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 #ifdef USE_PTHREADS
  62 #include        <pthread.h>
  63 #else
  64 #include        <sched.h>
  65 #endif
  66
  67 #include        "mdadm.h"
  68 #include        "mdmon.h"
  69
  70 struct active_array *discard_this;
  71 struct active_array *pending_discard;
  72
  73 int mon_tid, mgr_tid;
  74
  75 int sigterm;
  76
  77 #ifdef USE_PTHREADS
  78 static void *run_child(void *v)
  79 {
  80         struct supertype *c = v;
  81
  82         mon_tid = syscall(SYS_gettid);
  83         do_monitor(c);
  84         return 0;
  85 }
  86
  87 static int clone_monitor(struct supertype *container)
  88 {
  89         pthread_attr_t attr;
  90         pthread_t thread;
  91         int rc;
  92
  93         mon_tid = -1;
  94         pthread_attr_init(&attr);
  95         pthread_attr_setstacksize(&attr, 4096);
  96         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
  97         rc = pthread_create(&thread, &attr, run_child, container);
  98         if (rc)
  99                 return rc;
 100         while (mon_tid == -1)
 101                 usleep(10);
 102         pthread_attr_destroy(&attr);
 103
 104         mgr_tid = syscall(SYS_gettid);
 105
 106         return mon_tid;
 107 }
 108 #else /* USE_PTHREADS */
 109 static int run_child(void *v)
 110 {
 111         struct supertype *c = v;
 112
 113         do_monitor(c);
 114         return 0;
 115 }
 116
 117 #ifdef __ia64__
 118 int __clone2(int (*fn)(void *),
 119             void *child_stack_base, size_t stack_size,
 120             int flags, void *arg, ...
 121          /* pid_t *pid, struct user_desc *tls, pid_t *ctid */ );
 122 #endif
 123 static int clone_monitor(struct supertype *container)
 124 {
 125         static char stack[4096];
 126
 127 #ifdef __ia64__
 128         mon_tid = __clone2(run_child, stack, sizeof(stack),
 129                    CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD,
 130                    container);
 131 #else
 132         mon_tid = clone(run_child, stack+4096-64,
 133                    CLONE_FS|CLONE_FILES|CLONE_VM|CLONE_SIGHAND|CLONE_THREAD,
 134                    container);
 135 #endif
 136
 137         mgr_tid = syscall(SYS_gettid);
 138
 139         return mon_tid;
 140 }
 141 #endif /* USE_PTHREADS */
 142
 143 static int make_pidfile(char *devname)
 144 {
 145         char path[100];
 146         char pid[10];
 147         int fd;
 148         int n;
 149
 150         if (mkdir(MDMON_DIR, 0755) < 0 &&
 151             errno != EEXIST)
 152                 return -errno;
 153         sprintf(path, "%s/%s.pid", MDMON_DIR, devname);
 154
 155         fd = open(path, O_RDWR|O_CREAT|O_EXCL, 0600);
 156         if (fd < 0)
 157                 return -errno;
 158         sprintf(pid, "%d\n", getpid());
 159         n = write(fd, pid, strlen(pid));
 160         close(fd);
 161         if (n < 0)
 162                 return -errno;
 163         return 0;
 164 }
 165
 166 static void try_kill_monitor(pid_t pid, char *devname, int sock)
 167 {
 168         char buf[100];
 169         int fd;
 170         int n;
 171         long fl;
 172
 173         /* first rule of survival... don't off yourself */
 174         if (pid == getpid())
 175                 return;
 176
 177         /* kill this process if it is mdmon */
 178         sprintf(buf, "/proc/%lu/cmdline", (unsigned long) pid);
 179         fd = open(buf, O_RDONLY);
 180         if (fd < 0)
 181                 return;
 182
 183         n = read(fd, buf, sizeof(buf)-1);
 184         buf[sizeof(buf)-1] = 0;
 185         close(fd);
 186
 187         /* Note that if started with --offroot, the name
 188          * might be "@dmon"
 189          */
 190         if (n < 0 || !(strstr(buf, "mdmon") ||
 191                        strstr(buf, "@dmon")))
 192                 return;
 193
 194         kill(pid, SIGTERM);
 195
 196         if (sock < 0)
 197                 return;
 198
 199         /* Wait for monitor to exit by reading from the socket, after
 200          * clearing the non-blocking flag */
 201         fl = fcntl(sock, F_GETFL, 0);
 202         fl &= ~O_NONBLOCK;
 203         fcntl(sock, F_SETFL, fl);
 204         n = read(sock, buf, 100);
 205         /* Ignore result, it is just the wait that
 206          * matters
 207          */
