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 <stdbool.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 <poll.h>
  38 #include <linux/types.h>
  39 #include <linux/netlink.h>
  40 #include <pthread.h>
  41 #include <sys/mman.h>
  42 #include <sys/time.h>
  43 #include <libudev.h>
  44 #include <errno.h>
  45
  46 #include "memory.h"
  47 #include "debug.h"
  48 #include "list.h"
  49 #include "uevent.h"
  50 #include "vector.h"
  51 #include "structs.h"
  52 #include "util.h"
  53 #include "config.h"
  54 #include "blacklist.h"
  55 #include "devmapper.h"
  56
  57 #define MAX_ACCUMULATION_COUNT 2048
  58 #define MAX_ACCUMULATION_TIME 30*1000
  59 #define MIN_BURST_SPEED 10
  60
  61 typedef int (uev_trigger)(struct uevent *, void * trigger_data);
  62
  63 LIST_HEAD(uevq);
  64 pthread_mutex_t uevq_lock = PTHREAD_MUTEX_INITIALIZER;
  65 pthread_mutex_t *uevq_lockp = &uevq_lock;
  66 pthread_cond_t uev_cond = PTHREAD_COND_INITIALIZER;
  67 pthread_cond_t *uev_condp = &uev_cond;
  68 uev_trigger *my_uev_trigger;
  69 void * my_trigger_data;
  70 int servicing_uev;
  71
  72 int is_uevent_busy(void)
  73 {
  74         int empty;
  75
  76         pthread_mutex_lock(uevq_lockp);
  77         empty = list_empty(&uevq);
  78         pthread_mutex_unlock(uevq_lockp);
  79         return (!empty || servicing_uev);
  80 }
  81
  82 struct uevent * alloc_uevent (void)
  83 {
  84         struct uevent *uev = MALLOC(sizeof(struct uevent));
  85
  86         if (uev) {
  87                 INIT_LIST_HEAD(&uev->node);
  88                 INIT_LIST_HEAD(&uev->merge_node);
  89         }
  90
  91         return uev;
  92 }
  93
  94 void
  95 uevq_cleanup(struct list_head *tmpq)
  96 {
  97         struct uevent *uev, *tmp;
  98
  99         list_for_each_entry_safe(uev, tmp, tmpq, node) {
 100                 list_del_init(&uev->node);
 101
 102                 if (uev->udev)
 103                         udev_device_unref(uev->udev);
 104                 FREE(uev);
 105         }
 106 }
 107
 108 static const char* uevent_get_env_var(const struct uevent *uev,
 109                                       const char *attr)
 110 {
 111         int i, len;
 112         const char *p = NULL;
 113
 114         if (attr == NULL)
 115                 goto invalid;
 116
 117         len = strlen(attr);
 118         if (len == 0)
 119                 goto invalid;
 120
 121         for (i = 0; uev->envp[i] != NULL; i++) {
 122                 const char *var = uev->envp[i];
 123
 124                 if (strlen(var) > len &&
 125                     !memcmp(var, attr, len) && var[len] == '=') {
 126                         p = var + len + 1;
 127                         break;
 128                 }
 129         }
 130
 131         condlog(4, "%s: %s -> '%s'", __func__, attr, p);
 132         return p;
 133
 134 invalid:
 135         condlog(2, "%s: empty variable name", __func__);
 136         return NULL;
 137 }
 138
 139 static int uevent_get_env_positive_int(const struct uevent *uev,
 140                                        const char *attr)
 141 {
 142         const char *p = uevent_get_env_var(uev, attr);
 143         char *q;
 144         int ret;
 145
 146         if (p == NULL || *p == '\0')
 147                 return -1;
 148
 149         ret = strtoul(p, &q, 10);
 150         if (*q != '\0' || ret < 0) {
 151                 condlog(2, "%s: invalid %s: '%s'", __func__, attr, p);
 152                 return -1;
 153         }
 154         return ret;
 155 }
 156
 157 void
 158 uevent_get_wwid(struct uevent *uev)
 159 {
 160         char *uid_attribute;
 161         const char *val;
 162         struct config * conf;
 163
 164         conf = get_multipath_config();
 165         pthread_cleanup_push(put_multipath_config, conf);
