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 "monitor.h"
  50 #include "state.h"
  51 #include "utils.h"
  52
  53 #ifndef HAVE_STRLCPY
  54 #include "include/strlcpy.h"
  55 #endif
  56
  57 lxc_log_define(monitor, lxc);
  58
  59 /* routines used by monitor publishers (containers) */
  60 int lxc_monitor_fifo_name(const char *lxcpath, char *fifo_path, size_t fifo_path_sz,
  61                           int do_mkdirp)
  62 {
  63         int ret;
  64         char *rundir;
  65
  66         rundir = get_rundir();
  67         if (!rundir)
  68                 return -1;
  69
  70         if (do_mkdirp) {
  71                 ret = snprintf(fifo_path, fifo_path_sz, "%s/lxc/%s", rundir, lxcpath);
  72                 if (ret < 0 || (size_t)ret >= fifo_path_sz) {
  73                         ERROR("rundir/lxcpath (%s/%s) too long for monitor fifo.", rundir, lxcpath);
  74                         free(rundir);
  75                         return -1;
  76                 }
  77                 ret = mkdir_p(fifo_path, 0755);
  78                 if (ret < 0) {
  79                         ERROR("Unable to create monitor fifo directory %s.", fifo_path);
  80                         free(rundir);
  81                         return ret;
  82                 }
  83         }
  84         ret = snprintf(fifo_path, fifo_path_sz, "%s/lxc/%s/monitor-fifo", rundir, lxcpath);
  85         if (ret < 0 || (size_t)ret >= fifo_path_sz) {
  86                 ERROR("rundir/lxcpath (%s/%s) too long for monitor fifo.", rundir, lxcpath);
  87                 free(rundir);
  88                 return -1;
  89         }
  90         free(rundir);
  91         return 0;
  92 }
  93
  94 static void lxc_monitor_fifo_send(struct lxc_msg *msg, const char *lxcpath)
  95 {
  96         int fd,ret;
  97         char fifo_path[PATH_MAX];
  98
  99         BUILD_BUG_ON(sizeof(*msg) > PIPE_BUF); /* write not guaranteed atomic */
 100
 101         ret = lxc_monitor_fifo_name(lxcpath, fifo_path, sizeof(fifo_path), 0);
 102         if (ret < 0)
 103                 return;
 104
 105         /* Open the fifo nonblock in case the monitor is dead, we don't want the
 106          * open to wait for a reader since it may never come.
 107          */
 108         fd = open(fifo_path, O_WRONLY | O_NONBLOCK);
 109         if (fd < 0) {
 110                 /* It is normal for this open() to fail with ENXIO when there is
 111                  * no monitor running, so we don't log it.
 112                  */
 113                 if (errno == ENXIO || errno == ENOENT)
 114                         return;
 115
 116                 SYSWARN("Failed to open fifo to send message");
 117                 return;
 118         }
 119
 120         if (fcntl(fd, F_SETFL, O_WRONLY) < 0) {
 121                 close(fd);
 122                 return;
 123         }
 124
 125         ret = lxc_write_nointr(fd, msg, sizeof(*msg));
 126         if (ret != sizeof(*msg)) {
 127                 close(fd);
 128                 SYSERROR("Failed to write to monitor fifo \"%s\".", fifo_path);
 129                 return;
 130         }
 131
 132         close(fd);
 133 }
 134
 135 void lxc_monitor_send_state(const char *name, lxc_state_t state,
 136                             const char *lxcpath)
 137 {
 138         struct lxc_msg msg = {.type = lxc_msg_state, .value = state};
 139
 140         (void)strlcpy(msg.name, name, sizeof(msg.name));
 141         lxc_monitor_fifo_send(&msg, lxcpath);
 142 }
 143
 144 void lxc_monitor_send_exit_code(const char *name, int exit_code,
 145                                 const char *lxcpath)
 146 {
 147         struct lxc_msg msg = {.type = lxc_msg_exit_code, .value = exit_code};
 148
 149         (void)strlcpy(msg.name, name, sizeof(msg.name));
 150         lxc_monitor_fifo_send(&msg, lxcpath);
 151 }
 152
 153 /* routines used by monitor subscribers (lxc-monitor) */
 154 int lxc_monitor_close(int fd)
 155 {
 156         return close(fd);
 157 }
 158
 159 /* Enforces \0-termination for the abstract unix socket. This is not required
 160  * but allows us to print it out.
