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 <sys/mman.h>
  41 #include <libudev.h>
  42 #include <errno.h>
  43
  44 #include "memory.h"
  45 #include "debug.h"
  46 #include "list.h"
  47 #include "uevent.h"
  48 #include "vector.h"
  49
  50 typedef int (uev_trigger)(struct uevent *, void * trigger_data);
  51
  52 LIST_HEAD(uevq);
  53 pthread_mutex_t uevq_lock = PTHREAD_MUTEX_INITIALIZER;
  54 pthread_mutex_t *uevq_lockp = &uevq_lock;
  55 pthread_cond_t uev_cond = PTHREAD_COND_INITIALIZER;
  56 pthread_cond_t *uev_condp = &uev_cond;
  57 uev_trigger *my_uev_trigger;
  58 void * my_trigger_data;
  59 int servicing_uev;
  60
  61 int is_uevent_busy(void)
  62 {
  63         int empty;
  64
  65         pthread_mutex_lock(uevq_lockp);
  66         empty = list_empty(&uevq);
  67         pthread_mutex_unlock(uevq_lockp);
  68         return (!empty || servicing_uev);
  69 }
  70
  71 struct uevent * alloc_uevent (void)
  72 {
  73         struct uevent *uev = MALLOC(sizeof(struct uevent));
  74
  75         if (uev)
  76                 INIT_LIST_HEAD(&uev->node);
  77
  78         return uev;
  79 }
  80
  81 void
  82 service_uevq(struct list_head *tmpq)
  83 {
  84         struct uevent *uev, *tmp;
  85
  86         list_for_each_entry_safe(uev, tmp, tmpq, node) {
  87                 list_del_init(&uev->node);
  88
  89                 if (my_uev_trigger && my_uev_trigger(uev, my_trigger_data))
  90                         condlog(0, "uevent trigger error");
  91
  92                 if (uev->udev)
  93                         udev_device_unref(uev->udev);
  94                 FREE(uev);
  95         }
  96 }
  97
  98 static void uevent_cleanup(void *arg)
  99 {
 100         struct udev *udev = arg;
 101
 102         condlog(3, "Releasing uevent_listen() resources");
 103         udev_unref(udev);
 104 }
 105
 106 void
 107 uevq_cleanup(struct list_head *tmpq)
 108 {
 109         struct uevent *uev, *tmp;
 110
 111         list_for_each_entry_safe(uev, tmp, tmpq, node) {
 112                 list_del_init(&uev->node);
 113                 FREE(uev);
 114         }
 115 }
 116
 117 /*
 118  * Service the uevent queue.
 119  */
 120 int uevent_dispatch(int (*uev_trigger)(struct uevent *, void * trigger_data),
 121                     void * trigger_data)
 122 {
 123         my_uev_trigger = uev_trigger;
 124         my_trigger_data = trigger_data;
 125
 126         mlockall(MCL_CURRENT | MCL_FUTURE);
 127
 128         while (1) {
 129                 LIST_HEAD(uevq_tmp);
 130
 131                 pthread_mutex_lock(uevq_lockp);
 132                 servicing_uev = 0;
 133                 /*
 134                  * Condition signals are unreliable,
 135                  * so make sure we only wait if we have to.
 136                  */
 137                 if (list_empty(&uevq)) {
 138                         pthread_cond_wait(uev_condp, uevq_lockp);
 139                 }
 140                 servicing_uev = 1;
 141                 list_splice_init(&uevq, &uevq_tmp);
 142                 pthread_mutex_unlock(uevq_lockp);
 143                 if (!my_uev_trigger)
 144                         break;
 145                 service_uevq(&uevq_tmp);
 146         }
 147         condlog(3, "Terminating uev service queue");
 148         uevq_cleanup(&uevq);
 149         return 0;
 150 }
 151
 152 struct uevent *uevent_from_buffer(char *buf, ssize_t buflen)
 153 {
 154         struct uevent *uev;
 155         char *buffer;
 156         size_t bufpos;
 157         int i;
 158         char *pos;
 159
 160         uev = alloc_uevent();
 161         if (!uev) {
 162                 condlog(1, "lost uevent, oom");
 163                 return NULL;
 164         }
 165
 166         if ((size_t)buflen > sizeof(buf)-1)
 167                 buflen = sizeof(buf)-1;
 168
 169         /*
 170          * Copy the shared receive buffer contents to buffer private
 171          * to this uevent so we can immediately reuse the shared buffer.