 166         uid_attribute = get_uid_attribute_by_attrs(conf, uev->kernel);
 167         pthread_cleanup_pop(1);
 168
 169         val = uevent_get_env_var(uev, uid_attribute);
 170         if (val)
 171                 uev->wwid = val;
 172 }
 173
 174 bool
 175 uevent_need_merge(void)
 176 {
 177         struct config * conf;
 178         bool need_merge = false;
 179
 180         conf = get_multipath_config();
 181         if (VECTOR_SIZE(&conf->uid_attrs) > 0)
 182                 need_merge = true;
 183         put_multipath_config(conf);
 184
 185         return need_merge;
 186 }
 187
 188 bool
 189 uevent_can_discard(struct uevent *uev)
 190 {
 191         int invalid = 0;
 192         struct config * conf;
 193
 194         /*
 195          * do not filter dm devices by devnode
 196          */
 197         if (!strncmp(uev->kernel, "dm-", 3))
 198                 return false;
 199         /*
 200          * filter paths devices by devnode
 201          */
 202         conf = get_multipath_config();
 203         pthread_cleanup_push(put_multipath_config, conf);
 204         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 205                            uev->kernel) > 0)
 206                 invalid = 1;
 207         pthread_cleanup_pop(1);
 208
 209         if (invalid)
 210                 return true;
 211         return false;
 212 }
 213
 214 bool
 215 uevent_can_filter(struct uevent *earlier, struct uevent *later)
 216 {
 217
 218         /*
 219          * filter earlier uvents if path has removed later. Eg:
 220          * "add path1 |chang path1 |add path2 |remove path1"
 221          * can filter as:
 222          * "add path2 |remove path1"
 223          * uevents "add path1" and "chang path1" are filtered out
 224          */
 225         if (!strcmp(earlier->kernel, later->kernel) &&
 226                 !strcmp(later->action, "remove") &&
 227                 strncmp(later->kernel, "dm-", 3)) {
 228                 return true;
 229         }
 230
 231         /*
 232          * filter change uvents if add uevents exist. Eg:
 233          * "change path1| add path1 |add path2"
 234          * can filter as:
 235          * "add path1 |add path2"
 236          * uevent "chang path1" is filtered out
 237          */
 238         if (!strcmp(earlier->kernel, later->kernel) &&
 239                 !strcmp(earlier->action, "change") &&
 240                 !strcmp(later->action, "add") &&
 241                 strncmp(later->kernel, "dm-", 3)) {
 242                 return true;
 243         }
 244
 245         return false;
 246 }
 247
 248 bool
 249 merge_need_stop(struct uevent *earlier, struct uevent *later)
 250 {
 251         /*
 252          * dm uevent do not try to merge with left uevents
 253          */
 254         if (!strncmp(later->kernel, "dm-", 3))
 255                 return true;
 256
 257         /*
 258          * we can not make a jugement without wwid,
 259          * so it is sensible to stop merging
 260          */
 261         if (!earlier->wwid || !later->wwid)
 262                 return true;
 263         /*
 264          * uevents merging stopped
 265          * when we meet an opposite action uevent from the same LUN to AVOID
 266          * "add path1 |remove path1 |add path2 |remove path2 |add path3"
 267          * to merge as "remove path1, path2" and "add path1, path2, path3"
 268          * OR
 269          * "remove path1 |add path1 |remove path2 |add path2 |remove path3"
 270          * to merge as "add path1, path2" and "remove path1, path2, path3"
 271          * SO
 272          * when we meet a non-change uevent from the same LUN
 273          * with the same wwid and different action
 274          * it would be better to stop merging.
