libmultipath/uevent.c

   1 /*
   2  * uevent.c - trigger upon netlink uevents from the kernel
   3  *
   4  *      Only kernels from version 2.6.10* on provide the uevent netlink socket.
   5  *      Until the libc-kernel-headers are updated, you need to compile with:
   6  *
   7  *        gcc -I /lib/modules/`uname -r`/build/include -o uevent_listen uevent_listen.c
   8  *
   9  * Copyright (C) 2004 Kay Sievers <kay.sievers@vrfy.org>
  10  *
  11  *      This program is free software; you can redistribute it and/or modify it
  12  *      under the terms of the GNU General Public License as published by the
  13  *      Free Software Foundation version 2 of the License.
  14  *
  15  *      This program is distributed in the hope that it will be useful, but
  16  *      WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *      MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  *      General Public License for more details.
  19  *
  20  *      You should have received a copy of the GNU General Public License along
  21  *      with this program; if not, write to the Free Software Foundation, Inc.,
  22  *      675 Mass Ave, Cambridge, MA 02139, USA.
  23  *
  24  */
  25
  26 #include <unistd.h>
  27 #include <stdio.h>
  28 #include <errno.h>
  29 #include <stdlib.h>
  30 #include <stddef.h>
  31 #include <string.h>
  32 #include <fcntl.h>
  33 #include <time.h>
  34 #include <sys/socket.h>
  35 #include <sys/user.h>
  36 #include <sys/un.h>
  37 #include <linux/types.h>
  38 #include <linux/netlink.h>
  39 #include <pthread.h>
  40 #include <limits.h>
  41 #include <sys/mman.h>
  42 #include <libudev.h>
  43 #include <errno.h>
  44
  45 #include "memory.h"
  46 #include "debug.h"
  47 #include "list.h"
  48 #include "uevent.h"
  49 #include "vector.h"
  50 #include "config.h"
  51
  52 typedef int (uev_trigger)(struct uevent *, void * trigger_data);
  53
  54 pthread_t uevq_thr;
  55 LIST_HEAD(uevq);
  56 pthread_mutex_t uevq_lock = PTHREAD_MUTEX_INITIALIZER;
  57 pthread_mutex_t *uevq_lockp = &uevq_lock;
  58 pthread_cond_t uev_cond = PTHREAD_COND_INITIALIZER;
  59 pthread_cond_t *uev_condp = &uev_cond;
  60 uev_trigger *my_uev_trigger;
  61 void * my_trigger_data;
  62 int servicing_uev;
  63
  64 int is_uevent_busy(void)
  65 {
  66         int empty;
  67
  68         pthread_mutex_lock(uevq_lockp);
  69         empty = list_empty(&uevq);
  70         pthread_mutex_unlock(uevq_lockp);
  71         return (!empty || servicing_uev);
  72 }
  73
  74 struct uevent * alloc_uevent (void)
  75 {
  76         struct uevent *uev = MALLOC(sizeof(struct uevent));
  77
  78         if (uev)
  79                 INIT_LIST_HEAD(&uev->node);
  80
  81         return uev;
  82 }
  83
  84 void
  85 setup_thread_attr(pthread_attr_t *attr, size_t stacksize, int detached)
  86 {
  87         if (pthread_attr_init(attr)) {
  88                 fprintf(stderr, "can't initialize thread attr: %s\n",
  89                         strerror(errno));
  90                 exit(1);
  91         }
  92         if (stacksize < PTHREAD_STACK_MIN)
  93                 stacksize = PTHREAD_STACK_MIN;
  94
  95         if (pthread_attr_setstacksize(attr, stacksize)) {
  96                 fprintf(stderr, "can't set thread stack size to %lu: %s\n",
  97                         (unsigned long)stacksize, strerror(errno));
  98                 exit(1);
  99         }
 100         if (detached && pthread_attr_setdetachstate(attr,
 101                                                     PTHREAD_CREATE_DETACHED)) {
 102                 fprintf(stderr, "can't set thread to detached: %s\n",
 103                         strerror(errno));
 104                 exit(1);
 105         }
 106 }
 107
 108 /*
 109  * Called with uevq_lockp held
 110  */
 111 void
 112 service_uevq(struct list_head *tmpq)
 113 {
 114         struct uevent *uev, *tmp;
 115
 116         list_for_each_entry_safe(uev, tmp, tmpq, node) {
 117                 list_del_init(&uev->node);
 118
 119                 if (my_uev_trigger && my_uev_trigger(uev, my_trigger_data))
 120                         condlog(0, "uevent trigger error");
 121
 122                 if (uev->udev)
 123                         udev_device_unref(uev->udev);
 124                 FREE(uev);
 125         }
 126 }
 127
 128 static void uevq_stop(void *arg)
 129 {
 130         condlog(3, "Stopping uev queue");
 131         pthread_mutex_lock(uevq_lockp);
 132         my_uev_trigger = NULL;
 133         pthread_cond_signal(uev_condp);
 134         pthread_mutex_unlock(uevq_lockp);
 135 }
 136
 137 void
 138 uevq_cleanup(struct list_head *tmpq)
 139 {
 140         struct uevent *uev, *tmp;
 141
 142         list_for_each_entry_safe(uev, tmp, tmpq, node) {
 143                 list_del_init(&uev->node);
 144                 FREE(uev);
 145         }
 146 }
 147
 148 /*
 149  * Service the uevent queue.
