mdmon.c

   1 /*
   2  * mdmon - monitor external metadata arrays
   3  *
   4  * Copyright (C) 2007-2008 Neil Brown <neilb@suse.de>
   5  * Copyright (C) 2007-2008 Intel Corporation
   6  *
   7  * This program is free software; you can redistribute it and/or modify it
   8  * under the terms and conditions of the GNU General Public License,
   9  * version 2, as published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope it will be useful, but WITHOUT
  12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  14  * more details.
  15  *
  16  * You should have received a copy of the GNU General Public License along with
  17  * this program; if not, write to the Free Software Foundation, Inc.,
  18  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  19  */
  20
  21 /*
  22  * md array manager.
  23  * When md arrays have user-space managed metadata, this is the program
  24  * that does the managing.
  25  *
  26  * Given one argument: the name of the array (e.g. /dev/md0) that is
  27  * the container.
  28  * We fork off a helper that runs high priority and mlocked.  It responds to
  29  * device failures and other events that might stop writeout, or that are
  30  * trivial to deal with.
  31  * The main thread then watches for new arrays being created in the container
  32  * and starts monitoring them too ... along with a few other tasks.
  33  *
  34  * The main thread communicates with the priority thread by writing over
  35  * a pipe.
  36  * Separate programs can communicate with the main thread via Unix-domain
  37  * socket.
  38  * The two threads share address space and open file table.
  39  *
  40  */
  41
  42 #ifndef _GNU_SOURCE
  43 #define _GNU_SOURCE
  44 #endif
  45
  46 #include        <unistd.h>
  47 #include        <stdlib.h>
  48 #include        <sys/types.h>
  49 #include        <sys/stat.h>
  50 #include        <sys/socket.h>
  51 #include        <sys/un.h>
  52 #include        <sys/mman.h>
  53 #include        <sys/syscall.h>
  54 #include        <sys/wait.h>
  55 #include        <stdio.h>
  56 #include        <errno.h>
  57 #include        <string.h>
  58 #include        <fcntl.h>
  59 #include        <signal.h>
  60 #include        <dirent.h>
  61
  62 #include        <sched.h>
  63
  64 #include        "mdadm.h"
  65 #include        "mdmon.h"
  66
  67 struct active_array *discard_this;
  68 struct active_array *pending_discard;
  69
  70 int mon_tid, mgr_tid;
  71
  72 int sigterm;
  73
  74 int run_child(void *v)
  75 {
  76         struct supertype *c = v;
  77
  78         do_monitor(c);
  79         return 0;
  80 }
  81
  82 int clone_monitor(struct supertype *container)
  83 {
  84         static char stack[4096];
  85
  86         mon_tid = clone(run_child, stack+4096-64,
  87                    CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD,
  88                    container);
  89
  90         mgr_tid = syscall(SYS_gettid);
  91
  92         return mon_tid;
  93 }
  94
  95
  96 int make_pidfile(char *devname, int o_excl)
  97 {
  98         char path[100];
  99         char pid[10];
 100         int fd;
 101         int n;
 102
 103         if (sigterm)
 104                 return -1;
 105
 106         sprintf(path, "/var/run/mdadm/%s.pid", devname);
 107
 108         fd = open(path, O_RDWR|O_CREAT|o_excl, 0600);
 109         if (fd < 0)
 110                 return -errno;
 111         sprintf(pid, "%d\n", getpid());
 112         n = write(fd, pid, strlen(pid));
 113         close(fd);
 114         if (n < 0)
 115                 return -errno;
 116         return 0;
 117 }
 118
 119 int is_container_member(struct mdstat_ent *mdstat, char *container)
 120 {
 121         if (mdstat->metadata_version == NULL ||
 122             strncmp(mdstat->metadata_version, "external:", 9) != 0 ||
 123             !is_subarray(mdstat->metadata_version+9) ||
 124             strncmp(mdstat->metadata_version+10, container, strlen(container)) != 0 ||
 125             mdstat->metadata_version[10+strlen(container)] != '/')
 126                 return 0;
 127
 128         return 1;
 129 }
 130
 131 void remove_pidfile(char *devname);
 132 static void try_kill_monitor(char *devname)
 133 {
 134         char buf[100];
 135         int fd;
 136         pid_t pid;
 137         struct mdstat_ent *mdstat;
 138
 139         sprintf(buf, "/var/run/mdadm/%s.pid", devname);
 140         fd = open(buf, O_RDONLY);
 141         if (fd < 0)
 142                 return;
 143
 144         if (read(fd, buf, sizeof(buf)) < 0) {
 145                 close(fd);
 146                 return;
 147         }
 148
 149         close(fd);
 150         pid = strtoul(buf, NULL, 10);
 151
 152         /* first rule of survival... don't off yourself */
 153         if (pid == getpid())
 154                 return;
 155
 156         /* kill this process if it is mdmon */
 157         sprintf(buf, "/proc/%lu/cmdline", (unsigned long) pid);
 158         fd = open(buf, O_RDONLY);
 159         if (fd < 0)
 160                 return;
 161
 162         if (read(fd, buf, sizeof(buf)) < 0) {
 163                 close(fd);
 164                 return;
 165         }
 166
 167         if (!strstr(buf, "mdmon"))
 168                 return;
 169
 170         kill(pid, SIGTERM);
 171
 172         mdstat = mdstat_read(0, 0);
 173         for ( ; mdstat; mdstat = mdstat->next)
 174                 if (is_container_member(mdstat, devname)) {
 175                         sprintf(buf, "/dev/%s", mdstat->dev);
 176                         WaitClean(buf, 0);
 177                 }
 178         free_mdstat(mdstat);
 179         remove_pidfile(devname);
 180 }
 181
 182 void remove_pidfile(char *devname)
 183 {
 184         char buf[100];
 185
 186         if (sigterm)
 187                 return;
 188
 189         sprintf(buf, "/var/run/mdadm/%s.pid", devname);
 190         unlink(buf);
 191         sprintf(buf, "/var/run/mdadm/%s.sock", devname);
 192         unlink(buf);
 193 }
 194