 208 }
 209
 210 void remove_pidfile(char *devname)
 211 {
 212         char buf[100];
 213
 214         sprintf(buf, "%s/%s.pid", MDMON_DIR, devname);
 215         unlink(buf);
 216         sprintf(buf, "%s/%s.sock", MDMON_DIR, devname);
 217         unlink(buf);
 218 }
 219
 220 static int make_control_sock(char *devname)
 221 {
 222         char path[100];
 223         int sfd;
 224         long fl;
 225         struct sockaddr_un addr;
 226
 227         if (sigterm)
 228                 return -1;
 229
 230         sprintf(path, "%s/%s.sock", MDMON_DIR, devname);
 231         unlink(path);
 232         sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
 233         if (sfd < 0)
 234                 return -1;
 235
 236         addr.sun_family = PF_LOCAL;
 237         strcpy(addr.sun_path, path);
 238         if (bind(sfd, &addr, sizeof(addr)) < 0) {
 239                 close(sfd);
 240                 return -1;
 241         }
 242         listen(sfd, 10);
 243         fl = fcntl(sfd, F_GETFL, 0);
 244         fl |= O_NONBLOCK;
 245         fcntl(sfd, F_SETFL, fl);
 246         return sfd;
 247 }
 248
 249 static void term(int sig)
 250 {
 251         sigterm = 1;
 252 }
 253
 254 static void wake_me(int sig)
 255 {
 256
 257 }
 258
 259 /* if we are debugging and starting mdmon by hand then don't fork */
 260 static int do_fork(void)
 261 {
 262         #ifdef DEBUG
 263         if (check_env("MDADM_NO_MDMON"))
 264                 return 0;
 265         #endif
 266
 267         return 1;
 268 }
 269
 270 void usage(void)
 271 {
 272         fprintf(stderr,
 273 "Usage: mdmon [options] CONTAINER\n"
 274 "\n"
 275 "Options are:\n"
 276 "  --help        -h   : This message\n"
 277 "  --all              : All devices\n"
 278 "  --takeover    -t   : Takeover container\n"
 279 "  --offroot          : Set first character of argv[0] to @ to indicate the\n"
 280 "                       application was launched from initrd/initramfs and\n"
 281 "                       should not be shutdown by systemd as part of the\n"
 282 "                       regular shutdown process.\n"
 283 );
 284         exit(2);
 285 }
 286
 287 static int mdmon(char *devname, int devnum, int must_fork, int takeover);
 288
 289 int main(int argc, char *argv[])
 290 {
 291         char *container_name = NULL;
 292         int devnum;
 293         char *devname;
 294         int status = 0;
 295         int opt;
 296         int all = 0;
 297         int takeover = 0;
 298         static struct option options[] = {
 299                 {"all", 0, NULL, 'a'},
 300                 {"takeover", 0, NULL, 't'},
 301                 {"help", 0, NULL, 'h'},
 302                 {"offroot", 0, NULL, OffRootOpt},
 303                 {NULL, 0, NULL, 0}
 304         };
 305
 306         while ((opt = getopt_long(argc, argv, "tha", options, NULL)) != -1) {
 307                 switch (opt) {
 308                 case 'a':
 309                         container_name = argv[optind-1];
 310                         all = 1;
 311                         break;
 312                 case 't':
 313                         takeover = 1;
 314                         break;
 315                 case OffRootOpt:
 316                         argv[0][0] = '@';
 317                         break;
 318                 case 'h':
 319                 default:
 320                         usage();
 321                         break;
 322                 }
 323         }
 324
 325         if (all == 0 && container_name == NULL) {
 326                 if (argv[optind])
 327                         container_name = argv[optind];
 328         }
 329
 330         if (container_name == NULL)
 331                 usage();
 332
 333         if (argc - optind > 1)
 334                 usage();
 335
 336         if (strcmp(container_name, "/proc/mdstat") == 0)
 337                 all = 1;
 338
 339         if (all) {
 340                 struct mdstat_ent *mdstat, *e;
 341                 int container_len = strlen(container_name);
 342
 343                 /* launch an mdmon instance for each container found */
 344                 mdstat = mdstat_read(0, 0);
 345                 for (e = mdstat; e; e = e->next) {
 346                         if (e->metadata_version &&
 347                             strncmp(e->metadata_version, "external:", 9) == 0 &&