 172          */
 173         memcpy(uev->buffer, buf, HOTPLUG_BUFFER_SIZE + OBJECT_SIZE);
 174         buffer = uev->buffer;
 175         buffer[buflen] = '\0';
 176
 177         /* save start of payload */
 178         bufpos = strlen(buffer) + 1;
 179
 180         /* action string */
 181         uev->action = buffer;
 182         pos = strchr(buffer, '@');
 183         if (!pos) {
 184                 condlog(3, "bad action string '%s'", buffer);
 185                 FREE(uev);
 186                 return NULL;
 187         }
 188         pos[0] = '\0';
 189
 190         /* sysfs path */
 191         uev->devpath = &pos[1];
 192
 193         /* hotplug events have the environment attached - reconstruct envp[] */
 194         for (i = 0; (bufpos < (size_t)buflen) && (i < HOTPLUG_NUM_ENVP-1); i++) {
 195                 int keylen;
 196                 char *key;
 197
 198                 key = &buffer[bufpos];
 199                 keylen = strlen(key);
 200                 uev->envp[i] = key;
 201                 /* Filter out sequence number */
 202                 if (strncmp(key, "SEQNUM=", 7) == 0) {
 203                         char *eptr;
 204
 205                         uev->seqnum = strtoul(key + 7, &eptr, 10);
 206                         if (eptr == key + 7)
 207                                 uev->seqnum = -1;
 208                 }
 209                 bufpos += keylen + 1;
 210         }
 211         uev->envp[i] = NULL;
 212
 213         condlog(3, "uevent %ld '%s' from '%s'", uev->seqnum,
 214                 uev->action, uev->devpath);
 215         uev->kernel = strrchr(uev->devpath, '/');
 216         if (uev->kernel)
 217                 uev->kernel++;
 218
 219         /* print payload environment */
 220         for (i = 0; uev->envp[i] != NULL; i++)
 221                 condlog(5, "%s", uev->envp[i]);
 222
 223         return uev;
 224 }
 225
 226 int failback_listen(void)
 227 {
 228         int sock;
 229         struct sockaddr_nl snl;
 230         struct sockaddr_un sun;
 231         socklen_t addrlen;
 232         int retval;
 233         int rcvbufsz = 128*1024;
 234         int rcvsz = 0;
 235         int rcvszsz = sizeof(rcvsz);
 236         unsigned int *prcvszsz = (unsigned int *)&rcvszsz;
 237         const int feature_on = 1;
 238         /*
 239          * First check whether we have a udev socket
 240          */
 241         memset(&sun, 0x00, sizeof(struct sockaddr_un));
 242         sun.sun_family = AF_LOCAL;
 243         strcpy(&sun.sun_path[1], "/org/kernel/dm/multipath_event");
 244         addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(sun.sun_path+1) + 1;
 245
 246         sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
 247         if (sock >= 0) {
 248
 249                 condlog(3, "reading events from udev socket.");
 250
 251                 /* the bind takes care of ensuring only one copy running */
 252                 retval = bind(sock, (struct sockaddr *) &sun, addrlen);
 253                 if (retval < 0) {
 254                         condlog(0, "bind failed, exit");
 255                         goto exit;
 256                 }
 257
 258                 /* enable receiving of the sender credentials */
 259                 retval = setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 260                                     &feature_on, sizeof(feature_on));
 261                 if (retval < 0) {
 262                         condlog(0, "failed to enable credential passing, exit");
 263                         goto exit;
 264                 }
 265
 266         } else {
 267                 /* Fallback to read kernel netlink events */
 268                 memset(&snl, 0x00, sizeof(struct sockaddr_nl));
 269                 snl.nl_family = AF_NETLINK;
 270                 snl.nl_pid = getpid();
 271                 snl.nl_groups = 0x01;
 272
 273                 sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
 274                 if (sock == -1) {
 275                         condlog(0, "error getting socket, exit");
 276                         return 1;
 277                 }
 278
 279                 condlog(3, "reading events from kernel.");
 280
 281                 /*
 282                  * try to avoid dropping uevents, even so, this is not a guarantee,
 283                  * but it does help to change the netlink uevent socket's
 284                  * receive buffer threshold from the default value of 106,496 to
 285                  * the maximum value of 262,142.
