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, see <http://www.gnu.org/licenses/>.
  22  *
  23  */
  24
  25 #include <unistd.h>
  26 #include <stdio.h>
  27 #include <errno.h>
  28 #include <stdlib.h>
  29 #include <stddef.h>
  30 #include <string.h>
  31 #include <fcntl.h>
  32 #include <time.h>
  33 #include <sys/socket.h>
  34 #include <sys/user.h>
  35 #include <sys/un.h>
  36 #include <poll.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
  51 typedef int (uev_trigger)(struct uevent *, void * trigger_data);
  52
  53 pthread_t uevq_thr;
  54 LIST_HEAD(uevq);
  55 pthread_mutex_t uevq_lock = PTHREAD_MUTEX_INITIALIZER;
  56 pthread_mutex_t *uevq_lockp = &uevq_lock;
  57 pthread_cond_t uev_cond = PTHREAD_COND_INITIALIZER;
  58 pthread_cond_t *uev_condp = &uev_cond;
  59 uev_trigger *my_uev_trigger;
  60 void * my_trigger_data;
  61 int servicing_uev;
  62
  63 int is_uevent_busy(void)
  64 {
  65         int empty;
  66
  67         pthread_mutex_lock(uevq_lockp);
  68         empty = list_empty(&uevq);
  69         pthread_mutex_unlock(uevq_lockp);
  70         return (!empty || servicing_uev);
  71 }
  72
  73 struct uevent * alloc_uevent (void)
  74 {
  75         struct uevent *uev = MALLOC(sizeof(struct uevent));
  76
  77         if (uev)
  78                 INIT_LIST_HEAD(&uev->node);
  79
  80         return uev;
  81 }
  82
  83 void
  84 setup_thread_attr(pthread_attr_t *attr, size_t stacksize, int detached)
  85 {
  86         if (pthread_attr_init(attr)) {
  87                 fprintf(stderr, "can't initialize thread attr: %s\n",
  88                         strerror(errno));
  89                 exit(1);
  90         }
  91         if (stacksize < PTHREAD_STACK_MIN)
  92                 stacksize = PTHREAD_STACK_MIN;
  93
  94         if (pthread_attr_setstacksize(attr, stacksize)) {
  95                 fprintf(stderr, "can't set thread stack size to %lu: %s\n",
  96                         (unsigned long)stacksize, strerror(errno));
  97                 exit(1);
  98         }
  99         if (detached && pthread_attr_setdetachstate(attr,
 100                                                     PTHREAD_CREATE_DETACHED)) {
 101                 fprintf(stderr, "can't set thread to detached: %s\n",
 102                         strerror(errno));
 103                 exit(1);
 104         }
 105 }
 106
 107 /*
 108  * Called with uevq_lockp held
 109  */
 110 void
 111 service_uevq(struct list_head *tmpq)
 112 {
 113         struct uevent *uev, *tmp;
 114
 115         list_for_each_entry_safe(uev, tmp, tmpq, node) {
 116                 list_del_init(&uev->node);
 117
 118                 if (my_uev_trigger && my_uev_trigger(uev, my_trigger_data))
 119                         condlog(0, "uevent trigger error");
 120
 121                 if (uev->udev)
 122                         udev_device_unref(uev->udev);
 123                 FREE(uev);
 124         }
 125 }
 126
 127 static void uevq_stop(void *arg)
 128 {
 129         struct udev *udev = arg;
 130
 131         condlog(3, "Stopping uev queue");
 132         pthread_mutex_lock(uevq_lockp);
 133         my_uev_trigger = NULL;
 134         pthread_cond_signal(uev_condp);
 135         pthread_mutex_unlock(uevq_lockp);
 136         udev_unref(udev);
 137 }
 138
 139 void
 140 uevq_cleanup(struct list_head *tmpq)
 141 {
 142         struct uevent *uev, *tmp;
 143
 144         list_for_each_entry_safe(uev, tmp, tmpq, node) {
 145                 list_del_init(&uev->node);
 146                 FREE(uev);
 147         }
 148 }
 149
 150 /*
 151  * Service the uevent queue.
 152  */
 153 int uevent_dispatch(int (*uev_trigger)(struct uevent *, void * trigger_data),
 154                     void * trigger_data)
 155 {
 156         my_uev_trigger = uev_trigger;
 157         my_trigger_data = trigger_data;
 158
 159         mlockall(MCL_CURRENT | MCL_FUTURE);
 160
 161         while (1) {
 162                 LIST_HEAD(uevq_tmp);
 163
 164                 pthread_mutex_lock(uevq_lockp);
 165                 servicing_uev = 0;
 166                 /*
 167                  * Condition signals are unreliable,
 168                  * so make sure we only wait if we have to.
