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 #ifndef _GNU_SOURCE
  26 #define _GNU_SOURCE 1
  27 #endif
  28 #include <errno.h>
  29 #include <fcntl.h>
  30 #include <inttypes.h>
  31 #include <net/if.h>
  32 #include <netinet/in.h>
  33 #include <poll.h>
  34 #include <stddef.h>
  35 #include <stdint.h>
  36 #include <stdio.h>
  37 #include <stdlib.h>
  38 #include <string.h>
  39 #include <sys/param.h>
  40 #include <sys/socket.h>
  41 #include <sys/stat.h>
  42 #include <sys/types.h>
  43 #include <sys/wait.h>
  44 #include <unistd.h>
  45
  46 #include "af_unix.h"
  47 #include "config.h"
  48 #include "error.h"
  49 #include "log.h"
  50 #include "lxclock.h"
  51 #include "macro.h"
  52 #include "monitor.h"
  53 #include "state.h"
  54 #include "utils.h"
  55
  56 #ifndef HAVE_STRLCPY
  57 #include "include/strlcpy.h"
  58 #endif
  59
  60 lxc_log_define(monitor, lxc);
  61
  62 /* routines used by monitor publishers (containers) */
  63 int lxc_monitor_fifo_name(const char *lxcpath, char *fifo_path, size_t fifo_path_sz,
  64                           int do_mkdirp)
  65 {
  66         int ret;
  67         char *rundir;
  68
  69         rundir = get_rundir();
  70         if (!rundir)
  71                 return -1;
  72
  73         if (do_mkdirp) {
  74                 ret = snprintf(fifo_path, fifo_path_sz, "%s/lxc/%s", rundir, lxcpath);
  75                 if (ret < 0 || (size_t)ret >= fifo_path_sz) {
  76                         ERROR("rundir/lxcpath (%s/%s) too long for monitor fifo", rundir, lxcpath);
  77                         free(rundir);
  78                         return -1;
  79                 }
  80                 ret = mkdir_p(fifo_path, 0755);
  81                 if (ret < 0) {
  82                         ERROR("Unable to create monitor fifo directory %s.", fifo_path);
  83                         free(rundir);
  84                         return ret;
  85                 }
  86         }
  87         ret = snprintf(fifo_path, fifo_path_sz, "%s/lxc/%s/monitor-fifo", rundir, lxcpath);
  88         if (ret < 0 || (size_t)ret >= fifo_path_sz) {
  89                 ERROR("rundir/lxcpath (%s/%s) too long for monitor fifo", rundir, lxcpath);
  90                 free(rundir);
  91                 return -1;
  92         }
  93         free(rundir);
  94         return 0;
  95 }
  96
  97 static void lxc_monitor_fifo_send(struct lxc_msg *msg, const char *lxcpath)
  98 {
  99         int fd,ret;
 100         char fifo_path[PATH_MAX];
 101
 102         BUILD_BUG_ON(sizeof(*msg) > PIPE_BUF); /* write not guaranteed atomic */
 103
 104         ret = lxc_monitor_fifo_name(lxcpath, fifo_path, sizeof(fifo_path), 0);
 105         if (ret < 0)
 106                 return;
 107
 108         /* Open the fifo nonblock in case the monitor is dead, we don't want the
 109          * open to wait for a reader since it may never come.
 110          */
 111         fd = open(fifo_path, O_WRONLY | O_NONBLOCK);
 112         if (fd < 0) {
 113                 /* It is normal for this open() to fail with ENXIO when there is
 114                  * no monitor running, so we don't log it.
 115                  */
 116                 if (errno == ENXIO || errno == ENOENT)
 117                         return;
 118
 119                 SYSWARN("Failed to open fifo to send message");
 120                 return;
 121         }
 122
 123         if (fcntl(fd, F_SETFL, O_WRONLY) < 0) {
 124                 close(fd);
 125                 return;
 126         }
 127
 128         ret = lxc_write_nointr(fd, msg, sizeof(*msg));
 129         if (ret != sizeof(*msg)) {
 130                 close(fd);
 131                 SYSERROR("Failed to write to monitor fifo \"%s\"", fifo_path);
 132                 return;
 133         }
 134
 135         close(fd);
 136 }
 137
 138 void lxc_monitor_send_state(const char *name, lxc_state_t state,
 139                             const char *lxcpath)
 140 {
 141         struct lxc_msg msg = {.type = lxc_msg_state, .value = state};
 142
 143         (void)strlcpy(msg.name, name, sizeof(msg.name));
 144         lxc_monitor_fifo_send(&msg, lxcpath);
 145 }
 146
 147 void lxc_monitor_send_exit_code(const char *name, int exit_code,
 148                                 const char *lxcpath)
 149 {
 150         struct lxc_msg msg = {.type = lxc_msg_exit_code, .value = exit_code};
 151
 152         (void)strlcpy(msg.name, name, sizeof(msg.name));
 153         lxc_monitor_fifo_send(&msg, lxcpath);
 154 }
 155
 156 /* routines used by monitor subscribers (lxc-monitor) */
 157 int lxc_monitor_close(int fd)
 158 {
 159         return close(fd);
 160 }
 161
 162 /* Enforces \0-termination for the abstract unix socket. This is not required
 163  * but allows us to print it out.