 150  */
 151 int uevent_dispatch(int (*uev_trigger)(struct uevent *, void * trigger_data),
 152                     void * trigger_data)
 153 {
 154         my_uev_trigger = uev_trigger;
 155         my_trigger_data = trigger_data;
 156
 157         mlockall(MCL_CURRENT | MCL_FUTURE);
 158
 159         while (1) {
 160                 LIST_HEAD(uevq_tmp);
 161
 162                 pthread_mutex_lock(uevq_lockp);
 163                 servicing_uev = 0;
 164                 /*
 165                  * Condition signals are unreliable,
 166                  * so make sure we only wait if we have to.
 167                  */
 168                 if (list_empty(&uevq)) {
 169                         pthread_cond_wait(uev_condp, uevq_lockp);
 170                 }
 171                 servicing_uev = 1;
 172                 list_splice_init(&uevq, &uevq_tmp);
 173                 pthread_mutex_unlock(uevq_lockp);
 174                 if (!my_uev_trigger)
 175                         break;
 176                 service_uevq(&uevq_tmp);
 177         }
 178         condlog(3, "Terminating uev service queue");
 179         uevq_cleanup(&uevq);
 180         return 0;
 181 }
 182
 183 int failback_listen(void)
 184 {
 185         int sock;
 186         struct sockaddr_nl snl;
 187         struct sockaddr_un sun;
 188         socklen_t addrlen;
 189         int retval;
 190         int rcvbufsz = 128*1024;
 191         int rcvsz = 0;
 192         int rcvszsz = sizeof(rcvsz);
 193         unsigned int *prcvszsz = (unsigned int *)&rcvszsz;
 194         const int feature_on = 1;
 195         /*
 196          * First check whether we have a udev socket
 197          */
 198         memset(&sun, 0x00, sizeof(struct sockaddr_un));
 199         sun.sun_family = AF_LOCAL;
 200         strcpy(&sun.sun_path[1], "/org/kernel/dm/multipath_event");
 201         addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(sun.sun_path+1) + 1;
 202
 203         sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
 204         if (sock >= 0) {
 205
 206                 condlog(3, "reading events from udev socket.");
 207
 208                 /* the bind takes care of ensuring only one copy running */
 209                 retval = bind(sock, (struct sockaddr *) &sun, addrlen);
 210                 if (retval < 0) {
 211                         condlog(0, "bind failed, exit");
 212                         goto exit;
 213                 }
 214
 215                 /* enable receiving of the sender credentials */
 216                 setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 217                            &feature_on, sizeof(feature_on));
 218
 219         } else {
 220                 /* Fallback to read kernel netlink events */
 221                 memset(&snl, 0x00, sizeof(struct sockaddr_nl));
 222                 snl.nl_family = AF_NETLINK;
 223                 snl.nl_pid = getpid();
 224                 snl.nl_groups = 0x01;
 225
 226                 sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
 227                 if (sock == -1) {
 228                         condlog(0, "error getting socket, exit");
 229                         return 1;
 230                 }
 231
 232                 condlog(3, "reading events from kernel.");
 233
 234                 /*
 235                  * try to avoid dropping uevents, even so, this is not a guarantee,
 236                  * but it does help to change the netlink uevent socket's
 237                  * receive buffer threshold from the default value of 106,496 to
 238                  * the maximum value of 262,142.
