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(pid_dir, 0700) < 0 &&
 151             errno != EEXIST)
 152                 return -errno;
 153         sprintf(path, "%s/%s.pid", pid_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         if (n < 0 || !strstr(buf, "mdmon"))
 188                 return;
 189
 190         kill(pid, SIGTERM);
 191
 192         /* Wait for monitor to exit by reading from the socket, after
 193          * clearing the non-blocking flag */
 194         fl = fcntl(sock, F_GETFL, 0);
 195         fl &= ~O_NONBLOCK;
 196         fcntl(sock, F_SETFL, fl);
 197         n = read(sock, buf, 100);
 198         /* Ignore result, it is just the wait that
 199          * matters
 200          */
 201 }
 202
 203 void remove_pidfile(char *devname)
 204 {
 205         char buf[100];
 206
 207         sprintf(buf, "%s/%s.pid", pid_dir, devname);
 208         unlink(buf);
 209         sprintf(buf, "%s/%s.sock", pid_dir, devname);
 210         unlink(buf);
 211         if (strcmp(pid_dir, ALT_RUN) == 0)
 212                 /* try to clean up when we are finished with this dir */
 213                 rmdir(pid_dir);
 214 }
 215
 216 static 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, "%s/%s.sock", pid_dir, 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 static void term(int sig)
 246 {
 247         sigterm = 1;
 248 }
 249
 250 static void wake_me(int sig)
 251 {
 252
 253 }
 254
 255 /* if we are debugging and starting mdmon by hand then don't fork */
 256 static int do_fork(void)
 257 {
 258         #ifdef DEBUG
 259         if (check_env("MDADM_NO_MDMON"))
 260                 return 0;
 261         #endif
 262
 263         return 1;
 264 }
 265
 266 void usage(void)
 267 {
 268         fprintf(stderr, "Usage: mdmon [--all] [--takeover] CONTAINER\n");
 269         exit(2);
 270 }
 271
 272 static int mdmon(char *devname, int devnum, int must_fork, int takeover);
 273
 274 int main(int argc, char *argv[])
 275 {
 276         char *container_name = NULL;
 277         int devnum;
 278         char *devname;
 279         int status = 0;
 280         int arg;
 281         int all = 0;
 282         int takeover = 0;
 283
 284         for (arg = 1; arg < argc; arg++) {
 285                 if (strncmp(argv[arg], "--all",5) == 0 ||
 286                     strcmp(argv[arg], "/proc/mdstat") == 0) {
 287                         container_name = argv[arg];
 288                         all = 1;
 289                 } else if (strcmp(argv[arg], "--takeover") == 0)
 290                         takeover = 1;
 291                 else if (container_name == NULL)
 292                         container_name = argv[arg];
 293                 else
 294                         usage();
 295         }
 296         if (container_name == NULL)
 297                 usage();
 298
 299         if (all) {
 300                 struct mdstat_ent *mdstat, *e;
 301                 int container_len = strlen(container_name);
 302
 303                 /* launch an mdmon instance for each container found */
 304                 mdstat = mdstat_read(0, 0);
 305                 for (e = mdstat; e; e = e->next) {
 306                         if (strncmp(e->metadata_version, "external:", 9) == 0 &&
 307                             !is_subarray(&e->metadata_version[9])) {
 308                                 devname = devnum2devname(e->devnum);
 309                                 /* update cmdline so this mdmon instance can be
 310                                  * distinguished from others in a call to ps(1)
 311                                  */
 312                                 if (strlen(devname) <= container_len) {
 313                                         memset(container_name, 0, container_len);
 314                                         sprintf(container_name, "%s", devname);
 315                                 }
 316                                 status |= mdmon(devname, e->devnum, 1,
 317                                                 takeover);
 318                         }
 319                 }
 320                 free_mdstat(mdstat);
 321
 322                 return status;
 323         } else if (strncmp(container_name, "md", 2) == 0) {
 324                 devnum = devname2devnum(container_name);
 325                 devname = devnum2devname(devnum);
 326                 if (strcmp(container_name, devname) != 0)
 327                         devname = NULL;
 328         } else {
 329                 struct stat st;
 330
 331                 devnum = NoMdDev;
 332                 if (stat(container_name, &st) == 0)
 333                         devnum = stat2devnum(&st);
 334                 if (devnum == NoMdDev)
 335                         devname = NULL;
 336                 else
 337                         devname = devnum2devname(devnum);
 338         }
 339
 340         if (!devname) {
 341                 fprintf(stderr, "mdmon: %s is not a valid md device name\n",
 342                         container_name);
 343                 exit(1);
 344         }
 345         return mdmon(devname, devnum, do_fork(), takeover);
 346 }
 347
 348 static int mdmon(char *devname, int devnum, int must_fork, int takeover)
 349 {
 350         int mdfd;
 351         struct mdinfo *mdi, *di;
 352         struct supertype *container;
 353         sigset_t set;
 354         struct sigaction act;
 355         int pfd[2];
 356         int status;
 357         int ignore;
 358         pid_t victim = -1;
 359         int victim_sock = -1;
 360
 361         dprintf("starting mdmon for %s\n", devname);
 362
 363         mdfd = open_dev(devnum);
 364         if (mdfd < 0) {
 365                 fprintf(stderr, "mdmon: %s: %s\n", devname,
 366                         strerror(errno));
 367                 return 1;
 368         }
 369         if (md_get_version(mdfd) < 0) {
 370                 fprintf(stderr, "mdmon: %s: Not an md device\n",
 371                         devname);
 372                 return 1;
 373         }
 374
 375         /* Fork, and have the child tell us when they are ready */
 376         if (must_fork) {