 286                  */
 287                 retval = setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsz,
 288                                     sizeof(rcvbufsz));
 289
 290                 if (retval < 0) {
 291                         condlog(0, "error setting receive buffer size for socket, exit");
 292                         exit(1);
 293                 }
 294                 retval = getsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvsz, prcvszsz);
 295                 if (retval < 0) {
 296                         condlog(0, "error setting receive buffer size for socket, exit");
 297                         exit(1);
 298                 }
 299                 condlog(3, "receive buffer size for socket is %u.", rcvsz);
 300
 301                 /* enable receiving of the sender credentials */
 302                 if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 303                                &feature_on, sizeof(feature_on)) < 0) {
 304                         condlog(0, "error on enabling credential passing for socket");
 305                         exit(1);
 306                 }
 307
 308                 retval = bind(sock, (struct sockaddr *) &snl,
 309                               sizeof(struct sockaddr_nl));
 310                 if (retval < 0) {
 311                         condlog(0, "bind failed, exit");
 312                         goto exit;
 313                 }
 314         }
 315
 316         while (1) {
 317                 size_t bufpos;
 318                 ssize_t buflen;
 319                 struct uevent *uev;
 320                 struct msghdr smsg;
 321                 struct iovec iov;
 322                 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
 323                 struct cmsghdr *cmsg;
 324                 struct ucred *cred;
 325                 static char buf[HOTPLUG_BUFFER_SIZE + OBJECT_SIZE];
 326
 327                 memset(buf, 0x00, sizeof(buf));
 328                 iov.iov_base = &buf;
 329                 iov.iov_len = sizeof(buf);
 330                 memset (&smsg, 0x00, sizeof(struct msghdr));
 331                 smsg.msg_iov = &iov;
 332                 smsg.msg_iovlen = 1;
 333                 smsg.msg_control = cred_msg;
 334                 smsg.msg_controllen = sizeof(cred_msg);
 335
 336                 buflen = recvmsg(sock, &smsg, 0);
 337                 if (buflen < 0) {
 338                         if (errno != EINTR)
 339                                 condlog(0, "error receiving message, errno %d", errno);
 340                         continue;
 341                 }
 342
 343                 cmsg = CMSG_FIRSTHDR(&smsg);
 344                 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
 345                         condlog(3, "no sender credentials received, message ignored");
 346                         continue;
 347                 }
 348
 349                 cred = (struct ucred *)CMSG_DATA(cmsg);
 350                 if (cred->uid != 0) {
 351                         condlog(3, "sender uid=%d, message ignored", cred->uid);
 352                         continue;
 353                 }
 354
 355                 /* skip header */
 356                 bufpos = strlen(buf) + 1;
 357                 if (bufpos < sizeof("a@/d") || bufpos >= sizeof(buf)) {
 358                         condlog(3, "invalid message length");
 359                         continue;
 360                 }
 361
 362                 /* check message header */
 363                 if (strstr(buf, "@/") == NULL) {
 364                         condlog(3, "unrecognized message header");
 365                         continue;
 366                 }
 367                 if ((size_t)buflen > sizeof(buf)-1) {
 368                         condlog(2, "buffer overflow for received uevent");
 369                         buflen = sizeof(buf)-1;
 370                 }
 371
 372                 uev = uevent_from_buffer(buf, buflen);
 373                 if (!uev)
 374                         continue;
 375                 /*
 376                  * Queue uevent and poke service pthread.