 275          */
 276         if (!strcmp(earlier->wwid, later->wwid) &&
 277             strcmp(earlier->action, later->action) &&
 278             strcmp(earlier->action, "change") &&
 279             strcmp(later->action, "change"))
 280                 return true;
 281
 282         return false;
 283 }
 284
 285 bool
 286 uevent_can_merge(struct uevent *earlier, struct uevent *later)
 287 {
 288         /* merge paths uevents
 289          * whose wwids exsit and are same
 290          * and actions are same,
 291          * and actions are addition or deletion
 292          */
 293         if (earlier->wwid && later->wwid &&
 294             !strcmp(earlier->wwid, later->wwid) &&
 295             !strcmp(earlier->action, later->action) &&
 296             strncmp(earlier->action, "change", 6) &&
 297             strncmp(earlier->kernel, "dm-", 3)) {
 298                 return true;
 299         }
 300
 301         return false;
 302 }
 303
 304 void
 305 uevent_prepare(struct list_head *tmpq)
 306 {
 307         struct uevent *uev, *tmp;
 308
 309         list_for_each_entry_reverse_safe(uev, tmp, tmpq, node) {
 310                 if (uevent_can_discard(uev)) {
 311                         list_del_init(&uev->node);
 312                         if (uev->udev)
 313                                 udev_device_unref(uev->udev);
 314                         FREE(uev);
 315                         continue;
 316                 }
 317
 318                 if (strncmp(uev->kernel, "dm-", 3) &&
 319                     uevent_need_merge())
 320                         uevent_get_wwid(uev);
 321         }
 322 }
 323
 324 void
 325 uevent_filter(struct uevent *later, struct list_head *tmpq)
 326 {
 327         struct uevent *earlier, *tmp;
 328
 329         list_for_some_entry_reverse_safe(earlier, tmp, &later->node, tmpq, node) {
 330                 /*
 331                  * filter unnessary earlier uevents
 332                  * by the later uevent
 333                  */
 334                 if (uevent_can_filter(earlier, later)) {
 335                         condlog(3, "uevent: %s-%s has filtered by uevent: %s-%s",
 336                                 earlier->kernel, earlier->action,
 337                                 later->kernel, later->action);
 338
 339                         list_del_init(&earlier->node);
 340                         if (earlier->udev)
 341                                 udev_device_unref(earlier->udev);
 342                         FREE(earlier);
 343                 }
 344         }
 345 }
 346
 347 void
 348 uevent_merge(struct uevent *later, struct list_head *tmpq)
 349 {
 350         struct uevent *earlier, *tmp;
 351
 352         list_for_some_entry_reverse_safe(earlier, tmp, &later->node, tmpq, node) {
 353                 if (merge_need_stop(earlier, later))
 354                         break;
 355                 /*
 356                  * merge earlier uevents to the later uevent
 357                  */
 358                 if (uevent_can_merge(earlier, later)) {
 359                         condlog(3, "merged uevent: %s-%s-%s with uevent: %s-%s-%s",
 360                                 earlier->action, earlier->kernel, earlier->wwid,
 361                                 later->action, later->kernel, later->wwid);
 362
 363                         list_move(&earlier->node, &later->merge_node);
 364                 }
 365         }
 366 }
 367
 368 void
 369 merge_uevq(struct list_head *tmpq)
 370 {
 371         struct uevent *later;
 372
 373         uevent_prepare(tmpq);
 374         list_for_each_entry_reverse(later, tmpq, node) {
 375                 uevent_filter(later, tmpq);
 376                 if(uevent_need_merge())
 377                         uevent_merge(later, tmpq);
 378         }
 379 }
 380
 381 void
 382 service_uevq(struct list_head *tmpq)
 383 {
 384         struct uevent *uev, *tmp;
 385
 386         list_for_each_entry_safe(uev, tmp, tmpq, node) {
 387                 list_del_init(&uev->node);
 388
 389                 if (my_uev_trigger && my_uev_trigger(uev, my_trigger_data))
 390                         condlog(0, "uevent trigger error");
 391
 392                 uevq_cleanup(&uev->merge_node);
 393
 394                 if (uev->udev)
 395                         udev_device_unref(uev->udev);
 396                 FREE(uev);
 397         }
 398 }
 399
 400 static void uevent_cleanup(void *arg)
 401 {
 402         struct udev *udev = arg;
 403
 404         condlog(3, "Releasing uevent_listen() resources");
 405         udev_unref(udev);
 406 }
 407
 408 static void monitor_cleanup(void *arg)
 409 {
 410         struct udev_monitor *monitor = arg;
 411
 412         condlog(3, "Releasing uevent_monitor() resources");
 413         udev_monitor_unref(monitor);
 414 }
 415
 416 /*
 417  * Service the uevent queue.