 161  *
 162  * Older version of liblxc only allowed for 105 bytes to be used for the
 163  * abstract unix domain socket name because the code for our abstract unix
 164  * socket handling performed invalid checks. Since we \0-terminate we could now
 165  * have a maximum of 106 chars. But to not break backwards compatibility we keep
 166  * the limit at 105.
 167  */
 168 int lxc_monitor_sock_name(const char *lxcpath, struct sockaddr_un *addr) {
 169         size_t len;
 170         int ret;
 171         char *path;
 172         uint64_t hash;
 173
 174         /* addr.sun_path is only 108 bytes, so we hash the full name and
 175          * then append as much of the name as we can fit.
 176          */
 177         memset(addr, 0, sizeof(*addr));
 178         addr->sun_family = AF_UNIX;
 179
 180         /* strlen("lxc/") + strlen("/monitor-sock") + 1 = 18 */
 181         len = strlen(lxcpath) + 18;
 182         path = alloca(len);
 183         ret = snprintf(path, len, "lxc/%s/monitor-sock", lxcpath);
 184         if (ret < 0 || (size_t)ret >= len) {
 185                 ERROR("failed to create name for monitor socket");
 186                 return -1;
 187         }
 188
 189         /* Note: snprintf() will \0-terminate addr->sun_path on the 106th byte
 190          * and so the abstract socket name has 105 "meaningful" characters. This
 191          * is absolutely intentional. For further info read the comment for this
 192          * function above!
 193          */
 194         len = sizeof(addr->sun_path) - 1;
 195         hash = fnv_64a_buf(path, ret, FNV1A_64_INIT);
 196         ret = snprintf(addr->sun_path, len, "@lxc/%016" PRIx64 "/%s", hash, lxcpath);
 197         if (ret < 0) {
 198                 ERROR("failed to create hashed name for monitor socket");
 199                 return -1;
 200         }
 201
 202         /* replace @ with \0 */
 203         addr->sun_path[0] = '\0';
 204         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);
 205
 206         return 0;
 207 }
 208
 209 int lxc_monitor_open(const char *lxcpath)
 210 {
 211         struct sockaddr_un addr;
 212         int fd;
 213         size_t retry;
 214         size_t len;
 215         int backoff_ms[] = {10, 50, 100};
 216
 217         if (lxc_monitor_sock_name(lxcpath, &addr) < 0)
 218                 return -1;
 219
 220         len = strlen(&addr.sun_path[1]);
 221         DEBUG("opening monitor socket %s with len %zu", &addr.sun_path[1], len);
 222         if (len >= sizeof(addr.sun_path) - 1) {
 223                 errno = ENAMETOOLONG;
 224                 SYSERROR("The name of monitor socket too long (%zu bytes)", len);
 225                 return -1;
 226         }
 227
 228         for (retry = 0; retry < sizeof(backoff_ms) / sizeof(backoff_ms[0]); retry++) {
 229                 fd = lxc_abstract_unix_connect(addr.sun_path);
 230                 if (fd != -1 || errno != ECONNREFUSED)
 231                         break;
 232
 233                 SYSERROR("Failed to connect to monitor socket. Retrying in %d ms", backoff_ms[retry]);
 234                 usleep(backoff_ms[retry] * 1000);
 235         }
 236
 237         if (fd < 0) {
 238                 SYSERROR("Failed to connect to monitor socket");
 239                 return -1;
 240         }
 241
 242         return fd;
 243 }
 244
 245 int lxc_monitor_read_fdset(struct pollfd *fds, nfds_t nfds, struct lxc_msg *msg,
 246                            int timeout)
 247 {
 248         long i;
 249         int ret;
 250
 251         ret = poll(fds, nfds, timeout * 1000);
 252         if (ret == -1)
 253                 return -1;
 254         else if (ret == 0)
 255                 return -2;  /* timed out */
 256
 257         /* Only read from the first ready fd, the others will remain ready for
 258          * when this routine is called again.