 169                  */
 170                 if (list_empty(&uevq)) {
 171                         pthread_cond_wait(uev_condp, uevq_lockp);
 172                 }
 173                 servicing_uev = 1;
 174                 list_splice_init(&uevq, &uevq_tmp);
 175                 pthread_mutex_unlock(uevq_lockp);
 176                 if (!my_uev_trigger)
 177                         break;
 178                 service_uevq(&uevq_tmp);
 179         }
 180         condlog(3, "Terminating uev service queue");
 181         uevq_cleanup(&uevq);
 182         return 0;
 183 }
 184
 185 struct uevent *uevent_from_buffer(char *buf, ssize_t buflen)
 186 {
 187         struct uevent *uev;
 188         char *buffer;
 189         size_t bufpos;
 190         int i;
 191         char *pos;
 192
 193         uev = alloc_uevent();
 194         if (!uev) {
 195                 condlog(1, "lost uevent, oom");
 196                 return NULL;
 197         }
 198
 199         if ((size_t)buflen > sizeof(buf)-1)
 200                 buflen = sizeof(buf)-1;
 201
 202         /*
 203          * Copy the shared receive buffer contents to buffer private
 204          * to this uevent so we can immediately reuse the shared buffer.
 205          */
 206         memcpy(uev->buffer, buf, HOTPLUG_BUFFER_SIZE + OBJECT_SIZE);
 207         buffer = uev->buffer;
 208         buffer[buflen] = '\0';
 209
 210         /* save start of payload */
 211         bufpos = strlen(buffer) + 1;
 212
 213         /* action string */
 214         uev->action = buffer;
 215         pos = strchr(buffer, '@');
 216         if (!pos) {
 217                 condlog(3, "bad action string '%s'", buffer);
 218                 FREE(uev);
 219                 return NULL;
 220         }
 221         pos[0] = '\0';
 222
 223         /* sysfs path */
 224         uev->devpath = &pos[1];
 225
 226         /* hotplug events have the environment attached - reconstruct envp[] */
 227         for (i = 0; (bufpos < (size_t)buflen) && (i < HOTPLUG_NUM_ENVP-1); i++) {
 228                 int keylen;
 229                 char *key;
 230
 231                 key = &buffer[bufpos];
 232                 keylen = strlen(key);
 233                 uev->envp[i] = key;
 234                 /* Filter out sequence number */
 235                 if (strncmp(key, "SEQNUM=", 7) == 0) {
 236                         char *eptr;
 237
 238                         uev->seqnum = strtoul(key + 7, &eptr, 10);
 239                         if (eptr == key + 7)
 240                                 uev->seqnum = -1;
 241                 }
 242                 bufpos += keylen + 1;
 243         }
 244         uev->envp[i] = NULL;
 245
 246         condlog(3, "uevent %ld '%s' from '%s'", uev->seqnum,
 247                 uev->action, uev->devpath);
 248         uev->kernel = strrchr(uev->devpath, '/');
 249         if (uev->kernel)
 250                 uev->kernel++;
 251
 252         /* print payload environment */
 253         for (i = 0; uev->envp[i] != NULL; i++)
 254                 condlog(5, "%s", uev->envp[i]);
 255
 256         return uev;
 257 }
 258
 259 int failback_listen(void)
 260 {
 261         int sock;
 262         struct sockaddr_nl snl;
 263         struct sockaddr_un sun;
 264         socklen_t addrlen;
 265         int retval;
 266         int rcvbufsz = 128*1024;
 267         int rcvsz = 0;
 268         int rcvszsz = sizeof(rcvsz);
 269         unsigned int *prcvszsz = (unsigned int *)&rcvszsz;
 270         const int feature_on = 1;
 271         /*
 272          * First check whether we have a udev socket
 273          */
 274         memset(&sun, 0x00, sizeof(struct sockaddr_un));
 275         sun.sun_family = AF_LOCAL;
 276         strcpy(&sun.sun_path[1], "/org/kernel/dm/multipath_event");
 277         addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(sun.sun_path+1) + 1;
 278
 279         sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
 280         if (sock >= 0) {
 281
 282                 condlog(3, "reading events from udev socket.");
 283
 284                 /* the bind takes care of ensuring only one copy running */
 285                 retval = bind(sock, (struct sockaddr *) &sun, addrlen);
 286                 if (retval < 0) {
 287                         condlog(0, "bind failed, exit");
 288                         goto exit;
 289                 }
 290
 291                 /* enable receiving of the sender credentials */
 292                 retval = setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 293                                     &feature_on, sizeof(feature_on));
 294                 if (retval < 0) {
 295                         condlog(0, "failed to enable credential passing, exit");
 296                         goto exit;
 297                 }
 298
 299         } else {
 300                 /* Fallback to read kernel netlink events */
 301                 memset(&snl, 0x00, sizeof(struct sockaddr_nl));
 302                 snl.nl_family = AF_NETLINK;
 303                 snl.nl_pid = getpid();
 304                 snl.nl_groups = 0x01;
 305
 306                 sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
 307                 if (sock == -1) {
 308                         condlog(0, "error getting socket, exit");
 309                         return 1;
 310                 }
 311
 312                 condlog(3, "reading events from kernel.");
 313
 314                 /*
 315                  * try to avoid dropping uevents, even so, this is not a guarantee,
 316                  * but it does help to change the netlink uevent socket's
 317                  * receive buffer threshold from the default value of 106,496 to
 318                  * the maximum value of 262,142.
