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 <stdio.h>
  26 #include <errno.h>
  27 #include <unistd.h>
  28 #include <string.h>
  29 #include <stdlib.h>
  30 #include <stddef.h>
  31 #include <fcntl.h>
  32 #include <inttypes.h>
  33 #include <stdint.h>
  34 #include <sys/types.h>
  35 #include <sys/stat.h>
  36 #include <sys/param.h>
  37 #include <sys/socket.h>
  38 #include <sys/wait.h>
  39 #include <netinet/in.h>
  40 #include <net/if.h>
  41
  42 #include "error.h"
  43 #include "af_unix.h"
  44 #include "log.h"
  45 #include "lxclock.h"
  46 #include "state.h"
  47 #include "monitor.h"
  48 #include "utils.h"
  49
  50 lxc_log_define(lxc_monitor, lxc);
  51
  52 /* routines used by monitor publishers (containers) */
  53 int lxc_monitor_fifo_name(const char *lxcpath, char *fifo_path, size_t fifo_path_sz,
  54                           int do_mkdirp)
  55 {
  56         int ret;
  57         char *rundir;
  58
  59         rundir = get_rundir();
  60         if (!rundir)
  61                 return -1;
  62
  63         if (do_mkdirp) {
  64                 ret = snprintf(fifo_path, fifo_path_sz, "%s/lxc/%s", rundir, lxcpath);
  65                 if (ret < 0 || ret >= fifo_path_sz) {
  66                         ERROR("rundir/lxcpath (%s/%s) too long for monitor fifo", rundir, lxcpath);
  67                         free(rundir);
  68                         return -1;
  69                 }
  70                 ret = mkdir_p(fifo_path, 0755);
  71                 if (ret < 0) {
  72                         ERROR("unable to create monitor fifo dir %s", fifo_path);
  73                         free(rundir);
  74                         return ret;
  75                 }
  76         }
  77         ret = snprintf(fifo_path, fifo_path_sz, "%s/lxc/%s/monitor-fifo", rundir, lxcpath);
  78         if (ret < 0 || ret >= fifo_path_sz) {
  79                 ERROR("rundir/lxcpath (%s/%s) too long for monitor fifo", rundir, lxcpath);
  80                 free(rundir);
  81                 return -1;
  82         }
  83         free(rundir);
  84         return 0;
  85 }
  86
  87 static void lxc_monitor_fifo_send(struct lxc_msg *msg, const char *lxcpath)
  88 {
  89         int fd,ret;
  90         char fifo_path[PATH_MAX];
  91
  92         BUILD_BUG_ON(sizeof(*msg) > PIPE_BUF); /* write not guaranteed atomic */
  93
  94         ret = lxc_monitor_fifo_name(lxcpath, fifo_path, sizeof(fifo_path), 0);
  95         if (ret < 0)
  96                 return;
  97
  98         /* open the fifo nonblock in case the monitor is dead, we don't want
  99          * the open to wait for a reader since it may never come.
