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 int make_pidfile(char *devname)
 108 {
 109         char path[100];
 110         char pid[10];
 111         int fd;
 112         int n;
 113
 114         if (mkdir(pid_dir, 0700) < 0 &&
 115             errno != EEXIST)
 116                 return -errno;
 117         sprintf(path, "%s/%s.pid", pid_dir, devname);
 118
 119         fd = open(path, O_RDWR|O_CREAT|O_EXCL, 0600);
 120         if (fd < 0)
 121                 return -errno;
 122         sprintf(pid, "%d\n", getpid());
 123         n = write(fd, pid, strlen(pid));
 124         close(fd);
 125         if (n < 0)
 126                 return -errno;
 127         return 0;
 128 }
 129
 130 static void try_kill_monitor(pid_t pid, char *devname, int sock)
 131 {
 132         char buf[100];
 133         int fd;
 134         int n;
 135         long fl;
 136
 137         /* first rule of survival... don't off yourself */
 138         if (pid == getpid())
 139                 return;
 140
 141         /* kill this process if it is mdmon */
 142         sprintf(buf, "/proc/%lu/cmdline", (unsigned long) pid);
 143         fd = open(buf, O_RDONLY);
 144         if (fd < 0)
 145                 return;
 146
 147         n = read(fd, buf, sizeof(buf)-1);
 148         buf[sizeof(buf)-1] = 0;
 149         close(fd);
 150
 151         if (n < 0 || !strstr(buf, "mdmon"))
 152                 return;
 153
 154         kill(pid, SIGTERM);
 155
 156         /* Wait for monitor to exit by reading from the socket, after
 157          * clearing the non-blocking flag */
 158         fl = fcntl(sock, F_GETFL, 0);
 159         fl &= ~O_NONBLOCK;
 160         fcntl(sock, F_SETFL, fl);
 161         n = read(sock, buf, 100);
 162         /* Ignore result, it is just the wait that
 163          * matters
 164          */
 165 }
 166
 167 void remove_pidfile(char *devname)
 168 {
 169         char buf[100];
 170
 171         sprintf(buf, "%s/%s.pid", pid_dir, devname);
 172         unlink(buf);
 173         sprintf(buf, "%s/%s.sock", pid_dir, devname);
 174         unlink(buf);
 175         if (strcmp(pid_dir, ALT_RUN) == 0)
 176                 /* try to clean up when we are finished with this dir */
 177                 rmdir(pid_dir);
 178 }
 179
 180 static int make_control_sock(char *devname)
 181 {
 182         char path[100];
 183         int sfd;
 184         long fl;
 185         struct sockaddr_un addr;
 186
 187         if (sigterm)
 188                 return -1;
 189
 190         sprintf(path, "%s/%s.sock", pid_dir, devname);
 191         unlink(path);
 192         sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
 193         if (sfd < 0)
 194                 return -1;
 195
 196         addr.sun_family = PF_LOCAL;
 197         strcpy(addr.sun_path, path);
 198         if (bind(sfd, &addr, sizeof(addr)) < 0) {
 199                 close(sfd);
 200                 return -1;
 201         }
 202         listen(sfd, 10);
 203         fl = fcntl(sfd, F_GETFL, 0);
 204         fl |= O_NONBLOCK;
 205         fcntl(sfd, F_SETFL, fl);
 206         return sfd;
 207 }
 208
 209 static void term(int sig)
 210 {
 211         sigterm = 1;
 212 }
 213
 214 static void wake_me(int sig)
 215 {
 216
 217 }
 218
 219 /* if we are debugging and starting mdmon by hand then don't fork */
 220 static int do_fork(void)
 221 {
 222         #ifdef DEBUG
 223         if (check_env("MDADM_NO_MDMON"))
 224                 return 0;
 225         #endif
 226
 227         return 1;
 228 }
 229
 230 void usage(void)
 231 {
 232         fprintf(stderr, "Usage: mdmon [--all] [--takeover] CONTAINER\n");
 233         exit(2);
 234 }
 235
 236 static int mdmon(char *devname, int devnum, int must_fork, int takeover);
 237
 238 int main(int argc, char *argv[])
 239 {
 240         char *container_name = NULL;
 241         int devnum;
 242         char *devname;
 243         int status = 0;
 244         int arg;
 245         int all = 0;
 246         int takeover = 0;
 247
 248         for (arg = 1; arg < argc; arg++) {
 249                 if (strncmp(argv[arg], "--all",5) == 0 ||
 250                     strcmp(argv[arg], "/proc/mdstat") == 0) {
 251                         container_name = argv[arg];
 252                         all = 1;
 253                 } else if (strcmp(argv[arg], "--takeover") == 0)
 254                         takeover = 1;
 255                 else if (container_name == NULL)
 256                         container_name = argv[arg];
 257                 else
 258                         usage();
 259         }
 260         if (container_name == NULL)
 261                 usage();
 262
 263         if (all) {
 264                 struct mdstat_ent *mdstat, *e;
 265                 int container_len = strlen(container_name);
 266
 267                 /* launch an mdmon instance for each container found */
 268                 mdstat = mdstat_read(0, 0);
 269                 for (e = mdstat; e; e = e->next) {
 270                         if (strncmp(e->metadata_version, "external:", 9) == 0 &&
 271                             !is_subarray(&e->metadata_version[9])) {
 272                                 devname = devnum2devname(e->devnum);
 273                                 /* update cmdline so this mdmon instance can be
 274                                  * distinguished from others in a call to ps(1)
 275                                  */
 276                                 if (strlen(devname) <= container_len) {