 195 int make_control_sock(char *devname)
 196 {
 197         char path[100];
 198         int sfd;
 199         long fl;
 200         struct sockaddr_un addr;
 201
 202         if (sigterm)
 203                 return -1;
 204
 205         sprintf(path, "/var/run/mdadm/%s.sock", devname);
 206         unlink(path);
 207         sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
 208         if (sfd < 0)
 209                 return -1;
 210
 211         addr.sun_family = PF_LOCAL;
 212         strcpy(addr.sun_path, path);
 213         if (bind(sfd, &addr, sizeof(addr)) < 0) {
 214                 close(sfd);
 215                 return -1;
 216         }
 217         listen(sfd, 10);
 218         fl = fcntl(sfd, F_GETFL, 0);
 219         fl |= O_NONBLOCK;
 220         fcntl(sfd, F_SETFL, fl);
 221         return sfd;
 222 }
 223
 224 int socket_hup_requested;
 225 static void hup(int sig)
 226 {
 227         socket_hup_requested = 1;
 228 }
 229
 230 static void term(int sig)
 231 {
 232         sigterm = 1;
 233 }
 234
 235 static void wake_me(int sig)
 236 {
 237
 238 }
 239
 240 /* if we are debugging and starting mdmon by hand then don't fork */
 241 static int do_fork(void)
 242 {
 243         #ifdef DEBUG
 244         if (env_no_mdmon())
 245                 return 0;
 246         #endif
 247
 248         return 1;
 249 }
 250
 251 void usage(void)
 252 {
 253         fprintf(stderr, "Usage: mdmon [--switch-root dir] /device/name/for/container\n");
 254         exit(2);
 255 }
 256
 257 int main(int argc, char *argv[])
 258 {
 259         int mdfd;
 260         struct mdinfo *mdi, *di;
 261         struct supertype *container;
 262         sigset_t set;
 263         struct sigaction act;
 264         int pfd[2];
 265         int status;
 266         int ignore;
 267         char *container_name = NULL;
 268         char *switchroot = NULL;
 269
 270         switch (argc) {
 271         case 2:
 272                 container_name = argv[1];
 273                 break;
 274         case 4:
 275                 if (strcmp(argv[1], "--switch-root") != 0) {
 276                         fprintf(stderr, "mdmon: unknown argument %s\n", argv[1]);
 277                         usage();
 278                 }
 279                 switchroot = argv[2];
 280                 container_name = argv[3];
 281                 break;
 282         default:
 283                 usage();
 284         }
 285
 286         mdfd = open(container_name, O_RDWR);
 287         if (mdfd < 0) {
 288                 fprintf(stderr, "mdmon: %s: %s\n", container_name,
 289                         strerror(errno));
 290                 exit(1);
 291         }
 292         if (md_get_version(mdfd) < 0) {
 293                 fprintf(stderr, "mdmon: %s: Not an md device\n",
 294                         container_name);
 295                 exit(1);
 296         }
 297
 298         /* Fork, and have the child tell us when they are ready */
 299         if (do_fork()) {
 300                 if (pipe(pfd) != 0) {
 301                         fprintf(stderr, "mdmon: failed to create pipe\n");
 302                         exit(1);
 303                 }
 304                 switch(fork()) {
 305                 case -1:
 306                         fprintf(stderr, "mdmon: failed to fork: %s\n",
 307                                 strerror(errno));
 308                         exit(1);
 309                 case 0: /* child */
 310                         close(pfd[0]);
 311                         break;
 312                 default: /* parent */
 313                         close(pfd[1]);
 314                         if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) {
 315                                 wait(&status);
 316                                 status = WEXITSTATUS(status);
 317                         }
 318                         exit(status);
 319                 }
 320         } else
 321                 pfd[0] = pfd[1] = -1;
 322
 323         container = malloc(sizeof(*container));
 324         container->devnum = fd2devnum(mdfd);
 325         container->devname = devnum2devname(container->devnum);
 326         container->device_name = container_name;
 327         container->arrays = NULL;
 328
 329         if (!container->devname) {
 330                 fprintf(stderr, "mdmon: failed to allocate container name string\n");
 331                 exit(3);
 332         }
 333
 334         mdi = sysfs_read(mdfd, container->devnum,
 335                          GET_VERSION|GET_LEVEL|GET_DEVS);
 336
 337         if (!mdi) {
 338                 fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
 339                         container->devname);
 340                 exit(3);
 341         }
 342         if (mdi->array.level != UnSet) {
 343                 fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
 344                         container_name);
 345                 exit(3);
 346         }
 347         if (mdi->array.major_version != -1 ||
 348             mdi->array.minor_version != -2) {
 349                 fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
 350                         container_name);
 351                 exit(3);
 352         }
 353
 354         container->ss = find_metadata_methods(mdi->text_version);
 355         if (container->ss == NULL) {
 356                 fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
 357                         container_name, mdi->text_version);
 358                 exit(3);
 359         }
 360
 361         container->devs = NULL;
 362         for (di = mdi->devs; di; di = di->next) {
 363                 struct mdinfo *cd = malloc(sizeof(*cd));
 364                 *cd = *di;
 365                 cd->next = container->devs;
 366                 container->devs = cd;
 367         }
 368         sysfs_free(mdi);
 369
 370         /* SIGUSR is sent between parent and child.  So both block it
 371          * and enable it only with pselect.