 239                  */
 240                 retval = setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsz,
 241                                     sizeof(rcvbufsz));
 242
 243                 if (retval < 0) {
 244                         condlog(0, "error setting receive buffer size for socket, exit");
 245                         exit(1);
 246                 }
 247                 retval = getsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvsz, prcvszsz);
 248                 if (retval < 0) {
 249                         condlog(0, "error setting receive buffer size for socket, exit");
 250                         exit(1);
 251                 }
 252                 condlog(3, "receive buffer size for socket is %u.", rcvsz);
 253
 254                 /* enable receiving of the sender credentials */
 255                 setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 256                            &feature_on, sizeof(feature_on));
 257
 258                 retval = bind(sock, (struct sockaddr *) &snl,
 259                               sizeof(struct sockaddr_nl));
 260                 if (retval < 0) {
 261                         condlog(0, "bind failed, exit");
 262                         goto exit;
 263                 }
 264         }
 265
 266         while (1) {
 267                 int i;
 268                 char *pos;
 269                 size_t bufpos;
 270                 ssize_t buflen;
 271                 struct uevent *uev;
 272                 char *buffer;
 273                 struct msghdr smsg;
 274                 struct iovec iov;
 275                 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
 276                 struct cmsghdr *cmsg;
 277                 struct ucred *cred;
 278                 static char buf[HOTPLUG_BUFFER_SIZE + OBJECT_SIZE];
 279
 280                 memset(buf, 0x00, sizeof(buf));
 281                 iov.iov_base = &buf;
 282                 iov.iov_len = sizeof(buf);
 283                 memset (&smsg, 0x00, sizeof(struct msghdr));
 284                 smsg.msg_iov = &iov;
 285                 smsg.msg_iovlen = 1;
 286                 smsg.msg_control = cred_msg;
 287                 smsg.msg_controllen = sizeof(cred_msg);
 288
 289                 buflen = recvmsg(sock, &smsg, 0);
 290                 if (buflen < 0) {
 291                         if (errno != EINTR)
 292                                 condlog(0, "error receiving message, errno %d", errno);
 293                         continue;
 294                 }
 295
 296                 cmsg = CMSG_FIRSTHDR(&smsg);
 297                 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
 298                         condlog(3, "no sender credentials received, message ignored");
 299                         continue;
 300                 }
 301
 302                 cred = (struct ucred *)CMSG_DATA(cmsg);
 303                 if (cred->uid != 0) {
 304                         condlog(3, "sender uid=%d, message ignored", cred->uid);
 305                         continue;
 306                 }
 307
 308                 /* skip header */
 309                 bufpos = strlen(buf) + 1;
 310                 if (bufpos < sizeof("a@/d") || bufpos >= sizeof(buf)) {
 311                         condlog(3, "invalid message length");
 312                         continue;
 313                 }
 314
 315                 /* check message header */
 316                 if (strstr(buf, "@/") == NULL) {
 317                         condlog(3, "unrecognized message header");
 318                         continue;
 319                 }
 320                 if ((size_t)buflen > sizeof(buf)-1) {
 321                         condlog(2, "buffer overflow for received uevent");
 322                         buflen = sizeof(buf)-1;
 323                 }
 324
 325                 uev = alloc_uevent();
 326
 327                 if (!uev) {
 328                         condlog(1, "lost uevent, oom");
 329                         continue;
 330                 }
 331
 332                 if ((size_t)buflen > sizeof(buf)-1)
 333                         buflen = sizeof(buf)-1;
 334
 335                 /*
 336                  * Copy the shared receive buffer contents to buffer private
 337                  * to this uevent so we can immediately reuse the shared buffer.