 277                                         memset(container_name, 0, container_len);
 278                                         sprintf(container_name, "%s", devname);
 279                                 }
 280                                 status |= mdmon(devname, e->devnum, 1,
 281                                                 takeover);
 282                         }
 283                 }
 284                 free_mdstat(mdstat);
 285
 286                 return status;
 287         } else if (strncmp(container_name, "md", 2) == 0) {
 288                 devnum = devname2devnum(container_name);
 289                 devname = devnum2devname(devnum);
 290                 if (strcmp(container_name, devname) != 0)
 291                         devname = NULL;
 292         } else {
 293                 struct stat st;
 294
 295                 devnum = NoMdDev;
 296                 if (stat(container_name, &st) == 0)
 297                         devnum = stat2devnum(&st);
 298                 if (devnum == NoMdDev)
 299                         devname = NULL;
 300                 else
 301                         devname = devnum2devname(devnum);
 302         }
 303
 304         if (!devname) {
 305                 fprintf(stderr, "mdmon: %s is not a valid md device name\n",
 306                         container_name);
 307                 exit(1);
 308         }
 309         return mdmon(devname, devnum, do_fork(), takeover);
 310 }
 311
 312 static int mdmon(char *devname, int devnum, int must_fork, int takeover)
 313 {
 314         int mdfd;
 315         struct mdinfo *mdi, *di;
 316         struct supertype *container;
 317         sigset_t set;
 318         struct sigaction act;
 319         int pfd[2];
 320         int status;
 321         int ignore;
 322         pid_t victim = -1;
 323         int victim_sock = -1;
 324
 325         dprintf("starting mdmon for %s\n", devname);
 326
 327         mdfd = open_dev(devnum);
 328         if (mdfd < 0) {
 329                 fprintf(stderr, "mdmon: %s: %s\n", devname,
 330                         strerror(errno));
 331                 return 1;
 332         }
 333         if (md_get_version(mdfd) < 0) {
 334                 fprintf(stderr, "mdmon: %s: Not an md device\n",
 335                         devname);
 336                 return 1;
 337         }
 338
 339         /* Fork, and have the child tell us when they are ready */
 340         if (must_fork) {
 341                 if (pipe(pfd) != 0) {
 342                         fprintf(stderr, "mdmon: failed to create pipe\n");
 343                         return 1;
 344                 }
 345                 switch(fork()) {
 346                 case -1:
 347                         fprintf(stderr, "mdmon: failed to fork: %s\n",
 348                                 strerror(errno));
 349                         return 1;
 350                 case 0: /* child */
 351                         close(pfd[0]);
 352                         break;
 353                 default: /* parent */
 354                         close(pfd[1]);
 355                         if (read(pfd[0], &status, sizeof(status)) != sizeof(status)) {
 356                                 wait(&status);
 357                                 status = WEXITSTATUS(status);
 358                         }
 359                         return status;
 360                 }
 361         } else
 362                 pfd[0] = pfd[1] = -1;
 363
 364         container = calloc(1, sizeof(*container));
 365         container->devnum = devnum;
 366         container->devname = devname;
 367         container->arrays = NULL;
 368         container->subarray[0] = 0;
 369         container->sock = -1;
 370
 371         if (!container->devname) {
 372                 fprintf(stderr, "mdmon: failed to allocate container name string\n");
 373                 exit(3);
 374         }
 375
 376         mdi = sysfs_read(mdfd, container->devnum,
 377                          GET_VERSION|GET_LEVEL|GET_DEVS|SKIP_GONE_DEVS);
 378
 379         if (!mdi) {
 380                 fprintf(stderr, "mdmon: failed to load sysfs info for %s\n",
 381                         container->devname);
 382                 exit(3);
 383         }
 384         if (mdi->array.level != UnSet) {
 385                 fprintf(stderr, "mdmon: %s is not a container - cannot monitor\n",
 386                         devname);
 387                 exit(3);
 388         }
 389         if (mdi->array.major_version != -1 ||
 390             mdi->array.minor_version != -2) {
 391                 fprintf(stderr, "mdmon: %s does not use external metadata - cannot monitor\n",
 392                         devname);
 393                 exit(3);
 394         }
 395
 396         container->ss = version_to_superswitch(mdi->text_version);
 397         if (container->ss == NULL) {
 398                 fprintf(stderr, "mdmon: %s uses unsupported metadata: %s\n",
 399                         devname, mdi->text_version);
 400                 exit(3);
 401         }
 402
 403         container->devs = NULL;
 404         for (di = mdi->devs; di; di = di->next) {
 405                 struct mdinfo *cd = malloc(sizeof(*cd));
 406                 *cd = *di;
 407                 cd->next = container->devs;
 408                 container->devs = cd;
 409         }
 410         sysfs_free(mdi);
 411
 412         /* SIGUSR is sent between parent and child.  So both block it
 413          * and enable it only with pselect.
