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