 418  */
 419 int uevent_dispatch(int (*uev_trigger)(struct uevent *, void * trigger_data),
 420                     void * trigger_data)
 421 {
 422         my_uev_trigger = uev_trigger;
 423         my_trigger_data = trigger_data;
 424
 425         mlockall(MCL_CURRENT | MCL_FUTURE);
 426
 427         while (1) {
 428                 LIST_HEAD(uevq_tmp);
 429
 430                 pthread_mutex_lock(uevq_lockp);
 431                 servicing_uev = 0;
 432                 /*
 433                  * Condition signals are unreliable,
 434                  * so make sure we only wait if we have to.
 435                  */
 436                 if (list_empty(&uevq)) {
 437                         pthread_cond_wait(uev_condp, uevq_lockp);
 438                 }
 439                 servicing_uev = 1;
 440                 list_splice_init(&uevq, &uevq_tmp);
 441                 pthread_mutex_unlock(uevq_lockp);
 442                 if (!my_uev_trigger)
 443                         break;
 444                 merge_uevq(&uevq_tmp);
 445                 service_uevq(&uevq_tmp);
 446         }
 447         condlog(3, "Terminating uev service queue");
 448         uevq_cleanup(&uevq);
 449         return 0;
 450 }
 451
 452 struct uevent *uevent_from_buffer(char *buf, ssize_t buflen)
 453 {
 454         struct uevent *uev;
 455         char *buffer;
 456         size_t bufpos;
 457         int i;
 458         char *pos;
 459
 460         uev = alloc_uevent();
 461         if (!uev) {
 462                 condlog(1, "lost uevent, oom");
 463                 return NULL;
 464         }
 465
 466         if ((size_t)buflen > sizeof(buf)-1)
 467                 buflen = sizeof(buf)-1;
 468
 469         /*
 470          * Copy the shared receive buffer contents to buffer private
 471          * to this uevent so we can immediately reuse the shared buffer.
 472          */
 473         memcpy(uev->buffer, buf, HOTPLUG_BUFFER_SIZE + OBJECT_SIZE);
 474         buffer = uev->buffer;
 475         buffer[buflen] = '\0';
 476
 477         /* save start of payload */
 478         bufpos = strlen(buffer) + 1;
 479
 480         /* action string */
 481         uev->action = buffer;
 482         pos = strchr(buffer, '@');
 483         if (!pos) {
 484                 condlog(3, "bad action string '%s'", buffer);
 485                 FREE(uev);
 486                 return NULL;
 487         }
 488         pos[0] = '\0';
 489
 490         /* sysfs path */
 491         uev->devpath = &pos[1];
 492
 493         /* hotplug events have the environment attached - reconstruct envp[] */
 494         for (i = 0; (bufpos < (size_t)buflen) && (i < HOTPLUG_NUM_ENVP-1); i++) {
 495                 int keylen;
 496                 char *key;
 497
 498                 key = &buffer[bufpos];
 499                 keylen = strlen(key);
 500                 uev->envp[i] = key;
 501                 /* Filter out sequence number */
 502                 if (strncmp(key, "SEQNUM=", 7) == 0) {
 503                         char *eptr;
 504
 505                         uev->seqnum = strtoul(key + 7, &eptr, 10);
 506                         if (eptr == key + 7)
 507                                 uev->seqnum = -1;
 508                 }
 509                 bufpos += keylen + 1;
 510         }
 511         uev->envp[i] = NULL;
 512
 513         condlog(3, "uevent %ld '%s' from '%s'", uev->seqnum,
 514                 uev->action, uev->devpath);
 515         uev->kernel = strrchr(uev->devpath, '/');
 516         if (uev->kernel)
 517                 uev->kernel++;
 518
 519         /* print payload environment */
 520         for (i = 0; uev->envp[i] != NULL; i++)
 521                 condlog(5, "%s", uev->envp[i]);
 522
 523         return uev;
 524 }
 525
 526 int failback_listen(void)
 527 {
 528         int sock;
 529         struct sockaddr_nl snl;
 530         struct sockaddr_un sun;
 531         socklen_t addrlen;
 532         int retval;
 533         int rcvbufsz = 128*1024;
 534         int rcvsz = 0;
 535         int rcvszsz = sizeof(rcvsz);
 536         unsigned int *prcvszsz = (unsigned int *)&rcvszsz;
 537         const int feature_on = 1;