 377                  */
 378                 pthread_mutex_lock(uevq_lockp);
 379                 list_add_tail(&uev->node, &uevq);
 380                 pthread_cond_signal(uev_condp);
 381                 pthread_mutex_unlock(uevq_lockp);
 382         }
 383
 384 exit:
 385         close(sock);
 386         return 1;
 387 }
 388
 389 struct uevent *uevent_from_udev_device(struct udev_device *dev)
 390 {
 391         struct uevent *uev;
 392         int i = 0;
 393         char *pos, *end;
 394         struct udev_list_entry *list_entry;
 395
 396         uev = alloc_uevent();
 397         if (!uev) {
 398                 udev_device_unref(dev);
 399                 condlog(1, "lost uevent, oom");
 400                 return NULL;
 401         }
 402         pos = uev->buffer;
 403         end = pos + HOTPLUG_BUFFER_SIZE + OBJECT_SIZE - 1;
 404         udev_list_entry_foreach(list_entry, udev_device_get_properties_list_entry(dev)) {
 405                 const char *name, *value;
 406                 int bytes;
 407
 408                 name = udev_list_entry_get_name(list_entry);
 409                 if (!name)
 410                         name = "(null)";
 411                 value = udev_list_entry_get_value(list_entry);
 412                 if (!value)
 413                         value = "(null)";
 414                 bytes = snprintf(pos, end - pos, "%s=%s", name, value);
 415                 if (pos + bytes >= end) {
 416                         condlog(2, "buffer overflow for uevent");
 417                         break;
 418                 }
 419                 uev->envp[i] = pos;
 420                 pos += bytes;
 421                 *pos = '\0';
 422                 pos++;
 423                 if (strcmp(name, "DEVPATH") == 0)
 424                         uev->devpath = uev->envp[i] + 8;
 425                 if (strcmp(name, "ACTION") == 0)
 426                         uev->action = uev->envp[i] + 7;
 427                 i++;
 428                 if (i == HOTPLUG_NUM_ENVP - 1)
 429                         break;
 430         }
 431         uev->udev = dev;
 432         uev->envp[i] = NULL;
 433
 434         condlog(3, "uevent '%s' from '%s'", uev->action, uev->devpath);
 435         uev->kernel = strrchr(uev->devpath, '/');
 436         if (uev->kernel)
 437                 uev->kernel++;
 438
 439         /* print payload environment */
 440         for (i = 0; uev->envp[i] != NULL; i++)
 441                 condlog(5, "%s", uev->envp[i]);
 442         return uev;
 443 }
 444
 445 int uevent_listen(struct udev *udev)
 446 {
 447         int err = 2;
 448         struct udev_monitor *monitor = NULL;
 449         int fd, socket_flags, events;
 450         int need_failback = 1;
 451         int timeout = 30;
 452         LIST_HEAD(uevlisten_tmp);
 453
 454         /*
 455          * Queue uevents for service by dedicated thread so that the uevent
 456          * listening thread does not block on multipathd locks (vecs->lock)
 457          * thereby not getting to empty the socket's receive buffer queue
 458          * often enough.
 459          */
 460         if (!udev) {
 461                 condlog(1, "no udev context");
 462                 return 1;
 463         }
 464         udev_ref(udev);
 465         pthread_cleanup_push(uevent_cleanup, udev);
 466
 467         monitor = udev_monitor_new_from_netlink(udev, "udev");
 468         if (!monitor) {
 469                 condlog(2, "failed to create udev monitor");
 470                 goto out;
 471         }
 472 #ifdef LIBUDEV_API_RECVBUF
 473         if (udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024))
 474                 condlog(2, "failed to increase buffer size");
 475 #endif
 476         fd = udev_monitor_get_fd(monitor);
 477         if (fd < 0) {
 478                 condlog(2, "failed to get monitor fd");
 479                 goto out;
 480         }
 481         socket_flags = fcntl(fd, F_GETFL);
 482         if (socket_flags < 0) {
 483                 condlog(2, "failed to get monitor socket flags : %s",
 484                         strerror(errno));
 485                 goto out;
 486         }
 487         if (fcntl(fd, F_SETFL, socket_flags & ~O_NONBLOCK) < 0) {
 488                 condlog(2, "failed to set monitor socket flags : %s",
 489                         strerror(errno));
 490                 goto out;
 491         }
 492         err = udev_monitor_filter_add_match_subsystem_devtype(monitor, "block",
 493                                                               NULL);
 494         if (err)
 495                 condlog(2, "failed to create filter : %s", strerror(-err));
 496         err = udev_monitor_enable_receiving(monitor);
 497         if (err) {
 498                 condlog(2, "failed to enable receiving : %s", strerror(-err));
 499                 goto out;
 500         }
 501
 502         events = 0;
 503         while (1) {
 504                 struct uevent *uev;
 505                 struct udev_device *dev;
 506                 struct pollfd ev_poll;
 507                 int poll_timeout;
 508                 int fdcount;
 509
 510                 memset(&ev_poll, 0, sizeof(struct pollfd));
 511                 ev_poll.fd = fd;
 512                 ev_poll.events = POLLIN;
 513                 poll_timeout = timeout * 1000;
 514                 errno = 0;
 515                 fdcount = poll(&ev_poll, 1, poll_timeout);
 516                 if (fdcount && ev_poll.revents & POLLIN) {
 517                         timeout = 0;
 518                         dev = udev_monitor_receive_device(monitor);
 519                         if (!dev) {
 520                                 condlog(0, "failed getting udev device");
 521                                 continue;
 522                         }
 523                         uev = uevent_from_udev_device(dev);
 524                         if (!uev)
 525                                 continue;
 526                         list_add_tail(&uev->node, &uevlisten_tmp);
 527                         events++;
 528                         continue;
 529                 }
 530                 if (fdcount < 0) {
 531                         if (errno == EINTR)
 532                                 continue;
 533
 534                         condlog(0, "error receiving "
 535                                 "uevent message: %m");
 536                         err = -errno;
 537                         break;
 538                 }
 539                 if (!list_empty(&uevlisten_tmp)) {
 540                         /*
 541                          * Queue uevents and poke service pthread.