 414          */
 415         sigemptyset(&set);
 416         sigaddset(&set, SIGUSR1);
 417         sigaddset(&set, SIGTERM);
 418         sigprocmask(SIG_BLOCK, &set, NULL);
 419         act.sa_handler = wake_me;
 420         act.sa_flags = 0;
 421         sigaction(SIGUSR1, &act, NULL);
 422         act.sa_handler = term;
 423         sigaction(SIGTERM, &act, NULL);
 424         act.sa_handler = SIG_IGN;
 425         sigaction(SIGPIPE, &act, NULL);
 426
 427         pid_dir = VAR_RUN;
 428         victim = mdmon_pid(container->devnum);
 429         if (victim < 0) {
 430                 pid_dir = ALT_RUN;
 431                 victim = mdmon_pid(container->devnum);
 432         }
 433         if (victim >= 0)
 434                 victim_sock = connect_monitor(container->devname);
 435
 436         ignore = chdir("/");
 437         if (!takeover && victim > 0 && victim_sock >= 0) {
 438                 if (fping_monitor(victim_sock) == 0) {
 439                         fprintf(stderr, "mdmon: %s already managed\n",
 440                                 container->devname);
 441                         exit(3);
 442                 }
 443                 close(victim_sock);
 444         }
 445         if (container->ss->load_super(container, mdfd, devname)) {
 446                 fprintf(stderr, "mdmon: Cannot load metadata for %s\n",
 447                         devname);
 448                 exit(3);
 449         }
 450         close(mdfd);
 451
 452         /* Ok, this is close enough.  We can say goodbye to our parent now.
 453          */
 454         if (victim > 0)
 455                 remove_pidfile(devname);
 456         pid_dir = VAR_RUN;
 457         if (make_pidfile(devname) < 0) {
 458                 /* Try the alternate */
 459                 pid_dir = ALT_RUN;
 460                 if (make_pidfile(devname) < 0) {
 461                         fprintf(stderr, "mdmon: Neither %s nor %s are writable\n"
 462                                 "       cannot create .pid or .sock files.  Aborting\n",
 463                                 VAR_RUN, ALT_RUN);
 464                         exit(3);
 465                 }
 466         }
 467         container->sock = make_control_sock(devname);
 468
 469         status = 0;
 470         if (write(pfd[1], &status, sizeof(status)) < 0)
 471                 fprintf(stderr, "mdmon: failed to notify our parent: %d\n",
 472                         getppid());
 473         close(pfd[1]);
 474
 475         mlockall(MCL_CURRENT | MCL_FUTURE);
 476
 477         if (clone_monitor(container) < 0) {
 478                 fprintf(stderr, "mdmon: failed to start monitor process: %s\n",
 479                         strerror(errno));
 480                 exit(2);
 481         }
 482
 483         if (victim > 0) {
 484                 try_kill_monitor(victim, container->devname, victim_sock);
 485                 close(victim_sock);
 486         }
 487
 488         setsid();
 489         close(0);
 490         open("/dev/null", O_RDWR);
 491         close(1);
 492         ignore = dup(0);
 493 #ifndef DEBUG
 494         close(2);
 495         ignore = dup(0);
 496 #endif
 497
 498         do_manager(container);
 499
 500         exit(0);
 501 }