 538         /*
 539          * First check whether we have a udev socket
 540          */
 541         memset(&sun, 0x00, sizeof(struct sockaddr_un));
 542         sun.sun_family = AF_LOCAL;
 543         strcpy(&sun.sun_path[1], "/org/kernel/dm/multipath_event");
 544         addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(sun.sun_path+1) + 1;
 545
 546         sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
 547         if (sock >= 0) {
 548
 549                 condlog(3, "reading events from udev socket.");
 550
 551                 /* the bind takes care of ensuring only one copy running */
 552                 retval = bind(sock, (struct sockaddr *) &sun, addrlen);
 553                 if (retval < 0) {
 554                         condlog(0, "bind failed, exit");
 555                         goto exit;
 556                 }
 557
 558                 /* enable receiving of the sender credentials */
 559                 retval = setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 560                                     &feature_on, sizeof(feature_on));
 561                 if (retval < 0) {
 562                         condlog(0, "failed to enable credential passing, exit");
 563                         goto exit;
 564                 }
 565
 566         } else {
 567                 /* Fallback to read kernel netlink events */
 568                 memset(&snl, 0x00, sizeof(struct sockaddr_nl));
 569                 snl.nl_family = AF_NETLINK;
 570                 snl.nl_pid = getpid();
 571                 snl.nl_groups = 0x01;
 572
 573                 sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
 574                 if (sock == -1) {
 575                         condlog(0, "error getting socket, exit");
 576                         return 1;
 577                 }
 578
 579                 condlog(3, "reading events from kernel.");
 580
 581                 /*
 582                  * try to avoid dropping uevents, even so, this is not a guarantee,
 583                  * but it does help to change the netlink uevent socket's
 584                  * receive buffer threshold from the default value of 106,496 to
 585                  * the maximum value of 262,142.
 586                  */
 587                 retval = setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsz,
 588                                     sizeof(rcvbufsz));
 589
 590                 if (retval < 0) {
 591                         condlog(0, "error setting receive buffer size for socket, exit");
 592                         exit(1);
 593                 }
 594                 retval = getsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvsz, prcvszsz);
 595                 if (retval < 0) {
 596                         condlog(0, "error setting receive buffer size for socket, exit");
 597                         exit(1);
 598                 }
 599                 condlog(3, "receive buffer size for socket is %u.", rcvsz);
 600
 601                 /* enable receiving of the sender credentials */
 602                 if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 603                                &feature_on, sizeof(feature_on)) < 0) {
 604                         condlog(0, "error on enabling credential passing for socket");
 605                         exit(1);
 606                 }
 607
 608                 retval = bind(sock, (struct sockaddr *) &snl,
 609                               sizeof(struct sockaddr_nl));
 610                 if (retval < 0) {
 611                         condlog(0, "bind failed, exit");
 612                         goto exit;
 613                 }
 614         }
 615
 616         while (1) {
 617                 size_t bufpos;
 618                 ssize_t buflen;
 619                 struct uevent *uev;
 620                 struct msghdr smsg;
 621                 struct iovec iov;
 622                 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
 623                 struct cmsghdr *cmsg;
 624                 struct ucred *cred;
 625                 static char buf[HOTPLUG_BUFFER_SIZE + OBJECT_SIZE];
 626