 377                 if (pipe(pfd) != 0) {
 378                         fprintf(stderr, "mdmon: failed to create pipe\n");
 379                         return 1;
 380                 }
 381                 switch(fork()) {
 382                 case -1:
 383                         fprintf(stderr, "mdmon: failed to fork: %s\n",
 384                                 strerror(errno));
 385                         return 1;
 386                 case 0: /* child */
 387                         close(pfd[0]);
 388                         break;
 389                 default: /* parent */
 390                         close(pfd[1]);
 391                         if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) {
 392                                 wait(&status);
 393                                 status = WEXITSTATUS(status);
 394                         }
 395                         return status;
 396                 }
 397         } else
 398                 pfd[0] = pfd[1] = -1;
 399
 400         container = calloc(1, sizeof(*container));
 401         container->devnum = devnum;
 402         container->devname = devname;
 403         container->arrays = NULL;
 404         container->subarray[0] = 0;
 405         container->sock = -1;
 406
 407         if (!container->devname) {
 408                 fprintf(stderr, "mdmon: failed to allocate container name string\n");
 409                 exit(3);
 410         }
 411
 412         mdi = sysfs_read(mdfd, container->devnum, GET_VERSION|GET_LEVEL|GET_DEVS);
 413
 414         if (!mdi) {
 415                 fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
 416                         container->devname);
 417                 exit(3);
 418         }
 419         if (mdi->array.level != UnSet) {
 420                 fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
 421                         devname);
 422                 exit(3);
 423         }
 424         if (mdi->array.major_version != -1 ||
 425             mdi->array.minor_version != -2) {
 426                 fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
 427                         devname);
 428                 exit(3);
 429         }
 430
 431         container->ss = version_to_superswitch(mdi->text_version);
 432         if (container->ss == NULL) {
 433                 fprintf(stderr, "mdmon: %s uses unsupported metadata: %s\n",
 434                         devname, mdi->text_version);
 435                 exit(3);
 436         }
 437
 438         container->devs = NULL;
 439         for (di = mdi->devs; di; di = di->next) {
 440                 struct mdinfo *cd = malloc(sizeof(*cd));
 441                 *cd = *di;
 442                 cd->next = container->devs;
 443                 container->devs = cd;
 444         }
 445         sysfs_free(mdi);
 446
 447         /* SIGUSR is sent between parent and child.  So both block it
 448          * and enable it only with pselect.
 449          */
 450         sigemptyset(&set);
 451         sigaddset(&set, SIGUSR1);
 452         sigaddset(&set, SIGTERM);
 453         sigprocmask(SIG_BLOCK, &set, NULL);
 454         act.sa_handler = wake_me;
 455         act.sa_flags = 0;
 456         sigaction(SIGUSR1, &act, NULL);
 457         act.sa_handler = term;
 458         sigaction(SIGTERM, &act, NULL);
 459         act.sa_handler = SIG_IGN;
 460         sigaction(SIGPIPE, &act, NULL);
 461
 462         pid_dir = VAR_RUN;
 463         victim = mdmon_pid(container->devnum);
 464         if (victim < 0) {
 465                 pid_dir = ALT_RUN;
 466                 victim = mdmon_pid(container->devnum);
 467         }
 468         if (victim >= 0)
 469                 victim_sock = connect_monitor(container->devname);
 470
 471         ignore = chdir("/");
 472         if (!takeover && victim > 0 && victim_sock >= 0) {
 473                 if (fping_monitor(victim_sock) == 0) {
 474                         fprintf(stderr, "mdmon: %s already managed\n",
 475                                 container->devname);
 476                         exit(3);
 477                 }
 478                 close(victim_sock);
 479         }
 480         if (container->ss->load_super(container, mdfd, devname)) {
 481                 fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
 482                         devname);
 483                 exit(3);
 484         }
 485         close(mdfd);
 486
 487         /* Ok, this is close enough.  We can say goodbye to our parent now.
 488          */
 489         if (victim > 0)
 490                 remove_pidfile(devname);
 491         pid_dir = VAR_RUN;
 492         if (make_pidfile(devname) < 0) {
 493                 /* Try the alternate */
 494                 pid_dir = ALT_RUN;
 495                 if (make_pidfile(devname) < 0) {
 496                         fprintf(stderr, "mdmon: Neither %s nor %s are writable\n"
 497                                 "       cannot create .pid or .sock files.  Aborting\n",
 498                                 VAR_RUN, ALT_RUN);
 499                         exit(3);
 500                 }
 501         }
 502         container->sock = make_control_sock(devname);
 503
 504         status = 0;
 505         if (write(pfd[1], &status, sizeof(status)) < 0)
 506                 fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
 507                         getppid());
 508         close(pfd[1]);
 509
 510         mlockall(MCL_CURRENT | MCL_FUTURE);
 511
 512         if (clone_monitor(container) < 0) {
 513                 fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
 514                         strerror(errno));
 515                 exit(2);
 516         }
 517
 518         if (victim > 0) {
 519                 try_kill_monitor(victim, container->devname, victim_sock);
 520                 close(victim_sock);
 521         }
 522
 523         setsid();
 524         close(0);
 525         open("/dev/null", O_RDWR);
 526         close(1);
 527         ignore = dup(0);
 528 #ifndef DEBUG
 529         close(2);
 530         ignore = dup(0);
 531 #endif
 532
 533         do_manager(container);
 534
 535         exit(0);
 536 }