mdmon.c

   1 /*
   2  * mdmon - monitor external metadata arrays
   3  *
   4  * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de>
   5  * Copyright (C) 2007-2009 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 #ifdef __ia64__
  83 int __clone2(int (*fn)(void *),
  84             void *child_stack_base, size_t stack_size,
  85             int flags, void *arg, ...
  86          /* pid_t *pid, struct user_desc *tls, pid_t *ctid */ );
  87 #endif
  88  int clone_monitor(struct supertype *container)
  89 {
  90         static char stack[4096];
  91
  92 #ifdef __ia64__
  93         mon_tid = __clone2(run_child, stack, sizeof(stack),
  94                    CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD,
  95                    container);
  96 #else
  97         mon_tid = clone(run_child, stack+4096-64,
  98                    CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD,
  99                    container);
 100 #endif
 101
 102         mgr_tid = syscall(SYS_gettid);
 103
 104         return mon_tid;
 105 }
 106
 107 static struct superswitch *find_metadata_methods(char *vers)
 108 {
 109         if (strcmp(vers, "ddf") == 0)
 110                 return &super_ddf;
 111         if (strcmp(vers, "imsm") == 0)
 112                 return &super_imsm;
 113         return NULL;
 114 }
 115
 116 int make_pidfile(char *devname, int o_excl)
 117 {
 118         char path[100];
 119         char pid[10];
 120         int fd;
 121         int n;
 122
 123         if (sigterm)
 124                 return -1;
 125
 126         sprintf(path, "/var/run/mdadm/%s.pid", devname);
 127
 128         fd = open(path, O_RDWR|O_CREAT|o_excl, 0600);
 129         if (fd < 0)
 130                 return -errno;
 131         sprintf(pid, "%d\n", getpid());
 132         n = write(fd, pid, strlen(pid));
 133         close(fd);
 134         if (n < 0)
 135                 return -errno;
 136         return 0;
 137 }
 138
 139 int is_container_member(struct mdstat_ent *mdstat, char *container)
 140 {
 141         if (mdstat->metadata_version == NULL ||
 142             strncmp(mdstat->metadata_version, "external:", 9) != 0 ||
 143             !is_subarray(mdstat->metadata_version+9) ||
 144             strncmp(mdstat->metadata_version+10, container, strlen(container)) != 0 ||
 145             mdstat->metadata_version[10+strlen(container)] != '/')
 146                 return 0;
 147
 148         return 1;
 149 }
 150
 151 pid_t devname2mdmon(char *devname)
 152 {
 153         char buf[100];
 154         pid_t pid = -1;
 155         int fd;
 156
 157         sprintf(buf, "/var/run/mdadm/%s.pid", devname);
 158         fd = open(buf, O_RDONLY|O_NOATIME);
 159         if (fd < 0)
 160                 return -1;
 161
 162         if (read(fd, buf, sizeof(buf)) > 0)
 163                 sscanf(buf, "%d\n", &pid);
 164         close(fd);
 165
 166         return pid;
 167 }
 168
 169 static void try_kill_monitor(pid_t pid, char *devname, int sock)
 170 {
 171         char buf[100];
 172         int fd;
 173         int n;
 174         long fl;
 175
 176         /* first rule of survival... don't off yourself */
 177         if (pid == getpid())
 178                 return;
 179
 180         /* kill this process if it is mdmon */
 181         sprintf(buf, "/proc/%lu/cmdline", (unsigned long) pid);
 182         fd = open(buf, O_RDONLY);
 183         if (fd < 0)
 184                 return;
 185
 186         n = read(fd, buf, sizeof(buf)-1);
 187         buf[sizeof(buf)-1] = 0;
 188         close(fd);
 189
 190         if (n < 0 || !strstr(buf, "mdmon"))
 191                 return;
 192
 193         kill(pid, SIGTERM);
 194
 195         /* Wait for monitor to exit by reading from the socket, after
 196          * clearing the non-blocking flag */
 197         fl = fcntl(sock, F_GETFL, 0);
 198         fl &= ~O_NONBLOCK;
 199         fcntl(sock, F_SETFL, fl);
 200         read(sock, buf, 100);
 201 }
 202
 203 void remove_pidfile(char *devname)
 204 {
 205         char buf[100];
 206
 207         if (sigterm)
 208                 return;
 209
 210         sprintf(buf, "/var/run/mdadm/%s.pid", devname);
 211         unlink(buf);
 212         sprintf(buf, "/var/run/mdadm/%s.sock", devname);
 213         unlink(buf);
 214 }
 215
 216 int make_control_sock(char *devname)
 217 {
 218         char path[100];
 219         int sfd;
 220         long fl;
 221         struct sockaddr_un addr;
 222
 223         if (sigterm)
 224                 return -1;
 225
 226         sprintf(path, "/var/run/mdadm/%s.sock", devname);
 227         unlink(path);
 228         sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