 627                 memset(buf, 0x00, sizeof(buf));
 628                 iov.iov_base = &buf;
 629                 iov.iov_len = sizeof(buf);
 630                 memset (&smsg, 0x00, sizeof(struct msghdr));
 631                 smsg.msg_iov = &iov;
 632                 smsg.msg_iovlen = 1;
 633                 smsg.msg_control = cred_msg;
 634                 smsg.msg_controllen = sizeof(cred_msg);
 635
 636                 buflen = recvmsg(sock, &smsg, 0);
 637                 if (buflen < 0) {
 638                         if (errno != EINTR)
 639                                 condlog(0, "error receiving message, errno %d", errno);
 640                         continue;
 641                 }
 642
 643                 cmsg = CMSG_FIRSTHDR(&smsg);
 644                 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
 645                         condlog(3, "no sender credentials received, message ignored");
 646                         continue;
 647                 }
 648
 649                 cred = (struct ucred *)CMSG_DATA(cmsg);
 650                 if (cred->uid != 0) {
 651                         condlog(3, "sender uid=%d, message ignored", cred->uid);
 652                         continue;
 653                 }
 654
 655                 /* skip header */
 656                 bufpos = strlen(buf) + 1;
 657                 if (bufpos < sizeof("a@/d") || bufpos >= sizeof(buf)) {
 658                         condlog(3, "invalid message length");
 659                         continue;
 660                 }
 661
 662                 /* check message header */
 663                 if (strstr(buf, "@/") == NULL) {
 664                         condlog(3, "unrecognized message header");
 665                         continue;
 666                 }
 667                 if ((size_t)buflen > sizeof(buf)-1) {
 668                         condlog(2, "buffer overflow for received uevent");
 669                         buflen = sizeof(buf)-1;
 670                 }
 671
 672                 uev = uevent_from_buffer(buf, buflen);
 673                 if (!uev)
 674                         continue;
 675                 /*
 676                  * Queue uevent and poke service pthread.
 677                  */
 678                 pthread_mutex_lock(uevq_lockp);
 679                 list_add_tail(&uev->node, &uevq);
 680                 pthread_cond_signal(uev_condp);
 681                 pthread_mutex_unlock(uevq_lockp);
 682         }
 683
 684 exit:
 685         close(sock);
 686         return 1;
 687 }
 688
 689 struct uevent *uevent_from_udev_device(struct udev_device *dev)
 690 {
 691         struct uevent *uev;
 692         int i = 0;
 693         char *pos, *end;
 694         struct udev_list_entry *list_entry;
 695
 696         uev = alloc_uevent();
 697         if (!uev) {
 698                 udev_device_unref(dev);
 699                 condlog(1, "lost uevent, oom");
 700                 return NULL;
 701         }
 702         pos = uev->buffer;
 703         end = pos + HOTPLUG_BUFFER_SIZE + OBJECT_SIZE - 1;
 704         udev_list_entry_foreach(list_entry, udev_device_get_properties_list_entry(dev)) {
 705                 const char *name, *value;
 706                 int bytes;
 707
 708                 name = udev_list_entry_get_name(list_entry);
 709                 if (!name)
 710                         name = "(null)";
 711                 value = udev_list_entry_get_value(list_entry);
 712                 if (!value)
 713                         value = "(null)";
 714                 bytes = snprintf(pos, end - pos, "%s=%s", name, value);
 715                 if (pos + bytes >= end) {
 716                         condlog(2, "buffer overflow for uevent");
 717                         break;
 718                 }
 719                 uev->envp[i] = pos;
 720                 pos += bytes;
 721                 *pos = '\0';
 722                 pos++;
 723                 if (strcmp(name, "DEVPATH") == 0)
 724                         uev->devpath = uev->envp[i] + 8;
 725                 if (strcmp(name, "ACTION") == 0)
 726                         uev->action = uev->envp[i] + 7;
 727                 i++;