 100          */
 101         fd = open(fifo_path, O_WRONLY|O_NONBLOCK);
 102         if (fd < 0) {
 103                 /* it is normal for this open to fail ENXIO when there is no
 104                  * monitor running, so we don't log it
 105                  */
 106                 return;
 107         }
 108
 109         if (fcntl(fd, F_SETFL, O_WRONLY) < 0) {
 110                 close(fd);
 111                 return;
 112         }
 113
 114         ret = write(fd, msg, sizeof(*msg));
 115         if (ret != sizeof(*msg)) {
 116                 close(fd);
 117                 SYSERROR("failed to write monitor fifo %s", fifo_path);
 118                 return;
 119         }
 120
 121         close(fd);
 122 }
 123
 124 void lxc_monitor_send_state(const char *name, lxc_state_t state, const char *lxcpath)
 125 {
 126         struct lxc_msg msg = { .type = lxc_msg_state,
 127                                .value = state };
 128         strncpy(msg.name, name, sizeof(msg.name));
 129         msg.name[sizeof(msg.name) - 1] = 0;
 130
 131         lxc_monitor_fifo_send(&msg, lxcpath);
 132 }
 133
 134 void lxc_monitor_send_exit_code(const char *name, int exit_code, const char *lxcpath)
 135 {
 136         struct lxc_msg msg = { .type = lxc_msg_exit_code,
 137                                .value = exit_code };
 138         strncpy(msg.name, name, sizeof(msg.name));
 139         msg.name[sizeof(msg.name) - 1] = 0;
 140
 141         lxc_monitor_fifo_send(&msg, lxcpath);
 142 }
 143
 144
 145 /* routines used by monitor subscribers (lxc-monitor) */
 146 int lxc_monitor_close(int fd)
 147 {
 148         return close(fd);
 149 }
 150
 151 int lxc_monitor_sock_name(const char *lxcpath, struct sockaddr_un *addr) {
 152         size_t len;
 153         int ret;
 154         char *sockname = &addr->sun_path[1];
 155         char *path;
 156         uint64_t hash;
 157
 158         /* addr.sun_path is only 108 bytes, so we hash the full name and
 159          * then append as much of the name as we can fit.
 160          */
 161         memset(addr, 0, sizeof(*addr));
 162         addr->sun_family = AF_UNIX;
 163         len = strlen(lxcpath) + 18;
 164         path = alloca(len);
 165         ret = snprintf(path, len, "lxc/%s/monitor-sock", lxcpath);
 166         if (ret < 0 || ret >= len) {
 167                 ERROR("memory error creating monitor path");
 168                 return -1;
 169         }
 170
 171         len = sizeof(addr->sun_path) - 1;
 172         hash = fnv_64a_buf(path, ret, FNV1A_64_INIT);
 173         ret = snprintf(sockname, len, "lxc/%016" PRIx64 "/%s", hash, lxcpath);
 174         if (ret < 0)
 175                 return -1;
 176         sockname[sizeof(addr->sun_path)-3] = '\0';
 177         INFO("using monitor sock name %s", sockname);
 178         return 0;
 179 }
 180
 181 int lxc_monitor_open(const char *lxcpath)
 182 {
 183         struct sockaddr_un addr;
 184         int fd,ret;
 185         int retry,backoff_ms[] = {10, 50, 100};
 186         size_t len;
 187
 188         if (lxc_monitor_sock_name(lxcpath, &addr) < 0)
 189                 return -1;
 190
 191         fd = socket(PF_UNIX, SOCK_STREAM, 0);
 192         if (fd < 0) {
 193                 ERROR("socket : %s", strerror(errno));
 194                 return -1;
 195         }
 196
 197         len = strlen(&addr.sun_path[1]) + 1;
 198         if (len >= sizeof(addr.sun_path) - 1) {
 199                 ret = -1;
 200                 errno = ENAMETOOLONG;
 201                 goto err1;
 202         }
 203
 204         for (retry = 0; retry < sizeof(backoff_ms)/sizeof(backoff_ms[0]); retry++) {
 205                 ret = connect(fd, (struct sockaddr *)&addr, offsetof(struct sockaddr_un, sun_path) + len);
 206                 if (ret == 0 || errno != ECONNREFUSED)
 207                         break;
 208                 ERROR("connect : backing off %d", backoff_ms[retry]);
 209                 usleep(backoff_ms[retry] * 1000);
 210         }
 211
 212         if (ret < 0) {
 213                 ERROR("connect : %s", strerror(errno));
 214                 goto err1;
 215         }
 216         return fd;
 217 err1:
 218         close(fd);
