msg.c

   1 /*
   2  * Copyright (C) 2008 Intel Corporation
   3  *
   4  *      mdmon socket / message handling
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms and conditions of the GNU General Public License,
   8  * version 2, as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along with
  16  * this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18  */
  19 #ifndef _GNU_SOURCE
  20 #define _GNU_SOURCE
  21 #endif
  22 #include <unistd.h>
  23 #include <stdio.h>
  24 #include <string.h>
  25 #include <stdlib.h>
  26 #include <errno.h>
  27 #include <fcntl.h>
  28 #include <sys/types.h>
  29 #include <sys/socket.h>
  30 #include <sys/un.h>
  31 #include "mdadm.h"
  32 #include "mdmon.h"
  33
  34 static const __u32 start_magic = 0x5a5aa5a5;
  35 static const __u32 end_magic = 0xa5a55a5a;
  36
  37 static int send_buf(int fd, const void* buf, int len, int tmo)
  38 {
  39         fd_set set;
  40         int rv;
  41         struct timeval timeout = {tmo, 0};
  42         struct timeval *ptmo = tmo ? &timeout : NULL;
  43
  44         while (len) {
  45                 FD_ZERO(&set);
  46                 FD_SET(fd, &set);
  47                 rv = select(fd+1, NULL, &set, NULL, ptmo);
  48                 if (rv <= 0)
  49                         return -1;
  50                 rv = write(fd, buf, len);
  51                 if (rv <= 0)
  52                         return -1;
  53                 len -= rv;
  54                 buf += rv;
  55         }
  56         return 0;
  57 }
  58
  59 static int recv_buf(int fd, void* buf, int len, int tmo)
  60 {
  61         fd_set set;
  62         int rv;
  63         struct timeval timeout = {tmo, 0};
  64         struct timeval *ptmo = tmo ? &timeout : NULL;
  65
  66         while (len) {
  67                 FD_ZERO(&set);
  68                 FD_SET(fd, &set);
  69                 rv = select(fd+1, &set, NULL, NULL, ptmo);
  70                 if (rv <= 0)
  71                         return -1;
  72                 rv = read(fd, buf, len);
  73                 if (rv <= 0)
  74                         return -1;
  75                 len -= rv;
  76                 buf += rv;
  77         }
  78         return 0;
  79 }
  80
  81
  82 int send_message(int fd, struct metadata_update *msg, int tmo)
  83 {
  84         __s32 len = msg->len;
  85         int rv;
  86
  87         rv = send_buf(fd, &start_magic, 4, tmo);
  88         rv = rv ?: send_buf(fd, &len, 4, tmo);
  89         if (len > 0)
  90                 rv = rv ?: send_buf(fd, msg->buf, msg->len, tmo);
  91         rv = send_buf(fd, &end_magic, 4, tmo);
  92
  93         return rv;
  94 }
  95
  96 int receive_message(int fd, struct metadata_update *msg, int tmo)
  97 {
  98         __u32 magic;
  99         __s32 len;
 100         int rv;
 101
 102         rv = recv_buf(fd, &magic, 4, tmo);
 103         if (rv < 0 || magic != start_magic)
 104                 return -1;
 105         rv = recv_buf(fd, &len, 4, tmo);
 106         if (rv < 0 || len > MSG_MAX_LEN)
 107                 return -1;
 108         if (len > 0) {
 109                 msg->buf = malloc(len);
 110                 if (msg->buf == NULL)
 111                         return -1;
 112                 rv = recv_buf(fd, msg->buf, len, tmo);
 113                 if (rv < 0) {
 114                         free(msg->buf);
 115                         return -1;
 116                 }
 117         } else
 118                 msg->buf = NULL;
 119         rv = recv_buf(fd, &magic, 4, tmo);
 120         if (rv < 0 || magic != end_magic) {
 121                 free(msg->buf);
 122                 return -1;
 123         }
 124         msg->len = len;
 125         return 0;
 126 }
 127
 128 int ack(int fd, int tmo)
 129 {
 130         struct metadata_update msg = { .len = 0 };
 131
 132         return send_message(fd, &msg, tmo);
 133 }
 134
 135 int wait_reply(int fd, int tmo)
 136 {
 137         struct metadata_update msg;
 138         int err = receive_message(fd, &msg, tmo);
 139
 140         /* mdmon sent extra data, but caller only cares that we got a
 141          * successful reply
 142          */
 143         if (err == 0 && msg.len > 0)
 144                 free(msg.buf);
 145
 146         return err;
 147 }
 148
 149 int connect_monitor(char *devname)
 150 {
 151         char path[100];
 152         int sfd;
 153         long fl;
 154         struct sockaddr_un addr;
 155         int pos;
 156         char *c;
 157
 158         pos = sprintf(path, "%s/", MDMON_DIR);
