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 #include <errno.h>
  26 #include <fcntl.h>
  27 #include <inttypes.h>
  28 #include <poll.h>
  29 #include <stddef.h>
  30 #include <stdint.h>
  31 #include <stdio.h>
  32 #include <stdlib.h>
  33 #include <string.h>
  34 #include <unistd.h>
  35 #include <net/if.h>
  36 #include <netinet/in.h>
  37 #include <sys/param.h>
  38 #include <sys/socket.h>
  39 #include <sys/stat.h>
  40 #include <sys/types.h>
  41 #include <sys/wait.h>
  42
  43 #include "config.h"
  44 #include "af_unix.h"
  45 #include "error.h"
  46 #include "log.h"
  47 #include "lxclock.h"
  48 #include "monitor.h"
  49 #include "state.h"
  50 #include "utils.h"
  51
  52 lxc_log_define(lxc_monitor, lxc);
  53
  54 /* routines used by monitor publishers (containers) */
  55 int lxc_monitor_fifo_name(const char *lxcpath, char *fifo_path, size_t fifo_path_sz,
  56                           int do_mkdirp)
  57 {
  58         int ret;
  59         char *rundir;
  60
  61         rundir = get_rundir();
  62         if (!rundir)
  63                 return -1;
  64
  65         if (do_mkdirp) {
  66                 ret = snprintf(fifo_path, fifo_path_sz, "%s/lxc/%s", rundir, lxcpath);
  67                 if (ret < 0 || (size_t)ret >= fifo_path_sz) {
  68                         ERROR("rundir/lxcpath (%s/%s) too long for monitor fifo.", rundir, lxcpath);
  69                         free(rundir);
  70                         return -1;
  71                 }
  72                 ret = mkdir_p(fifo_path, 0755);
  73                 if (ret < 0) {
  74                         ERROR("Unable to create monitor fifo directory %s.", fifo_path);
  75                         free(rundir);
  76                         return ret;
  77                 }
  78         }
  79         ret = snprintf(fifo_path, fifo_path_sz, "%s/lxc/%s/monitor-fifo", rundir, lxcpath);
  80         if (ret < 0 || (size_t)ret >= fifo_path_sz) {
  81                 ERROR("rundir/lxcpath (%s/%s) too long for monitor fifo.", rundir, lxcpath);
  82                 free(rundir);
  83                 return -1;
  84         }
  85         free(rundir);
  86         return 0;
  87 }
  88
  89 static void lxc_monitor_fifo_send(struct lxc_msg *msg, const char *lxcpath)
  90 {
  91         int fd,ret;
  92         char fifo_path[PATH_MAX];
  93
  94         BUILD_BUG_ON(sizeof(*msg) > PIPE_BUF); /* write not guaranteed atomic */
  95
  96         ret = lxc_monitor_fifo_name(lxcpath, fifo_path, sizeof(fifo_path), 0);
  97         if (ret < 0)
  98                 return;
  99
 100         /* Open the fifo nonblock in case the monitor is dead, we don't want the
 101          * open to wait for a reader since it may never come.
 102          */
 103         fd = open(fifo_path, O_WRONLY | O_NONBLOCK);
 104         if (fd < 0) {
 105                 /* It is normal for this open() to fail with ENXIO when there is
 106                  * no monitor running, so we don't log it.
 107                  */
 108                 if (errno == ENXIO || errno == ENOENT)
 109                         return;
 110
 111                 WARN("%s - Failed to open fifo to send message", strerror(errno));
 112                 return;
 113         }
 114
 115         if (fcntl(fd, F_SETFL, O_WRONLY) < 0) {
 116                 close(fd);
 117                 return;
 118         }
 119
 120         ret = write(fd, msg, sizeof(*msg));
 121         if (ret != sizeof(*msg)) {
 122                 close(fd);
 123                 SYSERROR("Failed to write to monitor fifo \"%s\".", fifo_path);
 124                 return;
 125         }
 126
 127         close(fd);
 128 }
 129
 130 void lxc_monitor_send_state(const char *name, lxc_state_t state,
 131                             const char *lxcpath)
 132 {
 133         struct lxc_msg msg = {.type = lxc_msg_state, .value = state};
 134         strncpy(msg.name, name, sizeof(msg.name));
 135         msg.name[sizeof(msg.name) - 1] = 0;
 136
 137         lxc_monitor_fifo_send(&msg, lxcpath);
 138 }
 139
 140 void lxc_monitor_send_exit_code(const char *name, int exit_code,
 141                                 const char *lxcpath)
 142 {
 143         struct lxc_msg msg = {.type = lxc_msg_exit_code, .value = exit_code};
 144         strncpy(msg.name, name, sizeof(msg.name));
 145         msg.name[sizeof(msg.name) - 1] = 0;
 146
 147         lxc_monitor_fifo_send(&msg, lxcpath);
 148 }
 149
 150 /* routines used by monitor subscribers (lxc-monitor) */
 151 int lxc_monitor_close(int fd)
 152 {
 153         return close(fd);
 154 }
 155
 156 /* Enforces \0-termination for the abstract unix socket. This is not required
 157  * but allows us to print it out.