 319                  */
 320                 retval = setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsz,
 321                                     sizeof(rcvbufsz));
 322
 323                 if (retval < 0) {
 324                         condlog(0, "error setting receive buffer size for socket, exit");
 325                         exit(1);
 326                 }
 327                 retval = getsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvsz, prcvszsz);
 328                 if (retval < 0) {
 329                         condlog(0, "error setting receive buffer size for socket, exit");
 330                         exit(1);
 331                 }
 332                 condlog(3, "receive buffer size for socket is %u.", rcvsz);
 333
 334                 /* enable receiving of the sender credentials */
 335                 if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 336                                &feature_on, sizeof(feature_on)) < 0) {
 337                         condlog(0, "error on enabling credential passing for socket");
 338                         exit(1);
 339                 }
 340
 341                 retval = bind(sock, (struct sockaddr *) &snl,
 342                               sizeof(struct sockaddr_nl));
 343                 if (retval < 0) {
 344                         condlog(0, "bind failed, exit");
 345                         goto exit;
 346                 }
 347         }
 348
 349         while (1) {
 350                 size_t bufpos;
 351                 ssize_t buflen;
 352                 struct uevent *uev;
 353                 struct msghdr smsg;
 354                 struct iovec iov;
 355                 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
 356                 struct cmsghdr *cmsg;
 357                 struct ucred *cred;
 358                 static char buf[HOTPLUG_BUFFER_SIZE + OBJECT_SIZE];
 359
 360                 memset(buf, 0x00, sizeof(buf));
 361                 iov.iov_base = &buf;
 362                 iov.iov_len = sizeof(buf);
 363                 memset (&smsg, 0x00, sizeof(struct msghdr));
 364                 smsg.msg_iov = &iov;
 365                 smsg.msg_iovlen = 1;
 366                 smsg.msg_control = cred_msg;
 367                 smsg.msg_controllen = sizeof(cred_msg);
 368
 369                 buflen = recvmsg(sock, &smsg, 0);
 370                 if (buflen < 0) {
 371                         if (errno != EINTR)
 372                                 condlog(0, "error receiving message, errno %d", errno);
 373                         continue;
 374                 }
 375
 376                 cmsg = CMSG_FIRSTHDR(&smsg);
 377                 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
 378                         condlog(3, "no sender credentials received, message ignored");
 379                         continue;
 380                 }
 381
 382                 cred = (struct ucred *)CMSG_DATA(cmsg);
 383                 if (cred->uid != 0) {
 384                         condlog(3, "sender uid=%d, message ignored", cred->uid);
 385                         continue;
 386                 }
 387
 388                 /* skip header */
 389                 bufpos = strlen(buf) + 1;
 390                 if (bufpos < sizeof("a@/d") || bufpos >= sizeof(buf)) {
 391                         condlog(3, "invalid message length");
 392                         continue;
 393                 }
 394
 395                 /* check message header */
 396                 if (strstr(buf, "@/") == NULL) {
 397                         condlog(3, "unrecognized message header");
 398                         continue;
 399                 }
 400                 if ((size_t)buflen > sizeof(buf)-1) {
 401                         condlog(2, "buffer overflow for received uevent");
 402                         buflen = sizeof(buf)-1;
 403                 }
 404
 405                 uev = uevent_from_buffer(buf, buflen);
 406                 if (!uev)
 407                         continue;
 408                 /*
 409                  * Queue uevent and poke service pthread.