 159         if (is_subarray(devname)) {
 160                 devname++;
 161                 c = strchr(devname, '/');
 162                 if (!c)
 163                         return -1;
 164                 snprintf(&path[pos], c - devname + 1, "%s", devname);
 165                 pos += c - devname;
 166         } else
 167                 pos += sprintf(&path[pos], "%s", devname);
 168         sprintf(&path[pos], ".sock");
 169
 170         sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
 171         if (sfd < 0)
 172                 return -1;
 173
 174         addr.sun_family = PF_LOCAL;
 175         strcpy(addr.sun_path, path);
 176         if (connect(sfd, &addr, sizeof(addr)) < 0) {
 177                 close(sfd);
 178                 return -1;
 179         }
 180
 181         fl = fcntl(sfd, F_GETFL, 0);
 182         fl |= O_NONBLOCK;
 183         fcntl(sfd, F_SETFL, fl);
 184
 185         return sfd;
 186 }
 187
 188 int fping_monitor(int sfd)
 189 {
 190         int err = 0;
 191
 192         if (sfd < 0)
 193                 return sfd;
 194
 195         /* try to ping existing socket */
 196         if (ack(sfd, 20) != 0)
 197                 err = -1;
 198
 199         /* check the reply */
 200         if (!err && wait_reply(sfd, 20) != 0)
 201                 err = -1;
 202
 203         return err;
 204 }
 205
 206 /* give the monitor a chance to update the metadata */
 207 int ping_monitor(char *devname)
 208 {
 209         int sfd = connect_monitor(devname);
 210         int err = fping_monitor(sfd);
 211
 212         close(sfd);
 213         return err;
 214 }
 215
 216 /* ping monitor using device number */
 217 int ping_monitor_by_id(int devnum)
 218 {
 219         int err = -1;
 220         char *container = devnum2devname(devnum);
 221
 222         if (container) {
 223                 err = ping_monitor(container);
 224                 free(container);
 225         }
 226
 227         return err;
 228 }
 229
 230 static char *ping_monitor_version(char *devname)
 231 {
 232         int sfd = connect_monitor(devname);
 233         struct metadata_update msg;
 234         int err = 0;
 235
 236         if (sfd < 0)
 237                 return NULL;
 238
 239         if (ack(sfd, 20) != 0)
 240                 err = -1;
 241
 242         if (!err && receive_message(sfd, &msg, 20) != 0)
 243                 err = -1;
 244
 245         close(sfd);
 246
 247         if (err || !msg.len || !msg.buf)
 248                 return NULL;
 249         return msg.buf;
 250 }
 251
 252 int unblock_subarray(struct mdinfo *sra, const int unfreeze)
 253 {
 254         char buf[64];
 255         int rc = 0;
 256
 257         if (sra) {
 258                 sprintf(buf, "external:%s\n", sra->text_version);
 259                 buf[9] = '/';
 260         } else
 261                 buf[9] = '-';
 262
 263         if (buf[9] == '-' ||
 264             sysfs_set_str(sra, NULL, "metadata_version", buf) ||
 265             (unfreeze &&
 266              sysfs_attribute_available(sra, NULL, "sync_action") &&
 267              sysfs_set_str(sra, NULL, "sync_action", "idle")))
 268                 rc = -1;
 269         return rc;
 270 }
 271
 272 int block_subarray(struct mdinfo *sra)
 273 {
 274         char buf[64];
 275         int rc = 0;
 276
 277         sprintf(buf, "external:%s\n", sra->text_version);
 278         buf[9] = '-';
 279         if (sysfs_set_str(sra, NULL, "metadata_version", buf))
 280                 rc = -1;
 281
 282         return rc;
 283 }
 284 /**
 285  * block_monitor - prevent mdmon spare assignment
 286  * @container - container to block
 287  * @freeze - flag to additionally freeze sync_action
 288  *
 289  * This is used by the reshape code to freeze the container, and the
 290  * auto-rebuild implementation to atomically move spares.
 291  * In both cases we need to stop mdmon from assigning spares to replace
 292  * failed devices as we might have other plans for the spare.
 293  * For the reshape case we also need to 'freeze' sync_action so that
 294  * no recovery happens until we have fully prepared for the reshape.
 295  *
 296  * We tell mdmon that the array is frozen by marking the 'metadata' name
 297  * with a leading '-'.  The previously told mdmon "Don't make this array
 298  * read/write, leave it readonly".  Now it means a more general "Don't
 299  * reconfigure this array at all".
 300  * As older versions of mdmon (which might run from initrd) don't understand
 301  * this, we first check that the running mdmon is new enough.