 164  *
 165  * Older version of liblxc only allowed for 105 bytes to be used for the
 166  * abstract unix domain socket name because the code for our abstract unix
 167  * socket handling performed invalid checks. Since we \0-terminate we could now
 168  * have a maximum of 106 chars. But to not break backwards compatibility we keep
 169  * the limit at 105.
 170  */
 171 int lxc_monitor_sock_name(const char *lxcpath, struct sockaddr_un *addr) {
 172         size_t len;
 173         int ret;
 174         char *path;
 175         uint64_t hash;
 176
 177         /* addr.sun_path is only 108 bytes, so we hash the full name and
 178          * then append as much of the name as we can fit.
 179          */
 180         memset(addr, 0, sizeof(*addr));
 181         addr->sun_family = AF_UNIX;
 182
 183         /* strlen("lxc/") + strlen("/monitor-sock") + 1 = 18 */
 184         len = strlen(lxcpath) + 18;
 185         path = alloca(len);
 186         ret = snprintf(path, len, "lxc/%s/monitor-sock", lxcpath);
 187         if (ret < 0 || (size_t)ret >= len) {
 188                 ERROR("Failed to create name for monitor socket");
 189                 return -1;
 190         }
 191
 192         /* Note: snprintf() will \0-terminate addr->sun_path on the 106th byte
 193          * and so the abstract socket name has 105 "meaningful" characters. This
 194          * is absolutely intentional. For further info read the comment for this
 195          * function above!
 196          */
 197         len = sizeof(addr->sun_path) - 1;
 198         hash = fnv_64a_buf(path, ret, FNV1A_64_INIT);
 199         ret = snprintf(addr->sun_path, len, "@lxc/%016" PRIx64 "/%s", hash, lxcpath);
 200         if (ret < 0) {
 201                 ERROR("Failed to create hashed name for monitor socket");
 202                 return -1;
 203         }
 204
 205         /* replace @ with \0 */
 206         addr->sun_path[0] = '\0';
 207         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);
 208
 209         return 0;
 210 }
 211
 212 int lxc_monitor_open(const char *lxcpath)
 213 {
 214         struct sockaddr_un addr;
 215         int fd;
 216         size_t retry;
 217         size_t len;
 218         int backoff_ms[] = {10, 50, 100};
 219
 220         if (lxc_monitor_sock_name(lxcpath, &addr) < 0)
 221                 return -1;
 222
 223         len = strlen(&addr.sun_path[1]);
 224         DEBUG("Opening monitor socket %s with len %zu", &addr.sun_path[1], len);
 225         if (len >= sizeof(addr.sun_path) - 1) {
 226                 errno = ENAMETOOLONG;
 227                 SYSERROR("The name of monitor socket too long (%zu bytes)", len);
 228                 return -1;
 229         }
 230
 231         for (retry = 0; retry < sizeof(backoff_ms) / sizeof(backoff_ms[0]); retry++) {
 232                 fd = lxc_abstract_unix_connect(addr.sun_path);
 233                 if (fd != -1 || errno != ECONNREFUSED)
 234                         break;
 235
 236                 SYSERROR("Failed to connect to monitor socket. Retrying in %d ms", backoff_ms[retry]);
 237                 usleep(backoff_ms[retry] * 1000);
 238         }
 239
 240         if (fd < 0) {
 241                 SYSERROR("Failed to connect to monitor socket");
 242                 return -1;
 243         }
 244
 245         return fd;
 246 }
 247
 248 int lxc_monitor_read_fdset(struct pollfd *fds, nfds_t nfds, struct lxc_msg *msg,
 249                            int timeout)
 250 {
 251         long i;
 252         int ret;
 253
 254         ret = poll(fds, nfds, timeout * 1000);
 255         if (ret == -1)
 256                 return -1;
 257         else if (ret == 0)
 258                 return -2;  /* timed out */
 259
 260         /* Only read from the first ready fd, the others will remain ready for
 261          * when this routine is called again.