 348                             !is_subarray(&e->metadata_version[9])) {
 349                                 devname = devnum2devname(e->devnum);
 350                                 /* update cmdline so this mdmon instance can be
 351                                  * distinguished from others in a call to ps(1)
 352                                  */
 353                                 if (strlen(devname) <= (unsigned)container_len) {
 354                                         memset(container_name, 0, container_len);
 355                                         sprintf(container_name, "%s", devname);
 356                                 }
 357                                 status |= mdmon(devname, e->devnum, 1,
 358                                                 takeover);
 359                         }
 360                 }
 361                 free_mdstat(mdstat);
 362
 363                 return status;
 364         } else if (strncmp(container_name, "md", 2) == 0) {
 365                 devnum = devname2devnum(container_name);
 366                 devname = devnum2devname(devnum);
 367                 if (strcmp(container_name, devname) != 0)
 368                         devname = NULL;
 369         } else {
 370                 struct stat st;
 371
 372                 devnum = NoMdDev;
 373                 if (stat(container_name, &st) == 0)
 374                         devnum = stat2devnum(&st);
 375                 if (devnum == NoMdDev)
 376                         devname = NULL;
 377                 else
 378                         devname = devnum2devname(devnum);
 379         }
 380
 381         if (!devname) {
 382                 fprintf(stderr, "mdmon: %s is not a valid md device name\n",
 383                         container_name);
 384                 exit(1);
 385         }
 386         return mdmon(devname, devnum, do_fork(), takeover);
 387 }
 388
 389 static int mdmon(char *devname, int devnum, int must_fork, int takeover)
 390 {
 391         int mdfd;
 392         struct mdinfo *mdi, *di;
 393         struct supertype *container;
 394         sigset_t set;
 395         struct sigaction act;
 396         int pfd[2];
 397         int status;
 398         int ignore;
 399         pid_t victim = -1;
 400         int victim_sock = -1;
 401
 402         dprintf("starting mdmon for %s\n", devname);
 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 = xcalloc(1, sizeof(*container));
 442         container->devnum = devnum;
 443         container->devname = devname;
 444         container->arrays = NULL;
 445         container->sock = -1;
 446
 447         if (!container->devname) {
 448                 fprintf(stderr, "mdmon: failed to allocate container name string\n");
 449                 exit(3);
 450         }
 451
 452         mdi = sysfs_read(mdfd, container->devnum, GET_VERSION|GET_LEVEL|GET_DEVS);
 453
 454         if (!mdi) {
 455                 fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
 456                         container->devname);
 457                 exit(3);
 458         }
 459         if (mdi->array.level != UnSet) {
 460                 fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
 461                         devname);
 462                 exit(3);
 463         }
 464         if (mdi->array.major_version != -1 ||
 465             mdi->array.minor_version != -2) {
 466                 fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
 467                         devname);
 468                 exit(3);
 469         }
 470
 471         container->ss = version_to_superswitch(mdi->text_version);
 472         if (container->ss == NULL) {
 473                 fprintf(stderr, "mdmon: %s uses unsupported metadata: %s\n",
 474                         devname, mdi->text_version);
 475                 exit(3);
 476         }
 477
 478         container->devs = NULL;
 479         for (di = mdi->devs; di; di = di->next) {
 480                 struct mdinfo *cd = xmalloc(sizeof(*cd));
 481                 *cd = *di;
 482                 cd->next = container->devs;
 483                 container->devs = cd;
 484         }
 485         sysfs_free(mdi);
 486
 487         /* SIGUSR is sent between parent and child.  So both block it
 488          * and enable it only with pselect.