 338                  */
 339                 memcpy(uev->buffer, buf, HOTPLUG_BUFFER_SIZE + OBJECT_SIZE);
 340                 buffer = uev->buffer;
 341                 buffer[buflen] = '\0';
 342
 343                 /* save start of payload */
 344                 bufpos = strlen(buffer) + 1;
 345
 346                 /* action string */
 347                 uev->action = buffer;
 348                 pos = strchr(buffer, '@');
 349                 if (!pos) {
 350                         condlog(3, "bad action string '%s'", buffer);
 351                         continue;
 352                 }
 353                 pos[0] = '\0';
 354
 355                 /* sysfs path */
 356                 uev->devpath = &pos[1];
 357
 358                 /* hotplug events have the environment attached - reconstruct envp[] */
 359                 for (i = 0; (bufpos < (size_t)buflen) && (i < HOTPLUG_NUM_ENVP-1); i++) {
 360                         int keylen;
 361                         char *key;
 362
 363                         key = &buffer[bufpos];
 364                         keylen = strlen(key);
 365                         uev->envp[i] = key;
 366                         /* Filter out sequence number */
 367                         if (strncmp(key, "SEQNUM=", 7) == 0) {
 368                                 char *eptr;
 369
 370                                 uev->seqnum = strtoul(key + 7, &eptr, 10);
 371                                 if (eptr == key + 7)
 372                                         uev->seqnum = -1;
 373                         }
 374                         bufpos += keylen + 1;
 375                 }
 376                 uev->envp[i] = NULL;
 377
 378                 condlog(3, "uevent %ld '%s' from '%s'", uev->seqnum,
 379                         uev->action, uev->devpath);
 380                 uev->kernel = strrchr(uev->devpath, '/');
 381                 if (uev->kernel)
 382                         uev->kernel++;
 383
 384                 /* print payload environment */
 385                 for (i = 0; uev->envp[i] != NULL; i++)
 386                         condlog(5, "%s", uev->envp[i]);
 387
 388                 /*
 389                  * Queue uevent and poke service pthread.
 390                  */
 391                 pthread_mutex_lock(uevq_lockp);
 392                 list_add_tail(&uev->node, &uevq);
 393                 pthread_cond_signal(uev_condp);
 394                 pthread_mutex_unlock(uevq_lockp);
 395         }
 396
 397 exit:
 398         close(sock);
 399         return 1;
 400 }
 401
 402 int uevent_listen(void)
 403 {
 404         int err;
 405         struct udev_monitor *monitor = NULL;
 406         int fd, socket_flags;
 407         int need_failback = 1;
 408         /*
 409          * Queue uevents for service by dedicated thread so that the uevent
 410          * listening thread does not block on multipathd locks (vecs->lock)
 411          * thereby not getting to empty the socket's receive buffer queue
 412          * often enough.
 413          */
 414         pthread_cleanup_push(uevq_stop, NULL);
 415
 416         monitor = udev_monitor_new_from_netlink(conf->udev, "udev");
 417         if (!monitor) {
 418                 condlog(2, "failed to create udev monitor");
 419                 goto out;
 420         }
 421 #ifdef LIBUDEV_API_RECVBUF
 422         if (udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024))
 423                 condlog(2, "failed to increase buffer size");
 424 #endif
 425         fd = udev_monitor_get_fd(monitor);
 426         if (fd < 0) {
 427                 condlog(2, "failed to get monitor fd");
 428                 goto out;
 429         }
 430         socket_flags = fcntl(fd, F_GETFL);
 431         if (socket_flags < 0) {
 432                 condlog(2, "failed to get monitor socket flags : %s",
 433                         strerror(errno));
 434                 goto out;
 435         }
 436         if (fcntl(fd, F_SETFL, socket_flags & ~O_NONBLOCK) < 0) {
 437                 condlog(2, "failed to set monitor socket flags : %s",
 438                         strerror(errno));
 439                 goto out;
 440         }
 441         err = udev_monitor_filter_add_match_subsystem_devtype(monitor, "block",
 442                                                               NULL);
 443         if (err)
 444                 condlog(2, "failed to create filter : %s", strerror(-err));
 445         err = udev_monitor_enable_receiving(monitor);
 446         if (err) {
 447                 condlog(2, "failed to enable receiving : %s", strerror(-err));
 448                 goto out;
 449         }
 450         while (1) {
 451                 int i = 0;
 452                 char *pos, *end;
 453                 struct uevent *uev;
 454                 struct udev_device *dev;
 455                 struct udev_list_entry *list_entry;
 456
 457                 dev = udev_monitor_receive_device(monitor);
 458                 if (!dev) {
 459                         condlog(0, "failed getting udev device");
 460                         continue;
 461                 }
 462
 463                 uev = alloc_uevent();
 464                 if (!uev) {
 465                         udev_device_unref(dev);
 466                         condlog(1, "lost uevent, oom");
 467                         continue;
 468                 }
 469                 pos = uev->buffer;
 470                 end = pos + HOTPLUG_BUFFER_SIZE + OBJECT_SIZE - 1;
 471                 udev_list_entry_foreach(list_entry, udev_device_get_properties_list_entry(dev)) {
 472                         const char *name, *value;
 473                         int bytes;
 474
 475                         name = udev_list_entry_get_name(list_entry);
 476                         if (!name)
 477                                 name = "(null)";
 478                         value = udev_list_entry_get_value(list_entry);
 479                         if (!value)
 480                                 value = "(null)";
 481                         bytes = snprintf(pos, end - pos, "%s=%s", name,
 482                                         value);
 483                         if (pos + bytes >= end) {
 484                                 condlog(2, "buffer overflow for uevent");
 485                                 break;
 486                         }
 487                         uev->envp[i] = pos;
 488                         pos += bytes;
 489                         *pos = '\0';
 490                         pos++;
 491                         if (strcmp(name, "DEVPATH") == 0)
 492                                 uev->devpath = uev->envp[i] + 8;
 493                         if (strcmp(name, "ACTION") == 0)
 494                                 uev->action = uev->envp[i] + 7;
 495                         i++;
 496                         if (i == HOTPLUG_NUM_ENVP - 1)
 497                                 break;
 498                 }
 499                 uev->udev = dev;
 500                 uev->envp[i] = NULL;
 501
 502                 condlog(3, "uevent '%s' from '%s'", uev->action, uev->devpath);
 503                 uev->kernel = strrchr(uev->devpath, '/');
 504                 if (uev->kernel)
 505                         uev->kernel++;
 506
 507                 /* print payload environment */
 508                 for (i = 0; uev->envp[i] != NULL; i++)
 509                         condlog(5, "%s", uev->envp[i]);
 510
 511                 /*
 512                  * Queue uevent and poke service pthread.