 728                 if (i == HOTPLUG_NUM_ENVP - 1)
 729                         break;
 730         }
 731         if (!uev->devpath || ! uev->action) {
 732                 udev_device_unref(dev);
 733                 condlog(1, "uevent missing necessary fields");
 734                 FREE(uev);
 735                 return NULL;
 736         }
 737         uev->udev = dev;
 738         uev->envp[i] = NULL;
 739
 740         condlog(3, "uevent '%s' from '%s'", uev->action, uev->devpath);
 741         uev->kernel = strrchr(uev->devpath, '/');
 742         if (uev->kernel)
 743                 uev->kernel++;
 744
 745         /* print payload environment */
 746         for (i = 0; uev->envp[i] != NULL; i++)
 747                 condlog(5, "%s", uev->envp[i]);
 748         return uev;
 749 }
 750
 751 bool uevent_burst(struct timeval *start_time, int events)
 752 {
 753         struct timeval diff_time, end_time;
 754         unsigned long speed;
 755         unsigned long eclipse_ms;
 756
 757         if(events > MAX_ACCUMULATION_COUNT) {
 758                 condlog(2, "burst got %u uevents, too much uevents, stopped", events);
 759                 return false;
 760         }
 761
 762         gettimeofday(&end_time, NULL);
 763         timersub(&end_time, start_time, &diff_time);
 764
 765         eclipse_ms = diff_time.tv_sec * 1000 + diff_time.tv_usec / 1000;
 766
 767         if (eclipse_ms == 0)
 768                 return true;
 769
 770         if (eclipse_ms > MAX_ACCUMULATION_TIME) {
 771                 condlog(2, "burst continued %lu ms, too long time, stopped", eclipse_ms);
 772                 return false;
 773         }
 774
 775         speed = (events * 1000) / eclipse_ms;
 776         if (speed > MIN_BURST_SPEED)
 777                 return true;
 778
 779         return false;
 780 }
 781
 782 int uevent_listen(struct udev *udev)
 783 {
 784         int err = 2;
 785         struct udev_monitor *monitor = NULL;
 786         int fd, socket_flags, events;
 787         struct timeval start_time;
 788         int need_failback = 1;
 789         int timeout = 30;
 790         LIST_HEAD(uevlisten_tmp);
 791
 792         /*
 793          * Queue uevents for service by dedicated thread so that the uevent
 794          * listening thread does not block on multipathd locks (vecs->lock)
 795          * thereby not getting to empty the socket's receive buffer queue
 796          * often enough.
 797          */
 798         if (!udev) {
 799                 condlog(1, "no udev context");
 800                 return 1;
 801         }
 802         udev_ref(udev);
 803         pthread_cleanup_push(uevent_cleanup, udev);
 804
 805         monitor = udev_monitor_new_from_netlink(udev, "udev");
 806         if (!monitor) {
 807                 condlog(2, "failed to create udev monitor");
 808                 goto failback;
 809         }
 810         pthread_cleanup_push(monitor_cleanup, monitor);
 811 #ifdef LIBUDEV_API_RECVBUF
 812         if (udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024))
 813                 condlog(2, "failed to increase buffer size");
 814 #endif
 815         fd = udev_monitor_get_fd(monitor);
 816         if (fd < 0) {
 817                 condlog(2, "failed to get monitor fd");
 818                 goto out;
 819         }
 820         socket_flags = fcntl(fd, F_GETFL);
 821         if (socket_flags < 0) {
 822                 condlog(2, "failed to get monitor socket flags : %s",
 823                         strerror(errno));
 824                 goto out;
 825         }
 826         if (fcntl(fd, F_SETFL, socket_flags & ~O_NONBLOCK) < 0) {
 827                 condlog(2, "failed to set monitor socket flags : %s",
 828                         strerror(errno));
 829                 goto out;
 830         }
 831         err = udev_monitor_filter_add_match_subsystem_devtype(monitor, "block",
 832                                                               "disk");
 833         if (err)
 834                 condlog(2, "failed to create filter : %s", strerror(-err));
 835         err = udev_monitor_enable_receiving(monitor);
 836         if (err) {