 229         if (sfd < 0)
 230                 return -1;
 231
 232         addr.sun_family = PF_LOCAL;
 233         strcpy(addr.sun_path, path);
 234         if (bind(sfd, &addr, sizeof(addr)) < 0) {
 235                 close(sfd);
 236                 return -1;
 237         }
 238         listen(sfd, 10);
 239         fl = fcntl(sfd, F_GETFL, 0);
 240         fl |= O_NONBLOCK;
 241         fcntl(sfd, F_SETFL, fl);
 242         return sfd;
 243 }
 244
 245 int socket_hup_requested;
 246 static void hup(int sig)
 247 {
 248         socket_hup_requested = 1;
 249 }
 250
 251 static void term(int sig)
 252 {
 253         sigterm = 1;
 254 }
 255
 256 static void wake_me(int sig)
 257 {
 258
 259 }
 260
 261 /* if we are debugging and starting mdmon by hand then don't fork */
 262 static int do_fork(void)
 263 {
 264         #ifdef DEBUG
 265         if (check_env("MDADM_NO_MDMON"))
 266                 return 0;
 267         #endif
 268
 269         return 1;
 270 }
 271
 272 void usage(void)
 273 {
 274         fprintf(stderr, "Usage: mdmon /device/name/for/container [target_dir]\n");
 275         exit(2);
 276 }
 277
 278 static int mdmon(char *devname, int devnum, int must_fork, char *switchroot);
 279
 280 int main(int argc, char *argv[])
 281 {
 282         char *container_name = NULL;
 283         char *switchroot = NULL;
 284         int devnum;
 285         char *devname;
 286         int status = 0;
 287
 288         switch (argc) {
 289         case 3:
 290                 switchroot = argv[2];
 291         case 2:
 292                 container_name = argv[1];
 293                 break;
 294         default:
 295                 usage();
 296         }
 297
 298         if (strcmp(container_name, "/proc/mdstat") == 0) {
 299                 struct mdstat_ent *mdstat, *e;
 300
 301                 /* launch an mdmon instance for each container found */
 302                 mdstat = mdstat_read(0, 0);
 303                 for (e = mdstat; e; e = e->next) {
 304                         if (strncmp(e->metadata_version, "external:", 9) == 0 &&
 305                             !is_subarray(&e->metadata_version[9])) {
 306                                 devname = devnum2devname(e->devnum);
 307                                 /* update cmdline so this mdmon instance can be
 308                                  * distinguished from others in a call to ps(1)
 309                                  */
 310                                 if (strlen(devname) <= strlen(container_name)) {
 311                                         memset(container_name, 0, strlen(container_name));
 312                                         sprintf(container_name, "%s", devname);
 313                                 }
 314                                 status |= mdmon(devname, e->devnum, 1,
 315                                                 switchroot);
 316                         }
 317                 }
 318                 free_mdstat(mdstat);
 319
 320                 return status;
 321         } else if (strncmp(container_name, "md", 2) == 0) {
 322                 devnum = devname2devnum(container_name);
 323                 devname = devnum2devname(devnum);
 324                 if (strcmp(container_name, devname) != 0)
 325                         devname = NULL;
 326         } else {
 327                 struct stat st;
 328
 329                 devnum = NoMdDev;
 330                 if (stat(container_name, &st) == 0)
 331                         devnum = stat2devnum(&st);
 332                 if (devnum == NoMdDev)
 333                         devname = NULL;
 334                 else
 335                         devname = devnum2devname(devnum);
 336         }
 337
 338         if (!devname) {
 339                 fprintf(stderr, "mdmon: %s is not a valid md device name\n",
 340                         container_name);
 341                 exit(1);
 342         }
 343         return mdmon(devname, devnum, do_fork(), switchroot);
 344 }
 345
 346 static int mdmon(char *devname, int devnum, int must_fork, char *switchroot)
 347 {
 348         int mdfd;
 349         struct mdinfo *mdi, *di;
 350         struct supertype *container;
 351         sigset_t set;
 352         struct sigaction act;
 353         int pfd[2];
 354         int status;
 355         int ignore;
 356         pid_t victim = -1;
 357         int victim_sock = -1;
 358
 359         dprintf("starting mdmon for %s in %s\n",
 360                 devname, switchroot ? : "/");
 361
 362         /* switchroot is either a path name starting with '/', or a
 363          * pid of the original mdmon (we have already done the chroot).