 219         return ret;
 220 }
 221
 222 int lxc_monitor_read_fdset(fd_set *rfds, int nfds, struct lxc_msg *msg,
 223                            int timeout)
 224 {
 225         struct timeval tval,*tv = NULL;
 226         int ret,i;
 227
 228         if (timeout != -1) {
 229                 tv = &tval;
 230                 tv->tv_sec = timeout;
 231                 tv->tv_usec = 0;
 232         }
 233
 234         ret = select(nfds, rfds, NULL, NULL, tv);
 235         if (ret == -1)
 236                 return -1;
 237         else if (ret == 0)
 238                 return -2;  // timed out
 239
 240         /* only read from the first ready fd, the others will remain ready
 241          * for when this routine is called again
 242          */
 243         for (i = 0; i < nfds; i++) {
 244                 if (FD_ISSET(i, rfds)) {
 245                         ret = recv(i, msg, sizeof(*msg), 0);
 246                         if (ret <= 0) {
 247                                 SYSERROR("client failed to recv (monitord died?) %s",
 248                                          strerror(errno));
 249                                 return -1;
 250                         }
 251                         return ret;
 252                 }
 253         }
 254         SYSERROR("no ready fd found?");
 255         return -1;
 256 }
 257
 258 int lxc_monitor_read_timeout(int fd, struct lxc_msg *msg, int timeout)
 259 {
 260         fd_set rfds;
 261
 262         FD_ZERO(&rfds);
 263         FD_SET(fd, &rfds);
 264
 265         return lxc_monitor_read_fdset(&rfds, fd+1, msg, timeout);
 266 }
 267
 268 int lxc_monitor_read(int fd, struct lxc_msg *msg)
 269 {
 270         return lxc_monitor_read_timeout(fd, msg, -1);
 271 }
 272
 273
 274 #define LXC_MONITORD_PATH LIBEXECDIR "/lxc/lxc-monitord"
 275
 276 /* used to spawn a monitord either on startup of a daemon container, or when
 277  * lxc-monitor starts
 278  */
 279 int lxc_monitord_spawn(const char *lxcpath)
 280 {
 281         pid_t pid1,pid2;
 282         int pipefd[2];
 283         char pipefd_str[11];
 284
 285         char * const args[] = {
 286                 LXC_MONITORD_PATH,
 287                 (char *)lxcpath,
 288                 pipefd_str,
 289                 NULL,
 290         };
 291
 292         /* double fork to avoid zombies when monitord exits */
 293         pid1 = fork();
 294         if (pid1 < 0) {
 295                 SYSERROR("failed to fork");
 296                 return -1;
 297         }
 298
 299         if (pid1) {
 300                 if (waitpid(pid1, NULL, 0) != pid1)
 301                         return -1;
 302                 return 0;
 303         }
 304
 305         if (pipe(pipefd) < 0) {
 306                 SYSERROR("failed to create pipe");
 307                 exit(EXIT_FAILURE);
 308         }
 309
 310         pid2 = fork();
 311         if (pid2 < 0) {
 312                 SYSERROR("failed to fork");
 313                 exit(EXIT_FAILURE);
 314         }
 315         if (pid2) {
 316                 char c;
 317                 /* wait for daemon to create socket */
 318                 close(pipefd[1]);
 319                 /* sync with child, we're ignoring the return from read
 320                  * because regardless if it works or not, either way we've
 321                  * synced with the child process. the if-empty-statement
 322                  * construct is to quiet the warn-unused-result warning.
 323                  */
 324                 if (read(pipefd[0], &c, 1))
 325                         ;
 326                 close(pipefd[0]);
 327                 exit(EXIT_SUCCESS);
 328         }
 329
 330         if (setsid() < 0) {
 331                 SYSERROR("failed to setsid");
 332                 exit(EXIT_FAILURE);
 333         }
 334         lxc_check_inherited(NULL, pipefd[1]);
 335         close(0);
 336         close(1);
 337         close(2);
 338         open("/dev/null", O_RDONLY);
 339         open("/dev/null", O_RDWR);
 340         open("/dev/null", O_RDWR);
 341         close(pipefd[0]);
 342         sprintf(pipefd_str, "%d", pipefd[1]);
 343         execvp(args[0], args);
 344         exit(EXIT_FAILURE);
 345 }