 837                 condlog(2, "failed to enable receiving : %s", strerror(-err));
 838                 goto out;
 839         }
 840
 841         events = 0;
 842         gettimeofday(&start_time, NULL);
 843         while (1) {
 844                 struct uevent *uev;
 845                 struct udev_device *dev;
 846                 struct pollfd ev_poll;
 847                 int poll_timeout;
 848                 int fdcount;
 849
 850                 memset(&ev_poll, 0, sizeof(struct pollfd));
 851                 ev_poll.fd = fd;
 852                 ev_poll.events = POLLIN;
 853                 poll_timeout = timeout * 1000;
 854                 errno = 0;
 855                 fdcount = poll(&ev_poll, 1, poll_timeout);
 856                 if (fdcount && ev_poll.revents & POLLIN) {
 857                         timeout = uevent_burst(&start_time, events + 1) ? 1 : 0;
 858                         dev = udev_monitor_receive_device(monitor);
 859                         if (!dev) {
 860                                 condlog(0, "failed getting udev device");
 861                                 continue;
 862                         }
 863                         uev = uevent_from_udev_device(dev);
 864                         if (!uev)
 865                                 continue;
 866                         list_add_tail(&uev->node, &uevlisten_tmp);
 867                         events++;
 868                         continue;
 869                 }
 870                 if (fdcount < 0) {
 871                         if (errno == EINTR)
 872                                 continue;
 873
 874                         condlog(0, "error receiving "
 875                                 "uevent message: %m");
 876                         err = -errno;
 877                         break;
 878                 }
 879                 if (!list_empty(&uevlisten_tmp)) {
 880                         /*
 881                          * Queue uevents and poke service pthread.
 882                          */
 883                         condlog(3, "Forwarding %d uevents", events);
 884                         pthread_mutex_lock(uevq_lockp);
 885                         list_splice_tail_init(&uevlisten_tmp, &uevq);
 886                         pthread_cond_signal(uev_condp);
 887                         pthread_mutex_unlock(uevq_lockp);
 888                         events = 0;
 889                 }
 890                 gettimeofday(&start_time, NULL);
 891                 timeout = 30;
 892         }
 893         need_failback = 0;
 894 out:
 895         pthread_cleanup_pop(1);
 896 failback:
 897         if (need_failback)
 898                 err = failback_listen();
 899         pthread_cleanup_pop(1);
 900         return err;
 901 }
 902
 903 int uevent_get_major(const struct uevent *uev)
 904 {
 905         return uevent_get_env_positive_int(uev, "MAJOR");
 906 }
 907
 908 int uevent_get_minor(const struct uevent *uev)
 909 {
 910         return uevent_get_env_positive_int(uev, "MINOR");
 911 }
 912
 913 int uevent_get_disk_ro(const struct uevent *uev)
 914 {
 915         return uevent_get_env_positive_int(uev, "DISK_RO");
 916 }
 917
 918 static char *uevent_get_dm_str(const struct uevent *uev, char *attr)
 919 {
 920         const char *tmp = uevent_get_env_var(uev, attr);
 921
 922         if (tmp == NULL)
 923                 return NULL;
 924         return strdup(tmp);
 925 }
 926
 927 char *uevent_get_dm_name(const struct uevent *uev)
 928 {
 929         return uevent_get_dm_str(uev, "DM_NAME");
 930 }
 931
 932 char *uevent_get_dm_path(const struct uevent *uev)
 933 {
 934         return uevent_get_dm_str(uev, "DM_PATH");
 935 }
 936
 937 char *uevent_get_dm_action(const struct uevent *uev)
 938 {
 939         return uevent_get_dm_str(uev, "DM_ACTION");
 940 }
 941
 942 bool uevent_is_mpath(const struct uevent *uev)
 943 {
 944         const char *uuid = uevent_get_env_var(uev, "DM_UUID");
 945
 946         if (uuid == NULL)
 947                 return false;
 948         if (strncmp(uuid, UUID_PREFIX, UUID_PREFIX_LEN))
 949                 return false;
 950         return uuid[UUID_PREFIX_LEN] != '\0';
 951 }