 302  */
 303 int block_monitor(char *container, const int freeze)
 304 {
 305         int devnum = devname2devnum(container);
 306         struct mdstat_ent *ent, *e, *e2;
 307         struct mdinfo *sra = NULL;
 308         char *version = NULL;
 309         char buf[64];
 310         int rv = 0;
 311
 312         if (!mdmon_running(devnum)) {
 313                 /* if mdmon is not active we assume that any instance that is
 314                  * later started will match the current mdadm version, if this
 315                  * assumption is violated we may inadvertantly rebuild an array
 316                  * that was meant for reshape, or start rebuild on a spare that
 317                  * was to be moved to another container
 318                  */
 319                 /* pass */;
 320         } else {
 321                 int ver;
 322
 323                 version = ping_monitor_version(container);
 324                 ver = version ? mdadm_version(version) : -1;
 325                 free(version);
 326                 if (ver < 3002000) {
 327                         fprintf(stderr, Name
 328                                 ": mdmon instance for %s cannot be disabled\n",
 329                                 container);
 330                         return -1;
 331                 }
 332         }
 333
 334         ent = mdstat_read(0, 0);
 335         if (!ent) {
 336                 fprintf(stderr, Name
 337                         ": failed to read /proc/mdstat while disabling mdmon\n");
 338                 return -1;
 339         }
 340
 341         /* freeze container contents */
 342         for (e = ent; e; e = e->next) {
 343                 if (!is_container_member(e, container))
 344                         continue;
 345                 sysfs_free(sra);
 346                 sra = sysfs_read(-1, e->devnum, GET_VERSION);
 347                 if (!sra) {
 348                         fprintf(stderr, Name
 349                                 ": failed to read sysfs for subarray%s\n",
 350                                 to_subarray(e, container));
 351                         break;
 352                 }
 353                 /* can't reshape an array that we can't monitor */
 354                 if (sra->text_version[0] == '-')
 355                         break;
 356
 357                 if (freeze && sysfs_freeze_array(sra) < 1)
 358                         break;
 359                 /* flag this array to not be modified by mdmon (close race with
 360                  * takeover in reshape case and spare reassignment in the
 361                  * auto-rebuild case)
 362                  */
 363                 if (block_subarray(sra))
 364                         break;
 365                 ping_monitor(container);
 366
 367                 /* check that we did not race with recovery */
 368                 if ((freeze &&
 369                      !sysfs_attribute_available(sra, NULL, "sync_action")) ||
 370                     (freeze &&
 371                      sysfs_attribute_available(sra, NULL, "sync_action") &&
 372                      sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0 &&
 373                      strcmp(buf, "frozen\n") == 0))
 374                         /* pass */;
 375                 else {
 376                         unblock_subarray(sra, 0);
 377                         break;
 378                 }
 379                 /* Double check against races - there should be no spares
 380                  * or part-spares
 381                  */
 382                 sysfs_free(sra);
 383                 sra = sysfs_read(-1, e->devnum, GET_DEVS | GET_STATE);
 384                 if (sra && sra->array.spare_disks > 0) {
 385                         unblock_subarray(sra, freeze);
 386                         break;
 387                 }
 388         }
 389
 390         if (e) {
 391                 fprintf(stderr, Name ": failed to freeze subarray%s\n",
 392                         to_subarray(e, container));
 393
 394                 /* thaw the partially frozen container */
 395                 for (e2 = ent; e2 && e2 != e; e2 = e2->next) {
 396                         if (!is_container_member(e2, container))
 397                                 continue;
 398                         sysfs_free(sra);
 399                         sra = sysfs_read(-1, e2->devnum, GET_VERSION);
 400                         if (unblock_subarray(sra, freeze))
 401                                 fprintf(stderr, Name ": Failed to unfreeze %s\n", e2->dev);
 402                 }
 403
 404                 ping_monitor(container); /* cleared frozen */
 405                 rv = -1;
 406         }
 407
 408         sysfs_free(sra);
 409         free_mdstat(ent);
 410
 411         return rv;
 412 }
 413
 414 void unblock_monitor(char *container, const int unfreeze)
 415 {
 416         struct mdstat_ent *ent, *e;
 417         struct mdinfo *sra = NULL;
 418         int to_ping = 0;
 419
 420         ent = mdstat_read(0, 0);
 421         if (!ent) {
 422                 fprintf(stderr, Name
 423                         ": failed to read /proc/mdstat while unblocking container\n");
 424                 return;
 425         }
 426
 427         /* unfreeze container contents */
 428         for (e = ent; e; e = e->next) {
 429                 if (!is_container_member(e, container))
 430                         continue;
 431                 sysfs_free(sra);
 432                 sra = sysfs_read(-1, e->devnum, GET_VERSION|GET_LEVEL);
 433                 if (sra->array.level > 0)
 434                         to_ping++;
 435                 if (unblock_subarray(sra, unfreeze))
 436                         fprintf(stderr, Name ": Failed to unfreeze %s\n", e->dev);
 437         }
 438         if (to_ping)
 439                 ping_monitor(container);
 440
 441         sysfs_free(sra);
 442         free_mdstat(ent);
 443 }
 444
 445
 446
 447 /* give the manager a chance to view the updated container state.  This
 448  * would naturally happen due to the manager noticing a change in
 449  * /proc/mdstat; however, pinging encourages this detection to happen
 450  * while an exclusive open() on the container is active
 451  */
 452 int ping_manager(char *devname)
 453 {
 454         int sfd = connect_monitor(devname);
 455         struct metadata_update msg = { .len = -1 };
 456         int err = 0;
 457
 458         if (sfd < 0)
 459                 return sfd;
 460
 461         err = send_message(sfd, &msg, 20);
 462
 463         /* check the reply */
 464         if (!err && wait_reply(sfd, 20) != 0)
 465                 err = -1;
 466
 467         close(sfd);
 468         return err;
 469 }