 364          * In the latter case, stdin is a socket connected to the original
 365          * mdmon.
 366          */
 367
 368         /* try to spawn mdmon instances from the target file system */
 369         if (switchroot && switchroot[0] == '/' &&
 370             strcmp(switchroot, "/") != 0) {
 371                 pid_t pid;
 372                 char buf[20];
 373
 374                 switch (fork()) {
 375                 case 0:
 376                         victim = devname2mdmon(devname);
 377                         victim_sock = connect_monitor(devname);
 378                         if (chroot(switchroot) != 0) {
 379                                 fprintf(stderr, "mdmon: failed to chroot to '%s': %s\n",
 380                                         switchroot, strerror(errno));
 381                                 exit(4);
 382                         }
 383                         ignore = chdir("/");
 384                         sprintf(buf, "%d", victim);
 385                         if (victim_sock) {
 386                                 close(0);
 387                                 dup(victim_sock);
 388                                 close(victim_sock);
 389                         }
 390                         execl("/sbin/mdmon", "mdmon", devname, buf, NULL);
 391                         exit(1);
 392                 case -1:
 393                         return 1;
 394                 default:
 395                         pid = wait(&status);
 396                         if (pid > -1 && WIFEXITED(status) &&
 397                             WEXITSTATUS(status) == 0)
 398                                 return 0;
 399                         else
 400                                 return 1;
 401                 }
 402         }
 403
 404         mdfd = open_dev(devnum);
 405         if (mdfd < 0) {
 406                 fprintf(stderr, "mdmon: %s: %s\n", devname,
 407                         strerror(errno));
 408                 return 1;
 409         }
 410         if (md_get_version(mdfd) < 0) {
 411                 fprintf(stderr, "mdmon: %s: Not an md device\n",
 412                         devname);
 413                 return 1;
 414         }
 415
 416         /* Fork, and have the child tell us when they are ready */
 417         if (must_fork) {
 418                 if (pipe(pfd) != 0) {
 419                         fprintf(stderr, "mdmon: failed to create pipe\n");
 420                         return 1;
 421                 }
 422                 switch(fork()) {
 423                 case -1:
 424                         fprintf(stderr, "mdmon: failed to fork: %s\n",
 425                                 strerror(errno));
 426                         return 1;
 427                 case 0: /* child */
 428                         close(pfd[0]);
 429                         break;
 430                 default: /* parent */
 431                         close(pfd[1]);
 432                         if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) {
 433                                 wait(&status);
 434                                 status = WEXITSTATUS(status);
 435                         }
 436                         return status;
 437                 }
 438         } else
 439                 pfd[0] = pfd[1] = -1;
 440
 441         container = calloc(1, sizeof(*container));
 442         container->devnum = devnum;
 443         container->devname = devname;
 444         container->arrays = NULL;
 445         container->subarray[0] = 0;
 446         container->sock = -1;
 447
 448         if (!container->devname) {
 449                 fprintf(stderr, "mdmon: failed to allocate container name string\n");
 450                 exit(3);
 451         }
 452
 453         mdi = sysfs_read(mdfd, container->devnum,
 454                          GET_VERSION|GET_LEVEL|GET_DEVS|SKIP_GONE_DEVS);
 455
 456         if (!mdi) {
 457                 fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
 458                         container->devname);
 459                 exit(3);
 460         }
 461         if (mdi->array.level != UnSet) {
 462                 fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