 542                          */
 543                         condlog(3, "Forwarding %d uevents", events);
 544                         pthread_mutex_lock(uevq_lockp);
 545                         list_splice_tail_init(&uevlisten_tmp, &uevq);
 546                         pthread_cond_signal(uev_condp);
 547                         pthread_mutex_unlock(uevq_lockp);
 548                         events = 0;
 549                 }
 550                 timeout = 30;
 551         }
 552         need_failback = 0;
 553 out:
 554         if (monitor)
 555                 udev_monitor_unref(monitor);
 556         if (need_failback)
 557                 err = failback_listen();
 558         pthread_cleanup_pop(1);
 559         return err;
 560 }
 561
 562 extern int
 563 uevent_get_major(struct uevent *uev)
 564 {
 565         char *p, *q;
 566         int i, major = -1;
 567
 568         for (i = 0; uev->envp[i] != NULL; i++) {
 569                 if (!strncmp(uev->envp[i], "MAJOR", 5) && strlen(uev->envp[i]) > 6) {
 570                         p = uev->envp[i] + 6;
 571                         major = strtoul(p, &q, 10);
 572                         if (p == q) {
 573                                 condlog(2, "invalid major '%s'", p);
 574                                 major = -1;
 575                         }
 576                         break;
 577                 }
 578         }
 579         return major;
 580 }
 581
 582 extern int
 583 uevent_get_minor(struct uevent *uev)
 584 {
 585         char *p, *q;
 586         int i, minor = -1;
 587
 588         for (i = 0; uev->envp[i] != NULL; i++) {
 589                 if (!strncmp(uev->envp[i], "MINOR", 5) && strlen(uev->envp[i]) > 6) {
 590                         p = uev->envp[i] + 6;
 591                         minor = strtoul(p, &q, 10);
 592                         if (p == q) {
 593                                 condlog(2, "invalid minor '%s'", p);
 594                                 minor = -1;
 595                         }
 596                         break;
 597                 }
 598         }
 599         return minor;
 600 }
 601
 602 extern int
 603 uevent_get_disk_ro(struct uevent *uev)
 604 {
 605         char *p, *q;
 606         int i, ro = -1;
 607
 608         for (i = 0; uev->envp[i] != NULL; i++) {
 609                 if (!strncmp(uev->envp[i], "DISK_RO", 6) && strlen(uev->envp[i]) > 7) {
 610                         p = uev->envp[i] + 8;
 611                         ro = strtoul(p, &q, 10);
 612                         if (p == q) {
 613                                 condlog(2, "invalid read_only setting '%s'", p);
 614                                 ro = -1;
 615                         }
 616                         break;
 617                 }
 618         }
 619         return ro;
 620 }
 621
 622 extern char *
 623 uevent_get_dm_name(struct uevent *uev)
 624 {
 625         char *p = NULL;
 626         int i;
 627
 628         for (i = 0; uev->envp[i] != NULL; i++) {
 629                 if (!strncmp(uev->envp[i], "DM_NAME", 6) &&
 630                     strlen(uev->envp[i]) > 7) {
 631                         p = MALLOC(strlen(uev->envp[i] + 8) + 1);
 632                         strcpy(p, uev->envp[i] + 8);
 633                         break;
 634                 }
 635         }
 636         return p;
 637 }