 372          */
 373         sigemptyset(&set);
 374         sigaddset(&set, SIGUSR1);
 375         sigaddset(&set, SIGHUP);
 376         sigaddset(&set, SIGALRM);
 377         sigaddset(&set, SIGTERM);
 378         sigprocmask(SIG_BLOCK, &set, NULL);
 379         act.sa_handler = wake_me;
 380         act.sa_flags = 0;
 381         sigaction(SIGUSR1, &act, NULL);
 382         sigaction(SIGALRM, &act, NULL);
 383         act.sa_handler = hup;
 384         sigaction(SIGHUP, &act, NULL);
 385         act.sa_handler = term;
 386         sigaction(SIGTERM, &act, NULL);
 387         act.sa_handler = SIG_IGN;
 388         sigaction(SIGPIPE, &act, NULL);
 389
 390         if (switchroot) {
 391                 /* we assume we assume that /sys /proc /dev are available in
 392                  * the new root (see nash:setuproot)
 393                  *
 394                  * kill any monitors in the current namespace and change
 395                  * to the new one
 396                  */
 397                 try_kill_monitor(container->devname);
 398                 if (chroot(switchroot) != 0) {
 399                         fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n",
 400                                 switchroot, strerror(errno));
 401                         exit(4);
 402                 }
 403         }
 404
 405         /* If this fails, we hope it already exists
 406          * pid file lives in /var/run/mdadm/mdXX.pid
 407          */
 408         mkdir("/var", 0600);
 409         mkdir("/var/run", 0600);
 410         mkdir("/var/run/mdadm", 0600);
 411         ignore = chdir("/");
 412         if (make_pidfile(container->devname, O_EXCL) < 0) {
 413                 if (ping_monitor(container->devname) == 0) {
 414                         fprintf(stderr, "mdmon: %s already managed\n",
 415                                 container->devname);
 416                         exit(3);
 417                 } else {
 418                         int err;
 419
 420                         /* cleanup the old monitor, this one is taking over */
 421                         try_kill_monitor(container->devname);
 422                         err = make_pidfile(container->devname, 0);
 423                         if (err < 0) {
 424                                 fprintf(stderr, "mdmon: %s Cannot create pidfile\n",
 425                                         container->devname);
 426                                 if (err == -EROFS) {
 427                                         /* FIXME implement a mechanism to
 428                                          * prevent duplicate monitor instances
 429                                          */
 430                                         fprintf(stderr,
 431                                                 "mdmon: continuing on read-only file system\n");
 432                                 } else
 433                                         exit(3);
 434                         }
 435                 }
 436         }
 437         container->sock = make_control_sock(container->devname);
 438
 439         if (container->ss->load_super(container, mdfd, container_name)) {
 440                 fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
 441                         container_name);
 442                 exit(3);
 443         }
 444
 445         /* Ok, this is close enough.  We can say goodbye to our parent now.
 446          */
 447         status = 0;
 448         if (write(pfd[1], &status, sizeof(status)) < 0)
 449                 fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
 450                         getppid());
 451         close(pfd[1]);
 452
 453         setsid();
 454         close(0);
 455         open("/dev/null", O_RDWR);
 456         close(1);
 457         ignore = dup(0);
 458 #ifndef DEBUG
 459         close(2);
 460         ignore = dup(0);
 461 #endif
 462
 463         mlockall(MCL_FUTURE);
 464
 465         if (clone_monitor(container) < 0) {
 466                 fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
 467                         strerror(errno));
 468                 exit(2);
 469         }
 470
 471         do_manager(container);
 472
 473         exit(0);
 474 }