 463                         devname);
 464                 exit(3);
 465         }
 466         if (mdi->array.major_version != -1 ||
 467             mdi->array.minor_version != -2) {
 468                 fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
 469                         devname);
 470                 exit(3);
 471         }
 472
 473         container->ss = find_metadata_methods(mdi->text_version);
 474         if (container->ss == NULL) {
 475                 fprintf(stderr, "mdmon: %s uses unknown metadata: %s\n",
 476                         devname, mdi->text_version);
 477                 exit(3);
 478         }
 479
 480         container->devs = NULL;
 481         for (di = mdi->devs; di; di = di->next) {
 482                 struct mdinfo *cd = malloc(sizeof(*cd));
 483                 *cd = *di;
 484                 cd->next = container->devs;
 485                 container->devs = cd;
 486         }
 487         sysfs_free(mdi);
 488
 489         /* SIGUSR is sent between parent and child.  So both block it
 490          * and enable it only with pselect.
 491          */
 492         sigemptyset(&set);
 493         sigaddset(&set, SIGUSR1);
 494         sigaddset(&set, SIGHUP);
 495         sigaddset(&set, SIGALRM);
 496         sigaddset(&set, SIGTERM);
 497         sigprocmask(SIG_BLOCK, &set, NULL);
 498         act.sa_handler = wake_me;
 499         act.sa_flags = 0;
 500         sigaction(SIGUSR1, &act, NULL);
 501         sigaction(SIGALRM, &act, NULL);
 502         act.sa_handler = hup;
 503         sigaction(SIGHUP, &act, NULL);
 504         act.sa_handler = term;
 505         sigaction(SIGTERM, &act, NULL);
 506         act.sa_handler = SIG_IGN;
 507         sigaction(SIGPIPE, &act, NULL);
 508
 509         if (switchroot) {
 510                 /* we assume we assume that /sys /proc /dev are available in
 511                  * the new root
 512                  */
 513                 if (switchroot[0] == '/') {
 514                         victim = devname2mdmon(container->devname);
 515                         victim_sock = connect_monitor(container->devname);
 516                 } else {
 517                         victim = atoi(switchroot);
 518                         victim_sock = 0;
 519                 }
 520         }
 521
 522         ignore = chdir("/");
 523         if (victim < 0) {
 524                 if (ping_monitor(container->devname) == 0) {
 525                         fprintf(stderr, "mdmon: %s already managed\n",
 526                                 container->devname);
 527                         exit(3);
 528                 }
 529                 /* if there is a pid file, kill whoever is there just in case */
 530                 victim = devname2mdmon(container->devname);
 531         }
 532         if (container->ss->load_super(container, mdfd, devname)) {
 533                 fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
 534                         devname);
 535                 exit(3);
 536         }
 537         close(mdfd);
 538
 539         /* Ok, this is close enough.  We can say goodbye to our parent now.
 540          */
 541         status = 0;
 542         if (write(pfd[1], &status, sizeof(status)) < 0)
 543                 fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
 544                         getppid());
 545         close(pfd[1]);
 546
 547         mlockall(MCL_CURRENT | MCL_FUTURE);
 548
 549         if (clone_monitor(container) < 0) {
 550                 fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
 551                         strerror(errno));
 552                 exit(2);
 553         }
 554
 555         if (victim > -1) {
 556                 try_kill_monitor(victim, container->devname, victim_sock);
 557                 close(victim_sock);
 558         }
 559
 560         setsid();
 561         close(0);
 562         open("/dev/null", O_RDWR);
 563         close(1);
 564         ignore = dup(0);
 565 #ifndef DEBUG
 566         close(2);
 567         ignore = dup(0);
 568 #endif
 569
 570         do_manager(container);
 571
 572         exit(0);
 573 }