 410                  */
 411                 pthread_mutex_lock(uevq_lockp);
 412                 list_add_tail(&uev->node, &uevq);
 413                 pthread_cond_signal(uev_condp);
 414                 pthread_mutex_unlock(uevq_lockp);
 415         }
 416
 417 exit:
 418         close(sock);
 419         return 1;
 420 }
 421
 422 struct uevent *uevent_from_udev_device(struct udev_device *dev)
 423 {
 424         struct uevent *uev;
 425         int i = 0;
 426         char *pos, *end;
 427         struct udev_list_entry *list_entry;
 428
 429         uev = alloc_uevent();
 430         if (!uev) {
 431                 udev_device_unref(dev);
 432                 condlog(1, "lost uevent, oom");
 433                 return NULL;
 434         }
 435         pos = uev->buffer;
 436         end = pos + HOTPLUG_BUFFER_SIZE + OBJECT_SIZE - 1;
 437         udev_list_entry_foreach(list_entry, udev_device_get_properties_list_entry(dev)) {
 438                 const char *name, *value;
 439                 int bytes;
 440
 441                 name = udev_list_entry_get_name(list_entry);
 442                 if (!name)
 443                         name = "(null)";
 444                 value = udev_list_entry_get_value(list_entry);
 445                 if (!value)
 446                         value = "(null)";
 447                 bytes = snprintf(pos, end - pos, "%s=%s", name, value);
 448                 if (pos + bytes >= end) {
 449                         condlog(2, "buffer overflow for uevent");
 450                         break;
 451                 }
 452                 uev->envp[i] = pos;
 453                 pos += bytes;
 454                 *pos = '\0';
 455                 pos++;
 456                 if (strcmp(name, "DEVPATH") == 0)
 457                         uev->devpath = uev->envp[i] + 8;
 458                 if (strcmp(name, "ACTION") == 0)
 459                         uev->action = uev->envp[i] + 7;
 460                 i++;
 461                 if (i == HOTPLUG_NUM_ENVP - 1)
 462                         break;
 463         }
 464         uev->udev = dev;
 465         uev->envp[i] = NULL;
 466
 467         condlog(3, "uevent '%s' from '%s'", uev->action, uev->devpath);
 468         uev->kernel = strrchr(uev->devpath, '/');
 469         if (uev->kernel)
 470                 uev->kernel++;
 471
 472         /* print payload environment */
 473         for (i = 0; uev->envp[i] != NULL; i++)
 474                 condlog(5, "%s", uev->envp[i]);
 475         return uev;
 476 }
 477
 478 int uevent_listen(struct udev *udev)
 479 {
 480         int err = 2;
 481         struct udev_monitor *monitor = NULL;
 482         int fd, socket_flags, events;
 483         int need_failback = 1;
 484         int timeout = 30;
 485         LIST_HEAD(uevlisten_tmp);
 486
 487         /*
 488          * Queue uevents for service by dedicated thread so that the uevent
 489          * listening thread does not block on multipathd locks (vecs->lock)
 490          * thereby not getting to empty the socket's receive buffer queue
 491          * often enough.
 492          */
 493         if (!udev) {
 494                 condlog(1, "no udev context");
 495                 return 1;
 496         }
 497         udev_ref(udev);
 498         pthread_cleanup_push(uevq_stop, udev);
 499
 500         monitor = udev_monitor_new_from_netlink(udev, "udev");
 501         if (!monitor) {
 502                 condlog(2, "failed to create udev monitor");
 503                 goto out;
 504         }
 505 #ifdef LIBUDEV_API_RECVBUF
 506         if (udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024))
 507                 condlog(2, "failed to increase buffer size");
 508 #endif
 509         fd = udev_monitor_get_fd(monitor);
 510         if (fd < 0) {
 511                 condlog(2, "failed to get monitor fd");
 512                 goto out;
 513         }
 514         socket_flags = fcntl(fd, F_GETFL);
 515         if (socket_flags < 0) {
 516                 condlog(2, "failed to get monitor socket flags : %s",
 517                         strerror(errno));
 518                 goto out;
 519         }
 520         if (fcntl(fd, F_SETFL, socket_flags & ~O_NONBLOCK) < 0) {
 521                 condlog(2, "failed to set monitor socket flags : %s",
 522                         strerror(errno));
 523                 goto out;
 524         }
 525         err = udev_monitor_filter_add_match_subsystem_devtype(monitor, "block",
 526                                                               NULL);
 527         if (err)
 528                 condlog(2, "failed to create filter : %s", strerror(-err));
 529         err = udev_monitor_enable_receiving(monitor);
 530         if (err) {
 531                 condlog(2, "failed to enable receiving : %s", strerror(-err));
 532                 goto out;
 533         }
 534
 535         events = 0;
 536         while (1) {
 537                 struct uevent *uev;
 538                 struct udev_device *dev;
 539                 struct pollfd ev_poll;
 540                 int poll_timeout;
 541                 int fdcount;
 542
 543                 memset(&ev_poll, 0, sizeof(struct pollfd));
 544                 ev_poll.fd = fd;
 545                 ev_poll.events = POLLIN;
 546                 poll_timeout = timeout * 1000;
 547                 errno = 0;
 548                 fdcount = poll(&ev_poll, 1, poll_timeout);
 549                 if (fdcount && ev_poll.revents & POLLIN) {
 550                         timeout = 0;
 551                         dev = udev_monitor_receive_device(monitor);
 552                         if (!dev) {
 553                                 condlog(0, "failed getting udev device");
 554                                 continue;
 555                         }
 556                         uev = uevent_from_udev_device(dev);
 557                         if (!uev)
 558                                 continue;
 559                         list_add_tail(&uev->node, &uevlisten_tmp);
 560                         events++;
 561                         continue;
 562                 }
 563                 if (fdcount < 0) {
 564                         if (errno == EINTR)
 565                                 continue;
 566
 567                         condlog(0, "error receiving "
 568                                 "uevent message: %m");
 569                         err = -errno;
 570                         break;
 571                 }
 572                 if (!list_empty(&uevlisten_tmp)) {
 573                         /*
 574                          * Queue uevents and poke service pthread.