 489          */
 490         sigemptyset(&set);
 491         sigaddset(&set, SIGUSR1);
 492         sigaddset(&set, SIGTERM);
 493         sigprocmask(SIG_BLOCK, &set, NULL);
 494         act.sa_handler = wake_me;
 495         act.sa_flags = 0;
 496         sigaction(SIGUSR1, &act, NULL);
 497         act.sa_handler = term;
 498         sigaction(SIGTERM, &act, NULL);
 499         act.sa_handler = SIG_IGN;
 500         sigaction(SIGPIPE, &act, NULL);
 501
 502         victim = mdmon_pid(container->devnum);
 503         if (victim >= 0)
 504                 victim_sock = connect_monitor(container->devname);
 505
 506         ignore = chdir("/");
 507         if (!takeover && victim > 0 && victim_sock >= 0) {
 508                 if (fping_monitor(victim_sock) == 0) {
 509                         fprintf(stderr, "mdmon: %s already managed\n",
 510                                 container->devname);
 511                         exit(3);
 512                 }
 513                 close(victim_sock);
 514                 victim_sock = -1;
 515         }
 516         if (container->ss->load_container(container, mdfd, devname)) {
 517                 fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
 518                         devname);
 519                 exit(3);
 520         }
 521         close(mdfd);
 522
 523         /* Ok, this is close enough.  We can say goodbye to our parent now.
 524          */
 525         if (victim > 0)
 526                 remove_pidfile(devname);
 527         if (make_pidfile(devname) < 0) {
 528                 exit(3);
 529         }
 530         container->sock = make_control_sock(devname);
 531
 532         status = 0;
 533         if (write(pfd[1], &status, sizeof(status)) < 0)
 534                 fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
 535                         getppid());
 536         close(pfd[1]);
 537
 538         mlockall(MCL_CURRENT | MCL_FUTURE);
 539
 540         if (clone_monitor(container) < 0) {
 541                 fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
 542                         strerror(errno));
 543                 exit(2);
 544         }
 545
 546         if (victim > 0) {
 547                 try_kill_monitor(victim, container->devname, victim_sock);
 548                 if (victim_sock >= 0)
 549                         close(victim_sock);
 550         }
 551
 552         setsid();
 553         close(0);
 554         open("/dev/null", O_RDWR);
 555         close(1);
 556         ignore = dup(0);
 557 #ifndef DEBUG
 558         close(2);
 559         ignore = dup(0);
 560 #endif
 561
 562         /* This silliness is to stop the compiler complaining
 563          * that we ignore 'ignore'
 564          */
 565         if (ignore)
 566                 ignore++;
 567
 568         do_manager(container);
 569
 570         exit(0);
 571 }
 572
 573 /* Some stub functions so super-* can link with us */
 574 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
 575                   struct supertype *st, unsigned long blocks,
 576                   int *fds, unsigned long long *offsets,
 577                   int dests, int *destfd, unsigned long long *destoffsets)
 578 {
 579         return 0;
 580 }
 581
 582 int restore_stripes(int *dest, unsigned long long *offsets,
 583                     int raid_disks, int chunk_size, int level, int layout,
 584                     int source, unsigned long long read_offset,
 585                     unsigned long long start, unsigned long long length,
 586                     char *src_buf)
 587 {
 588         return 1;
 589 }
 590
 591 void abort_reshape(struct mdinfo *sra)
 592 {
 593         return;
 594 }
 595
 596 int save_stripes(int *source, unsigned long long *offsets,
 597                  int raid_disks, int chunk_size, int level, int layout,
 598                  int nwrites, int *dest,
 599                  unsigned long long start, unsigned long long length,
 600                  char *buf)
 601 {
 602         return 0;
 603 }