 262          */
 263         for (i = 0; i < nfds; i++) {
 264                 if (fds[i].revents != 0) {
 265                         fds[i].revents = 0;
 266                         ret = recv(fds[i].fd, msg, sizeof(*msg), 0);
 267                         if (ret <= 0) {
 268                                 SYSERROR("Failed to receive message. Did monitord die?");
 269                                 return -1;
 270                         }
 271                         return ret;
 272                 }
 273         }
 274
 275         SYSERROR("No ready fd found");
 276
 277         return -1;
 278 }
 279
 280 int lxc_monitor_read_timeout(int fd, struct lxc_msg *msg, int timeout)
 281 {
 282         struct pollfd fds;
 283
 284         fds.fd = fd;
 285         fds.events = POLLIN | POLLPRI;
 286         fds.revents = 0;
 287
 288         return lxc_monitor_read_fdset(&fds, 1, msg, timeout);
 289 }
 290
 291 int lxc_monitor_read(int fd, struct lxc_msg *msg)
 292 {
 293         return lxc_monitor_read_timeout(fd, msg, -1);
 294 }
 295
 296 #define LXC_MONITORD_PATH LIBEXECDIR "/lxc/lxc-monitord"
 297
 298 /* Used to spawn a monitord either on startup of a daemon container, or when
 299  * lxc-monitor starts.
 300  */
 301 int lxc_monitord_spawn(const char *lxcpath)
 302 {
 303         int ret;
 304         int pipefd[2];
 305         char pipefd_str[INTTYPE_TO_STRLEN(int)];
 306         pid_t pid1, pid2;
 307
 308         char *const args[] = {
 309                 LXC_MONITORD_PATH,
 310                 (char *)lxcpath,
 311                 pipefd_str,
 312                 NULL,
 313         };
 314
 315         /* double fork to avoid zombies when monitord exits */
 316         pid1 = fork();
 317         if (pid1 < 0) {
 318                 SYSERROR("Failed to fork()");
 319                 return -1;
 320         }
 321
 322         if (pid1) {
 323                 DEBUG("Going to wait for pid %d", pid1);
 324
 325                 if (waitpid(pid1, NULL, 0) != pid1)
 326                         return -1;
 327
 328                 DEBUG("Finished waiting on pid %d", pid1);
 329                 return 0;
 330         }
 331
 332         if (pipe(pipefd) < 0) {
 333                 SYSERROR("Failed to create pipe");
 334                 _exit(EXIT_FAILURE);
 335         }
 336
 337         pid2 = fork();
 338         if (pid2 < 0) {
 339                 SYSERROR("Failed to fork()");
 340                 _exit(EXIT_FAILURE);
 341         }
 342
 343         if (pid2) {
 344                 DEBUG("Trying to sync with child process");
 345                 char c;
 346                 /* Wait for daemon to create socket. */
 347                 close(pipefd[1]);
 348
 349                 /* Sync with child, we're ignoring the return from read
 350                  * because regardless if it works or not, either way we've
 351                  * synced with the child process. the if-empty-statement
 352                  * construct is to quiet the warn-unused-result warning.
 353                  */
 354                 if (lxc_read_nointr(pipefd[0], &c, 1))
 355                         ;
 356
 357                 close(pipefd[0]);
 358
 359                 DEBUG("Successfully synced with child process");
 360                 _exit(EXIT_SUCCESS);
 361         }
 362
 363         if (setsid() < 0) {
 364                 SYSERROR("Failed to setsid()");
 365                 _exit(EXIT_FAILURE);
 366         }
 367
 368         lxc_check_inherited(NULL, true, &pipefd[1], 1);
 369         if (null_stdfds() < 0) {
 370                 SYSERROR("Failed to dup2() standard file descriptors to /dev/null");
 371                 _exit(EXIT_FAILURE);
 372         }
 373
 374         close(pipefd[0]);
 375
 376         ret = snprintf(pipefd_str, sizeof(pipefd_str), "%d", pipefd[1]);
 377         if (ret < 0 || ret >= sizeof(pipefd_str)) {
 378                 ERROR("Failed to create pid argument to pass to monitord");
 379                 _exit(EXIT_FAILURE);
 380         }
 381
 382         DEBUG("Using pipe file descriptor %d for monitord", pipefd[1]);
 383
 384         execvp(args[0], args);
 385         SYSERROR("Failed to exec lxc-monitord");
 386
 387         _exit(EXIT_FAILURE);
 388 }