 513                  */
 514                 pthread_mutex_lock(uevq_lockp);
 515                 list_add_tail(&uev->node, &uevq);
 516                 pthread_cond_signal(uev_condp);
 517                 pthread_mutex_unlock(uevq_lockp);
 518         }
 519         need_failback = 0;
 520 out:
 521         if (monitor)
 522                 udev_monitor_unref(monitor);
 523         if (need_failback)
 524                 err = failback_listen();
 525         pthread_cleanup_pop(1);
 526         return err;
 527 }
 528
 529 extern int
 530 uevent_get_major(struct uevent *uev)
 531 {
 532         char *p, *q;
 533         int i, major = -1;
 534
 535         for (i = 0; uev->envp[i] != NULL; i++) {
 536                 if (!strncmp(uev->envp[i], "MAJOR", 5) && strlen(uev->envp[i]) > 6) {
 537                         p = uev->envp[i] + 6;
 538                         major = strtoul(p, &q, 10);
 539                         if (p == q) {
 540                                 condlog(2, "invalid major '%s'", p);
 541                                 major = -1;
 542                         }
 543                         break;
 544                 }
 545         }
 546         return major;
 547 }
 548
 549 extern int
 550 uevent_get_minor(struct uevent *uev)
 551 {
 552         char *p, *q;
 553         int i, minor = -1;
 554
 555         for (i = 0; uev->envp[i] != NULL; i++) {
 556                 if (!strncmp(uev->envp[i], "MINOR", 5) && strlen(uev->envp[i]) > 6) {
 557                         p = uev->envp[i] + 6;
 558                         minor = strtoul(p, &q, 10);
 559                         if (p == q) {
 560                                 condlog(2, "invalid minor '%s'", p);
 561                                 minor = -1;
 562                         }
 563                         break;
 564                 }
 565         }
 566         return minor;
 567 }
 568
 569 extern int
 570 uevent_get_disk_ro(struct uevent *uev)
 571 {
 572         char *p, *q;
 573         int i, ro = -1;
 574
 575         for (i = 0; uev->envp[i] != NULL; i++) {
 576                 if (!strncmp(uev->envp[i], "DISK_RO", 6) && strlen(uev->envp[i]) > 7) {
 577                         p = uev->envp[i] + 8;
 578                         ro = strtoul(p, &q, 10);
 579                         if (p == q) {
 580                                 condlog(2, "invalid read_only setting '%s'", p);
 581                                 ro = -1;
 582                         }
 583                         break;
 584                 }
 585         }
 586         return ro;
 587 }
 588
 589 extern char *
 590 uevent_get_dm_name(struct uevent *uev)
 591 {
 592         char *p = NULL;
 593         int i;
 594
 595         for (i = 0; uev->envp[i] != NULL; i++) {
 596                 if (!strncmp(uev->envp[i], "DM_NAME", 6) &&
 597                     strlen(uev->envp[i]) > 7) {
 598                         p = MALLOC(strlen(uev->envp[i] + 8) + 1);
 599                         strcpy(p, uev->envp[i] + 8);
 600                         break;
 601                 }
 602         }
 603         return p;
 604 }