 575                          */
 576                         condlog(3, "Forwarding %d uevents", events);
 577                         pthread_mutex_lock(uevq_lockp);
 578                         list_splice_tail_init(&uevlisten_tmp, &uevq);
 579                         pthread_cond_signal(uev_condp);
 580                         pthread_mutex_unlock(uevq_lockp);
 581                         events = 0;
 582                 }
 583                 timeout = 30;
 584         }
 585         need_failback = 0;
 586 out:
 587         if (monitor)
 588                 udev_monitor_unref(monitor);
 589         if (need_failback)
 590                 err = failback_listen();
 591         pthread_cleanup_pop(1);
 592         return err;
 593 }
 594
 595 extern int
 596 uevent_get_major(struct uevent *uev)
 597 {
 598         char *p, *q;
 599         int i, major = -1;
 600
 601         for (i = 0; uev->envp[i] != NULL; i++) {
 602                 if (!strncmp(uev->envp[i], "MAJOR", 5) && strlen(uev->envp[i]) > 6) {
 603                         p = uev->envp[i] + 6;
 604                         major = strtoul(p, &q, 10);
 605                         if (p == q) {
 606                                 condlog(2, "invalid major '%s'", p);
 607                                 major = -1;
 608                         }
 609                         break;
 610                 }
 611         }
 612         return major;
 613 }
 614
 615 extern int
 616 uevent_get_minor(struct uevent *uev)
 617 {
 618         char *p, *q;
 619         int i, minor = -1;
 620
 621         for (i = 0; uev->envp[i] != NULL; i++) {
 622                 if (!strncmp(uev->envp[i], "MINOR", 5) && strlen(uev->envp[i]) > 6) {
 623                         p = uev->envp[i] + 6;
 624                         minor = strtoul(p, &q, 10);
 625                         if (p == q) {
 626                                 condlog(2, "invalid minor '%s'", p);
 627                                 minor = -1;
 628                         }
 629                         break;
 630                 }
 631         }
 632         return minor;
 633 }
 634
 635 extern int
 636 uevent_get_disk_ro(struct uevent *uev)
 637 {
 638         char *p, *q;
 639         int i, ro = -1;
 640
 641         for (i = 0; uev->envp[i] != NULL; i++) {
 642                 if (!strncmp(uev->envp[i], "DISK_RO", 6) && strlen(uev->envp[i]) > 7) {
 643                         p = uev->envp[i] + 8;
 644                         ro = strtoul(p, &q, 10);
 645                         if (p == q) {
 646                                 condlog(2, "invalid read_only setting '%s'", p);
 647                                 ro = -1;
 648                         }
 649                         break;
 650                 }
 651         }
 652         return ro;
 653 }
 654
 655 extern char *
 656 uevent_get_dm_name(struct uevent *uev)
 657 {
 658         char *p = NULL;
 659         int i;
 660
 661         for (i = 0; uev->envp[i] != NULL; i++) {
 662                 if (!strncmp(uev->envp[i], "DM_NAME", 6) &&
 663                     strlen(uev->envp[i]) > 7) {
 664                         p = MALLOC(strlen(uev->envp[i] + 8) + 1);
 665                         strcpy(p, uev->envp[i] + 8);
 666                         break;
 667                 }
 668         }
 669         return p;
 670 }