 259          */
 260         for (i = 0; i < nfds; i++) {
 261                 if (fds[i].revents != 0) {
 262                         fds[i].revents = 0;
 263                         ret = recv(fds[i].fd, msg, sizeof(*msg), 0);
 264                         if (ret <= 0) {
 265                                 SYSERROR("Failed to receive message. Did monitord die?");
 266                                 return -1;
 267                         }
 268                         return ret;
 269                 }
 270         }
 271
 272         SYSERROR("No ready fd found.");
 273
 274         return -1;
 275 }
 276
 277 int lxc_monitor_read_timeout(int fd, struct lxc_msg *msg, int timeout)
 278 {
 279         struct pollfd fds;
 280
 281         fds.fd = fd;
 282         fds.events = POLLIN | POLLPRI;
 283         fds.revents = 0;
 284
 285         return lxc_monitor_read_fdset(&fds, 1, msg, timeout);
 286 }
 287
 288 int lxc_monitor_read(int fd, struct lxc_msg *msg)
 289 {
 290         return lxc_monitor_read_timeout(fd, msg, -1);
 291 }
 292
 293 #define LXC_MONITORD_PATH LIBEXECDIR "/lxc/lxc-monitord"
 294
 295 /* Used to spawn a monitord either on startup of a daemon container, or when
 296  * lxc-monitor starts.
 297  */
 298 int lxc_monitord_spawn(const char *lxcpath)
 299 {
 300         int ret;
 301         int pipefd[2];
 302         char pipefd_str[LXC_NUMSTRLEN64];
 303         pid_t pid1, pid2;
 304
 305         char *const args[] = {
 306                 LXC_MONITORD_PATH,
 307                 (char *)lxcpath,
 308                 pipefd_str,
 309                 NULL,
 310         };
 311
 312         /* double fork to avoid zombies when monitord exits */
 313         pid1 = fork();
 314         if (pid1 < 0) {
 315                 SYSERROR("Failed to fork().");
 316                 return -1;
 317         }
 318
 319         if (pid1) {
 320                 DEBUG("Going to wait for pid %d.", pid1);
 321
 322                 if (waitpid(pid1, NULL, 0) != pid1)
 323                         return -1;
 324
 325                 DEBUG("Finished waiting on pid %d.", pid1);
 326                 return 0;
 327         }
 328
 329         if (pipe(pipefd) < 0) {
 330                 SYSERROR("Failed to create pipe.");
 331                 _exit(EXIT_FAILURE);
 332         }
 333
 334         pid2 = fork();
 335         if (pid2 < 0) {
 336                 SYSERROR("Failed to fork().");
 337                 _exit(EXIT_FAILURE);
 338         }
 339
 340         if (pid2) {
 341                 DEBUG("Trying to sync with child process.");
 342                 char c;
 343                 /* Wait for daemon to create socket. */
 344                 close(pipefd[1]);
 345
 346                 /* Sync with child, we're ignoring the return from read
 347                  * because regardless if it works or not, either way we've
 348                  * synced with the child process. the if-empty-statement
 349                  * construct is to quiet the warn-unused-result warning.
 350                  */
 351                 if (lxc_read_nointr(pipefd[0], &c, 1))
 352                         ;
 353
 354                 close(pipefd[0]);
 355
 356                 DEBUG("Successfully synced with child process.");
 357                 _exit(EXIT_SUCCESS);
 358         }
 359
 360         if (setsid() < 0) {
 361                 SYSERROR("Failed to setsid().");
 362                 _exit(EXIT_FAILURE);
 363         }
 364
 365         lxc_check_inherited(NULL, true, &pipefd[1], 1);
 366         if (null_stdfds() < 0) {
 367                 SYSERROR("Failed to dup2() standard file descriptors to /dev/null.");
 368                 _exit(EXIT_FAILURE);
 369         }
 370
 371         close(pipefd[0]);
 372
 373         ret = snprintf(pipefd_str, LXC_NUMSTRLEN64, "%d", pipefd[1]);
 374         if (ret < 0 || ret >= LXC_NUMSTRLEN64) {
 375                 ERROR("Failed to create pid argument to pass to monitord.");
 376                 _exit(EXIT_FAILURE);
 377         }
 378
 379         DEBUG("Using pipe file descriptor %d for monitord.", pipefd[1]);
 380
 381         execvp(args[0], args);
 382         SYSERROR("failed to exec lxc-monitord");
 383
 384         _exit(EXIT_FAILURE);
 385 }