 158  *
 159  * Older version of liblxc only allowed for 105 bytes to be used for the
 160  * abstract unix domain socket name because the code for our abstract unix
 161  * socket handling performed invalid checks. Since we \0-terminate we could now
 162  * have a maximum of 106 chars. But to not break backwards compatibility we keep
 163  * the limit at 105.
 164  */
 165 int lxc_monitor_sock_name(const char *lxcpath, struct sockaddr_un *addr) {
 166         size_t len;
 167         int ret;
 168         char *path;
 169         uint64_t hash;
 170
 171         /* addr.sun_path is only 108 bytes, so we hash the full name and
 172          * then append as much of the name as we can fit.
 173          */
 174         memset(addr, 0, sizeof(*addr));
 175         addr->sun_family = AF_UNIX;
 176
 177         /* strlen("lxc/") + strlen("/monitor-sock") + 1 = 18 */
 178         len = strlen(lxcpath) + 18;
 179         path = alloca(len);
 180         ret = snprintf(path, len, "lxc/%s/monitor-sock", lxcpath);
 181         if (ret < 0 || (size_t)ret >= len) {
 182                 ERROR("failed to create name for monitor socket");
 183                 return -1;
 184         }
 185
 186         /* Note: snprintf() will \0-terminate addr->sun_path on the 106th byte
 187          * and so the abstract socket name has 105 "meaningful" characters. This
 188          * is absolutely intentional. For further info read the comment for this
 189          * function above!
 190          */
 191         len = sizeof(addr->sun_path) - 1;
 192         hash = fnv_64a_buf(path, ret, FNV1A_64_INIT);
 193         ret = snprintf(addr->sun_path, len, "@lxc/%016" PRIx64 "/%s", hash, lxcpath);
 194         if (ret < 0) {
 195                 ERROR("failed to create hashed name for monitor socket");
 196                 return -1;
 197         }
 198
 199         /* replace @ with \0 */
 200         addr->sun_path[0] = '\0';
 201         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);
 202
 203         return 0;
 204 }
 205
 206 int lxc_monitor_open(const char *lxcpath)
 207 {
 208         struct sockaddr_un addr;
 209         int fd;
 210         size_t retry;
 211         size_t len;
 212         int backoff_ms[] = {10, 50, 100};
 213
 214         if (lxc_monitor_sock_name(lxcpath, &addr) < 0)
 215                 return -1;
 216
 217         fd = socket(PF_UNIX, SOCK_STREAM, 0);
 218         if (fd < 0) {
 219                 ERROR("Failed to create socket: %s.", strerror(errno));
 220                 return -1;
 221         }
 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                 ERROR("name of monitor socket too long (%zu bytes): %s", len, strerror(errno));
 228                 close(fd);
 229                 return -1;
 230         }
 231
 232         for (retry = 0; retry < sizeof(backoff_ms) / sizeof(backoff_ms[0]); retry++) {
 233                 fd = lxc_abstract_unix_connect(addr.sun_path);
 234                 if (fd != -1 || errno != ECONNREFUSED)
 235                         break;
 236                 ERROR("Failed to connect to monitor socket. Retrying in %d ms: %s", backoff_ms[retry], strerror(errno));
 237                 usleep(backoff_ms[retry] * 1000);
 238         }
 239
 240         if (fd < 0) {
 241                 ERROR("Failed to connect to monitor socket: %s.", strerror(errno));
 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?: %s.", strerror(errno));
 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[LXC_NUMSTRLEN64];
 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                 if (waitpid(pid1, NULL, 0) != pid1)
 325                         return -1;
 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 (read(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, LXC_NUMSTRLEN64, "%d", pipefd[1]);
 375         if (ret < 0 || ret >= LXC_NUMSTRLEN64) {
 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 }