 int lxc_monitor_read_timeout(int fd, struct lxc_msg *msg, int timeout)
 279 {
 280         struct pollfd fds;
 281
 282         fds.fd = fd;
 283         fds.events = POLLIN | POLLPRI;
 284         fds.revents = 0;
 285
 286         return lxc_monitor_read_fdset(&fds, 1, msg, timeout);
 287 }
 288
 289 int lxc_monitor_read(int fd, struct lxc_msg *msg)
 290 {
 291         return lxc_monitor_read_timeout(fd, msg, -1);
 292 }
 293
 294 #define LXC_MONITORD_PATH LIBEXECDIR "/lxc/lxc-monitord"
 295
 296 /* Used to spawn a monitord either on startup of a daemon container, or when
 297  * lxc-monitor starts.
 298  */
 299 int lxc_monitord_spawn(const char *lxcpath)
 300 {
 301         int ret;
 302         int pipefd[2];
 303         char pipefd_str[INTTYPE_TO_STRLEN(int)];
 304         pid_t pid1, pid2;
 305
 306         char *const args[] = {
 307                 LXC_MONITORD_PATH,
 308                 (char *)lxcpath,
 309                 pipefd_str,
 310                 NULL,
 311         };
 312
 313         /* double fork to avoid zombies when monitord exits */
 314         pid1 = fork();
 315         if (pid1 < 0) {
 316                 SYSERROR("Failed to fork().");
 317                 return -1;
 318         }
 319
 320         if (pid1) {
 321                 DEBUG("Going to wait for pid %d.", pid1);
 322
 323                 if (waitpid(pid1, NULL, 0) != pid1)
 324                         return -1;
 325
 326                 DEBUG("Finished waiting on pid %d.", pid1);
 327                 return 0;
 328         }
 329
 330         if (pipe(pipefd) < 0) {
 331                 SYSERROR("Failed to create pipe.");
 332                 _exit(EXIT_FAILURE);
 333         }
 334
 335         pid2 = fork();
 336         if (pid2 < 0) {
 337                 SYSERROR("Failed to fork().");
 338                 _exit(EXIT_FAILURE);
 339         }
 340
 341         if (pid2) {
 342                 DEBUG("Trying to sync with child process.");
 343                 char c;
 344                 /* Wait for daemon to create socket. */
 345                 close(pipefd[1]);
 346
 347                 /* Sync with child, we're ignoring the return from read
 348                  * because regardless if it works or not, either way we've
 349                  * synced with the child process. the if-empty-statement
 350                  * construct is to quiet the warn-unused-result warning.
 351                  */
 352                 if (lxc_read_nointr(pipefd[0], &c, 1))
 353                         ;
 354
 355                 close(pipefd[0]);
 356
 357                 DEBUG("Successfully synced with child process.");
 358                 _exit(EXIT_SUCCESS);
 359         }
 360
 361         if (setsid() < 0) {
 362                 SYSERROR("Failed to setsid().");
 363                 _exit(EXIT_FAILURE);
 364         }
 365
 366         lxc_check_inherited(NULL, true, &pipefd[1], 1);
 367         if (null_stdfds() < 0) {
 368                 SYSERROR("Failed to dup2() standard file descriptors to /dev/null.");
 369                 _exit(EXIT_FAILURE);
 370         }
 371
 372         close(pipefd[0]);
 373
 374         ret = snprintf(pipefd_str, sizeof(pipefd_str), "%d", pipefd[1]);
 375         if (ret < 0 || ret >= sizeof(pipefd_str)) {
 376                 ERROR("Failed to create pid argument to pass to monitord.");
 377                 _exit(EXIT_FAILURE);
 378         }
 379
 380         DEBUG("Using pipe file descriptor %d for monitord.", pipefd[1]);
 381
 382         execvp(args[0], args);
 383         SYSERROR("failed to exec lxc-monitord");
 384
 385         _exit(EXIT_FAILURE);
 386 }